Professional Documents
Culture Documents
RAIMO ANTTILA
University of California, Los Angeles
the languages students are most likely to know. Works of great historical im-
portance are often listed in translation, or in a recent reprint; thus the dates in
the bibliography lie mainly in the 1960s despite the concern with the history of
linguistics. I have listed works that have actually influenced this text by their
relation to my personal background, other elementary material, and some more
advanced works for further study.
When the "working date" used in a cross reference to any publication has
been superseded by the actual publication date during the production of this
book, the actual date may be found appended to the bibliography entry in
parentheses; brackets indicate an alternative publication of a work.
The omission of an author index is alleviated by the fact that mention of
authors is avoided as much as possible in the text itself. Most names are gathered
into the reference sections at the end of the chapters.
The influence on my thinking of over a century and a half of scholarship in
the area of historical linguistics is obvious and freely admitted. But of particular
importance has been the instruction and inspiration drawn from many teachers
in many countries, above all Bernard Bloch, Warren Cowgill, Isidore Dyen,
Floyd Lounsbury (all at Yale), and Robert Austerlitz. Colleagues who have
given me guidance and encouragement in connection with this book are Henning
Andersen, William Bright, Mati Hint, Erkki Itkonen (whose book Kieli ja sen
tutkimus profoundly influenced the following pages), Guy Jucquois, Kostas
Kazazis, J. Peter Maher, Hanns-Peter Schmidt, and Michael Shapiro. Andrew
Sihler, of the University of Wisconsin, provided invaluable advice and assistance,
especially in the matter of style. I am indebted to a great number of students
who over the years have seen my ideas develop and who have influenced them-in
particular the class at UCLA in the spring of 1970, the first one to have this
manuscript in their hands, and especially Lyle Campbell and Bruce Pearson, who
have responded beyond the call of duty. The valuable qualities of my undertaking
have been enhanced by these sources of influence; whatever is of doubtful value
is solely my own responsibility.
R.A.
CONTENTS
List of Illustrations ix
Unnumbered Illustrations xi
List of Exercises xi
PART I
BACKGROUND: GENETIC LINGUISTICS IN RELATION
TO GENERAL LINGUISTICS AND RELATED
FIELDS
1. Language and Linguistics 3
2. Writing and Language 31
3. Linguistic Variation 47
PART II
HISTORICAL LINGUISTICS: HOW DOES LANGUAGE
CHANGE?
4. Sound Change 57
5. Grammar Change: Analogy 88
6. Rule Change 109
7. Semantic Change 133
8. External Change: Borrowing 154
9. Why Does Language Change? Social and Linguistic Factors 179
PART III
COMPARATIVE LINGUISTICS (GENERAL NOTIONS
AND STRUCTURE): HOW CAN CHANGE BE
REVERSED?
10. Preliminaries to the Historical Methods 207
11. The Comparative Method (the Central Concept) 229
vii
viii CONTENTS
PART IV
LINGUISTIC RECONSTRUCTION: A SYNTHESIS OF
VARIOUS LINGUISTIC AND CULTURAL NOTIONS
14. Dialect Geography 289
15. Alternative Relationship Models 300
16. Classification of Languages 310
17. Philology and Etymology 323
18. Reconstructing Phonology 335
19. Reconstructing Grammar 351
20. Reconstructing Semology/Semantics 364
PART V
CONCLUSION: LINGU I STICS AS PART OF
ANTHROPOLOGY
21. Change and Reconstruction in Culture and Linguistics 377
22. Genetic Linguistics and Biological Genetics 389
APPENDIXES
I. Syllabus to Some Introductions to Historical Linguistics 399
II. Reading List for an Advanced Course 400
BIBLIOGRAPHY 401
INDEX 429
LIST OF ILLUSTRATIONS
Classification of signs 18
Structural classification of sound change 69
Articulatory classification of sound change 71
Proportional analogy 89
The meaning-form linkups 100
Grimm's and Verner's Jaws in Germanic Ill
Subtractive reordering and restructuring and loss 130
Onomasiological change in French 134
Semantic shift in Latin 147
Numerical parameters for diphthongs in Martha's Vineyeard 192
Abduction, deduction, and language learning 197
Inventory of reconstructed Proto-Indo-European consonants 271
Diphthongs in the Atlantic states 293
Subgrouping and phonetic reconstruction 344
Table for retention of shared vocabulary 397
LIST OF EXERCISES
Borrowing 178
Morphophonemic analysis/internal reconstruction 228
The comparative method 256
xi
For Selja, Selene, and Matti
PART I
BACKGROUND:
GENETIC LINGUISTICS
IN RELATION TO
GENERAL LINGUISTICS AND
RELATED FIELDS
CHAPTER 1
Nonlinguistic real
or imagined world
that make comparative linguistics possible. Figure 1-1 should be kept in mind
throughout this book, because it diagrams the basic structure of language. But
this graph shows only that language is a funnel with two wide ends connecting
sounds with meaning and vice versa. In addition, it should be remembered that
the speech sounds are actually as much a part of the same nonlinguistic universe
as the things we talk about; after all, one can talk about speech sounds them-
selves even without being a linguist or a phonetician. Furthermore, linguists
have found it useful and necessary to divide the monolithic funnel 'language'
into various subsections. Taking these two facts into consideration we may now
revise our graph as shown in Figure 1-2. Language is divided into three main
subsystems: (1) semology, which is connected with all of the real world (through
universal semantics); (2) a central section here called grammar, comprising
morphology and syntax; and (3) phonology, which completes the bridge back
into part of the real world, the speech sounds. Semology is linguistically organ-
ized semantics, semantics being the link between semology and the real world.
The relation between phonology and phonetics is the same. Phonology is the
linguistic counterpart of phonetics (Chapter 10). The arrows still indicate the
two directions of travel within the mechanism. Further divisions can be made:
it is, for example, quite common to distinguish at least two levels within the
phonological part (i.e., morphophonemic and phonemic), both of which are
intimately connected to the phonetic level. Usually one refers to all the levels
above the speech sounds as abstract. The notion of abstractness, however, breaks
down in the link between semology and semantics, because semantics is as con-
crete as the phonetic end of the system. Linguists are still not equipped to talk
LANGUAGE AND LINGUISTICS 5
FIGURE 1-2. Figure 1-1 drawn in more detail. Experience is funneled into
semology, and the speech sounds, which are part of this experience, into
phonology. [Modified with permission from Wallace L. Chafe, "Phonetics,
semantics, and language," Language, 38, 335-344 (1962).]
about semantics, and the term 'abstract' has been used as a justification for
ignorance. Positing separate subsystems does not imply clear-cut boundaries;
they are still integrally connected with adjoining sections, as shown in Figure 1-2
by the interlocking arrows. Of all these levels, we know the phonological one
best; this means that we know its changes best, too. The higher we go on the
scale, the less definite our knowledge is, and the results of our historical work
clearly reflect this defect in our knowledge.
[1.3 Language and Its Units] Semology, grammar, and phonology all
have units. Thus phonology is usually divided into the different degrees of
abstraction corresponding to phones, phonemes, and morphophonemes. (The unit
one chooses as the foundation of the historical treatment determines the nature
of the changes themselves.) Traditionally the phoneme has been defined as a
class of sounds in a given language that operate as one and to which the speakers
react as one sound. The members of this class are allophones, which occur in
mutually exclusive phonetic environments, and they share, in addition, at least
one concrete phonetic feature (e.g., velar or labial, above and beyond shared
negative features like [-vocalic]). Two different sounds contrast only if they
occupy analogous positions in two different morphemes or words (Chapter 10).
We shall see that for sound change (allo-)phones are of primary importance,
and these are the sound units closest to phonetics.
6 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
Tongue advancement
Lip
I
position \. Front Central Back
u R u R u R
H i i.i=y i i=ru u
High
L I tJ I u
....
.c
.~
~
H e 0=0 0
.c Mid ~
~
L e 5=re
=
eJ)
=
A ~
~ H re=a
Low
L a a D
Point in
Activity Apical Frontal Dorsal Faucal tongue
Manner of of vocal Labio- Dental Point in
articulation chords Bilabial Dental Alveolar Palatal Velar Glottal 'mouth'
stop voiceless p t k q=?
(closure) voiced b d g
voiceless P<P pf ts = c tS=c (kx)
affricate
voiced dz dz = J
.-;:: voiceless <P f 9=1> (( X (h)
r;; voiced v 6 y y (li)
spirant= B
II)
00 fricative >
0
voiceless s s
...0
00
voiced z z
nasal voiced m n fi=jl lJ
"0
·s lateral voiced 1 !i.. =I' L, l=JI
g trill
......
s:: voiced r R
flap .D
...
ro
voiced o=r
·;;:
glide voiced w y w
FIGURE 1-4. Basic consonant chart defined by three articulatory parameters. A row for glides is added to emphasize the
frequent patterning of [y] and [w]. Note that both occur twice in the diagram. Note that [q] will be used for the glottal
stop and not for a back velar.
LANGUAGE AND LINGUISTICS 9
liquids the voiceless counterparts have not been written in, although they are
by no means infrequent sounds in the languages of the world.
The two tables now roughly characterize the possibilities of all sounds made
by human articulating organs. They map in more detail the part of the real
world labeled 'sounds' in Figure 1-2. Any one language "chooses" only a small
part of all the possible speech sounds defined or definable in the tables. And
even if two languages use a p-sound, it does not mean that the p's would be
phonetically identical. The tables define cardinal points only, because the phonet-
ic truth is that the sounds are indefinitely varied, even in the pronunciation of
a single speaker, not to speak of different languages.
where cub again covers the young of certain mammals only. Languages differ
greatly in the semantic combinations represented by one formal unit. This fact
leads to various ethnocentric value judgments. One often reads in early treat-
ments that in a certain "primitive" language there are different words for' black
cow', 'brown cow', and so on, but no generic word for 'cow' (Zulu). Similar
situations are quoted for 'potato' in Aymara, 'snow' in Eskimo, and 'camel'
in Arabic. 'Primitive' here really means 'different' (often it just refers to the
nontechnological culture of the speakers using a certain language), since the
link-up between the semantic and formal configurations is different. No language
is primitive; they are all of the shape portrayed in Figure 1-2 (see§ 22.5). English
would also be "primitive," looked at from the point of view of the very languages
that are branded that way. It does not have a single generic term for 'bovine
animal' and what is "worse" by far, it has almost a hundred animal group
names (not necessarily American or Modern English), restricted to one or a
few species only. In addition to the rather general (but still not generic) flock,
herd, pack, and so on, there are also cast (of hawks), husk (of hares),jesnyng
(of ferrets), gaggle (of geese, on ground or water), skein (of geese, in the air),
shrewdness (of apes), skulk (of foxes), sleuth (of bears), wisp (of snipe), and
so on.
In general, a term whose extension is larger than that of another is abstract.
In contrast, the latter term is concrete. Thus the notions 'abstract' and' concrete'
are relative notions.
[1.8 Form, Meaning, and Classification] The particular nature of the link
between form and meaning has been an important criterion for typological
language classification. A language in which words contain one morpheme each
and have a one-to-one correspondence between formal and semantic units is
called isolating. If the words of a language are built up of a number of clearly
segmentable morphemes, which again have a one-to-one correspondence be-
tween form and meaning, the language is agglutinating. When the relation be-
tween form and meaning is one-to-many, which is the general case, the language
is synthetic. If words of a language represent an unusually high number of
semantic units, the language is called polysynthetic. These again are cardinal
points only, since there are no pure types (Chapter 16). They just show that
different languages make the meaning-form link-up differently-but they all
make it. This clearly shows, also, that meaning is basically independent of form
(one-to-many relation). It is curious that this kind of typological classification
was popular at a time when morphemes were defined semantically, that is, when
linguists operated with the assumption of a one-to-one relation between form
and meaning. Certain idioms, for example, to kick the bucket, 'to die', obviously
invalidated the assumption of one-to-one relation, but they were somehow
squeezed into the model anyway.
[1.9 Units and Rules] Linguists differ in the number of language levels
or subsystems they posit, as well as in the number of units they assign to each
level. The framework we have been characterizing here is only the common
core of various conceptions of the particular shape of linguistic structure. We
have referred to distinctive features, phones, phonemes, morphophonemes,
morphs, morphemes, and sememes or semantic components. One must remem-
ber that the morpheme is not the only grammatical unit; there are others of
different rank, where we also have various conglomerations of different size and
rank (i.e., word, phrase, clause, and sentence). Each level has various rules (now
often called transformations) for the combination of the units. These determine
the types of constructions that occur or could occur. There are also rules that
link the different levels with the adjoining ones. All these, taken together with
the units, give us the sum total of language as represented in Figure 1-1.
But a linguist is a mechanic of language who tries to take the machinery
apart: Figure 1-2 was a first crude step, which we followed by the establishment
I2 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
of units (both discussed earlier and mentioned here in connection with rules).
Units without rules and rules without units cannot exist in language. Both are
necessary, and consequently they cannot be hierarchically ordered; they are
both equally important in the actual functioning of language. Both the units and
the rules are subject to change, as we shall see, no matter what the linguist's
predilections are in his assumptions about the structure of the language mecha-
nism. But the way in which the change itself is described is, of course, dependent
on the kinds of rules and units the linguist posited in the first place. Absolute
change may be acknowledged differently by different linguists, therefore, because
they may have disagreed on the starting point and on the exact details posited
for the structure of language.
To emphasize how a change and its results can to a great degree depend on
the structure we adopt as our framework, let us refer to two hypothetical ex-
amples. Let us assume that two linguists describe one and the same language,
each according to his own principles. One arrives at a grammar that is a square,
the other posits a circle; but both get a definite form (which is always the target
of the linguist). The two grammars may be equally workable. Now comes a
chang~ that divides the grammar exactly in half. The square grammar gives
either two triangles or two trapezoids of various shapes, whereas the round
grammar always gives two half-circles. The impact of the change depended on
the configuration on which the change operated.
For the second example, suppose that one linguist posits more units of a
certain kind than another does. Say that we have two tables, one with four legs
and the other with three. Both are recognizable as tables, exactly as two different
grammars might be recognizable as grammars of the same language (as they
should be, of course). Then a historical change removes one leg from both.
The change is exactly the same in both tables, but the result is quite different
indeed; again, the results depend on the structure of the starting point. (These
metaphors can be translated into the grammars of different schools; see§ 6.6.)
This brief characterization of language as a bridge between "reality" and
speech sounds has proceeded from the more general toward the more detailed.
But this is only a presentation of the findings of descriptive linguistics and in no
way reflects the order in which the facts have been or should be discovered.
arbitrary symbol, brings us back to the linguistic sign and its importance for
genetic linguistics. Only a full understanding of the notion 'linguistic sign'
makes both change and reconstruction comprehensible and theoretically
explainable.
A given entity is a sign of a given referent (thing meant) if it elicits at least
some of the responses elicited by the referent, or as normally put, a sign stands
for a thing. There are three typt:s of signs:
the outer shape and the semantic range, and it is clearly seen in looking at
different languages. That is, it is arbitrary that different languages have such and
such formal shapes for, say, the meaning 'horse' (e.g., English horse, German
Pferd, Swedish hast, Finnish hevonen, French cheval, and so on). But when we
look at linguistic signs in terms of one language, we can see that there is a
tendency for a speaker to assume complete sameness between linguistic form
and reality. The sign captures and controls reality; in fact, it is reality in the
extreme case (nomen est omen, verbal magic, and so on). This is one source of
conflict between the speaker (who has not been linguistically trained) and the
linguist (who knows better) (see§ 18.7). Later we shall see how the obligatory
(nonarbitrary) aspect of the linguistic sign creates change (Part II) and how the
arbitrary side makes reconstruction possible (Part Ill). But before that we have
to go back to the other signs.
[1.11] The three types of signs (icons, indexes, and symbols) are only foot-
holds in the hierarchy of signs. In this sense, they are just cardinal points, not
unlike the cardinal vowels in the notation of the International Phonetic Associ-
ation (IPA). Photos and animal cries are also heavily indexical in relation to the
object and the source of the cry. The best signs are mixtures of all the ingredients,
a situation that is often clearest in poetry, where a symbol with associative power
(indexical) and sound symbolism (iconic) is very effective. It should be noted
that 'sound symbolism' means exactly the opposite of the technical term 'sym-
bol'. Onomatopoeic words, words that imitate nature sounds, are often naively
thought to be completely iconic, that is, perfect replicas of the actual sounds
they refer to. But that they are also symbolic is immediately evident in a com-
parison among different languages (third person singular endings separated from
the root/stem):
[1.12] The same is true of art. Most art in most cultures is highly repre-
sentative (except for music, perhaps), or what could be called nonabstract-in
other words, very iconic-because the necessary physical resemblance between
the referent and its formal marker (the signified and the signifier) exists. But
art is also based on many conventions, depending on the time and place of its
origin. In medieval art, for example, certain colors carry symbolic meaning.
The representation of, say, perspective depends on various conventions. In short,
for a full interpretation of a work of art one has to know certain conventions,
and these are symbolic. In general, one could say that the more "stylized" art
is, the more symbolic elements it contains. A" normal" piece of art is a symbolic
icon, exactly like the onomatopoeic words in a language. In both cases the
imitation of nature (the real world) is not perfect. Art gives the most concrete
example of the range from iconic to symbolic representation. It shows the
hierarchical character of the three basic types of signs-first, purely iconic rela-
tions between the picture and its topic, second, more abstract and stylized pic-
tures, and, finally, completely conventional forms of decorations that still show
definite meaning. But this meaning has to be specifically learned; it cannot be
deduced from the physical shape of the picture. This level is, then, largely sym-
bolic, exactly like the linguistic sign. The latter must also be learned, because
the meaning and the form have to be connected by a special rule which holds
for one particular case only. In Chapter 2 we shall see examples of this in the
development of writing.
One area of (artistic) signs showing various degrees of symbolicity and
iconicity is heraldry. It had its own rules in addition to the conventions of medi-
eval art in general. A form of heraldry has survived in trademarks. Many com-
panies have devised signs that blend all three sign elements, and if the mixture
is successful, the trademark is very effective. A frequent form of design is one
in which the initial letters of the name of a company are shaped so that together
they form a picture of the product of the company, its main tools of manu-
facturing, or its raw materials (all these standing in indexical relation to the
company). Thus the abbreviation IP, the registered trademark of the International
Paper Company, is a kind of index of the total name (Figure 1-6). The shapes of
the letters are completely symbolic in English; there is nothing intrinsic in the
fact that I is Ji/ and P Jpf and not vice versa. These two letters are written and
arranged so that they form a very stylized picture of a tree. Thus two symbolic
letters represent iconically a tree, which itself is an index of the company; the
principles involved hold for the other cases in Figure 1-6.
16 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
(1.13 Subdivision oflcons] The word icon means literally 'picture'. The
discussion of artistic representation showed that there are different kinds of
pictures (as everybody knows); it is therefore necessary to divide icons into three
distinct subclasses: images, diagrams, and metaphors. Images are characterized
by a relation between form and meaning in which the former contains the simple
qualities of the latter; this is the kind of icon we were referring to above. Dia-
grams are characterized by a similarity between form and meaning that is
constituted solely by the relations of their parts (Chapter 5). And metaphors
embody the representative character of the form by exhibiting a parallelism in
the meaning (Chapter 7). Linguistics, like any other science, has to rely heavily
on both diagrams and metaphors. A diagram is predominantly an icon of rela-
tion, and to interpret such icons one needs conventions. Thus two rectangles or
circles of different size, used to show a quantitative comparison of steel or coffee
production in two countries, make up a diagram in that the relations in the form
correspond to the relations in the meaning. The symbolic (conventional) aspect
here is that one has to know that it is steel production we are comparing and
not, say, coffee. A diagram is a symbolic icon. As far as language is concerned,
such diagrammatic relations are its essence. Language is very frequently referred
to as a system (or network) of relations. Linguistic units are more important as
end points in various relations than as entities in themselves.
In fact, we have already seen the usefulness of diagrammatic icons in this
background treatment. Figure 1-1 is a crude diagram of language, but it was
made more precise in Figure 1-2, which shows many relational aspects. It would
be impossible to draw an image of language, but it is quite feasible to have an
icon of the experience of the linguists. Such icons are often called models, and
they are important frames of research in all sciences. Parts of Figure 1-2 were
drawn as even more detailed diagrams (Figures 1-3-1-5). The purpose of these
phonetic and semantic tabular arrangements is precisely to show relationships;
LANGUAGE AND LINGUISTICS 17
for example,.the relation of dto ois the same as that of g toy, and gander:gosling
as ram: lamb. Many more diagrams will be given later.
[1.15) What we have seen here is that symbols need not be iconic, but that
symbol complexes are arranged in iconic relations with object complexes (see
also the trademark example). This situation is much better known in algebra,
where every equation is a diagrammatic icon. In algebra one has to define certain
symbols, but after that, the rules take care of the iconic arrangements. !conicity
is particularly clear in, say, analytic geometry. But remember that language is
often referred to as a kind of algebra, and, in fact, linguistics has strived to
develop algebraic notations to represent the relational aspects of a grammar.
This approach has been very successful in the description of syntax, which can
be almost totally represented with graphs. Thus we are able to separate the
iconic (diagrammatic) forms from the conventional (symbolic) features of one
and the same system.
Metaphors need not be discussed any further at this point. They are also
I8 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
relational, but rather than exhibiting mere part--whole relations, they concentrate
on the similarity of function (e.g., the foot of the mountain, the leg of a table,
and so on) (see Chapter 7).
This quotation also delineates the two necessary aspects of historical linguistics,
facts and general laws. Both are likely to be controversial in any particular case,
but an understanding of signs makes the situation more comprehensible.
In semiotics, the study of signs and sign systems, one calls the relation of the
signs to their referents semantic, that of signs to other signs in the code syntactic,
and the relation of signs to their users pragmatic. Linguistics is but a part of
semiotics (the dominating one), but it is clear that semantics has a correlate in
language, syntactics has a linguistic counterpart in syntax, and pragmatics points
to the social setting of language (e.g., sociolinguistics [Chapter 3], psycholin-
guistics, ethnolinguistics, and so on) (see § 21.2).
[1.19] A serious terminological difficulty has arisen from the fact that
genetic linguistics has preempted the term 'comparative'. Any kind of compari-
LANGUAGE AND LINGUISTICS 21
son for any other purpose (e.g., translation, classification of languages by their
structural characteristics, language teaching, language universals, and so on)
has to be referred to by a different term (see§§ 21.14, 22.8). Sometimes the words
'typological' or 'contrastive' serve this purpose (nonhistorical comparison), but
often they are not inclusive enough. The Greek counterpart to 'contrast' and
'comparison', syncrisis, has been proposed for this task. Others, in order to
avoid confusion, use the compound 'historical-comparative' for the highly
technical notion of 'comparative' in genetic linguistics.
(1.20] Comparative linguistics has also often been called comparative
philology, especially in England. The reason for this is that comparative lin-
guistics was practiced mainly with regard to the older Indo-European languages,
where philological screening of the material was a prerequisite (Chapter 17).
FIGURE 1-7. The frames of reference for linguistics. Shaded areas show that
all are involved in genetic linguistics. Row 3 is an important basis or pre-
requisite of row 2. [Based on Dell Hymes, "Linguistics: the field," modified
with permission of the Publisher from the International encyclopedia of the
social sciences (David L. Sills, ed., 9, 358b © 1968 Crowell-Collier and
Macmillan, Inc.)]
22 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
It is clear that the languages in the two columns are not equally dead; those in
the second one are still spoken in an altered form, whereas those in the first
column are actually stone dead. A dead language is no longer spoken in any
form; it has no speakers. This is only an accidental fact and is in itself not
directly connected with historical linguistics. Of course, it has serious conse-
quences in that our knowledge of such languages tends to be quite fragmentary,
because no further fieldwork or checking is possible. Dead languages can be the
object of descriptive linguistics, with the same limitations, of course. (All this
was implied in Figure 1-7.)
1 facts chronicle
2 hypothesis history
All historical sciences are very complicated and present extra procedural prob-
lems for the philosophy of science. General laws do not exist or, at least, are not
within the range of our ability to observe; one has to work with individual
events. One has to make the "right" selection of the documents, make state-
ments about the facts yielded by this first step, and only then proceed to ex-
planatory statements. Selection, interpretation, and historical criticism precede
the ultimate explanation; by comparison, in natural sciences the ultimate ex-
planation is connected almost directly to the facts. Further, the historical ex-
planation contains at least three features that make it different from explanations
in the natural sciences:
[1.25) It has been said that genetic linguistics is actually an art rather than
a doctrine of exact science. But all scientists must exhibit imagination and intel-
lectual sensitivity as well as self-discipline and exactness in their work. Conse-
quently, no two scientists are identical, and the outcome of their studies will
differ accordingly. It should also be clear that the ability of different individuals
to make historical inferences varies a great deal. Different individuals are simply
not equal in any activity. Thus, of those who know the rules of chess, some are
able to carry out the game markedly better than others, and only a few truly
excel. The same obtains in the use of language. All speakers know much the
same basic grammar, but some use language better than others. This is true
not only of such areas as literature, but also, say, speech perception: in noisy
situations, adverse to easy communication, some people understand better what
is said than others.
As in anthropology, archaeology, paleontology, and the like, which also try
to reconstruct the past, a successful worker in genetic linguistics must possess a
flair for piecing together fragmentary bits of information. As in any other em-
pirical science, we do have a body of exact procedures and guidelines, but this
does not in itself ensure proper or correct application. The actual practice of
genetic linguistics may seem like a special skill, an art; and it is perhaps more so
than descriptive linguistics is because of its more lacunary material and other
difficulties encountered in the study of history, as we saw above. But, by the
same token, some degree of" art" is required in the successful practice of other
scien.ces as well. The difference may still be very great, but it is one of degree
only.
It is universally agreed that regardless of the linguist's future area of special-
ization, his training should include the principles of genetic linguistics. He
might never actively apply them, but they will deepen his understanding of lan-
guage. Although everybody may not contribute to genetic linguistics, genetic
linguistics contributes to every linguist. This book will provide basic concepts
26 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
(1.26 The Origin of Language] A prime challenge for man has been
speculation about the origin of a communication system like that in Figure 1-2.
The topic has long been in disrepute, and once (1866) it was even prohibited in
French learned societies, but in quite recent times it has been discussed anew
by linguists together with anthropologists and evolutionary biologists (Chapters
21, 22). Although the area lies outside genetic linguistics, a brief paraphrase of
the newest speculation illuminates some of the basic characteristics of human
language as well as the independent bases of semantic and phonological change.
Animal communication, the transfer of information (a message) from one
nervous system to another, occurs through various media and various channels
(i.e., sound, sight, smell, taste, and touch). Although man receives and conveys
information through all these channels, sound is the most essential for language.
According to Hockett there are thirteen characteristic design features that are
used in one kind of animal communication or another:
1. Possession of a vocal/auditory channel.
2. Broadcast transmission and directional reception, which enables the signal
to travel a certain distance in all directions and be localized by binaural
direction finding.
3. Rapid fading of the signal so that it does not clog the channels (unlike
animal tracks or writing).
4. Interchangeability between sender and receiver; the sender can also
receive what he sends and vice versa; that is, the user of the communica-
tion system is a speaker-hearer.
5. Total feedback of the message sent; the sender hears his own message.
These five features characterize all mammal communication; the next
three are peculiar to man and other primates.
6. Specialization in sending sound waves just for the purpose of communi-
cation, in contrast to sounds that animals emit in performing biological
functions (e.g., the panting of a dog).
7. Semanticity, a relatively fixed association between the sound in the
message and the situation in the real world.
8. Arbitrariness (noniconicity) of the association; we have discussed this
before(§ 1.10).
The rest of the design features are common to all human languages.
9. Discreteness, in that any two linguistic units selected at random are either
'the same' or 'different'. There is no intermediate grading. For example,
the sound units have a fixed range. In English the initial sounds in gap
and cap are sharply and functionally different from each other. There is
nothing in between (e.g., no word *~ap 'a cap with gaps [holes]'). Gradi-
ence is only possible in paralinguistic features (e.g., different loudness of
LANGUAGE AND LINGUISTICS 27'
shouting as an index of anger; gibbons are also said to possess this
feature).
10. Displacement, in that the message is not tied to time and place, as we
have seen in connection with symbols.
1I. Productivity, the ability to produce and understand utterances which are
novel, a feature often called creative.
12. Traditional transmission of the system from earlier to later generations.
Only the capacity to learn language is biologically inherited (§ 21.3).
13. Duality of patterning. This last one is the most important single feature;
for example, the thousands of words in any language are built up by
arrangements of a relatively small number of sound units, which them-
selves do not mean anything.
[1.27] The hierarchical order of the above design features and man's bio-
logical characteristics point toward the primates as a source of information for
the origin of language. Primates possess call systems, which contain some kind
of holistic "calls" for a limited number of units of experience. Such units are,
for example, 'threat, danger, desire for group contact', and the like; the corre-
sponding holistic sound contours are describable as roars, barks, grunts, and so
on. There is only one-to-one symbolization: a limited number of meanings
corresponds to a limited number of sound units, each being monopolized by
one meaning only. If one indicates meanings by capitals and sounds by lower-
case, the signs are A/x, Bfy, Cfz . .. , and the primate call systems contain only
from a half dozen to two dozen of them. Their use is very indexical (i.e., here
and now); there seems to be no displacement.
The protoprimates or the protohominids must have had a call system of this
kind. Thanks to environmental changes (i.e., living away from the trees) bipedal
motion became possible. This, in turn, left the hands free for carrying, with
enormous cultural consequences. At the same time, the articulatory apparatus
was changing, because upright posture separated the glottis from the velum,
making room for more flexible manipulation of speech organs. The expansion
of human experience within the "real world" required more and more formal
sound contours. One way to ease the pressure was to resort to composite calls
(i.e., blends of the already existing calls). These composite parts would gradually
get a special meaning of their own, and the initial stages of syntax (word order)
would be in the making. The system was now on its way toward productivity.
Hockett calls this stage "pre-language." Some change was already possible,
'
28 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
and thereafter new blends might have increased in number. Speech perception
was, by necessity, directed toward receiving the total calls, because the end parts
of the calls could not be predicted by their beginnings. This system had to be
learned by tradition, and thus required a firmer cultural setting. This is a mecha-
nistic approach to the problem, even though it is quite clear that the mechanistic
position cannot explain all of evolution (Chapter 22).
REFERENCES
Because the bibliography is only a short selection, all works in the category of
Paull920 and Wartburg 1969 are not listed. For practical reasons, many anthol-
ogies or collections will be mentioned under the editor's name only, in spite of
the injustice to the actual authors.
The order in which authors are listed in these reference notes is basically
chronological (serving thus the history of linguistics as well), but it is not strictly
so. Other factors that influence it are the order of treatment in the text itself,
the degree of a work's influence on this book, the degree of difficulty of the
work, and the language the work is written in.
When the "working date" used in a cross reference to any publication has
been superseded by the actual publication date during the production of this
book, the actual date may be found appended to the bibliography entry in
parentheses.
CHAPTER 2
2 Pictography
Semology' Ideography
Semasiography
Q)
b.O
3
~
::s
bJ) Morphology
!=:
~
~
4 Alphabetic
b.O
Phonology writing .s
.t::
Graphemics ...
~
Graphic
channel
one wants to memorize or identify an object with a picture, one wants to retrieve
the linguistic counterpart of the sign. The connection may be quite loose; for
example, a picture with a tree and an ax might be read off as the felling of a tree,
to hit the tree with an ax, cutting wood, and so on. In other words, the semantic
experience is much more definite than the corresponding formal expression in
language. This situation, if it exists, can appropriately be called semasiography
(writing meanings), but it is still only a forerunner of writing, because the reader
has so much leeway (as does the writer in encoding his messages, but to a lesser
degree). Semasiography is roughly represented by link 2 of Figure 2-1. In prac-
tice, it is impossible to distinguish between ideography and semasiography,
because the semological units in language are hard to identify. Only when the
connection between the graphic sign and the linguistic unit is definite and con-
stant do we have real writing. The graphic sign has to be connected with sound.
If the sign holistically covers a sequence of sound units (e.g., a word), we have
logography (which is often mistakenly called ideography). The important differ-
WRITING AND LANGUAGE 33
ence between link 2 and link 3 is the strict conventionalization of the picture and
its reading. If the graphic sign represents a syllable, we have syllabic writing,
and if individual sound-units, alphabetic writing (the study of which is often
called graphemics), as represented by link 4. All the nonsemasiographic systems
are called phonographic, because they represent the phonological side of the
linguistic sign. The theoretically possible link 5 does not seem to be realizable:
phonological units often co-occur in one (temporal) segment, whereas writing
must convert this simultaneity into two-dimensional space. Thus special tactic
conventions are necessary, and these must be handled in a separate graphic
system (link 4, graphemics). In graphemics, the procedures are parallel to
phonemics (§§ 10.1-10.5). An invariant grapheme may have positional allo-
graphs; for example, the Greek sigma, J.:, has lowercase variants u and s (word-
finally); and instead of the sequences TT-u and K-u, or TT-s and K-s, unit symbols
,P and g occur, respectively.
[2.2] Most writing systems include signs that are not phonographic; for
example, in Europe, a star preceding a date means 'birth', and a cross, 'death'.
The sign 1 is not peculiar to any given language, but can be read as one, eins,
yksi, and so on. A writing system may have 'determinatives' that label abstract
semantic or syntactic categories, as in Hittite: appropriate words are prefixed
with signs meaning god, wood, country, man's name, nation, and the like; these
are purely graphic and do not reflect any trait of the spoken language. In English,
similarly, names are written with capital initials; in German, all nouns are; and
so on. Many systems combine both logographic and alphabetic (or syllabic)
elements. Thus English orthography is mostly alphabetic but is to some extent
logographic, since the relation between sound and graphic symbol is not one-to-
one; different ways of writing the same sound differentiate homophonous mor-
phemes. For example, the diphthong fay/ can be written (among other ways)
igh and iCe (where C represents any consonant), and the letter w is not pro-
nounced in an initial cluster with r. Possibilities of this kind are not used whim-
sically, but with specific linguistic signs: right, rite, wright, and write, all spell
/rayt/. The alphabetic constant is r-i-t; the rest is determined by the particular
word. The system is haphazard and accidental, being the result of etymological
spelling, and does not represent any conscious effort to reduce ambiguity. Note
that logographic writing represents words and morphemes, as the name implies;
logograms are linguistic signs that embody both the formal morpheme and the
semantic unit it represents. Basically, all four words (right, and so on) are the
same morpheme /rayt/. In general, one can know when orthographic i represents
/i/ as in win, and when it is fay/; there are a few exceptions like wind, which is
either /wind/ or fwaynd/. Compared with other European orthographies, the
English one is heavily logographic, although always on an alphabetic skeleton
(e.g., pane-pain, rain-J•ein-reign, hare-hear, cite-site-sight, and so on). This
system has definite advantages in communication, because it differentiates be-
tween homophones, and it accommodates easily great dialect diversity, but, of
course, learning the system is a burden. In writing systems, where the logo graphic
34 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
symbols are holistic (not built up from alphabetic parts), phonetic indicators are
used to distinguish between morphemes. By way of illustration, let us assume
that there were an English logographic writing system which used the picture of
the sun for the morphemes sun, day, and bright. Pronunciation could now be
specified with phonetic markers n, i and t, for example, SUNn 'sun', svNi 'day',
and SUN! 'bright'. In principle this is the system of Sumerian cuneiform writing,
for example. It is more logographic than, but structurally similar to, the Modern
English system. Both systems lie between morphology and phonology, if mor-
phology is taken as a collection of words or linguistic signs. Real morphemic
writing is seen when a logographic sign is transferred to other linguistic signs
that use the same phonological shape, for example, when the picture of the sun
(in our imaginary English) is not only the sign of sun but also of son, or when the
picture of the eye represents both the noun eye and the pronoun I. Such signs
write sequences of sounds rather than words or meanings.
[2.4] Alphabets can refer to different levels of phonology, that is, to phones,
phonemes, or morpho phonemes-in other words, to various segmental units. In
Middle High German the syllable-final devoicing of stops was generally written
as follows: grapfgrabes, lop/lobes, tacftages, sicfsiges, leitfleides, or zeigenfzeicte,
and so on (glosses below), although toward the end of the thirteenth century
g replaces c [k] in these words. Thus the writing of these stops represents the
phonetic/phonemic level. In Modern High German the nominatives are written
Grab 'grave', Lob 'praise', Tag 'day', Sieg 'victory', Leid 'pain', and the past
tense zeigte 'showed', although the pronunciation of the stops has not changed.
The stops are now written morphophonemically; that is, morphophonemically
WRITING AND LANGUAGE 35
voiced stops are always written with invariant symbols, because the voiceless
variants can be predicted from their environments. The link between sound and
writing has moved up the hierarchical scale; this is supplemented by the capital
letters, which indicate the category 'noun', a category even more abstract than
the morphophoneme. These orthographic customs do not reflect any change in
the language itself.
In English, morphophonemic spelling generally arose from a different source:
the orthography remained the same, whereas the language changed. The stressed
vowels of pairs like sane/sanity, divine/divinity, and serene/serenity were once
pronounced as well as written the same (see § 1.3). When the vowels changed,
the orthography remained unchanged and thus became morphophonemic (that
the vowels are contrastive, and not allophones, is proved by pairs like vain: van,
fine :fin, feed :fed). The same outcome has two histories: in the German case,
orthography was pulled up to a higher structural level; in the English case, the
sound system dropped away from the orthography, which remained higher up in
the hierarchy (see Figures 1-2, 2-1). (As in all other matters, English orthography
is not quite consistent and has respelled a few items, for example, vain/vanity
and Spain/Spaniard.)
(Her breast and her bright throat, displayed bare, shone clearer than snow that
falls [is shed] on hills ... That nothing was bare of that lady but the black
eyebrows, the two eyes and the nose, the naked lips ... A worshipful lady on
earth one may call her, by God. Her body was short and thick, her buttocks
smoothly swelling and broad, more delicious to taste was what she had with
her [on load].)
(Now we shall praise the guardian of the kingdom of heaven, the might of the
creator, and his intelligence, the work of the father of glory, as he, eternal lord,
established the beginning of each of the wonders.) The situation is exactly the
same in other languages (e.g., Old High German, Classical Greek, and so on),
but the Old English example will suffice.
end product could be chosen for the symbol (§ I. 12). These developments bring
us to logographic signs, which combine iconic, indexical, and symbolic elements.
[2.8] Pure logographic systems seem never to have existed. Even if in-
dexical relations made the drawing of pictures possible for many items, there
still remained words like to be, proper names, and the grammatical morphemes.
What happened was that the graphic sign was used independently of meaning
and started to represent form alone. This happened through homophones.
Thus the Sumerian word ti 'arrow' was written with a picture of an arrow, but
the same sign was used also forti 'life'. In English the verb be could be written
with the same symbol as the insect bee, and we have already seen another
hypothetical example in English, sun-son. Now the symbol does not represent
a word (linguistic sign), but a form, a morpheme. The step that produced this
new relation is called the rebus principle, or the principle of phonetization. It is
interesting to note that this step depends on an iconic relation between sound
and its graphic symbol. Similarity or sameness on the phonic side establishes
similarity or sameness also on the visual side. This is thus a diagrammatic rela-
tion, which underlies one of the most powerful forces of linguistic change,
analogy (Chapter 5). Here, then, an iconic first stage produces a more conven-
tionalized symbol, whose iconic elements may become totally obscure. Similarly,
in language, when metaphors become fossilized, they become symbols without
analyzable parts.
Once phonetization had taken place, semantically empty syllabic signs arose
automatically, because many of the morphemes thus written would be mono-
syllabic, in fact heavily so in Sumerian (some 56 per cent of the vocabulary),
Chinese, frequently also in Egyptian, and Hieroglyphic Hittite (Luwian). All
these languages had syllabic signs in addition to logographic signs. Thus the
so-called logographic writing systems are actually logosyllabic. A purely logo-
graphic writing system would be very unwieldy indeed, because each word would
require its own symbol. One other way in which the pressure arising from the
need for so many symbols was alleviated was the use of compound symbols (i.e.,
iconic arrangements of the existing symbols). For example, the symbol for
Chinese wu 'military' combines the characters of chih 'to stop' and ko 'arms'.
This is exactly what language does with its relatively limited number of mor-
phemes, combining them into compounds, idioms, and other arrangements (e.g.,
man of war).
(2.10] When this system was passed on to the Greeks, they reinterpreted
a sequence like bq the "logical" way. As the two signs stood for the sequence of
two sounds, ba, it was reasonable to make the segmentation directly b-a rather
than the original ba(-a). The Semitic syllabic sign for a glottal stop plus vowel
became thus the Greek letter a, and the Semitic sign for bi, be, and so on, was
stripped of its vowels to give the Greek letter b. The alphabet was born. Now
each letter represented a single sound segment, either a consonant or a vowel.
Single symbols for a cluster of two consonants did exist (e.g., 1/J = ps, exactly
like Roman x = ks), as they still do in modern alphabets, but these are peripheral
"aberrations." Only the Greeks are known to have developed an alphabet, and
it has now spread all over the world, mainly through the Roman adaptation.
(The impetus for the rise of the Korean system is not known; it may be original.)
In more recent times, linguists have applied the alphabetic principle to develop
universal phonetic notations, which are, of course, known to the readers of this
text from introductory descriptive linguistics (a basic sketch has already been
given in §§ 1.4-1.5). Suprasegmentals are the last phonological features to be
accommodated by alphabets; their inclusion is mainly due to modern linguists,
although there were antecedents in Greek Alexandria and India. For languages
in which tone and pitch play an important phonological role, there is the addi-
tional possibility of writing the suprasegmentals to the exclusion of the segmental
features. In a way, the drum and whistle 'speech' reported from West Africa
and Mexico convey suprasegmental units only. But such systems are not instances
of writing; rather they are parallel to radio broadcasting with a poor signal,
which lets only parts of the sounds through.
value it has in the Roman tradition, although it is also used for indicating length
(German Lehm), spirants (English this, alphabet, shin), affricates (English chat),
or nonaffricates (Italian che [ke]). In Classical Greek H symbolizes [e·], in
Modern Greek [i], and in Russian [n]. Then again the sounds [h, e·, i, n] are
written with completely different shapes in the Indic Devanagari, which also
derives from Semitic sources. This clearly shows that the modern letters are
symbols; but their arrangement within any particular language is iconic. Writing
is a map of sounds or morphemes. Chartographic maps contain symbols that
stand for woods, forests, bridges, swamps, mountains, houses, and so on; these
symbols represent diagrammatically (iconically) the relations between these
features, once we know the rules to read them off (e.g., where north is, and so
on). Similarly, once we know the rules for interpreting letter sequences, we can
read the relations between the sounds. Letter order represents the sequence of
sounds iconically. This is like language, as we have seen, where the symbolic
units are arranged iconically. It is interesting that alphabets use compound
symbols (e.g., sh, th, and ch) exactly like logography or language in general.
The development from iconic to symbolic shapes can also be dictated, in part,
by writing materials. Thus the use of clay tablets and the stylus transformed the
iconic pictures of the earliest records into completely symbolic arrangements of
dents. The Sumerian symbol for UMBIN 'wheel ' is a stretch of some twenty im-
pressions in the clay, having no similarity whatsoever to the original picture,
although, of course, the large number of wedges still represents the spokes and
the rim, in an unassembled or dismembered state, as it were. Similarly, the
Germanic runes have no curved strokes, because they were incised in wood with
a knife. In China the use of paper and brush resulted in graceful characters,
and, in fact, calligraphy is an important aspect of Chinese writing. Again, there
are parallels in language, where external influences also cause change-mainly
in the form of borrowing, but also because of changes in the material culture.
[2.14] The above was a brief reminder of the rather extensive influence
that writing exerts on language in various ways which are not obvious to the
42 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
[2.15] The spelling of particular words can also be influenced by the spell-
ing of other words. This is spelling spelling. Thus the "past" tenses of shall and
will write the I which was once pronounced: should, would. The past tense of
can has been influenced by these, being spelled could (compare delight § 2.6).
Similar things happen in pronunciation, where words in particular semantic
subsets influence each other (blends and contaminations). These kinds of changes
have also been called analogical; the analogy may come from another language
or an earlier stage of the same language. Thus the English word island never had
an [s] in it; the Old English form was iegland (German Eiland, Swedish Oland).
Similarly, Modern English debt comes from ME det(te), which was borrowed
from French without a [b] (and the French spelling is still dette). The prestige
of Latin, however, was strong enough to modify the English spellings, making
them more iconic with the Latin counterparts insula (which is altogether un-
related to English island) and debitum, decreasing the iconicity between the
English sound and the corresponding spelling. In Latin, of course, the words
did contain [s] and [b], respectively.
WRITING AND LANGUAGE 43
[2.16] It is important to note that in all three cases mentioned-spelling
reform or "pronunciation spelling"(§ 2.3), spelling pronunciation(§ 2.14), and
"spelling spelling" (could, debt§ 2.15)-the driving force is iconicity (for 'pro-
nunciation pronunciation' see§ 9.10). In the first two, it is the iconic relation
between sound and letter. The letter is a symbol of a certain sound, and the
iconic tendency would always have it as a symbol of the same sound. This drive
toward one-to-one correspondence between sound and letter can result in changes
at either end of the relationship. Spelling spelling transcends the relation between
sound and writing, because it refers back to semantics or other linguistic signs,
or to the wider social context of prestige dialects or languages. But such con-
vergence is still iconic: functional counterparts become more similar on the
formal (written) side, resulting in a better "fit." In spelling mistakes and inverse
spellings we see clearly how the conventionality of writing tends to be broken
in favor of iconic representation, and we get a spelling reform when the' mistake'
is accepted as the norm. 'Mistake' gives the point of view of tradition. On the
linguistic side, every change is a mistake of this kind, although linguists avoid
value judgments by speaking of' innovations'. Such innovations generally" make
sense"; that is, they are iconic.
[2.18] Thus manuscripts can be grouped into family trees showing daugh-
ter and sister relations. Three manuscripts can show five different relationships
(shown in Figure 2-2). The missing links have been labeled with asterisks. In
principle, the situation is the same for a greater number of manuscripts, although
then the bifurcations look more complicated, and in actual practice more work
will be involved.
In Figure 2-2 we see the model which was also useful for drawing family trees
for language families and which influenced and reinforced the Darwinian theory
of the origin of species (Chapters 15, 22). In language, the counterpart of the
mistake in textual criticism is a linguistic change, an innovation, which corre-
sponds in turn to mutation in evolutionary biology. Language and biology seem
to be parallel, in that small shifts can ultimately lead to great differences. Newest
results from both fields reveal extensive structural parallels between linguistic
and biological change; the danger of making false comparisons lies only in loose
terminology, or in taking the terms of such comparisons too literally (Chapter
22).
A *X *X
A
I
*X N.. *X
1\1\ I
B c B c B c A B
II\
C A B c
(a) (b) (c) (d) (e)
FIGURE 2-2. Five possible derivation trees showing the copying relationship
between three attested manuscripts A, B, and C.
Thus we have seen that writing does not only affect language, but that a study
of textual traditions has provided a useful model for comparative linguistics.
REFERENCES
General: Pulgram 1951, 1965, Gelb 1963, Diringer 1968, M. Cohen 1958, Jensen
1969; 2.2 Venezky 1970; 2.5 Frey 1966; 2.7 Wescott 1971; 2.8 Chao 1968, 2.10
Th. Stern 1957; 2.12-2.13 Malkie1 1968; 2.17-2.18 Maas 1958, Dearing 1959,
Hoenigswald 1960a, 1966, Stevick 1963.
CHAPTER 3
LINGUISTIC VARIATION
though the actual manifestations are utterly different (see § 3.5). Thus all the
dialects of English share more or less the same underlying system, although
their surface variety is immense. In American English alone the vowel of grass
can be [e;), re;), rer, re, a, a;)], not to speak of English on other continents. Such
differences do not impede communication at all, because people who come into
contact with any of these variants learn the proper correspondences; for ex-
ample, my [re;)] corresponds to X's [e;)] and Y's [a::J]. Every speaker builds such
correspondences into his grammar, depending on the dialects he hears, since
he is more likely to go on speaking his own variant (e.g., [re;)]) than to adopt the
variants of each interlocutor. (Often there is no noticeable difference between
two dialects, although, in fact, there must be.) For a striking example, let us
look at a few items from the so-called "Fox dialect" (spoken in the community
of Fox, north of Red Lodge, Montana) in juxtaposition with "General Ameri-
can" forms. Keeping the notation of the published report (even though it looks
very odd to professional linguists), we can emphasize those correspondences that
show difference3 between the dialects by boxing them in:
prelsfiliJ rn~
preruiJ rn~
'passing' 'because' 'busy'
The speakers in and around that locality know these correspondences, although
they are likely to speak one dialect only. The Fox dialect, extreme as it is, is a
derivative of General American in the sense that it can be mapped from General
American with a few rules: lengthening in certain environments, devoicing, the
merger of the dental slit spirants with dental stops, merging [v] and [w] in one
sound transcribed [vw], and so on. Merger-the situation where two or more
contrastive features of one dialect correspond to a single feature of another-
LINGUISTIC VARIATION 49
is well known from those Southern dialects where pen and pin have the same
vowel. In contact with this dialect and with others, speakers learn the proper
correspondences e-i, and i-i. The real sound substance of an actual grammar
are these interdialectal sets of correspondences (diaphone[me]s, see§ 14.3) and
not the segments in the speech of one speaker only(§§ 13.7, 13.8). Nobody speaks
only with himself; at the very outset-in the language learning process-one
needs other speakers.
[3.3] The examples have so far referred to regional variation. But any
speech community displays systematic variation on other scales-social layer
(occupation, ethnic background, and so on), age, sex, and social context. The last
is known as 'style', or more technically as 'register'. Most speech communities
have at least three varieties: the normal conversational, plus something above
it (formal) and below it (substandard, slang). All these factors are systematically
incorporated into the speaker's use of the language, and thus should be spelled
out in our grammatical descriptions. Practically no descriptions satisfy this
requirement. Of course, the notation and the amount of work necessary create
immediate practical problems. Figure 3-1 correlates the pronunciation of [9] in
Lower class-----------
Working class--------80
oL_~==::::::::t~~
A B c D
Casual Careful Reading Word
speech speech style lists
Contextual style
FiouRE 3-1. Class stratification diagram forth in New York City. [Reprinted
with modification from William Labov, "Phonological correlates of social
stratification," American Anthropologist, 66: 6, pt. 2, 169 (1964) (©American
Anthropological Association).]
50 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
New York City (where it alternates with an affricate [t9] and a Ienis (unaspirated)
stop [t]) with different social strata in different contexts. The upper middle class
is closest to having spirantal pronunciation (the prestige form) in every style.
The lower middle class comprises two groups: those who are able to shift con-
siderably toward uniform spirantal pronunciation in careful speech and those
who cannot. The lower working classes are clearly distinct from these by using
fewer purely spirantal pronunciations. But within each class pure spirantal
pronunciation increases with greater formalization of style and context. The
same kind of orderly variation can exist between sexes and age groups.
Pronunciation/language is one of the strongest social indicators; it tells about
the speaker's occupation, income, education, and social attitudes and aspirations.
Speakers who want to climb the social ladder must know quite well the sets of
correspondences that obtain between their speech and that of the prestigious
class. Hypercorrect forms clearly show that speakers are using correspondences
rather than individually learned items (e.g., in the case of the little girl who
moved from an r-dialect into an r-less one, where her playmates said [ya·d]
instead of [ya·rd]. She overdid this relation when she reported to her parents
about a giant she had heard about called [ga·rd]. The prestige factor is, of course,
rather special here, as the girl tried to use the parents' norm when talking to
them; but it is prestige, nevertheless). One single rule handles the corresponden-
ces automatically only when there is a two-way bidirectional mapping relation
between the dialects, as between the [t] of the lower class and the [9] of the
middle class in New York City. Once this relation is learned, no mistakes arise.
But when there is a one-way mapping relation into a dialect with fewer con-
trasts, speakers at the "impoverished" end have to learn every item separately.
Thus a Fox speaker learns that his [1, (,!, vw, tS] each take part in two corre-
spondences, with [t, 9; d, o; v, w; tS, dz], respectively; and an r-less speaker
finds out that his [ga·d] corresponds to both [ga·d] and [ga·rd]. This operation-
the matching of corresponding segments between two formal varieties of the
same meaning-is known as the comparative method. This method enables one
to discover the single sound system underlying the variations that show diagram-
matic relations of sounds against the same semantic background. In General
American and Fox, for example, the meanings 'because' and 'busy' show the
correspondence b-p for the first consonant and z- s for the last consonant. The
comparative method will be treated later; the notion of the correspondence is
introduced here because it is a relevant concept in a multidialectal synchronic
grammar, which is the general kind of grammar for any speaker(§§ 8.16f., 11.2).
Throughout the remainder of the text, two conventions of writing the sets of
correspondences will be used: horizontally, as b-p, or vertically, in a boxed-in
cartouche as above. The former saves space, but the latter brings out maximal
clarity for analysis.
[3.4] Variation exists not only in sounds, but also in morphology and
syntax . One of the most obvious features of slang or occupational jargons is of
course a special vocabulary. In some languages women use special vocabularies
LINGUISTIC VARIATION 51
among themselves in addition to special phonologies, and this has led to reports
that they can have a special language altogether; although this is doubtful, it
shows the extent of the differences. But in any language, women, at the very least,
would use certain items more often than men (e.g., terribly beautiful). Nursery
words are an exceptionally clear instance of the differences in age gradation.
Variation in morphology exists on exactly the same scale as the sounds. The
items can be completely unrelated, like pail vs. bucket, or can have some formal
relationship, as in I saw it vs. I seen it. Differences, of course, reach syntax
also, such as I haven't got any money vs. I ain't got no money. This is a general
situation; for example, in Brahmin Tutu the negative tenses of the verb
distinguish between gender, number, and person, whereas non-Brahmin Tulu
lacks all this in its negative forms. But we have to leave the matter with these
brief hints.
[3.5] Often variation appears to be more significant than it really is. Non-
standard Negro English has traditionally been heavily stigmatized, even to the
point of declaring that it is not language at all. Superficially, the differences from
General American are large: [bol] 'bold', [fayn] 'find', [rek] 'act', [fis] 'fist'
show phonological variation, with morphological consequences in forms like
[pres] 'pass; passed; past'. Syntactic variation includes things like [hi way!] 'he
is wild'. But the relation between General American and Negro English is as
orderly as among the dialects mentioned previously. The Negro English forms
mentioned here can be derived from the Standard English ones with rules that
apply in definite environments. The difference from the Fox situation is that the
rules here obey statistical laws (see § 9.8); there are no speakers who always
apply these rules, and none who never applies them. Some rules would escape
casual analysis; for example, in general, wherever Standard English can contract
the copula (He's wild), Negro English can delete it (He wild), and wherever
Standard English cannot contract, Negro English cannot delete (as in the last
word of That's what he is). Dialects are apt to differ chiefly in low-level rules,
and superficial differences tend to be greater than those found in their deep
structures, if there are any. On the other hand, it seems that some features of
Negro English are not derivable by one-way mapping from Standard English
(e.g., he done told me, he be sleeping, which involve aspectual contrasts absent
in Standard English).
Brahmin Non-Brahmin
A. son-in-law marumaha
maap!e
younger sister's husband
elder sister's husband attimbeer maccaa
wife's brother maccina
We have returned to the semiotic situation, especially its pragmatic side(§ 1.16).
Sociolinguists now hold that speech should be studied in terms of the whole
speech community, and that the object of such study is a sociolinguistic system,
not mere language.
If the variations in different social contexts in the speech of the same speakers
are related, but very different, one speaks of diglossia; if the context requires
the use of distinct languages by the same speakers, the situation is known as
bilingualism.
(3.8 Conclusion] This chapter has shown that synchronic dialectal diver-
sity must be treated by the same method as related languages, although the
method itself was left for later discussion (Chapter 11 ). It was briefly stated that
change and reconstruction are ultimately connected with variation, and it
became clear that Figure 1-7 does indeed represent cardinal points only. Chapters
1- 3 have delineated the place of genetic linguistics within other branches of the
total science of language, providing valuable guidelines for the following dis-
cussions of change and reconstruction. Going more deeply into synchronic
linguistics is not possible in this context, although synchronic linguistics is part
of the necessary background for genetic linguistics, as our constant references
have shown and will show.
54 BACKGROUND: GENETIC LINGUISTICS AND GENERAL LINGUISTICS
REFERENCES
General: Hymes (ed.) 1964, Bright (ed.) 1966, Lieberson (ed.) 1967, Fishman
(ed.) 1968, Alatis (ed.) 1969, Graur (ed.) 1.549- 773; Fischer 1958, Gumperz
1958, 1961, 1968, Bright and Ramanujan 1964, Hymes 1968b, Bolinger 1968,
Burling 1970, Katici<5 1970, Dahlstedt 1970; 3.2 Pilati 1969; 3.3 Joos 1962,
Labov 1964, 1966, Fonagy 1956-1957, Kazazis 1970; 3.4 Burling 1970, Loman
1970; 3.5 Labov 1969, 1970, Houston 1969, 1970, Wolfram 1969, 1970, Loflin
1969, Fasold 1969, 1970, Burling 1970; 3.6 Ferguson 1959, Ramanujan 1968,
Kazazis 1968.
PART II
HISTORICAL LINGUISTICS:
HOW DOES LANGUAGE CHANGE?
CHAPTER 4
SOUND CHANGE
The fact that most English dialects have replaced a trilled (initial) r with a
retroflex spirant r, as in red, is a case of mere phonetic change. Similarly, English
t and d used to be dental; now they are alveolar, and no structural change has
taken place. In Dutch all (or most) /'s are 'dark' (much like in sill, bottle), a
change of no systematic consequence even though most of the other Continental
Germanic languages have' bright' /'s. As in the case of the dialect correspond-
ences discussed in Chapter 3, there is regularity through time. For example, Old
English ii repeatedly corresponds to Modern English fowf (which can be spelled
three ways):
1 OE lip ligan biit biin hiim riid stan hal
t NE oath own boat bone home road stone whole
This regularity between the sounds of an earlier stage and a later stage is called
57
58 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
a phonetic (or sound) law. The term 'law ' is to be taken in a physical sense, as a
statement of regular behavior. We write this 'law', this correspondence between
two stages of one and the same language, as OE ii > NE fowj, where > means
'changes into'. (Note that this is different from the dialect correspondence, for
example, General American alveolar d-Fox dentalcj, where we use the dash for
the synchronic relation. The wedge is used only for variation through time,
for example, earlier English dentalcj > alveolar d.) The law ii > jowf does not
hold, however, if the ii occurs after a w-cluster : hwii > who, twii > two. But the
exception is regular, because it occurs in one environment only, where the rest
of the law does not apply. This is the general situation; for example, we have
the two correspondences in time
OE
or
cjwil
NE
~
which occur in mutually exclusive environments: the second (stated twice above,
the second time with conditioning shown) after a w-cluster in Old English, the
first elsewhere (i.e., ii > ow, except that Cwii > Cuw). Sound laws are not as
exact as physical laws, and linguists should not be startled to find exceptions
(e.g., swii > so; this change, however, is quite regular if w is first lost after s-,
whereafter *sii > so) (see§ 5.7).
[4.2 Allophonic Variation and Splits] Old English /f, p, sf were phonetically
[f, p, s] until about A.D. 700:/if'five', wuif'wolf', ofer 'over', porn, weorpan/
wearp 'become/became', ping, sen dan, nosu 'nose', wesan/w<£s 'to befwas '. About
A.D. 700 a mere change in pronunciation occurred in that a voiceless spirant
between voiced sounds after stressed vowel became voiced: [over, weoroan,
nozu]. Spelling remained the same, and the phonemic contrasts were not dis-
turbed; only the number of allophones doubled: /f/ [f, v], /l>/ [p, 5], and /sf
[s, z]. These voiced allophones became independent phonemes through later,
unrelated changes:
I. Borrowings from Kentish (and other Southern dialects), which had also
initial voiced spirants (readers may be familiar with the speech of Squire
Western in Tom Jones), replaced inherited items likejt£t 'barrel' (compare
German Pass) and fyxen (compare German Fuchsin) with vat and vixen.
This process was also reinforced by French loans, as in vile. Since fiett
(compare German Fett) gave fat, we have now minimal pairs vatffat,
vile/file. The speakers had to learn that some morphemes contained in-
herent v's which always remained as such, although the oldf's alternated
morphophonemically with v's in certain environments, as they still do.
2. Intervocalic long spirants /ff, pp, ss/ were simplified into [f, p, s], and these
of course contrasted with [v, o, z], which had been the intervocalic voiced
variants of the old short /f, p, sf.
SOUND CHANGE 59
3. Upon the loss of final -a in Middle English we get voiced spirants in final
position, e.g., ME [bao<l] > [baa] ( > fbeyof), contrasting with the noun
[brep] (see Figure 4-3, line 4). The -a is still written as -e, but now it means
the preceding spirant is voiced: bath/bathe, wreath/wreathe, tooth/teethe,
and so on.
Here then we see how allophonic variation, which initially had no impact on
the contrast system, was phonemicized by later developments: borrowing and
sound change. The number and distribution of sound units changed. English
developed morphophonemic alternation between voiceless and voiced spirants
after the original endings were dropped. In French the development is the reverse,
although the result is the same. Latin masculine novu(m) and feminine nova(m)
'new' started out with the similar v ('same' in French) but different endings.
When the masculine ending dropped, the feminine one was still retained as -a:
• masc. neuv, fern. neuva. Final v was devoiced, neuf, and then with the drop of the
-a we get [ncef, nce·v ], an exact parallel to the English grammatical alternation
safe/save, and so on. In French it is masculine versus feminine, in English, noun
versus verb; in French, v > v "'f, in English, f > f,..., v.
By about 600 B.c., Latin s between vowels became r (all s's between vowels
became r's).
To emphasize that sound laws are historical events that occur at a certain time
in a certain language under certain conditions, Edgar Sturtevant compared the
above statement about Latin rhotacism to another possible hypothetical his-
torical event, which he called "The Law of Waterloo":
All Prussian soldiers six feet tall were killed in the battle of Waterloo.
The Latin law as formulated does not seem to be completely true, since there
are still words with intervocalic s's: divisus 'divided', causa 'cause', caesus 'cut
down', visus 'seen'; nisi 'unless', desino 'desist', and so on. Morphophonemi-
cally the first four words haves's that are ld + tl (dlvidere 'divide', videre 'see' +
past passive participle ending -tus: scrip-Ius 'written', ama-tus 'loved', etc.),
and, in fact, Old Latin orthography shows a double ss in these words; for
60 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
example, caussa. That is, at the time when intervocalic s changed, these words
did not haves but ss, which later became short after long vowels and diphthongs.
As for nisi and desino, they are compounds that apparently had not yet been
formed at the time when s changed, and thus the s's in question were word-
initial: si 'if', sino 'allow'. These cases do not fall under the conditions of the
law; they are Prussians not yet born at the time of the battle. Another set of words
shows intervocalic s's after short vowels without these~ two possibilities of ex-
planation: asinus 'donkey', casa 'hut', rosa 'rose '; genesis, basis. However, we
can easily establish these as loans from other Italic languages and Greek. Hence
these did not exist in the language at the time of the change; they are Prussians
naturalized after the battle of Waterloo.
Thus the first formulation has withstood quite well the apparent discr pancies
in Latin words showing intervocalis s's. A more serious attack comes from two
words: miser 'miserable' and caesaries 'hair ' . But note that here the environ-
ment of s is different, in t. at it is followed by an r. Now our law needs an •
additional clause:
These are Prussians under six feet tall and hence not subject to the law of Water-
loo. According to this clause, the word for 'sister', soror, should be *sosor
(compare sister, Skt svasar-). Here one notices that the s, which has, in fact,
changed before a following r, is preceded by s earlier in the word. We must add
a further clause :
In other words, this Prussian is six feet tall after all, although his posture ap-
peared bent. Now the law covers the facts quite well. There is only one excep-
tion, nasus 'nose'. But one exception against hundreds of regular cases does not
invalidate the law. In historical linguistics one always finds a certain number of
unclear cases or irregular changes.
This and the English and French v "'f case, as well as many of the following
examples, will show how sound change creates morphophonemic alternation.
Looked at from the point of view of analysis, they are examples of internal re-
construction, because internal reconstruction is based totally on morphophone-
mic analysis (§§ 10.7-10.17, Chapter 12).
I II III
cor > coeur 'heart' centum > cent ' hundred' cantare > chanter' sing'
cliirus > clair 'clear' ceruus > cerf 'hart ' carbo > charbon 'coal'
quando > quand 'when' cinis > cendre 'ashes' causa > chose 'thing'
SOUND CHANGE 61
That is, k > k (k remains before o and in a cluster), k > s before front vowels,
and k > s before a. These are three correspondences through time, comple-
mentary in terms of their environments in the earlier stage:
Environments in Latin
~
E 1 Latin
French
fklk
~
5tJ fkls
~
it] ~s ~
~
I II III OE IV NE
*hal hal hal 'whole'
*ha/-ijJ *httli/J httl/J 'health'
*hal-jan *httljan httlan 'heal'
*dom dom dom 'doom'
*dom-jan *domjan doman > deman 'deem'
*gos gos gos 'goose'
*gos-i *gosi gos > ges 'geese'
*miis miis miis 'mouse'
*miis-i *mysi mys > mis 'mice'
*fiil fiil fiil 'foul'
*fiil-ijJ *fylijJ fyljJ > filjJ 'filth'
*fiil-jan *fyljan fylan > filan '(de)file'
Umlaut itself is thus the creation of front allophones [re, o, y] to the phonemes
/a, 6, fl/ (the step from column I to II). Such sounds did not exist earlier, and
even now they are just automatic variants, when l or j follow: /a/ = [a, re],
/6/ = [6, o], and /fl/ = [u, y]. This was a mere change in pronunciation, which
later acquired structural significance only through an unrelated phonetic change
-the loss of those high front vowels in unaccented syllables (i.e, the step from
62 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
The allophones have split up into ten phonemes. Th~:re was a clear hierarchy
to the catalysis. First the mid vowels served as catalysts to fa/ and then the high
vowels were catalysts to all the others, remaining intact themselves. But this is
not the whole story. When final -£ is dropped, a preceding :J gives 5: m:Jse-+
m5s 'to sleep', t:Jre -+ t5r 'to remain'; and similarly i, although disappearing,
SouND CHANGE 63
i ii u
(, 5 ., )
a
C_]J
FIGURE 4-2. Rotuman vowel triangle showing the splitting up of vowels
through raising umlaut (upward arrows) and fronting umlaut (left pointing
arrows).
umlauts a preceding o: mori -+ mor 'orange' and tori-+ tor 'to use extrava-
gantly'. Similarly, juri-+ fiir 'to turn', fuqi-+ fiiq 'thunder', and psi-+ ds 'to
go to see', andjpji-+ jdj 'to shape'. In other words, we end up with fourteen
vowels on the phonetic surface, although, again, the morphophonemic groupings
are not affected. Scholars differ on the number of phonemes they posit for Rotu-
man; the number has been variously posited as 5, 7, 10, 12, and Churchward's
14 (presented here). The derivation of the vowels can be summed up as in Figure
4-2, where the curved upward arrows indicate the first raising umlaut ( t) caused
by i and u and the straight arrows the fronting umlaut (+-) caused by i and e.
The splitting up of the vowels is still a synchronic mechanism in the language,
since eight of the vowels are unambiguously analyzable as sequences of two
morphophonemes: fa/ = Ia + el, fa/ = Ia + il, /o/ = Ia + t>l before h and q,
/Q/ = Ia + ul, fof = It> + ul, /o/ = I" + il, /5/ = I" + el, and /il/ = lu + il.
This analysis is strongly supported by the fact that the sequences with high and
mid vowels followed by the mid and low vowels, iCe, iCJ, iCa; uCe, uCJ, uCa;
eCa; JCa (where C represents any consonant), do not drop the final vowel at
all. Instead, the final vowel is transposed to the other side of the consonant:
ieC, ioC, iaC; ueC, uoC, uaC; eaC: JaC, for example, sib--+ siok 'to be untrue',
pija --+ piaj 'rat', pure --+ puer 'to rule', hula --+ hual 'moon', and hJsa --+ hJas
'flower'. These diphthongs represent single syllables, exactly as /il/ = lu + il
infiiq, and so on. Note further that when the mid vowels get next to the high
ones, the raising umlaut occurs (as it does also for a sometimes, but we have to
ignore it here). Because the front umlaut and the metathesis are so closely
related, one might try to interpret the vowel drop the same way (i.e., sere >
*seer> ser, and so on). No vowel transposition or drop occurs if the word ends
in two vowels, but, again, the second vowel from the end is affected, because it is
shortened in the "incomplete phase": for example, pupui--+ puptii 'floor', lelei
--+ lelei 'good', keu--+ keu 'to push', and jaJ --+ jiiJ 'spear'. Thus even here we
have ultimately a slight reduction in length, although not the loss of a whole
syllable as in the above cases.
64 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
All the above changes are still living morphophonemic processes in Rotuman,
and this is why the arrow~(§§ 6.7, 6.8) rather than the wedge > was used in
the examples. There are also a certain number of exceptions to the above changes/
rules, but such things are to be expected in every language. The overwhelming
regularity is the factor that counts, and it was presented as regular here. Two
aspects of the Rotuman case are unusual. It is noteworthy that the ..:-umlaut of
a raises the a much more than the (second) i-umlaut. The corresponding ;J- and
u-umlauts of a remain in the more usual proportion: the high vowel raises the
aone step higher than the mid vowel. The reason for the asymmetry is that the
first raising umlaut pulled a into the back vowels and only the second fronting
umlaut brought the a among the front vowels. This is how a lost ground it
the i-fronting. And when the latter occurred, the ..:-raising was a completed fac.t
[4.7) Even more curious is the fact that the fronting umlaut operates only
when the front vowel is dropped. In those environments where the final vowel
remains, the preceding vowel shows the shape at th1~ tail ends of the straight
arrows (Figure 4-2). That is, the drop of the vowel and the fronting are simultane-
ous processes; there is no gradual contact influence that would produce fronting
before the drop of the vowels. Linguists usually would like to posit a single step
at a time, for example, first metathesis, fuqi > *fuiq, and then *fuiq > filq, and
so on. We have no direct evidence for the middle term; rather, by all appearances,
the metathesis and the fronting could as well have been simultaneous. The same
paradoxical situation is known also in Old Norse i- and u-umlaut, as well as in
Livian (a Baltic Finnic language in Latvia). When a Livian final -i drops, it
causes fronting of a preceding back vowel, e.g., [nom. sg.] tammi > tiimm 'oak'
and [gen. sg.] tammen > tamm. No umlaut results if the front element remains
in the word, either as a vowel (e.g., vanhim > vanim 'oldest', vasikka > va's'ki,
t'a'iski 'calf'), or as palatalization (e.g., patja > pad'a'mattress'). A nominative
singular like nurmi drops its final -i and gives nilrm 'lawn', whereas the partitive
plural retains its i and remains as nurmidi, with a back vowel. Again we have
one apparent exception, kaksi > kaks '2 ', but s instead of s still reflects the i
as a feature of palatalization.
We have seen a typical case where, as the words of a language get shortened,
the number of sound units multiply; and mere changes in pronunciation were
given a different structural status through a different change, the loss of the
original conditioning factors.
[4.8 The English Vowel Shift) The great English vowel shift, which
affected all long vowels, had very little structural significance when it first hap-
pened, because the nuclei remained, on the whole, separate from each other, as
shown in Figure 4-3. "Long vowels" are here analyzed as long rather than
diphthongs to begin with, merely for the sake of simplicity. The important point
is that although the vowels shift places, they all remain distinct until the time
of Wordsworth, when there is a reduction and redistribution in the number of
"long" vowels. The shift is still going on in English. But before the time when
SOUND CHANGE 65
Modern
Chaucer Shakespeare Wordsworth English
the long vowels started to shift, long vowels were shortened when followed by
two consonants or one consonant and two unstressed vowels, thus:
-CC -CVCV
l filth divinity
e kept serenity
/£ health
ii cranberry sanity
j [<e] (see below)
6 gosling
u husband, hussy
Umlaut, unrounding, vowel shift, and vowel shortening produce forms like
whole "' health "' heal,foul "'filth "' (de)file, and sane/sanity. As always, such
vowels still tend to go together morphophonemically, although not to such an
automatic degree here as in Rotuman. When OE ii was shortened before it be-
came ME 5, the result was a complete separation of the variants: for example,
stone and staniel < stan-gel/a 'stone-yeller, a kind of hawk'; home and place-
names like Ham-den (compare [-::!m] at the end of names like Nottingham); and
goat and Gatton 'goat-town' in Surrey (compare, also, holy [which belongs to
whole] and hallow[e'en]). The single element *a in column I in§ 4.5, umlauted,
shortened, and otherwise conditioned, has the following outcomes in Modern
English: /ow/ in whole, stone, clothes; /ref in staniel, hallow; j'Jj in Nottingham;
/uw/ in swoop, two; fa/ in hot, holiday; fof in cloth, broad(this is a dialect borrow-
ing); /iy/ in heal, speech; fef in health, next; and /i/ in silly, nimble. Again there
are also "irregular" cases (e.g.,Jolly [fool], scholar [school], and zealous [zeal]),
where shortening occurs before one consonant only. But such an irregularity is
regular in the sense that there are many cases of it.
[4.9 The First Germanic Consonant Shift] The first Germanic consonant
shift, known as Grimm's law, is another good example of a sound 'law'. Note
that Germanic is not a directly attested "historical language" like Latin or Old
66 HISTORICAL LINGUISTICS: HOW DOES LANGUAGE CHANGE?
English, but inferred through reconstruction (Parts Ill and IV). Pre-Germanic
had the following stops: p t k, (b) d g, and bh dh gh (see § 11.12) of which the
last three might have been murmured (breathy phonation). The voiceless stops
were replaced by spirants: p t k > f p x (with h in some environments at the
attested stages; see§§ 11.3, 11.7) (e.g., Latinped-: Englishfoot; tres:three(and
tooth, below); and cord-: heart). The voiced stops shifted into the position
vacated by the above change: b d g > p t k (e.g., Lithuanian dubils: deep: Latin
dent-: tooth; andfoot, heart, above; and Latin ager:acre). And, finally, the last
series filled in the new vacant slot: bh dh gh > b d g (e.g., Sanskrit bhdriimi:bear;
Skt rudhird:red; and Greek khen:goose). Latin, Lithuanian, Sanskrit, and Greek
are standing in for Pre-Germanic merely for conveni(:nce-and Greek not too
well at that-because we have no direct attestation from the protolanguage,
although the comparative evidence is quite solid. English, for its part, is standing
in for Proto-Germanic here. So far nothing much has happened, because all
three series are still distinct; but this is the nucleus of Grimm's law. (Such" mere
phonetic" changes seem much less trivial from the point of view of distinctive
feature analysis, which concentrates exactly on this level of phonology; see§ 6.9.)
But, as usual, there is one environment in which th(: shift did not take place:
when the sounds p t k occurred second in a voiceless cluster, they remained as
stops, for example,
Latin stare Latin captivus
'stand' 'captive'
English stand OE hteft
Latin piscis Latin spuere
'fish' 'spit'
Gothic fisks OHG spiwan
Thus there was a change in distribution after all, because these unchanged p t k
were identified with the outcomes of older b d g.
One further set disturbs the operation of the law as developed so far: in some
cases the outcomes of p t k are not voiceless spirants but b d g:
[4.10) There is still some controversy about the order of the stages in the
consonant shift. In addition, we have treated the Germanic sounds b d g as
though they were simple structural entities that contrasted with the other two
series. Actually, they were spirants phonetically and changed into stops rather
late (if at all, in some Germanic languages; see§§ 6.3, 9.9). Let us now sum up
Grimm's and Verner's laws, taking the dentals as a model, and using a sequence
different from the above. Figure 4-4 is one possibility. We write one change at
each stage. First we have dh > d, then t > ]J (p > f); up until stage III no
structural change has occurred. We have the same phonemes, but number 2
now has two allophones: a stop in voiceless cluster, and p elsewhere. Between
III and IV Verner's law applies: the voiced outcome of unit 2 is the san .e as
unit I, and after d > t, the stop allophone of 2 goes together with unit 3. We
started out with three units and we end up with three units, but with a new dis-
tribution, and hence a change in structure (for further tabulation see § 6.3 and
Figure 6-2).
A B
[m] [1] [m] [b] [1]
Lips closed open closed -----:)- open
Tip of tongue down up down ~up
Velum down up down up-----:)-
FIGURE 4-5. Rise of excrescent [b] caused by readjUtstment of articulatory
movement.
I. Complete loss
fxf--0
is rather infrequent, but it occurred in the development of Latin /h/ into
nothing in Romance. A more frequent situation is (diagrams to be read
from left to right only)
2. Partial merger
OL Latin PIE Germanic ME NE
t~jJ
fxf:S,:fxf s~s
~~~i
fyf fyf r~r d~t l~l
with examples already discussed above. A subtype of this would be
3. Partial loss
OE NE
fxf~x/ k~-k- acknowledge
0 0 after # before n knowledge
(see § 2.14)
70 HISTORICAL LINGUISTICS: How DoES LANGUAGE CHANGE?
4. Complete merger
OE OE/ME OE OE
5. Split
Latin French
fxf k
Jxf
< Jyf
k~~s
English, Livian,
and Rotuman umlaut
The diagram includes a split in two only, but the examples have shown that
it can be multiple. Further, we saw that
6. Excrescence
0--fxf OE jJymle > NE thimble (and so on)
does not really come out of completely nothing, because the environment
is phonetically specified. One can also look at loss as merger with zero.
Thus every split becomes a result of merger, and excrescence is a split of
zero.
(4.16] In the cases mentioned so far the sound that changed leaned to a
sound that followed. This is called regressive assimilation. The opposite is ac-
cordingly progressive. The drawback of these terms, however, is that it is very
difficult to remember which is which. A simple remedy is to call the first type
anticipation, because one or more (even all) features of the following sound are
anticipated in the production of the preceding sound (whether in contact or at
a distance). The opposite is lag, in which some or all features of a sound persevere
into the next or following sound; for example, [sevQ] > [sevrp.] seven, in which
the labial articulation of the preceding labio-dental is maintained for the nasal.
Compensatory lengthening is a case of lag. In this event a consonant is lost as far
as the articulatory adjustments go, but its place is taken by (the bundle of
features of) the preceding vowel. A Pre-English nasal was lost before a voiceless
spirant, as in *tonjJ, *fimf, and *gons (compare Swedish tand, Gothic fimj, and
German Gans), giving its place over to the vowel and yielding OE top, gos, !if
(as it were, toojJ, goos, andfiif; that is, the assimilation is total, and at contact).
Lagging assimilation is common in inflectional and derivatory suffixes, which
lean backward toward the stem or root. The English past-tense marker /-d/
generally assimilates in voicing to the stem-final consonant of the verb, for
example: /kik-t/ kicked, slep-t, /r;)s-t/ rushed vs. /h;)g-d/ hugged, fstreb-d/ stabbed,
/st;)r-d/ stirred (compare, however, mean-t). We have here partial lagging contact
assimilation. The original vowel of the Finnish illative case was e, as it still is in
one word only, sii-hen 'into it'; otherwise thee assimilates to the preceding vowel,
for example, piiii-hiin 'head', piii-hin 'heads', puu-hun 'tree', maa-han 'land',
suo-hon 'bog', kyy-hyn 'viper', yo-htjn 'night', and tie-hen 'road'; we then have
total lagging assimilation at a distance. Whether total or not, this is what gen-
erally occurs in vowel harmony, as in Finnish, Hungarian, Turkish, and so on.
74 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
lag/anticipation
partial/total _____ I _ ___ contact/distance
ASSIMILATION
This does not explicitly treat articulatory space (Figures 1-3, 1-4), which is the
frame for other changes as well.
(4.20 Features and Segments in Change] It has become clear from the
survey and classification of sound changes that change applies to anything from
one feature to a whole segment, that is, a certain bundle of features (compare
the end of§ 4.12). Partial assimilation shows the former, total assimilation the
latter. Most frequently, however, features change one at a time. In one part of
the Germanic consonant shift, p t k become f jJ x; but this is one change only-
closure is replaced by spirantization in certain environments. When Rotuman
:J > oat the same time that & > e, or when Sanskrit ai > e [e•] and au > o [o·],
there was only a one-feature change in each language. On the other hand, if a
segment is added or lost, a whole bundle of features must be accounted for, as
in Latin scola > Old French escole. The symmetry of the above one-feature
changes has also been called analogy, that is, diagrammatic (p:fas t:p as k:x).
There is not always complete symmetry, however. The Latin medial stop dis
SOUND CHANGE 77
lost in Spanish, cadere > caer; but b is not, habere > haber, [b] > [B]. On the
whole, there is more play in the dental and velar areas than in labial articulation
(see§§ 11.5-11.8, 18.13 and Figure 18-2).
[4.25] It is interesting to note that the only relic in French of the Latin
accusative ending -m occurs in the pronominal rien, a grammatical particle. In
English, also, adverbs preserve endings that have been dropped from the nouns.
A well-known example is whilom 'in former days', which derives directly from
OE hwilum, dat. pl. of whil 'while, time'. This is, in fact, directly parallel to
riiiijgin, but a closer parallel is that English also used genitives as adverbs (e.g.,
whiles). A closure in the tongue position at the end of the word produced
whil(e)st (§§ 4.11, 4.14). The excrescence occurred only in adverbs (amidst,
against) and not if the -s was the marker of a "real" nominal genitive. The
Middle English adverbial genitives ones, twies, and j;ries had the same fate: they
were also cut off from the normal genitive. When the genitive ending assimilated
in voicing to the stem, only the productive genitive was involved (e.g., dog's fdogzf),
whereas the adverbial genitives remained voiceless and required a new ortho-
graphy to ensure the correct voiceless reading: once, twice, and thrice (these
now satisfy the environment for excrescence, and oncet fw<Jnstf, twicet /twayst/
occur). The new voiced genitive has also been introduced in some adverbs (e.g.,
Sundays, always, besides, betimes); but in any case, syntactic-semantic factors
were clearly at work. (Note that a collective pence also retains /s/ against the
productive plural pennies with /z/.)
way, and it keeps its voiceless stop. The noun from which it is derived, MHO
wecfwege 'road, way', did alternate, and accordingly the word-final fk/ was
replaced by fgf:
g
l
k>g
and the nominative became veg. Thus semantic specialization and the lack of
inflection in the adverbs cut them loose from their earlier morphophonemes. In
Standard German, also, the adverb was cut off from the noun, but with different
consequences. Starting from the MHG noun paradigm vekfvegesfvege we first
get lengthening in open syllables, which gives vekfvegesfvege. This was a normal
phonetic change which occurred in one phonetic environment; it increased
variation in the paradigm, because in addition to the k "' g alternation there was
also e "' e. Similar alternations occurred in innumerable paradigms where a
short vowel in closed syllable alternated with a long vowel in open syllable (e.g.,
takftiiges 'day'). In Standard German, all such short vowels which alternate
with long ones have been lengthened; that is,
e ii
l l etc. =
e> e a > ii
This gives us now nominatives fvek/ and ftak/ (Weg, Tag) (angular brackets
indicate spelling). The adverb ( weg) was outside the paradigm and did not
alternate, and remained with a short vowel, fvek/. When morphophonemic in-
formation plays a role in such changes, the result is always a surface sound which
is closer to the underlying unit. In all the cases given so far, the outcome was
actually identical to the "upper story" member of the morphophoneme (the
basic or most frequent alternant), but it need not be. In Ukrainian those z's
that alternate with d's become affricates, while others remain as z's, that is,
d
l
z > dz
The outcome is not identical to the "upper story" member, but it does ap-
propriate closure from the latter. The result is a compromise between the two
variants, and the old alternation d "' z is replaced by d "' dz. Normally, how-
ever, the alternation is eliminated altogether, as most of the above examples
have shown; in addition, these examples have shown the prevailing replacement
of the conditioned variant by the basic variant. There are exceptions to this,
however. In Estonian, the sequence ks which alternated with s has been replaced
by s:
ks > s teokse > teose (gen.)
l 'work'
s teos (nom.)(§ 6.14)
SouND CHANGE
Another case is the Latin replacement of the inherited stem-final s in the nomi-
native singular of s-stems by the r which was the outcome of earlier s between
vowels (§ 4.3). The inherited pattern is transmitted undisturbed in two classes of
nouns. The preservation of s injlosfjloris 'flower', mosjmoris 'custom', and rosj
roris 'dew' can be phonologically described: finals is preserved in monosyllables.
A morphological condition states that neuter nouns keep their original variation,
for example, corpusfcorporis 'body' and genusfgeneris 'kind'. (This type is very
numerous and, not surprisingly, there are a few exceptions: the neuters robus
'oak' and *fulgus 'lightening' appear as robur andfulgur in Classical Latin.)
Otherwise, that is, in polysyllabic masculines and feminines, the word-final s
has been universally replaced by the r of the oblique cases (e.g., amor 'love',
labor 'work', timor 'fear', and so on). Here too one can find a stray exception
among the numerous members of this form class: honos 'honor' is not uncom-
mon in Classical authors.
[4.29] The Finnish paradigm for 'water' includes (nom. sg.) vesi, (gen.
sg.) vede-n, (essive sg., another case) vete-nii 'as water', and so on. The stem-final
vowel is i in absolute final position, e elsewhere; and the preceding consonant is
din closed syllable, tin open syllable, and s before i (see§ 10.12). These alter-
nations are very regular and occur in a great number of nouns. The e ,. . , i
84 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
e
l
t>s before i
No assibilation occurs in words like neitifneidinfneitinii, where the vowel is an
unalternating i all the way. In some cases like this, however, the actual history
may be quite different (see §§ 6.10, 10.17, 11.17, 11.18); for example, many
scholars assume that the dental in neiti was earlier *o rather than *t. Cases like
this indicate that grammatical or morphophonemic conditioning in the syn-
chronic grammar need not mirror a grammatically conditioned sound change.
[4.31 Change and Variation] This chapter has been purely descriptive, as
is very often the case when we deal with linguistic change; this is a limitation
dictated by the historical circumstances. That is, we describe facts that (we be-
lieve) happened. This is, of course, a necessary prerequisite to explanation, which
will be treated later (Chapter 9). Even without taking the speakers themselves
into consideration yet, we have seen how variation and change are interrelated.
In most cases mentioned, synchronic variation was the source of change, and
change led to new synchronic variation. A description that ignored allophonic
variation would be very inadequate for our understanding of change. Most
changes remain evident in the synchronic workings of a language for quite some
time; a linguistic state is to a large degree a partial summary of the history of
the language. The result and source of the change generally continue to function
SOUND CHANGE ss
in the morphophonemic rules; only in a very few cases did change actually
eliminate variation at certain spots. The Romance developments of Latin k were
conditioned by its environmental variation; but in French, all outcomes occur
in exactly the same environments, kii, sii, and sii, and no synchronic alternation
remains between these(§ 4.4). In Yiddish the adverb vek fell outside the paradigm
of its nominal origin and did not undergo the change to veg. These two forms do
not alternate any longer, and perhaps for most speakers no mental connection
exists. The same happened in Standard German, but there, the separation oc-
curred through the vowels, fvek/ vs. fvek/. This does not mean that variation was
eliminated everywhere; e.g., the German syllable-final devoicing of voiced stops
(vekfveges) is still an automatic phenomenon in the language. New loans undergo
it, for example, Job [yop] and Trend [trent]. In most other cases change is vari-
ation and variation, change. Generally, then, change leads from variation to
variation. Sound change is largely unobserved, because speakers interpret it as
variation. Every speaker must be able to handle variation if he wants to com-
municate at all (Chapter 3), and speakers have no reason to know that one
aspect of variation is change.
genetic linguistics, no matter how much the slipped discs of sporadic change
may annoy the linguists.
Irregular sound change tends to occur in certain areas of grammar and
phonology more frequently than elsewhere. Iconic signs (onomatopoeia and
similar descriptive forms) resist regular phonetic change best. Although these
categories have their symbolic aspects-they depend on the particular language
-scholars, on the whole, agree as to where they expect such forms, for example,
in names for all kinds of noises, scraping, quick movements, slow movements,
tabu or unpleasant notions, and so on. The [i] in ME pipen 'to chirp' is expected
to turn out with [ai] as we saw (§ 4.8). But the vowel has not changed uniformly
and there is still a kind of [i] verb, spelled peep, although the phonetically regular
fpayp/ pipe also occurs and the instrument pipe also has the regular outcome. OE
-cwjisan, ME queisen 'to crush' should end up as /kwayz/ or /kweyz/ in Modern
English. The word, however, is squeeze with fiyf (not to speak of the extras-,
assumed to come from Old French es- < Latin ex-, as in espresser 'squeeze
out' and similar words: extract, extort). Expressive vocabulary does not invari-
ably resist change, but it can. In Classical Greek the sound that sheep gave was
appropriately something like [bre·] or [bs·]. This form has undergone the
"regular sound changes" in the modern reading of the classical word, ending up
as [vi], clearly a far less iconic shape; but the normal modern form is still [be·].
The Proto-Germanic word for 'cuckoo' was *gaukaz, which in due time gave
MHG gouch, OE geac, ON gaukr, and Swedish gok [y-]. In English and German
the words have again become more iconic, that is, cuckoo and Kuckuck. This is
obviously not regular sound change but "analogy" from the actual sound of
the bird. One should also note that in synchronic grammar descriptive vocabulary
often contains sounds not found in the rest of the language.
Frequent forms, such as pronouns and grammatical morphemes, are also
prone to undergo irregular changes. Without giving examples, we can note here
that the Rotuman changes delineated above (§ 4.6) do not take place so con-
sistently in the pronouns. On the other hand, we saw that in certain syntactic
or grammatical contexts a form can regularly remain unchanged (adverbs once
[§ 4.25], vek [§ 4.27], riiiijgin [§ 4.24], compound maantee [§ 4.24], and pronouns
rien [§ 4.25], siihen [§ 4.16]). Once more, there is regularity in the irregularity.
Then again, there are exceptions like niisus, folly, and scholar, although there
are many cases of the latter, and hence (incipient) regularity.
What aids this heterogeneity of sound change is the way it spreads. Speakers
adopt the changes at different times both in terms of social layers, individuals,
and vocabulary sets. This provides enough room for irregularities to spring up
(Chapters 6, 9).
REFERENCES
General: Hockett 1965, Postal 1968, Koch 1970; 4.1 Sturtevant 1947, Penzl
1957 ; 4.2 Penzl1957, Hoenigswald 1964a; 4.3-4.4 Sturtevant 1947; 4.5 H. Bennett
1969; 4.6 Churchward 1940, Biggs 1965; 4.7 Posti 1942, Wickman 1958-1960;
SOUND CHANGE
4.8 Wolfe 1969, R. Krohn 1969; 4.9 Foley 1970; 4.10 see§ 6.3; 4.11 E. Itkonen
1966; 4.12 Hill1936, Moulton 1967, Fairbanks 1969, Benediktsson 1970; Schane
1971, Lehmann 1971; 4.13 Lehmann 1964; 4.14 Kent 1936, Sturtevant 1947;
4.15 Wang 1969; 4.17 Sturtevant 1947, Dyen (private communication); 4.18
Hockett 1967, Thompson and Thompson 1969, Dyen (private communication);
4.19 Hockett 1967; 4.21 Hoenigswald 1964b, Hockett 1965, Bhat 1968,
Andersen 1972, T. Itkonen 1970; 4.22 Sapir 1921, S. Moore 1927, 1928,
Collinder 1937-1939, Pike 1947, Kiparsky 1965, Postal 1968; 4.23 Jakobson
1949, Janert 1961; 4.24 E. Itkonen 1966, Kettunen 1962; 4.25 Greenough and
Kittredge 1929; 4.27 Kiparsky 1965, 1968a, Newman 1968; 4.28 Kurylowicz
1968, Watkins 1970; 4.29 Anttila 1969a; 4.31 F6nagy 1956-1957, 1967; 4.32
Specht 1952, Senn 1953, Katicic 1970.
CHAPTER 5
analogies were used, giving what does, in fact, appear in grammatical construc-
tions. In the transformational approach, the intrasentential relations between
deep and surface structure provide the bases for analogies, with the extra
dimension of giving information on what might appear in grammatical construc-
tions. Readers who are not very well acquainted with formalized grammatical
description need only accept the assertion that all these different theoretical
frameworks use the same principle of analogy but on different terms and axes
(see§§ 5.21, 6.24). This is just a reminder that the basic structures of all formal
descriptions are, in fact, analogical. Thus it is no wonder that analogy operates
mainly in the structure of grammar.
Proportional analogy is, of course, diagrammatically iconic, an icon of relation
(§§ 1.13-1.15). Language has a general iconic tendency, whereby semantic
sameness is reflected also by formal sameness; this force underlies contamina-
tion. We often can predict the areas where analogy will enter, if it does enter,
by noting such things as formal imbalance in a semantically symmetric situation.
lost. Thus both dived and dove still exist, as well as an older brethren and a
newer brothers, with clear stylistic and social differentiation. On the other hand,
the original paradigm (sg.) book, (pl.) beech was given a new plural books; and
after a time the old one was lost. Because the word shifted into the majority
pattern, it is easy to give a proportion: pen :pens =book: X (see§ 5.19). Note
that the example is one of principle only, to avoid Middle English complexities
of spelling.
[5.3] Many of the iconic developments we saw in the first three chapters
show proportional analogy or at least can be described through it. The rebus
principle shows this in Sumerian orthography,
meaning 'arrow' 'life'
form ti ti
writing =-x
where the proportion exists between the last two rows and X was solved with a
spelling ~. This is a case of" spelling spellings " (§§ 2.6, 2.15). The 'past' tenses
of will and can were ME wolde and coude, in which the n had been lost already
in OE ciioe ( < *kunjJe; compare tooth < *tanjJ ; § 4.16). After the loss of I we
get (using modern forms)
pronunciation fwudf /kud/ /layt/ /di'layt/
spelling would= -x or light = X
This is very frequent in all languages. In Sicily medial ll had been replaced by
apical rJ4 (stella > stir/.r/.a 'star'). New immigrants into the area extended the rJr;l
also into initial position:
Dialect 1 stella luna
Dialect 2 stir/.r/.a = X
and we get Hyper-Sicilian #una 'moon', and so on. Such examples could be
multiplied by the hundred.
Hypercorrect forms show relations between regional and social variation,
but the same formal situation may obtain between variants in the same norm.
When British English lost the r in forms like better before pause or another
GRAMMAR CHANGE: ANALOGY 91
consonant, variation /bet~ "' VI resulted. This now serves as a model for
bet~r-
words with final a's:
Environment 1 beta ay'dia (before C)
Environment 2 betar = --y- (before V)
and phrases like the idea-r of it and Arnerica-r and England result. After Estonian
k had been lost medially at the beginning of closed syllables, as in kasket > kased
'birches', we get alternation, that is, sg. kask "' pl. kased. Words that originally
had a stem-finals look now the same in the plural, for example, kuused 'firs'.
Instead of the expected sg. kuus we have kuusk, arising from a proportion like
the following one:
Environment 1 kased kuused (nom. pl.)
Environment 2 kask = --x (nom. sg.)
In both English and Estonian, alternation has been extended into words
where it did not exist before. Such paradigmatic sets can even create new pho-
nemes. Russian nonstop consonants (continuants) were palatalized before front
vowels; when these vowels dropped, there was a split (e.g., v vs. v', r vs. r', and
so on), and both can alternate within paradigms. A stop like k was affricated
into c (ts) and later, in some new environments, into c (ts); this morpho-
phonemic alternation k "' c "' c remains (compare the Old French outcomes of
Latin k without paradigmatic alternation;§ 4.4). But paradigms in which v and
v' and so on alternate have called into being a new phoneme /k'/ for an expected
C:
,
1st sg. rv-u vr-u tk-u
2nd sg. rv'-6s vr'-6s tk'-6s
'tear' 'tell lies' 'weave'
Similarly, the instrumental of kto 'who' is k'ern, for an expected cern. The form
cern is found as the instrumental of cto 'what'; thus it appears that the k- of
the animate paradigm was restored (with automatic palatalization before e) as
an indirect marker of 'animate', while the original form was semantically
specialized as 'inanimate'. The analogical origin of k' in Russian is clearly
revealed by its restriction to position before a morpheme boundary (see§ 5.13),
at least in native vocabulary, although loans like k'in6 'cinema' have extended
its distribution into other positions. Here we have extension of an alternation,
and, at the same time, leveling of an alternation that would have been much
more pronounced if analogy had not occurred.
[5.7 Interplay Between Sound Change and Analogy] Typical for language
change is the constant tug of war between sound change and analogy. Sturtevant
phrased this as a paradox: sound change is regular and causes irregularity;
analogy is irregular and causes regularity. That is, the mainly regular sound
change can pull regular paradigms apart; analogy is generally irregular, in that
it does not occur in every case where it could, but when it does, the result is
greater regularity in morphology. In the case of morphophonemic conditioning
of sound change we have a case of analogy, which is sometimes even regular,
and, of course, sound change can be irregular. The paradox is not absolute, but
still accurate.
As a first example of how sound change destroys paradigmatic unity, let us
look at a Latin instance. For practical simplicity of handling examples, let us
confine ourselves to the nominative and genitive singular cases, because these
reveal the crux of the matter. A Pre-Latin paradigm (nom. sg.) *deiwos (gen. sg.)
*deiwi 'celestial' has a constant stem deiw-, and the case endings -os and -i, a
type that survived into Latin.
1. The diphthong changed into a long close vowel, *ei > *f, which had no
effect on the paradigm as such.
2. Now a *w before *o dropped, making the nominative *dfOS.
3. *DfOS is subject to another well-known Latin change: a long vowel is
shortened before another vowel; thus *deos.
4. o > u in final syllable.
5. *f > i, and the paradigm should end up as deusfdivi (in regularized Latin
orthography).
GRAMMAR CHANGE: ANALOGY 95
These five changes are regular sound changes in Latin, and they have produced
an irregular paradigm, where the stem now alternates between de- and div-.
This kind of unique alternation is a situation in which analogy might be expected
to restore balance (regularity), as it in fact did, because deus and divi do not
belong to the same paradigm in historical Latin. Analogy eliminated the
alternation by building complete paradigms to both alternants. The nominative
deus got a new genitive dei, and the genitive divi received a new nominative
divus. Now we have two regular paradigms, deusjdei 'god' and divus/divi 'god,
divine'. This is an eloquent example of Sturtevant's paradox. The situation is
parallel to the regeneration power of the planarian worm. When cut in half,
its front part grows a new rear end, and vice versa (see§ 22.1).
A paradigm need not split in two. Pre-Latin *ekwos 'horse' and *parwos
'little' should give *ekos and *paros (> *ecus, *parus) because of change 2
above, but the corresponding genitives *ekwi and *parwi (here again, of course,
representing the rest of the paradigm) prevailed and grew or maintained new
nominatives equus and parvus. The regular outcome is shown in the adverb
parum 'too little', which was no longer connected with the paradigm of parvus.
Such offshoots provide clear evidence for analogical interference. Another case
is *sekwondos > *sekondos > secundus 'second', developing regularly by the
sound laws after the word had been cut off from the paradigm of sequi 'to
follow ', which retained its [kw] in every position. English sword has also lost
its w in this position, and so should have swore, but it was restored/maintained
after the present swear. In Latin nouns the majority of the oblique stem generally
wins out, but in the third declension noun *wok-s (gen.) *wokw-is 'voice', the
alternation wok-jwokw- is eliminated in favor of the nominative wok-: vox [ks]/
vocis. This is the irregularity of analogy (one cannot predict the direction),
which may be quite regular, since Latin, after all, does not allow for an inter-
consonantal w, *wokws. On the other hand (nom.) *yekor (gen.) *yekwinis 'liver'
has also adopted the nominatival k: iecurjiecinoris, as well as the -or- from the
nominative. This is a clear case where proportional analogy is impossible but
where we have a complex contamination of the two stems.
English shows clearly the irregularity of the direction of analogical leveling
in the strong verb, where Old English had different vowels in the preterite
singular and plural:
The corresponding Modern English paradigms, like those of the weak verbs,
have just one form for the preterite. Alternation has been eliminated both ways:
in bite "' bit, the plural vocalism prevails, in ride "' rode, the singular, although
there is also an archaic rid (see§ 10.7).
96 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
[5.8] The following Old English paradigms (two representative forms have
been chosen-the minimum number, of course) gave Middle English:
OE ME
nom. sg.
(pl.)
stcef
stavas
sceadu mted
(obi.) sceadwe mtedwe
I staf schade
staves schadwe
mede
medwe
In the OE paradigm of stcef the nominative singular has a closed syllable (i.e.,
it ends in a consonant), but the first syllable is open in the plural (sta.vas ). The
syllable structure is reversed in scea.du (open)jscead.we (closed), and in mted
the root syllable is the same throughout the paradigm, a closed syllable but with
a long vowel. In the last case we have the same vocalic developments as in
htelan and hteljJ, that is, shortening before two consonants (§ 4.8)-mead exactly
like heal jiyj and meadow like health jej. This, of course, is the Modern English
result, but the short /e/ in meadow still shows the fact that the w was contiguous
to din Middle English. In ME staf and schade we have a reverse development,
equally regular: the lengthening of short vowels in open syllables. This effects
the plural of staf and the nominative of schade, giving us stiiL·es and schiide.
With the great vowel shift we get Modern English shapes staff/staves jstref "'
steyvz/ and shade/shadow jseyd "' sredow/. Now regular English sound changes
have produced the above forms as well as mead/meadow. All started from uniform
Old English vocalism and ended up as regular alternations, because such vowel
alternations occur in hundreds of English vocabulary items. But exactly as in
the Latin case of deusfdiws the paired English forms do not belong together
any more in Modern English, except perhaps for staff/staves (to a degree). The
resulting vowel alternations occur in different word classes, for example, adjec-
tive-noun sane/sanity, adjective-verb cleanfcleanse, and noun-verb grass/graze,
glass/glaze, and breath/breathe, but not within the same word. As in Latin the
variants have split into two words, and the missing parts have been supplied
analogically, that is, diagrammatically according to the regular patterns (rules)
of the language: staff/staffs (new), stave (new)fstaues (compare clothfclothes),
mead/meads (new), meadow (new)jmeadows, shade/shades (new), shadow (new)/
shadows (see § 7.9). As in Latin, semantic differentiation accompanies the
formal split; it is, in fact, a prerequisite of the survival of both forms (compare
Indian/Injun, § 2.14). Normally, only the oblique stem survives, for example,
in those words that had the w in Old English: yellow (geolu), fallow (fealu),
callow (calu), and arrow (earh). The oblique stem survived also in thimble
(§§ 4. I I, 4.12); today, when hardly any inflection is left, the nominative singular
has a strong position (e.g., fowozj being replaced by jow6s/ oaths after the
singular oath /6/; see §§ 10. I 6, I I .6). Formal vowel alternation survives in some
nouns only if the short-vowel variant occurs in fossilized derivatives (seam/
seamstress, goose/gosling) or compounds (crane/cranberry, vine/vineyard, house/
husband) which are independent words (not productive outputs of the "normal"
rules of the language). Actually seamstress is now generally jsiymstr<Js/, an
obvious analogical, partially productive form in relation to sempstress. The
GRAMMAR CHANGE: ANALOGY 97
original root vocalism is often better preserved in family names as in Webster/
weave and Baxter/bake.
[5.9] We saw above how Estonian k alternates with nothing (at the
beginning of a closed syllable; § 5.3). In intervocalic position this stop is written
withg, and the alternation is exemplified by the inf. piiga-ma next to the 1st pers.
sg. poa-n 'shear, cut (hair)'. Similarly, d alternates with nothing (among other
things), as in laadi-maflae-n 'load (gun)' and haudu-majhau-n 'brood, hatch'.
The alternation here is just one small aspect of the consonant gradation, which
was originally determined by the phonetic shape of the word (closed and open
syllables). This state of affairs is well preserved in Finnish (§§ 10.12, 10.13),
but Estonian has eliminated alternation on a large scale. In some cases the g
(and so on) has been generalized through an entire paradigm or through part
of it (e.g., the present); in others, the lack of the stop (nothing) has been general-
ized. And in part of the vocabulary, alternation remains. This lack of exact goals
is typical of the irregularity of analogical change, and we saw in the kuusk case
that alternation can be extended even to items that did not have it (§ 5.3). Thus
analogy levels out alternations by two means at the same time, either by general-
izing one of the variants or by creating new cases of an existing alternation.
The situation is very similar to the tug of war between the various classes of
English strong verbs and the weak verbs (e.g., doPe/dive). But the old and new
forms can both ultimately survive, if semantic difference is attached to them.
All three Estonian verbs mentioned developed analogical presents without
alternation, the leveling being in favor of the stop alternant. The new analogical
formspiiga-n 'cheat, swindle', laadi-n 'load (freight)', and haudu-n 'be hatched,
stew' coexist with the old ones because of the semantic differentiation, even
though the infinitives remain the same. (Actually the semantic differences are
not that clear for all speakers. There is a strong tendency for the new forms to
be generalized in both meanings.) Compare the English verb hang, which has
tolerated both a strong (hung) and a weak (hanged) inflection because of a
similar semantic difference, as well as the English examples above (i.e., sunk/
sunken, burnt/burned; shade/shadow, and so on).
kind of analogy that is regular. The regularity of change is the ultimate result.
While in progress, a change is not notably regular, because it spreads at different
times in different environments and speakers. When analogy levels out all
exceptions to a particular alternation, the result is perfect regularity, and it is
difficult to know whether we are dealing with sound change or analogy. In this
sense morphophonemic conditioning of sound change is both sound change and
analogy. English bite/bit and ridefrode exemplify two-way tendencies within a
category. This is also the case of Estonian consonant alternations, which are
eliminated here, extended there. In Lapp, however, the alternations have been
extended to every word(§§ 10.14, 13.3), and the result is perfect regularity.
[5.12] Greek has a general sound law whereby intervocalic s drops out.
In most dialects s is the sign for future, thus (with verbs in the 1st pers. sg.),
as is shown on the top of the next page.
The futures in group A are as expected, ass is not intervocalic here. Group B,
however, violates the law VsV > VV; but linguists have assumed that, in fact,
the s was lost in these futures also, giving *luo and *poMo. If these forms had
remained, they would have undergone a change whereby vowels are shortened
before other vowels, and would have ended up homophonous with the presents.
GRAMMAR CHANGE: ANALOGY 99
PRESENT FUTURE
trep-6 'turn' trep-s-6 A. root ends in consonant
deik-nu-mi 'point' deik-s-6
lu-6 'loosen' lii-s-6 B. root ends in vowel
poie-6 'do' poM-s-6
men-6 'remain' men-e-6 C. root ends in nasal or liquid
ste/-/6 'send' ste/-e-6
This was the destructive force of regular sound change, and analogy from the
consonant stems had to be invoked to reintroduce the characteristic s of the
future, that is, trep6:treps6 = /uo:X, where X gives liiso (a vowel before this s
is automatically lengthened). But we have no direct evidence of an s-less stage
in group B, and it has been suggested that the facts can equally well be covered
by grammatical conditioning of sound change, that is, "intervocalic s drops,
unless it means 'future"' (actually, some other grammatical markers are also
included: the aorist, the dative plural). This takes care of group B, but group C
shows that everything has not yet been considered. Here, after liquids and
nasals, the future morpheme was not s alone but es, and in this form the s
was, in fact, lost according to the sound law. The situation is the same as in
some of the Baltic Finnic cases (§ 4.24): if a morpheme could afford to lose
part of itself, it did, provided that something remained to mark the function.
In Greek the surviving e distinguishes the future from the p\esent, exactly like
-h < -hen in the Karelian illative. Thus we see that grammatical conditioning
of sound change and analogy can be explanations of one and the same thing;
this was true of morphophonemic conditioning of sound change as well (§ 4.21 f.).
What this teaches us is that analogy need not merely scavenge the debris of
sound change; it can prevent sound change from happening in tight-knit
morphological systems. That is, sometimes morphological iconicity is so strong
that sound change does not enter at all, although it may be quite general in
those areas where morphology is not directly involved.
[5.13] The Greek situation was presented first for historical reasons. It
is interesting to see how scholars have interpreted it and to note that there is a
wide margin for interpretation in historical situations not directly attested. But
similar cases can also be observed while they are happening. In Russian the
change of unstressed ii [;:,] > i after palatal (soft) consonants, for example,
p6jiis > p6jis 'belt', has been a living process for scores of years, although the
change has not yet ousted the earlier pronunciation, and both pronunciations
still occur. In the 1940s the change ii > i did not enter inflectional suffixes at all,
because in these the vowel in question sometimes occurs under stress. Thus
we have, for example,
gen. p61'-ii 'field' vs. zil'j-a 'dwelling'
dat. ust6j-iim 'foundations' kriij-am 'land'
(compare Greek ltiso Greek trepso)
IOO HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
We have a preventive analogy for the sound change ii > i based on the environ-
ment (stress) of the inflectional endings which are not subject to the change.
The net result of this analogy is that the conditions of the change "palatal
consonant plus unstressed ii" do not extend over a morpheme boundary in
front of inflectional suffixes (see § 5.3). This is how a grammatical limitation of
sound change is often analogical in origin, that is, alternation is actually pre-
vented from occurring and not merely leveled out by analogy. Greek s showed
the same situation : it was not dropped in certain grammatical morphemes,
because it was retained in some phonetic environments in any case. The Russian
situation has a further history; now the change/process ii > i has been extended
also to inflectional suffixes.
[5.14 Analogy and the Relation Between Meaning and Form] In the case
of Estonian -n 'I', we apparently have a situation where sound change proceeded
to completion before analogy became operative (§ 4.24). Final -n was lost in
preconsonantal position and preserved before a following vowel. At this stage
the change was a purely phonetic one, and it was only then that analogy entered.
It reestablished the-n in every environment in those dialects where its loss would
result in the same shape as the imperative. In the Southern dialects, where no
homonymy threatened, the sound change just continued, with the -n dropping
everywhere. In the Russian and Greek cases(§§ 5.12, 5.13), the driving force was
the prevention of variation (difference) within one morpheme, and in Estonian,
prevention of the same form from having two different meanings. But this is
actually the same force, prevention/elimination of one-to-many relations between
form and meaning:
meaning 1 meaning 2
~~ etc.
form 1 form 2
That is, both the 1\ (Russian and Greek, etc.) and V (Estonian) configurations
tend to be avoided by the iconic principle whose ideal is 'one meaning, one
form'. Of course, all languages do have such configurations, because semology
is, after all, independent of morphology, but such disparity is the characteristic
breeding ground of analogy. And if analogy comes into operation, it either
eliminates the alternation (i.e., establishes I -relations = one to one) or carries
the alternation into other parts of the vocabulary or morphology. The important
word is if, for it must be emphasized that nothing need happen. For example,
in English the morpheme -s fs "' z "' izj with variation represents the meanings
'3rd pers. sg.', 'possessive', and 'plural '. It can further be a variant of the
morphemes is and has, thereby representing at least two more meanings.
Again, we see how grammatical conditioning of sound change is structurally
parallel to analogy or the iconic tendency in that it also breaks up or forestalls
these one-to-many relations betwe~n form and meaning. We have a V-relation
in those instances where a case form represents also some adverbial element,
and we have at least two meanings for one form. When change does not touch
GRAMMAR CHANGE: ANALOGY 101
the adverbs, the meanings get forms of their own (e.g., once vs. one's, § 4.25).
And as for straightening out the 1\-relation, we have seen that morphophonemic
conditioning of sound change is this kind of analogy. If the /\-relation is based
on suppletion, we have simple analogy (e.g., gofwent ~ gofgoed). Sound change
can produce suppletion, for example, Latin oculus/oculi 'eye/eyes' gives French
f£iljyeux [reyjy0]. When morphophonemic rules get restricted (out of produc-
tivity) original alternation can change into a kind of suppletion: sit/seat, heat/
hot, cook/kitchen, tenf-teen, or for some speakers, even cases like opaque/opacity
(§§ 5.8, 6.21, 6.24 7.13, 10.7-10.9, 17.5, 18.17). The stronger the suppletive
element is, the more probable is the occurrence of analogy.
Throughout this chapter we have seen this tendency of 'one meaning, one
form' at work. Thus, in Yiddish (§ 4.27),
rv
'way' 'away' 'way' 'away'
veg-V vek
gave
I
veg
I
vek,
·possVerb'
and, in English(§ 4.25), we had (in certain cases)
yielding
'poss.'
I
'adv.'
s z s(t).
In both cases the end result was two linguistic signs with one-to-one correspond-
ence (!-relation) between form and meaning. Meaning is decisive here; two
meanings develop two linguistic signs. This is the regularity principle of analogy,
which restores what sound change and syntactic combinations had diversified.
Similarly, the irregular alternations gofwent and bad/worse are often straightened
out (by children) as
'go' 'past' 'bad' 'comparative' 'go' 'past' 'bad' 'comp.'
~~I
gowent -d bad worse -er go
I
-d bad
I
-er
(Again, this notation shows the simplification visually.) In these particular
cases the results (goed, badder) have not been generally accepted, because the
frequency of occurrence upholds the tradition, but in countless cases it has, for
example, book/beech~ book/books(§ 5.7). It was recognized early that there is
a strong correlation between analogy and frequency. A typical phrasing of this
principle would be that irregular (strong) forms stand outside the general rules
and have to be specially learned, thus burdening the memory; analogy is,
therefore, successful where memory fails; that is, infrequent forms are prone to
be changed first. This principle is generally valid, however it may be worded.
We have seen that the conflicts between sound change and grammatical analogy
102 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
often result in sound changes that are grammatically limited, or sound (::hanges
affecting only certain grammatical categories and not the general sound pattern
of the language. Or, in other words, a sound of certain grammatical/morpho-
logical value may resist sound laws. Grammatical conditioning of sound change
and analogy are very much two sides of the same coin.
[5.15 The Status of Old and Innovating Forms] When changes leave
behind old forms without ousting them completely, there is a universal tendency
for the innovating form to carry the primary semantic functioning of the old
linguistic sign. The old form is pushed aside for some peripheral or secondary
meaning. Most of the cases we have seen are clearly of this type, and it does
not matter whether the driving force is sound change or analogy. Thus:
II
OLD FORM: NEW FORM:
SECONDARY FUNCTION PRIMARY FUNCTION
adverb riiiijgin 'asunder' gen. riiiijgi 'hole' Lapp (§ 4.24)
once gen. one's English (§ 4.25)
vek 'away' nom. veg, vek 'way' German (§ 4.27)
parum 'too little' ace. parvum 'small' Latin (§ 5. 7)
compound cran(berry) crane English (§ 5.8)
hus(band), hus(sy) house English (§ 5.8)
shep(herd) sheep English (see § 4.8)
maan(tee) 'highway' gen. maa 'earth' Estonian (§ 4.24)
plural brethren brothers English (§ 5.2)
kine cows English (§ 5.4)
In every case the second column shows the regular, productive, stylistically
or syntactically unrestricted (unmarked) form. The situation is different when a
paradigm splits in two, because then there is a possibility that functions which
earlier shared a form can become independent signs (e.g., deusfdivus, shade/
shadow, and so on), but even here one offshoot may become stylistically re-
stricted, for example, mead/meadow, where the innovating oblique-stem form
meadow carries the "normal" functions of the word.
[5.16 Analogy and Syntax] We have seen how analogy works both in
phonology and morphology under semantic constraints. But syntax also has
been clearly involved both in sound change and analogy, for example, in the
form of adverbs and predicatives, and both mechanisms also change syntax.
Often they do this together. Greek had, for instance, the following forms in its
verbal paradigms:
The endings have been separated from the root by the hyphen. The infinitive
occurred in phrases like theli5 graphein 'I want to write' and thelei graphein 'he
wants to write'. Then the final -n of the infinitive dropped and its outer shape
became identical with the third singular: thelo *graphei, thelei graphei. The
former expression is "formally poor" for the meaning 'I want to write', because
it can also be interpreted 'I want, he writes' (V-relation). And the same applies
to all the other persons as well, except for the third singular thelei graphei. At
some point this sequence was reinterpreted as the 3rd sg. twice 'he wants, he
writes' with the same 'he', that is, 'he wants to write' in a new form. As the
reinterpretation of formation it would not show overtly here; this was an induc-
tive change, which did not alter the outer shape produced by the sound change.
The reinterpretation surfaced in the other persons; for example, thelei graphei =
the/a X, where the end result is thelo graphO 'I want to write' (formally also the
1st sg. twice). This deductive analogy restores the diagrammatic relation between
person and the corresponding form. Ultimately, the infinitive in Greek was lost
altogether. (The change shows also that infinitives are indeed underlying
sentences, or finite verbs; when sound change interfered with them they easily
reverted back to their basic form. We ignore here the subsequent modification
whereby the particle 'that' became obligatory, thus in Modern Greek: the/o na
grapho [literally] 'I want that I write'.)
Also the 1st sg. ending was -m. A sentence like 'I see the boy go' went niie-m
poja-m menevii-m (written here in a hybrid orthography where only the endings
reflect the earlier sounds). The last word menevii(m) is a participle of the verb
'to go', and because it is an attribute to pojam, it agrees in case and number
with it; that is, 'I see the boy, the going one' = 'I see the boy going'. The
corresponding plural object can be formed with cases given: niie-m poja-t
menevii-t 'I see the boys go'. A sound change -m > -n produced new endings: niien
pojan meneviin. The ace. sg. became homophonous with the gen. sg. (there was
no such merger in the verbal 1st sg. ending). As in the Greek example, sound
change made two forms identical, here pojan and meneviin (both ace. and gen.).
Note that, to start with, pojan is the head and meneviin an adjective attribute to
it. At some point the form pojan was reinterpreted as a genitive, and conse-
quently as an attribute to the following meneviin, which therewith became the
head to the genitive attribute. Again, such reinterpretation is not reflected in the
forms themselves; they remainpojan meneviin (compare theleigrdphei), although
104 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
the literal analysis is now '(the) going of the boy'. The new analysis is proved
by the plural, because there the accusative and the genitive are different, and the
original, unambiguous phrase has been replaced by the equally unambiguous
naen poikien meneviin 'I see the boys go'. Meneviin is now, unmistakably, an
uninflected head with the attribute poikien in the genitive plural (see § 9.16f.).
[5.18 Analogy and Speech Production] In the survey of the various types
of analogical changes, two ways of classifying them were occasionally referred
to: leveling and extension. When differences between two (related) forms are
reduced or eliminated, we have leveling. When a form or an alternation is
carried into a new environment, we have extension. All the examples we have
seen represent one of the cases or both. For example, the differences between
the originally unrelated linguistic signs ear and irrigate were partially leveled
by a new semantic identification and recutting irr-igate. The part -igate was then
subsequently extended to nosigate. The morphemes -ism and -able were borrowed
into English as parts of hundreds of loanwords (e.g., humanism and usable).
These endings have been extended to native stems or roots (e.g., token-ism and
think-able). Extension is similar to borrowing in that a form is lifted from one
environment into another, though, in borrowing, the source environment is
in a different language, dialect, or even idiolect, whereas, in extension, it is
within the same grammar in another grammatical environment or in another
part of the vocabulary (lexicon). The parallelism with borrowing has even led
to calling extension borrowing from within (the same grammar).
New analogical (deductive) forms are, by necessity, tied to speech production;
that is, a speaker must utter them according to his grammatical machinery. The
creation of such forms is independent of their subsequent fate, because they may
or may not become the new norms. One of the most mystifying characteristics
of human language is its productivity (§ 1.28). This is connected with man's
innate ability to learn a language. Such a capacity manifests itself very early
in the child's apprenticeship in speaking, as he can and does easily go beyond
the sentences he has heard. Each utterance is either a parroting or a new creation.
From the data he has been exposed to, the child is able to abstract regular
patterns or rules; he then extends his use of these into areas that are novel to
him, and maybe even to other speakers. Thus one aspect of extension of forms
or patterns is clearly a function of the use of the grammar, that is, speech
production.
Grammar is somehow internalized in the brain and is not directly observable
except for its product, the actual utterances. Of course it is a two-way affair, as
the regular patterns have to be abstracted from the utterances. But once they
have been established, they need not be reinforced by concrete instances. If
we heard a new English adjective glump, we would be automatically able to
form the comparative and superlative glumper, glumpest without referring to
another concrete instance like damper, dampest. If it were a noun, its plural
would be glumps, if a personal name, a genitive Glump's would follow. And a
GRAMMAR CHANGE: ANALOGY 105
verb would go he glumps, he glumped, and so on. These forms have now been
created by frequent productive patterns. Such patterns tend to prevail over
unproductive types. Instead of the unique good/better, the speaker may lapse
into a comparative gooder, or instead of an irregular weak brought, he may come
out with bringed. Adults usually quickly correct themselves, whereas children
tend to make an effort to stay with these. Only such irregularities as good/better
and badfworse have to be learned form by form, otherwise the patterns are
enough. Thus in highly inflected languages, speakers do not in every case store
hundreds of different forms for each word but create any form they need
according to the patterns at their disposal (see§ 18.17). Many forms are created
afresh for each occurrence rather than repeated from memory. This is even more
true in syntax than in morphology, because we speak and hear more different
sentences than different words. Language is one manifestation of the innate
faculty of analogizing, shown clearly by children even before they have acquired
language.
English, however, the umlaut plurals have become a tiny minority in relation
to the s-plurals. They are purposeless in this sense, and a natural target for
analogical realignment to the s-class (§ 5.14).
REFERENCES
RULE CHANGE
of certain short vowels(§ 5.8). These two changes are independent of each other,
because they apply to mutually exclusive environments, that is, long and short
vowels, respectively; but they interfere with the vowel shift, because the one
takes away long vowels that would have undergone the shift if they had remained
long, and the other provides new long vowel~ that subsequently do shift. The
relative chronology here is only partial, in that both the shortening and the
lengthening have to occur before the shift, but whether they themselves occurred
simultaneously or sequentially is unknown. (Direct historical attestation often
mixes with relative inferences; for example, we know that umlaut occurred
before shortening and lengthening.) This is summed up in Figure 6-1 :A. In
other words, both the shortening and lengthening occupy the same slot in the
relative chronology, although, in absolute historical chronology, they could have
been centuries apart. But that does not matter; only the interference with the
vowel shift is of interest. In the case of the shortening, this interference is called
subtractive, because it takes material away from the domain of the vowel shift;
in the second case, it is additive (Chafe), ~ecause the change adds new material
to the vowel shift (the terms 'bleeding' and 'feeding' are also used [Kiparsky]).
In Rotuman, we saw the same relationship between the raising umlaut and the
fronting umlaut. The former had to occur first to provide, for example, a closed
o, which then gave othrough the latter. Fronting umlaut itself is complementary
to the other two changes, metathesis and shortening, and all three thus occupy
the same position in the relative chronology, as shown in Figure 6-1 :B. Raising
umlaut is seen to be additive to the three other changes. In the passage from
Latin to French, we must first have the change k > ts (kantiire > tsanter),
and only then w > ftJ (kwando > kand), whereas the change ts > s is free in
regard to w > ftJ (but, of course, it must occur after the first change that feeds
into it, § 4.4). If w had been lost first, we would have had the sequence kwando >
kand > *tsand > */sa/, which is obviously contradicted by the attested form
/ka/ quand 'when'.
A. English vowel changes
~I
I. umlaut
2. shortening 3. lengthening
(-) (+)
4. vowel shift/diphthongization
A B c D E
1. dh > 0 t > p t > p, 0 t > 0
2. t>p d>t t > p
3. p>O p>O
4. d>t dh > 0 p>O
FIGURE 6-2. Possible relative chronologies of Grimm's and Verner's laws.
[6.3] When describing shifts one usually takes one change at a time.
Grimm's and Verner's laws(§§ 4.9, 4.10) thus require one of the relative chronolo-
gies in Figure 6-2. Possibility C lists p > oas the latest change, and what precedes
can follow either the order of A or B. Thus altogether four alternatives have been
tabulated in A-C, and D and E add further possibilities (these are not exhaustive).
This mirrors the indeterminacy or randomness which is characteristic of our
retrieval of history, but note that there is still a fair amount of relative chronology
in the arrangements. Change d > t must occur after t > p, because otherwise
it would add to the latter, which in its turn does, in fact, add to Verner's law
p > oand can occur before it, or simultaneously with it (D), and, indeed, even
E is a possibility. Change dh > o is free in respect to others, as long as we
interpret o as phonetically different from the starting point d. If, however, we
operated with dh > d, it would have to occur after d > t, because otherwise it
would add to the latter; that is, we would have basically (1) t > p, (2) d > t,
(3) dh > d, and (4) jJ > d, where the last two changes could be also reversed.
In this case, the relative chronology would be rather definite or strict.
Another way of presenting consonant shifts would be simultaneous chronology,
which gives also the desired result. The problem is, however, that such changes
have not been unambiguously attested in observable cases, especially when more
than two terms are involved in the shift. Thus it is very unlikely that in historical
fact the Indo-European consonantism was transformed into the Germanic at
one stroke:
Proto-Indo-Europeanl:
Jdh
~~ 0)
Germanic
(Of course, it is always possible to describe shifts at one stroke on paper.) Latin
medial stops on the whole develop as follows into Spanish: (1) d > @ (cadere >
caer 'to fall'), (2) t > d (tatum > todo 'all'), and (3) tt > t (gutta > gota
'drop'); the reverse order would give@ for everything. Shifts of this kind are
common in the languages of the world. Here one could devise, for example, a
feature 'closure' plus various degrees of it; writing one degree of 'closure'
with each dash, we have the following in Latin: d = [- ], t = [ = ], and tt =
[= ]. The Spanish shift can now be described in one step-loss of one degree
of closure-and the outcome is right.
II2 HISTORICAL LINGUISTICS: HOW DOES LANGUAGE CHANGE?
[6.4] Even though it is quite possible to use such tricks for presenting
shifts as simultaneous jumps, their value as historical indices is highly question-
able. We have presented the shifts above as chain reactions, that is, one sound
moves away and another takes its place. The question has also been raised
whether the shift can be caused as well by one sound invading the allophonic
range of another, thus pushing it out of its earlier place. These two ways of
looking at the shifts are known as pull chain and push chain changes, and in
general it seems that the former has more support in linguistic literature, although
the latter cannot be ruled out(§§ 9.3, 9.6, 9.7). It is interesting to note that push
chains would give a relative chronology with considerable overlapping between
the steps (see also§ 9.16).
A case of shift where two steps must be simultaneous is a switch between two
sounds, a change known also as a flip:fiop . Until quite recently, historical lin-
guists considered this kind of change impossible, and it is still controversial;
but evidence for it is believed to be accumulating. Whether this evidence repre-
sents the facts correctly cannot always be determined, which is a general defi-
ciency in historical explanation (compare the court procedure in which the
evidence may be quite clear, but the facts are not). However, a switch seems to
have taken place in the southwestern United States (e.g., Utah), where the vowels
in words like card and cord have switched places (also far/for, ardor/order).
Here the change is so recent that it seems to be true without intervening stages.
Often, of course, we simply do not have the total historical knowledge of a
shift, and even though the end points seem to point to a switch, it need not be
historically true. More than likely it never (or very seldom) is.
[6.6) It was already noted how change is dependent on the particular form
of grammar the linguist adopts (§ 1.9), and we saw an example of this in the
Spanish consonant shift(§ 6.3). We had either a sequence of steps, or with the
feature closure (in various degrees) we could describe the whole shift in one
stroke. In Chapter 4 we saw that most historical changes leave behind alterna-
tions that stay in the grammar indefinitely. But for this reason and because the
brain seems to be capable of hierarchical ordering (language acquisition/aphasia),
the practice has become prevalent of presenting phonological derivations from
the underlying invariant morpho phonemes with similar ordering. In other words,
the historical sequence of changes is supposed to be reflected, to a degree, in the
synchronic order of application of phonological rules. This synchronic order
("brain order") used to be called descriptive order in contrast to historical
order or relative chronology. Historical order thus tells us about the relative
history of how the language came to be as it is at the time of attestation (or
now), and descriptive order reflects what-perhaps-goes on in the brain of the
speaker every time he utters something. The former is a fact, unknown to most
speakers, of course, whereas the latter is a hypothesis; it is easily forgotten,
however, that it is a hypothesis and is often taken as a fact. The parallelism
between the history of a grammar and the present functioning of grammar is
attractive and reminds us of the similarities between phylogeny and ontogeny,
but here also the similarity is not absolute. As in culture, in general, all of
114 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
history cannot function forever, although a substantial part can, in some form
or other.
Let us return to the partial paradigm of Finnish vesi 'water', vede-n 'of
water', and vete-nii 'as water' (§ 4.29). The underlying invariant shape is vete,
to which the speakers apply the following rules to get the actually occurring
surface forms: (1) e--+ i in word-final position, (2) t--+ s before i, and (3) t--+ d
in closed syllable. Again, the arrow indicates a synchronic process. This rule
order could be the same as the historical order of the corresponding sound
changes. The same is true of the crucial ordering of raising umlaut first before
the other changes in Rotuman (§ 6.2). But we shall see how the synchronic
reflex of a change need not repeat history this well. This kind of a description
of phonology has led to new terminology, so that sound changes, too, are called
'rules', and relative chronology becomes' ordering of rules'. The term' rule' itself,
however, is rather old, although it was used only sporadically some hundred
years ago.
[6.7 Form of Rules] To be able to observe and describe rule change, one
must give the rules a definite form; and as has been stressed, this form has an
effect on the structure of the change itself (see§§ 1.1, 1.9). The general form of
a synchronic rule is sound a becomes sound b in environment c, that is a --+ bf-c.
The environment can be anything and need not be a segment following a, as
this general structure would seem to indicate. Such a rule can be read ac--+ be;
or if the environment flanks a, for example, ... .fd-c, we get dac--+ dbc. Sound
a is the input to the rule that operates in environment c, and the output is b.
Sections a and c belong together by virtue of the fact that they are the targets
that the rule seeks out to operate on, and b is the output. Thus a + c represents
a hierarchical upper level with respect to b, and it has been useful to have
separate names for this dichotomy, structural description (a + c) and structural
change (b) (Kiparsky). How a and care, in fact, related will be further clarified
by an example.
An English speaker must have a systematic way of distinguishing items like
chief/chiefs, and faith/faiths from pairs like knife/knives, and sheath/sheaths.
That is, in some words, spirants get voiced in the plural and, in others, they do
not. Thus the voicing rule must be able to discriminate between such items.
This must be done somewhere in the structural description. Traditionally, one
established different kinds of spirants, those that underwent voicing and those
that did not. For example, lf1l and l61l undergo voicing and lf2l and l62l do
not; in other words, the subscript 2 subtracts the spirants from the voicing
rule. Or we can put the conditioning into the environment by labeling those
words that do not voice by a feature like [-native]. Now voiceless spirant (a)--+
voiced spirant (b)/-[ +native] (c). By manipulating either a (units) or c (environ-
ment, with classificatory features) we get the same desired result. This shows
clearly the interdependency of units and features, and especially units and rules.
Both have to be adjusted to each other, and they are complementary to each
other (§ 1.9). Another way of blocking the voicing in words like chief and faith
RULE CHANGE 115
is to label them directly, with [-voicing rule], whereby the rule jumps over such
items. What was paraphrased here is, of course, morphophonemic analysis
(see§ 10.16), and we can leave it at this point. It showed, however, that different
paths can lead to the same results. After all, we do not really know what goes
on in the speaker's head, but we do know that he is able to handle words like
chiefvs. sheath correctly (i.e., differently).
[6.8] Because the rules must be able to describe social variation (Chapter
3), they can contain considerable depth, for example,
[6.9] It must be noted further that the parts a, b, and c of a rule must often
be decomposed into the relevant distinctive features, to get the maximal lin-
guistic generalizations (§ 4.20). Here again the exact shape of the rules depends
on the features used. Thus, using normal articulatory features, part of Grimm's
law would look like
l)
[ voiceless ~ voiceless [accent]- (after accent)
([
spirant
dental [ VOICe
j
spi.rant] [no. accent]-[voice]
VOICe
voiceless
Grimm's law is, of course, a historical change; but if the voiceless stops shifted
first, this would have been a synchronic rule as well, because such stops would
II6 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE'!
remain unshifted when second in a voiceless cluster. This particular formulation
takes Verner's law simultaneously with Grimm's law. Since the Indo-European
s also takes part in the former, it has been written in the rule separately. If we
ignore for the moment the so-called laryngeals (§§ 12.3-12.5), sand the voiceless
stops could be combined as [obstruent, voiceless]. Of course, the output of
the first subrule could be a later input for a voicing rule (§§ 6.3, 11.20).
By replacing > with -+ we get the corresponding synchronic rule. (It must be
again emphasized that this is characteristic of what linguists try to do with
their notation. The actual history shows quite different things. We have been
looking only at stem-final syllables in inflection. Elsewhere the change *ti > si
has indeed happened: *tina> sinii 'thou' [compare te 'ye'], silta 'bridge'
[borrowed from Baltic, compare Lithuanian ti!tas], morsian 'bride' [compare
Lithuanian marti; § 8.2], even in derivation: pit-kii 'long'-pite-mpi 'longer'-
pis-in 'longest' [compare§§ 1.14, 19.4 B]. All this shows that rule manipulation
can obscure the real history. Words that do not assibilate in a stem-final syllable
are somehow late [neiti § 11.18], for example, analogical offshoots like koti/
kodin/kotina 'home' [from kota 'hut'; compare§ 13.3], or loans [with original
spirants]: iiiti 'mother' [§ 8.2, compare § 4.29]. As in the case of Lachmann's
law [§ 4.28], the total evidence shows that such rules are largely achronological
restatements [of analogy and borrowing; compare§ 10.17].)
We have seen that information about the environment can be written in either
of the parts of structural description (a or c), but actually some of it can as well
be included next to the output (b) to specify, for example, social variation. This
would leave part c free for phonetic and grammatical environments, for example,
a-+ b( = f [style, age, etc.])/-c, that is, a becomes b, which is a function of
style (b'), age (b"), and so on, in the (grammatical) environment c. No matter
RULE CHANGE 117
how much we have to decompose any of the parts of the rule for our purposes,
its basic structure is always a-+ bj-c, a structure that is already familiar from
sound change (Chapter 4).
Having thus delineated the form of synchronic phonological rules, we can
go on to observe what happens to them through historical change. It is useful
to keep in mind that we took a similar position with regard to phonemes
(§ 4.12). Phonemic changes were not agents but results of sound change, given a
certain theoretical position on sound units (i.e., the phonemic principle,§§ 10.1-
10.5). Given a similar position on the structure of rules and their application,
we get a new vantage point for observing change.
[6.11 Change of Rules] In Chapter 4 we saw that most of the time sound
change did not disrupt the underlying invariant morphophonemes, although the
actual phonetics could change drastically. A perfect example is the Rotuman
case, which retained the five vowel morphophonemes all through the various
reshuffiings (Figure 4-2). The underlying structure of morphemes tends to
remain the same, although new paint jobs here and there modify the appearance.
Sound changes just add new layers of paint on top of the earlier ones. Rules
are accordingly modified only in that the final outputs (b) are different, or that
their number increases. When the earlier English realization rule of the dentals
(e.g., d-+ [dental ~W ... [everywhere]) was affected by the sound change t} >
alveolar d, the new rule came to be: d-+ [alveolar d]/ ... (§ 4.1). Such a rule
change is rather trivial, of course, but a sound change did, in fact, modify a
rule; we shall see more drastic changes below. As for additions of rules, all
those Rotuman changes added quite a few, the exact number depending on how
we write them (§ 4.6). Earliest German had a rule whereby voiced stops were
brought out as voiced stops, that is, a rule of the above English type: d-+ [d ]/ ....
Then a sound change devoiced syllable-final voiced stops, for example, [d.] >
[t.]. This meant an addition of a rule like d-+ [t]/-[.] to the phonology, and
the result of both the sound change and the interdependent rule change is, for
example, buntfbunde 'league'. Here dentals are just standing in for the real
change [voice]-+ [voiceless]/-[.] (or [voice]-+ 0, if loss of voicing would
automatically mean its opposite, voicelessness). These two simple cases show
that when a sound changes in all its environments the corresponding rule is
also modified; and when a sound splits, corresponding rules are created. Thus
the number of rules increases (see § 4.12). Rules increase to handle the more
complex relations between the surface sound units and their greater number.
The two cases show also that the historical change had a direct reflex on the rule;
in short, that history and synchrony were practically identical, whatever
theoretical philosophical considerations there might be.
[6.12) The Pre-English umlaut that fronted back vowels created alternation
and added its reflex as a synchronic rule [back]--+ [front]/-[high, front], for
example, *mus-i--+ *mys-i (§ 4.5). Again synchrony repeats history directly.
When the environment was lost, that is, [high, front] > 0, the change was to
118 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
have a repercussion in the rule, because the information on where the rule
must operate was wiped out. But since this was a mere phonetic change, the
rest of the grammar retained the syntactic and semantic relations as they had
been before, and the umlaut vowels were by necessity related to these facts. The
phonetic environment was at some point replaced by a grammatical one, that
is, [back]~ [front]/-[PLURAL, CAUSATIVE, and -p-NOUN], or the like. Not all
plurals had umlaut-only those that had *-i earlier; that is, such items had to be
specially learned, as before. The rule has now been modified. Its environment
needs a more complex specification in terms of the total grammar and not only
phonology. The umlaut rule in English has basically retained this structure,
although many other rules have piled up subsequently (unrounding, shortening,
vowel shift, §§ 4.5, 4.8, 6.2).
[6.13] Oldest English had only voiceless spirants; that is, all spirants were
realized as phonetically voiceless in every environment (e.g., f ~ [f]/ ... ;
§ 4.2). A change
spirant
. ] > [spirant]/[
. . ]- [voice
vmce . ]
[ VOICe 1ess VOICe
added its replica at the end of the already existing rules (replace > by~).
Borrowings from Kentish and French established initial voiced spirants, and
long consonants (and thus also voiceless spirants) were shortened. The balance
between units and rules was upset. For synchronic purposes, rules have a reason
only if they handle alternation. Now, some of those words that had had a short
intervocalic spirant, for example, always showed voice only, as did the loans.
There seemed to be no reason to carry this voicing rule in these words any
longer. It could be dropped by changing the underlying units into voiced ones,
for example, !vi in over. Such partial unit merger, of course, brings another
rule with it of the straightforward type !vi~ [v]/ ... which does not create or
carry alternation. Typical, in this case, is that when a rule is lost the underlying
units get rearranged. This process is called restructuring; that is, the end result
is that, for example, the old lofer! was restructured into lover!. The voicing
rule and old lfl remained, of course, in those items that preserved alternation
between /v/ and /f/ ([v] and [f)). This is one more example of the interrelations
between units and rules and how they adapt to each other as part of the tendency
toward 'one meaning, one form '. This tendency is hostile to alternation and
thus also to rules that carry alternation. Such rules are usually dropped, and
when rules drop, underlying units are also overhauled. We can continue with
this example, the material for which has already been given. By 1200, short
vowels had been lengthened in open syllables, and then a in final syllable
dropped. As in the umlaut case, a was the phonetic environment that induced
voicing for those spirants that were voiceless in other environments. And,
again, the conditioning of the new voicing rule had to refer to the grammatical
environments: voiceless spirant~ voiced spirant/-[ VERB, PLURAL] (e.g., bathe,
staves). Note that the voicing rule now always occurs with the lengthening
RULE CHANGE 119
rule (unless, of course, the base form already has a long vowel), and it might be
possible to combine them, for example, ~voiced spirant with preceding length •
or the like. Further, one could take the voiced variants as the basic units and
reverse the rule with the units, giving a devoicing in the environments [NOUN]
and [SINGULAR], because words where this occurs have to be especially learned
anyway. Now we see that the voicing rule starts to become historically dead,
although, of course, ultimately triggered by history. Synchronic rules can skip
historical stages or combine them into one. In cases like staff/staves the rule
has been dropped with concomitant restructuring of the earlier form lstafl into
ME lstafl and lstavl. Again the drop ensures invariance of morphemes. Perhaps
for many Modern English speakers this has happened also for lglresl and lgleyzl
or lgrresl and !greyz!, where the vowels also alternate and the verbs have a
technical meaning. The rule apparently still works fine in items like wreath/
wreathe. It is equally clear that alternation has been eliminated in cases like
who!efhal!owfheal, where indeed the meanings have also separated (§ 4.8). The
sequence of changes within the rule is, in a way, a kaleidoscopic process. In
these cases, the environmental part (c) of the rule shifts from phonetic to gram-
matical reference, and from there it jumps to the units (a), and the jump elimi-
nates the rule. This is another indication of the complementary nature of units
and rules, or a and c. Rules keep variants together within one morphophoneme.
When the variants are scattered enough they become independent, and unity
is achieved within each linguistic sign by dropping the rule. Paradoxically enough,
regularity is established by dropping the rule, but this is true only of unproductive
infrequent rules. The voicing rule started out as a most productive rule and is
now merely a relic. The reason is the grammar's adaptation to historical change.
have already operated on these forms; we are looking only at the k "' 0 alterna-
tion before s. We have already seen that Estonian loses a k which alternates
with zero(§§ 4.27, 6.10); k is lost from this item completely, and the underlying
form becomes teos- in contradistinction to the Finnish teoks- (ultimately some-
thing like jteko-kse- j and jteko-se- j, but this is irrelevant here). The Finnish
synchronic rules reflect history, Estonian no longer does at this point. Since this
change happened only in a clearly specified environment, an underlying jkj
remains in the language elsewhere (some dialects do retain -ks-). But both the
German and the Estonian cases show how a "sound change" leads to the drop
of a synchronic rule with concomitant restructuring.
[6.15] After the addition of the devoicing rule, German jdingJ 'thing'
was realized as dil;k, gen. dil;g-as. Then g is dropped after fJ, and the paradigm
goes dil]k/dil;as. This is the correct historical sequence and still the synchronic
state of affairs in the Northern dialects (note the similarity to the Finnish 0 "' k
alternation in teosfteoksen). In the standard dialect, a k that alternates with
nothing drops out, giving dif)/dif)aS (compare now the same kind of morpho-
phonemic change in Estonian yielding teosfteose). We can interpret this situation
so that jgj was now dropped after jnj, which itself became an independent !IJ!
in this position (i.e., the restructuring of jdingJ into Jdi1JJ); this item now skips
the devoicing rule completely, because there is no longer any stop to undergo
it. This is no doubt the correct synchronic state of affairs for most "innocent"
standard German speakers, exactly as there is no k for the Estonian speakers
in teos either. The problem is that linguists are not innocent speakers; they know
too much (see§ 18.17). Even without referring to the actual history, they would
notice that IIJ I has a very precarious domain in German, because normally [IJ]
occurs only before velars, for example, daf)k 'thank' +-Jdank J. They would
try to write rules which keep a velar stop there long enough to get JnJ __,.. [IJ]
and then drop it. The only free velar available in this position is g, so we would
come back to the underlying JdingJ purely on the basis of filling a gap in the
patterning. Thus we apply our rules in the order
1. n __,.. f)/-[velar, stop] (daf)k, dif)g)
2. g __,.. 0/f)- (dif))
3. devoicing (fo·p, bunt 'league', ta·k)
The order of the drop of g and devoicing is exactly the reverse of the historical
order (relative chronology), which was retained as such in the Northern syn-
chronic order:
1. devoicing (dif)k, fo·p, bunt)
2. drop of g (dif)as)
3. drop of k (when "' fJ) (dif)) [does not apply in the North]
A historical change of dropping a k when it alternated with nothing triggered
a reordering of rules. Here history and synchrony differ widely. Note again that
whatever we are not willing to do with separate units has to be done with rules.
RULE CHANGE 121
We do not know when speakers shift from rules to units, but it is clear that
they do it earlier than linguists. Reordering is largely due to the linguist's
reluctance to modify the underlying forms and, as we shall see, the reluctance
to modify the rules. Most cases of such reordering can be handled with loss
and concomitant restructuring or by modifying the environmental part of the
rule. The only mechanism that seems to be historically real is the addition of
rules (e.g., sound change, analogy); loss and reordering are effects in the
particular notation used (see § 22.9).
k-> cf-+{~}+
cf-( +)[front vowel]
kf- (elsewhere)
The environments (all of which are not exemplified here) can now be specified
with morpheme boundaries ( + ). The second palatalization takes place first in
the environments where front vowels i and eare the sole markers of a morpheme,
the first palatalization follows in other environments with front vowels, and,
finally, we get the constant subrule k-> kf in other environments. When subrules
are ordered, they specify less general cases first and give the most constant
part last. The subrules then show strict descriptive order. With the notion
of the generality of the constant rule among ordered subrules, descriptive order
is by expectation very different from relative chronology, as it is here. In the
case where all the realizations are phonetically different, for example, p t k
from the Greek labiovelar lkwl, the most general one is the one with widest
coverage (here p ).
[6.17] In Finnish, long mid vowels diphthongize, thus tee (still the
Estonian form, §§4.24, 5.15) >tie 'road'. Then comes consonant gradation,
122 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
Note that the synchronic rules may now quite well skip the intervening *y, at
least in this situation. In the Eastern dialects, there is another diphthongization,
so that teeq > tieq. Synchronically, however, both diphthongizations can be
combined by reversing the above order into
1. consonant gradation: tekeq-+ teeq
2. diphthongization: teeq-+ tieq, tee-+ tie
Structurally, this situation is exactly identical to the German Ding-case(§ 6.15):
in those dialects where an additional change modified the output of the earlier
shared rules, the new relation between the dialects could be described by
reordering. Description by this procedure has become popular in dialectology.
In the German case, there was another possibility of explanation, however:
loss of a rule with concomitant restructuring. In the Finnish case, there is no
such possibility. Clearly it could combine the diphthongizations, because in both
cases an e gives i. But we have already seen this situation in the paradigm
of vesi 'water' (§§ 4.29, 6.6). Synchronically, it is better to try to connect the
diphthongization to the raising. In Standard Finnish, we would now have
whereas articulatory changes were not, even though they triggered the structural
ones. Structural classification of analogy also embodied changes in distribution,
that is, leveling and extension, as well as analogies that were parallel to the
articulatory grouping: proportional and nonproportional (contamination, folk
etymology, and so on). Transformational grammar and related positions take a
similar stand in looking at changes, because only those changes are counted that
affect competence. Thus changes are of two kinds, those that affect performance
(not counted), and those that touch competence. This division is even more
arbitrary than the phoneme classification, because the two areas cannot be
divided. However, in science it has usually turned out to be more useful to have
some classification than none. After English d had become alveolar, the corre-
sponding feature [dental] was replaced by [alveolar] in the rule; this fact was
now part of "proper" English. A rule was modified, but since it changed a
surface feature into another surface feature without otherwise affecting the
grammar, it is not counted as change. This is exactly parallel to the position of
not counting allophonic changes.
did not involve leveling but extension of the diphthong ie into words that did
not have it before (tieq; § 6.17). In the Finnish case, a new environment is added
(fed) into the diphthongization; in the German case, an environment is subtracted
(bled) from the final devoicing and no alternation could result, although it was
there before.
Modification of the structural analysis of a rule (a + c) results in extension
of the rule, and shows again its unity. Thus, if a rule like
stop
[ voiceless
] ~ [spirant ]j
voiceless -
(p t k ~ f () x)
in a language that has voiced and voiceless stops drops the feature [voiceless],
the rule becomes more general. Fewer features need be specified, for example,
[stop]~ [spirant]/-(c), and the result is an extension of the domain of the rule,
because all stops undergo it: p t k b d g ~ f () x .P o y.
[6.20 Marking and Rule Change] Every language has both symbolic and
iconic/indexical elements. The lexicon is largely connected with the former, the
rules with the latter (see§ 1.14). Not everything in a language can be generated
with the regular rules; some things have to be learned as separate units. Thus
there is a clear difference between compounds like dog meat, which can be
accounted for in terms of a productive pattern, and nutmeat and sweetmeat,
which cannot. Because of semantic change, the compounding rule has been
dropped from the latter, making "the old compound a new single linguistic sign
which has to be learned as a unit (restructuring!). All languages show similar
"relics." All languages are composites of tradition (conventional symbolic
elements) and creation, and only the latter can be handled easily with rules. Of
course, every child learns the creation or rule aspect together with the unproduc-
tive conventions. Such unproductive relic mechanisms have to be especially
marked, which is another way of saying that they have to be learned separately,
that is, that they cannot be generated with the usual rules.
On all levels of grammar, one makes the distinction marked vs. unmarked,
when there is an opposition. For example, in English, the singular/plural
contrast is normally marked by some overt plural marker, as in boy: boy-s,
cat: cat-s, and so on. Within the plural itself, some patterns are further marked
formally against the regular-s, for example, mouse: mice, ox: oxen, and so on.
It is always the more general, less restricted member (in the case of two) which
is the unmarked one. Marking, however, need not be overt at all. Figure 1-5
contains semantically unmarked terms in the first column, for example, horse,
against words marked for sex (stallion, mare) and young age (foal). Horse is
clearly the most general term, because it can be used for the others if the marking
is eliminated, that is, if the situation does not warrant such detail.
If a language has many declensional and conjugational patterns, all words
have to be marked for one, and it might be difficult to have an unmarked type
at all (e.g., Latin and German). Lexical marking like that belongs to the diction-
ary; it acts as the address, which takes the word to the appropriate rules in each
126 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
case. In other words, the marking is a catalyst that triggers the rule. 'Irregular'
or 'strong' forms are the traditional terms for formally marked items in mor-
phology, and they mean exception from the most general rules. The English
verbal inflection includes the unmarked weak conjugation, the marked weak
irregular, and the various strong ones. Thus bring would have to be marked
[+weak irregular], although, from its phonetic shape, it might seem to belong
to the swim-type (see §§ 5.2, 5.19, 10.8), which we can call here the [+ablaut]
group. And, indeed, children try to introduce ablaut bring/brang into this verb.
In other words, the analogical shifts between classes can be spelled out with
features. The motivation is apparently the formal presence of a nasal in bring;
that is, like forms should have like treatment in the grammar. If, however,
marking is dropped altogether, the verb shifts to the unmarked class, as in
bringfbringed. The dropping of marking eliminates irregular forms by assigning
them to the productive types. Marking is no more than a restatement of tradi-
tional analogy, because marking notation does not add to our understanding
of such shifts; for example, it gives no insight into the reason why dreamed and
kneeled were reassigned to the pattern of dealt, so that we get dreamt, knelt.
This is a complication in the grammar, and a similar, more recent one is dove
for dived. On the other hand, languages seem not to go one way only, that is,
become simpler and simpler all the time. Thus the plural formation of OE cu
'cow' was marked as something like [+umlaut], giving cji. Later, apparently,
the semantic antonym ox influenced the feminine, and another overt marker
was piled onto the word, [ +n-pl.] = kine. Note that the mere additional
marking [ + n-pl.] does not "explain" the reason behind it, which is more to be
found in iconicity in the forms of this bovine set (§ 5.4). Now, when all the
marking has been dropped, the outcome is cows, according to the regular
pattern.
(6.21] Marking with respect to a rule can be positive or negative; that is,
some items are marked as subject to a rule, others as exceptions to a rule.
Morphemes that satisfy the environment of a rule without undergoing it are
marked negatively with respect to that rule. Loans often establish this situation;
for example, when French loans like chief, faith, and so on, were adopted into
English, they had to be marked [-voicing] for the plural formation. Similarly,
recent loans in Finnish are marked [-consonant gradation] (§§ 10.16, 10.17),
one such word being auto 'car', (gen.) auto-n. On the other hand, morphemes
subject to a rule that is not predictable from their shape are marked positively
for the rule. Thus, already in Old English, words like miis 'mouse' and hOc
'book' were marked [+umlaut], which ensured the corresponding plurals mjis
and bee. Elimination of negative marking expands the domain of the corre-
sponding rule, because exceptions to it diminish. Similarly, elimination of
positive marking restricts the domain of a rule, because all items positively
marked for a rule are exceptions in the total grammar. Marking is, as has been
said before, something that has to be learned by special effort, and children
in particular, while learning the language, try to omit it. Thus Finnish children
RULE CHANGE 127
[6.22 Cause and Effect] Even if we have a certain scoring board for
recognizing and counting changes, something like what we have observed in
this chapter, the question of cause and effect is still open. The point of view
128 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
that changes would first occur in the program, in competence, and only later
would change the utterance without having anything to do with performance is
clearly inadequate, although popular. This notion is a result of the scoring
mechanism for changes. Language is learned and hearers interpret utterances
largely from performance, that is, real concrete speech situations. Causes have
to be sought in the totality of language and the relation of its use with the
total culture and individual speech acts. If the surface forms are ambiguous, in
the sense that they can be produced in two different ways, we have a potential
source for reinterpretation, which may change the competence. Most changes
seem to be triggered by performance; only grammatical conditioning of sound
change and related phenomena are aided from higher up (in a hierarchy as in
Figure 1-2). And, indeed, one cannot separate performance from competence.
We have seen how ambiguous surface forms have been inductively reinterpreted.
This has changed the program, because the reanalysis comes to light deductively
in the outputs of this modified program(~~ 5.5, 5.6). Thus changes can originate
from both ends, which is indeed logical, because language is used from both
ends; that is, there are both hearers and speakers (Figures 1-1 and 1-2). Language
is a system that tries to keep an optimum balance between form and meaning.
Languages are there to be used; if a language is no longer used, it ceases to
change. Hence usage and the actual situation must play a role in the changes,
wherever the first impetus originates (see§ 9.16f.).
We have not yet mentioned changes in the social registers of the linguistic
rules, although linguistic rules are always society oriented (Chapter 3, § 6.8).
Age characteristics change when people grow up or die, or when a social group
emigrates as a whole. We have first the particular social or historical change in
the community, which has a reflex in the rules of the total grammar of the
community. It would be ridiculous to maintain that the rules changed first,
thereby killing off old people and exiling particular sects. In cases of this kind
we see clearly how rules are changed from the outside, by history in general,
and this is perhaps true in most cases.
Proto-Indo-European had no infinitive in its verbal paradigm, but most of
its daughter languages grew one. Thus we can describe the situation as the
addition of an infinitivization rule. In Greek, if the second half of the sentences
'he wants, he writes' is the object of the first, and he has the same referent in
both cases, a rule operates that deletes the second he and replaces the present
with the infinitive (which is the most unmarked neutral verbal form), for example,
thelei grdphein 'he wants to write'. Because infinitives are derived from underly-
ing full sentences, it is clear that we need this rule, and, historically, it was added
in Greek (see § 5.16). The same rule applies to the/a grdphein 'I want to write',
and so on, where there is more formal divergence between the verbs on the
surface. We could say that Modern Greek lost the infinitivization rule again
when the sentences come out as thelei (nd) grdphei, the!O (na) grdpho. But the
situation is not this simple, as we have seen. The motivation for the loss was
that sound change had eroded the output of the infinitivization rule, and this
triggered a reinterpretation of the surface ambiguity. The same is true of the
RULE CHANGE 129
Finnish head and attribute switch(§ 5.17), which as a change in a purely syntactic
rule would have hardly any motivation.
It is relatively easy to devise rules for mapping the transition from what was
earlier to what comes after-to say, for example, that Greek dropped its
infinitive rule and the rest followed. As we have seen, the historical changes are,
to a great degree, independent from the synchronic rules. What the rule changes
always describe, then, is the before-after relationship. They give a mechanism
for description, not a historical explanation, except in accidental cases. This
fact is often forgotten. Phonemic changes follow suit, because they, also, could
occur only between two stages (before-after), whereas phonetic change occurred
within the same system. If we count changes in competence only, we jump from
what was before to what comes after when we have a new grammar. And the
similarity between tallying phoneme changes and rule changes is no wonder,
since both use such rigid scoring rules. Moreover, the proportional formula is
heavily before-after oriented, because here one actually writes both parts in
the formula. All this is indication again of the basic unity of structural change,
whether presented as phonemic, analogical, or rule change. Every such concept
exemplifies a form of structural change.
The difference between gradual change through time and the linguist's
statements on before-after relations leads easily to misunderstanding. For
example, phonetic change is primarily gradual (see § 4.21), but the linguist's
phonological notations allow mostly for abrupt leaps. It is hasty to say, therefore,
that all change is abrupt, because that is a consequence of the notation. The
mistake is generally twofold. If, for example, a phonological notation is binary,
all features must manifest either as [+]or [-],and any change described through
this will look abrupt. This mistake is a deductive one. On the other hand, if
some changes are abrupt (e.g., metathesis), it is inductively wrong to argue that
all other changes must be the same. It is of utmost importance to distinguish
between the actual change and the before-after relations manifested in our nota-
tion and rules. The latter can be called diachronic correspondences; they need
not reflect the actual history at all, but they are always "abrupt" (see§ 9.14).
[6.24 Rules and Surface Forms] The question of the form of rules (and
their change) is still very much an open one. This chapter has delineated a
central position from which further investigation has to take off. Emphasis
was put on the obvious connections with the traditional analogy whose mecha-
nisms the rules make more precise, although not their motivation. Rules themselves
are posited on the basis of the (morphophonemic) alternations, or in syntax with
similar formal correspondences that show semantic similarity or identity
(transformational relations). The rules are there to bring out the actually
occurring surface forms. Thus the ultimate justification of rules includes very
much the same kind of surface linguistics as analogy, especially proportional
analogy (see § 5.1 ). Rules represent another hypothesis about the inner form of
grammar based on the actual surface alternations. In other words, rules belong
as much to competence as to performance, and after all, no boundary can
be drawn between the two. The material in this chapter, as in Chapter 5, has
shown the underlying unity of the attempt to elucidate the unknown. The rules
are also hypotheses and can quite well be wrong, exactly like, for example,
proportional analogy. Indeed, the number of mistakes made in the writing of
rules is perhaps as high as those made in applying other forms of analogy
earlier. The errors are chiefly seen in the general tendency to write rules for
unproductive fossilized connections of the type drink/drench, bake/batch,
hallowjwho!e, and so on. In short, regularity has been pushed into the irregular
RULE CHANGE 131
parts of grammar that are unproductive (i.e., symbolic rather than iconic
aspects).
REFERENCES
General. Halle 1962, Saporta 1965, Kiparsky 1965, 1968a, Sigurd 1966, Chafe
1968, Chomsky and Halle 1968, Postal 1968, Weinreich and Labov and Herzog
1968, Wang 1969, King 1969b, Leed 1970, Newton 1971; 6.1 Bremer 1894,
Gotze 1923, Hermann 1907, Hoenigswald 1960a, Horejsi 1964, Chen and
Hsieh 1971; 6.3 Voyles 1967, W. Bennett 1968; 6.4 Martinet 1958, 1964, King
1969a, Kiparsky 1965; 6.5 Jakobson 1968, Fudge 1969, Stampe 1969; 6.6
132 HISTORICAL LINGUISTICS: 1-low DOES LANGUAGE CHANGE?
Lenneberg 1967, Luria 1967, Oldfield and Marshall (eds.) 1968, Martinet (ed.)
1968, Whitaker 1969, Graur (ed.) 3.201-321, 683-779; 6.7 Kiparsky 1965,
Chomsky and Halle 1968; 6.8 Fischer 1958, Weinreich and Labov and Herzog
1968; 6.10 Weinreich and Labov and Herzog 1968; 6.12 Leed 1970; 6.13
Huntley 1968, Leed 1970; 6.15 Saporta 1965, Vennemann 1970ab; 6.16 Zeps
1967, D. Cohen 1969; 6.17 Anttila 1969a; 6.18 Andersen 1969, Vachek 1968,
Postal 1968, King 1969b; 6.19 Schuchardt 1928, Kiparsky 1965; 6.20 Lyons
1968, Maher 1969c; 6.21 Andersen 1969, Maher 1969c; 6.22 Grace 1969, T.
ltkonen 1970, Andersen 1972; 6.25 Leed 1970, Newton 1971.
CHAPTER 7
SEMANTIC CHANGE
[7.2 Pure Semantic Change] Changes in the linguistic sign are more
obvious than changes in the semantic structures alone. Figure 1-5 arranges the
semantic relations on exactly the same principles as those by which Figures 1-3
and 1-4 handle sounds, and such examples can be multiplied; see, for example,
Figure 7-1, in which Spanish semantic experience is divided in a different way
133
134 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
from Latin, and this is parallel to what happens in the relationships in a sound
system, as emphasized with the boxes of the diagrams. Some of the Latin
forms do survive in Spanish outside the system, for example, Latin novel/us >
novillo 'young steer, calf'. For pure semantic structure and change, one has to
observe the configurations of the boxes, as it were, and not the names of the
boxes. Thus the distinction between Latin 'horse' and 'mare ' has remained
semantically unchanged until Modern French, for example,
2. I
cheval ive French (early)
The meanings themselves remain the same, although the names change. However,
this kind of a situation is also called semantic change, because the word jument
meant 'pack horse', before it came to be connected with the meanings' female +
horse'. These meanings are connected with the form mare in English (Figure
I -5). These examples show the two possible ways of looking at the form-meaning
links, that is, we can concentrate on the name/form and see what meanings it
has represented through time, or we can take the meaning as the base and see
what forms have represented it at various periods of a language. The former is
called semasiology 'study of meanings', and the latter onomasiology 'study of
names', that is, forms. In both cases, the linkups of linguistic signs change.
When the form jument shifted its colligation from the meaning 'pack horse'
to 'mare', we had a case of semasiological change and the result was a new
linguistic sign, often loosely referred to as a new word. When the meanings
'horse + female' ('mare') replaced their form equa by ive, only sound change
had occurred, which might be regarded as a slight form of onomasiological
change. But when ive was ousted by cavale we have a normal case of name
change, and it is repeated in the shift from cavale to jument. Semasiological
changes are accompanied, at least somewhere in the grammar or lexicon, by
onomasiological ones, for example, overt increase or decrease in names, and
so on. This is indeed what makes discussion of pure semantic change so difficult,
and why one tends to remain with the total linguistic signs, that is, both semasio-
logy and onomasiology. In other words, the lexicalization rules, the rules that
link form to meaning, may change without any change in the semantic structure
itself. In addition to the Latin terms iiter 'black ' and a/bus 'white', there were
also two terms marked for the feature 'shining': niger 'shining black' and
candidus 'shining white'. When the semantic notion of 'lustre' is lost as an
obligatory feature, the units merge into one name for 'black' and 'white',
respectively, in Romance, for example, French noir 'black' (which continues
SEMANTIC CHANGE 135
Voiceless Voiced
Bilabial p b
Dental t d
Velar k g
Labial
Voiced b
Voiceless p I f
Stop fricative
'old' 'young'
'persons' senex iuvenis +human l ~
.§
'animals,
plants'
vetulus novellus -human ~ +
§
the old marked term) and blanc 'white' (with no formal connection with the
old terms). Of course, the possibility of speaking about 'lustre' or 'shine' was
not lost; only its obligatory indication was. Semantic merger was here accom-
panied by a reduction of forms as well. With semantic split, the number of
forms increases, at least in part of the vocabulary, for example, when Latin
avis 'bird' gives Spanish ave 'big bird' and pajaro 'little bird'. When we look
at this semasiologically, avis has restricted its semantic range, and pajaro has
expanded it, since the form continues Latin passer 'sparrow'(§ 7.12).
Latin had a generic term homo 'man', with sex marked in vir 'man' and
femina 'woman' (compare Figure 1-5). French and Italian have eliminated the
masculine marking, thereby giving a skewed system as shown in Figure 7-2:A.
136 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
homme
A. Latin
homo
>
uomo Ifemme I French
Ivir I femina I donna Italian
iiter
B. Latin > French
French thus continues the old Latin forms, but Italian has replaced the female
term with what was earlier a more restricted term domina •mistress'. Note that
Latin is of the German type: Mensch (Mann-Frau) vs. English man (woman),
which is just like French and Italian. This is, of course, parallel to a total loss
of marking like in the liter area (Figure 7-2 :B).
[7.4) A case where change was due to an ambiguous context is the meaning
of English bead. The word originally meant 'prayer' (compare bid, and German
beten 'pray'). Medieval and modern religious practice holds it important to
keep track of the number of prayers, and the scoring device is the rosary with
its small balls. Praying with this device was called literally counting one's beads,
that is, prayers. The balls just represented prayers as symbols, and praying and
counting the balls were contiguous activities, the former being the cause of
the latter. Now the more obvious referent to counting one's beads was the
physical activity of tallying the balls and the situation was interpreted this way,
whereby bead came to mean 'small ball' (also boon has shifted its meaning from
the 'prayer' [compare Swedish bOn 'prayer'] to the 'thing asked for, a welcome
138 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
benefit'). Without knowledge of this religious practice, the change from 'prayer'
to 'small ball' would be completely incomprehensible.
[7.6) Thus, with cultural expansion, there is a constant need for new names,
and this is often quite explicit. Mostly one resorts to perfectly iconic (motivated)
descriptions, for example, radio detecting and ranging(§ 2.13), lunar-exploration
module, drive-in theater, and so on. Often some kind of shortening is the result,
as in radar, LEM, and drive-in. Very seldom, indeed, does one coin something
out of the blue, for example, gas, kodak; even a word like gobbledygook has a
considerable onomatopoeic basis. But under the right sociolinguistic conditions
such creations are indeed possible; Estonian has incorporated dozens of them
since the 1910s. The following belong to the active vocabulary of contemporary
Estonian: laip 'corpse', relv 'weapon', laup 'forehead', roim 'crime', kahur
'cannon', and veenma 'to convince'.
The process of giving new names to either old or new things is called nomina-
tion, and it characteristically implies the extension of the machinery already
available in the language. It is, at the same time, both onomasiological and
semasiological change. Often nomination can be carried out with loanwords,
and we shall return to this both in this chapter and in Chapter 8. The need for
a new name need not be a consequence of a new meaning/thing. There are all
kinds of social and psychological reasons for renaming things, for example,
euphemism and tabu in general. All such restrictions depend on the particular
culture and the particular speakers. For instance, the old Indo-European name
for bear has been replaced by a euphemism (hunting tabu) 'the brown one'
in Germanic, and the Germanic word for wolf gave way to varg 'out-law' in
Swedish, varg, in turn, being replaced by descriptions like 'grey-foot'. Other
European languages have used circumlocutions like the 'honey-eater', th~..
'honey-paw', or the 'apple of the forest' for bear. The tabu of obscenity is well
known, although the notion itself varies quite arbitrarily. For example, at one
point the word leg acquired indelicate connotations in America, which would
clash with good taste at Thanksgiving dinners, for example. Euphemisms like
dark meat or drum stick saved the day. Sometimes the correction itself would
seem to be more objectionable than the original term; for example, in some
American dialects, the highly tabu word bull was replaced by top cow. All this
shows the unpredictable social forces at play. And man does indeed play with h1s
language (e.g., kisser for 'mouth'). The world wars immediately produced rich
soldier slangs, which were useful in releasing emotional pressure. Weapons were
called with household words, for example, 'coffee mill' or 'sewing machine' for
a machine gun, 'repair shop for comrades-at-arms' (Finnish asevelikorjaamo)
for the first aid depot in the field, and so on. On the other hand, the use of war
140 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
terms for familiar household items was also able to lessen the grim connotations
of the former, for example, 'wire obstacle' for beard and 'hand grenade' for
potato, and so on. The first steam drills in the Rocky Mountain mines were
known as widow makers, where again humor lessened the grim realities. Similar
euphemism exists also in those social layers which practice deceit in one form
or other, as in the underworld; but, linguistically, merchants and politicians also
fall into this class, with their advertising propaganda (e.g., 'ultimate solution'
for genocide or 'home' for house). Literary style also requires new names for
old notions; as Aristotle said, if one uses too little metaphor, language becomes
plain and dull, and if too much, language becomes enigmatic.
the meaning 'swiftly' by about 1300, whereas the adjective fast retained its
meaning 'firm' (compare German fest). Toward the end of the fourteenth
century, fast borrowed the meaning from the adverb, hence 'rapid, swift',
though the original meaning also survives, he is fast asleep (compare fasten).
Thus the borrowing can occur within the same language between different
categories which display formal similarity. Here also the principle at work is
analogical, that is, 'one form, one meaning', in other words, borrowing from
within is analogy (see§ 5.18).
Ieonicity lndexicality
(similarity) (contiguity)
Meaning (sense) metaphor metonymy
Form (name) folk etymology ellipsis
A. metaphor
metonymy
Form Form
'off' 'of'
'firm' 'swiftly'
('precipice') ('trick') 'modest' 'shameful'
('flying mammal') ('baseball stick') 'attacking tower' 'bell tower'
etymology
Form Form
dropping one part, for example, hliiford > lord, hussy, shepherd (see § 5.15).
These have to be learned as single signs; in other words, a motivated relation
1\ has yielded I, one symbol.
+
In polysemy (case C), meaning is the trigger of change, contrary to F. Depend-
ing on the different grammatical categories, the same form becomes differen-
tiated or leveled (see § 5.14), that is, V > 1. I or I· Thus the more emphatic
SEMANTIC .CHANGE 145
adverbs retain forms like thorough and off, whereas the less stressed prepositions
yield through and of (compare further convey-convoy, clothes-cloths, costume-
custom, conduct-conduit, alternate-alternative, masterful-masterly, and so on,
and similar cases in other languages, e.g., Swedish kliidning 'clothing'-kliinning
'woman's dress' and Latin religens and religiosus, both originally 'pious', in
religentem esse oportet, re/igiosum nefiis 'It is proper to be religious, wrong to be
superstitious'). Latin homo/ hominem yields both French on 'one' and homme
'man'. On the other hand, if a variant borrows its meaning from a different
category, we get V > I as in the case ofjast with its meaning 'rapid' borrowed
from the adverb, where it had developed creating elaboration of the type V
(ignoring here relics of the original, and derivatives like fasten; § 7.7). We have
already seen that this kind of borrowing from within is a type of analogy(§ 5.18).
Here, borrowing leads to simplification, in case B, to elaboration. The reason
is the different basis of simplification; in C, it is agreement (correspondence,
iconicity) within one and the same grammar, and in Bit is agreement between
two different grammars (languages), that is, cultural influence rather than purely
linguistic (compare spelling changes,§ 2.15).
Sometimes the split occurs in writing only, for example, flowerjjlour, born/
borne, metal/mettle, French dessein 'design, purpose' (God) vs. dessin 'design,
drawing' (artist), German Mann 'man' vs. man 'one', which also are different
in stress (compare French hommefon). This is one indication of the importance
of writing in communication, and the parallelism of changes in writing to those
in language.
When the phonetic shape of a morpheme leads to a new meaning (case D),
the innovation is likely to survive, since it increases iconicity. When the old
meaning vanishes, the situation is again simplified as V > 1. And this is the
configuration also in metaphor and metonymy (A), when their literal interpreta-
tion is lost and the transferred meaning becomes the normal unmarked one.
This process is called fading, and it occurs with all figures of speech (e.g.,
awfully nice, ne ... pas). Repetition itself is not enough for fading, because in
certain semantic spheres metaphors may stay alive. This has, in general, been
true of Western religious metaphors. The Latin metaphor pastor 'shepherd'
for the head of the congregation ('herd') has faded outside Latin and I tali an,
but, in all languages, the metaphor of a shepherd and sheep lives on. In com-
munication without emotional overtones, metaphors lose their association
with the primary meaning more easily if the referents do not normally have
emotive value for the speakers. As long as both meanings are present, the meta-
phor has both cognitive and emotive value. If the metaphor does not catch on,
there is really no change, but if it does, and this leads to the disuse of the original
meaning, only the marked reading remains. But a marked value without a
corresponding unmarked one is not possible, and hence the remaining connection
becomes the normal one. A metaphor for a weeping person was maudlin (Magda-
lene, because she was often depicted in that state). When this extralinguistic
connection was lost, maudlin became a normal adjective meaning 'weeping,
foolish, sentimental'. This is also how euphemisms tend to become the normal
146 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
symbols for the referent, and thus necessitate new euphemisms, for example,
Swedish ulv 'wolf'__,.. varg 'outlaw'__,.. graben 'grey-foot'. A new euphemism
could occur when the old metaphor or metonym had faded and become the
normal sign for 'wolf'. In Old English, the antecedent of wanton meant 'undisci-
plined, unruly' and the word was euphemistically used for 'lascivious, lewd',
which, indeed, became one of the habitual meanings of the word. Different
social layers handle euphemisms quite differently. In high style, a euphemism
can remain as such (e.g., outhouse), whereas in low style it easily becomes the
normal word for the referent; then it has to be avoided as a vulgarism even by
the higher circles (in favor of toilet, restroom, and so on).
All the cases show that fading is simplification in the meaning-form relations.
It is a shortcut between form and meaning in the colligation of the linguistic
sign. It is a by-product of the tendency 'one meaning, one form'.
The mechanisms presented here represent only the very basic core of semantic
change. The subject is enormous if all the possible distinctions are drawn. It
is easy to see why semantic change is so uncertain and unpredictable. Every
utterance can act as a metaphor in the right situation. All the faded figures of
speech can enter new living ones, which in their turn may again fade, and so
on. Thus pioneer meant originally 'foot soldier ' , whose task it was to clear the
terrain for the more prestigious troops. It was applied metaphorically to the
forerunners of colonialization and civilization, which is now the basic meaning
in English (but, in Russian, 'member of the communist youth organization').
A further metaphor, such as pioneer in linguistics, could therefore spring up.
Semantic change takes place in terms of the total cultural situation. Isolated
words do not change their meanings (unless they are so isolated that reanalysis
becomes easy), but words in clear syntactic contexts in sentences and the relation
of these sentences to the physical environment are often decisive. The diagrams
(Figure 7-4) emphasized again the similarity of semantic change to analogical
change (or rule change) (§§ 5.14, 6.19, 6.23). Both are different ends of one
and the same force, the adaptive power of language to meet the needs of com-
munication. Analogy and the rules of the grammar are exactly those mechanisms
that allow for metaphor, and so on, when the need arises. And the tendency
toward 'one meaning, one form' restores balance when either the formal or
semantic marking becomes highly loaded. Change is a function of language
as a system of communication; analogy and rules are the mechanism for it
but not the driving force.
[7.10 Semantic Shifts] Because linguistic units are not isolated, any
change in one item may lead to changes in other items within the same semantic
sphere (e.g., jaw used to mean 'cheek' and cheek meant 'jaw'). Such shifts
may be so regular and far-reaching that they look like shifts in the sound systems.
One of the best cases reported comes from Latin legal terminology. In the older
period, we have the following terms: damnum 'legal obligation, trust', noxia
'damage', culpa 'guilt', casus 'negligence', and fortuna 'chance'. In the later
period, the form- meaning links have shifted one notch:
SEMANTIC CHANGE 147
In short, the universal "laws" of semantic change are based on the figures
of speech in all languages (the same is true of other classifications to follow
below). In some cases, an item has obvious attributes after which it can be
called; for example, birds can be noted for eggs and flying. And, indeed, they
are rather commonly named after one or the other in the languages of the world.
German Vogel and English fo wl derive from 'fly/wing', although the latter is
no longer connected with flying; compare Finnish lintu 'bird' with lentiiii 'to
fly', and the Sanskrit words for bird a!Jrjaja 'egg-born' and dvija ' twice-born',
which refer beautifully to eggs.
switch to the one caused by sound change and reinterpretation(§ 5.17). Indeed,
Old English compounds with had 'state, quality' and lie 'body, form' give the
suffixes -hoodf-head and -ly (motherhood, tenderly, and so on). In Finnish, a
similar noun-forming suffix develops from the word vuosi 'year' > -uus;
compare uusi vuosi 'new year' vs. uut-uus 'novelty'. The French counterpart to
English -ly, -ment, was, in Latin, the nominal head; for example, dulce mente
'with a sweet mind' > doucement, and thus etymologically identical to ment
'mind'. Spanish still retains independent accent on the adjective, as in rdpida-
mente 'rapidly', although no space is written. German Drittel '3rd part' and
Viertel 'quarter, fourth part' contain a variant of Teil 'part'.
All this is quite parallel to the grammaticalization of the emphatic attributes
of the French negative. Forms that go together habitually become reinterpreted
as a unit, and semantics and forms adjust to the situation. Free nouns become
adverbs, and adverbs become affixes, prepositions, and postpositions, as well
as conjunctions. This is how grammatical morphemes often originate. We have
already seen the development of adverbs from the formal side (§§ 4.24-4.27,
5.7, 5.15), but even then semantics had to be considered as we are now consider-
ing form (belf-be, kerallaf-ke, /ikef-ly, Teilf-tel, vuosif-uus). In other words,
we see that sound change, analogy, and semantic change represent a whole,
which must be split up only for expository purposes (see Figure 9-1). And
syntactic change, also, forms a part of the whole-as a result of these other
changes; it is not an independent mechanism (see§ 19.5).
In Indo-European languages noun stem-forming suffixes rarely have any
clear meaning. But when inflectional endings are apocopated, their meaning
sometimes is reassigned to the (originally meaningless) stem formant that
remains. Thus PIE *ukwse 'ox' (nom. sg.) and *ukwsenes 'oxen' (nom. pl.) are
typical en-stems: *ukwsen- is the stem, and *-es is the ending for the nominative
plural. These forms appear in Old English as oxa, yxen and Old High German
as ohso, ohsen; the nominative singular without -n is an ancient form, but the
cases have disappeared from the ends of the other forms, leaving the -en to be
reinterpreted as a case or number marker (English ox, oxen shows a minor
adjustment of the vocalism-yxen would have given *ixen-but German Ochs,
Ochsen is quite regular). This also happened in the German weak feminines
like Kirche, Kirche-n 'church', where the bare stem (neutral with respect to
number) still occurs in compounds like Kirche-n-spiel 'parish'. In the same way,
the German plural ending -er is an originals-stem (as in Latin gen-usfgen-er-is,
N.B., by both Latin rhotacism and Verner's laws > r); and the Russian genitive
plural -ov represents the Indo-European u-stem from which the original Proto-
Indo-European genitive ending had been lost (see § 9.13). These cases are just
regrammaticalizations. In English, there is a tendency to reinterpret final -s fz/
as 'plural', especially if the meaning supports this. Thus cherry+- OF cerise,
where the mass noun was obviously plural ('inany berries'), and ME pese,
pees__,. pea, where the old form pease still lives on in certain dialects and archaic
contexts (e.g., nursery rhyme pease porridge). The result in all these cases is a
shift in morpheme boundary along the chain of sound units.
SEMANTIC CHANGE 151
in semantic change are iconic and indexical, exactly the same ones that operated
in analogy and rules, as well as in the development of the writing systems.
These are the forces of change in historical linguistics. Culture and society are
particularly clear in this respect, because the referents have so many connections
and similarities of function, and so on (see§ 9.16f.).
We have now seen that on all levels of grammar we have iconicity, part of
which is known under the term motivation. Onomatopoeia show phonetic
iconicity, and metaphors show semantic iconicity. The former show the nature
of images, the latter that of diagrams, exactly like analogy. Grammatical or
morphological iconicity resides in the rules of a grammar, for example, in the
motivated compounds like steam engine, raincoat, and armchair, which all show
indexical connections in addition; that is, steam is the energy for the engine, the
coat is against rain, and the chair has arms (itself a metaphor) (see Figure 1-6).
(7.14 The Role of History] The accidental nature of history and its
influence on language (semantic change) cannot be overstressed. Etymological
research has discovered many semantic leaps just by the availability of accidental
historical knowledge. Although the case of fiasco was chosen as a possible ex-
ample, it, of course, is not conclusive. But there are many others, although
perhaps not of such stark character. In 1956, in a Finnish army unit, one of the
women kitchen workers was known as Risteilija 'Cruiser'. Such a nickname
is metaphorical, as nicknames and names often are (if they are not indexical,
e.g., son of ... , names after the trade of the person, after individual physical
characteristics, which is also iconic, and so on). As the town had navy units, the
most obvious conclusion would have been that her promiscuous behavior had
been labeled by a direct metaphor from the naval vessel. But closer familiarity
with the total "cultural" situation proves that the assumption would be false.
Mess-hall tables had a bowl where one could deposit potato peels, bones, and
so on. This bowl cruised regularly between the eaters and was known as risteilija
'cruiser'. The nickname was a second metaphor based on this garbage bowl! In
another example, at the time of the first atom explosion on the atoll of Bikini, a
contemporary bathing suit was metaphorically named after it, that is, something
that has startling effects, leaves very little cloth there, lays things bare, or the like.
When, in the mid-1960s, the topless fashion set in, the word bikini was reinter-
preted as having the same bi- as, for example, binoculars, because the item did,
in fact, consist of two pieces (compare formation § 5.5), and the newcomer
could be named a monokini. It is perfectly irrelevant that bi- is originally Latin
and mono- Greek. This is how metaphors and other analogical changes combine
and pile up, when language adapts to the needs of communication. After the
passage of a few thousand years many such steps will necessarily always remain
beyond our grasp. Current history is being recorded in more detail than in the
past, so future etymologists will have it easier on this score.
(7.15 Semantic Change and Speech Production] We saw that both sound
change and analogy were intimately connected with speech synthesis. Change is
SEMANTIC CHANGE 153
a function of language use. In phonology, speaking means variation, which be-
comes socially interpreted, and, in social interplay, change results. Analogy is
based on the mechanism of the very rules of grammar. Change results when the
rules win over tradition (symbolic aspects). Semantic change has shown all this
in a much clearer profile, because the role of society and history is unmistakable.
The phenomena of culture, in general, elicit various responses to nomination,
for example, metaphor, metonymy, or other figures of speech, and, as a result,
synchronic variation increases. This variation is the basis of semantic change,
when fading in the marked values takes place. Fading is, of course, loss of
stylistic marking, style to be understood in its widest sense and not as literary
style only. Change is always a result of variation and speech production. A
language that is not used does not change.
Semantic change can be as abrupt mentally as other changes. The cultural
environment, of course, shows smooth gradience between situations and objects
(referents), but the semantic reinterpretation can be abrupt. It would be difficult
to imagine a gradual change from' prayer' to' small ball'. There was a contiguity
in the physical situation only. Characteristically, semantic change is gradual
for the speaker or the innovator, for example, in the formation of metaphors.
The shifts can be very small indeed. But the hearer may drastically reinterpret
such metaphors, and, in any case, the diachronic correspondences appear to
witness abrupt changes (see § 6.22). This exemplifies the two mechanisms of
change that a natural human language can undergo: evolution (more frequent)
and mutation. In contrast, artificial languages are subject to mutation only.
REFERENCES
General: Breal 1964, G. Stern 1931, Ogden and Richards 1923, Thorndike
1947, Ullmann 1959, 1962, S0rensen 1967, Leumann 1927, Sperber 1930,
Meisinger 1932, Ohman 1951, Kronasser 1952, Graur (ed.) 2.343-714; 7.2
Coseriu 1964, Hjelmslev 1963, Wiegand 1970; 7.3 Breal 1964, Shands 1970;
7.4 G. Stern 1931; 7.6 Tauli 1968; 7.8 G. Stern 1931, Sohngen 1962, Todorov
1967, Lausberg 1967, Lanham 1969, B. Campbell 1969, Reddy 1969; 7.9
Menner 1936, 1945; 7.10 Oertel 1901; 7.13 Tauli 1956, Ullmann 1962, E.
Itkonen 1966, Szemerenyi 1968, Stein 1970.
CHAPTER 8
[8.4 The Retailoring of Loans] Loanwords may be taken over with the
foreign morphemes unchanged, as in the above examples. Sometimes, part of
the word, especially if it is a compound or derivative or interpreted as such,
is substituted for by a corresponding morpheme from the borrowing language.
An often quoted example is the Pennsylvania Dutch substitution of the native
-ig for English -y, for example, bassig 'bossy ', fonnig 'funny', and tricksig
'tricky'. American Lithuanian, also, substitutes native suffixes, as in bossis
'boss ' ,Joniskas 'funny', and dotinas 'dirty'. This type can gradually grade into
a loanshift, for example, American Finnish lukkoruuma 'locker room', formally
a compound of lukko 'lock' and ruuma 'ship's hold' (cf. above in nonnautical
context). Both words are earlier Germanic loans in Finnish, but the new meaning
was triggered by the phonetic similarity to the English model (see § 7.7). The
scale and interrelationships between different types of borrowing can be repre-
sented roughly as in Figure 8-1. Sound substitution can range from zero to
such a degree that the original model is no longer discernible to the uninitiated.
English loans in Japanese are usually standard examples (bus--+ basu, taxi--+
takushii, and baseball-+ beisuboru), but similar situations obtain elsewhere,
for example, English loans in Hawaiian: laiki 'rice', palaki 'brush', and Me/e
Kalikimaka 'Merry Christmas'. Considerable sound substitution has occurred
in the following American Finnish loans from English: runnata rilisteettiii 'to
EXTERNAL CHANGE: BORROWING 157
run real estate', taippari 'diaper', and karpitsi 'garbage'. The amount of
sound substitution depends on the level of bilingualism (which is a prerequisite
for borrowing). In a prebilingual period substitution is heavy, but with succeed-
ing generations (as in the American immigrant situation) it becomes less and
less, and ultimately one switches to English altogether. Pronunciation borrowing
like /iy(y~r/ for fayo~r/ and /drens/ for /dans/ can be spotted when it is dependent
on particular words, but when a British English speaker switches completely
to the American pronunciation, he has borrowed the whole dialect (discarding
cases like lift-elevator, and so on). But even here there may be sound substitu-
tion; that is, the imitation is not perfect.
In some languages, phonetic substitution must be accompanied by inflectional
adjustment and gender alignment. Words ending in a consonant are given a
final vowel -i in Modern Finnish, as in posti 'mail', tulli 'customs', and jeeppi
'jeep'. In the last item, the long pp gradates in the inflection: the plural, for
example, is jeepit. An English plural may give the stem, for example, pointsi
'point', keksi 'cracker' (cakes), and American Finnish kukiiksia (partitive
plural) 'cookies'. Here the final /z/ of cookies was interpreted as /s/, which is
normal, and this further as an alternant of the -ks- stem, as in teos/teoksen (see
§ 6.14). Other immigrant loans show the same phenomenon, for example,
American Norwegian kars-er ' cars ' and American Italian pinozz-i 'peanuts'.
In these two, however, a native plural just overlays an English plural (compare
kine, and so on, §§ 5.4, 6.20, 6.21), whereas in kukiiksia there is a further stem
adaptation. As an extreme example, the Finnish word tituleerata 'to address
with a title' contains the original French infinitive ending as -eer-, then the
Swedish infinitive in -a-, and finally the Finnish infinitive in -ta. Such stratigraphy
tells part of the route the word has traveled.
These alignments are often impossible to predict though the language may
have set morphological classes where loanwords are accommodated. German
must assign one of its three genders to any loanword. Often the gender of the
lending language is retained, as in das Drama (Greek), die Mensa 'student
restaurant' (Latin), and die Chaise longue (French), but also das Chaiselongue.
Some other times, the gender of a native equivalent seems to prevail, as in
das Baby from English (das Kind), das Sofa from French (das Bett), and der
Smog from English (der Rauch 'smoke', der Nebel 'fog', der Dreck 'dirt'). But
why die Sauna from Finnish (against das Bad[ehaus]) and die Couch from English
(against das Bett, das Sofa)? In short, every alignment cannot be reasoned away,
and, anyway, such reasoning starts after the fact.
A striking example of substitution and adaptation is shown by Spanish loans
in Chiricahua (Apache), for example,jab6n ~ lu'ry6n 'soap', rico~ zi·go 'rich',
and loco ~ fo·go 'crazy'. Assignment of tones is obligatory in addition to the
various sound substitutions. Further, the distribution of sounds in Chiricahua is
changed, as i and I occurred only in medial and final position in native vocabu-
lary. Moreover, there are no adjectives of the Spanish type in Chiricahua, and
rico and loco were interpreted as third person verbs and equipped with paradigms
158 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
according to the native pattern of prefix and stem, for example (with partial
paradigms):
Similar things happen in other languages as well (see reanalysis § 5.5); for ex-
ample, English film fits into the Semitic triconsonantal pattern quite well, and
gets an Arabic plural 'aj/iim.
[8.5] Native speakers are aware of the distinctive features of their phonol-
ogy. Thus English doggerel and Mother Goose rhymes reflect the psychological
reality of componential features in phonology, witness nasality in a stitch in
time saves nine, one for my dame, and one for the little boy who lives in the lane,
or voiceless stops in Little Tommy Tucker sings for his supper (next : ... butter),
If the ocean was whisky and I was a duck, I'd go to the bottom and never come
up, and he catches fishes in other men's ditches, and so on. In sound substitution,
the borrowers apparently make a kind of distinctive feature analysis of the
foreign sounds and assign them to the closest native bundle. In oldest Germanic
loanwords, an f- is reflected by Baltic Finnic p- (§ 8.6), as [voiceless, labial].
There were no voiced stops in Finnic at that time. Later f is rendered by -hv-,
that is, [voiceless] = h and [labial, spirant] = v. Splitting a bundle of features
into two segments like that is not infrequent. French [i.i] was replaced by [u]
in English, but [u] gave a decomposite [iu] or [yii]. Similarly, Russian borrowings
render Baltic Finnic [ii] by [u], or more often with palatalization + [u], sysmii .-
s'uz'om [or s-] 'thicket'. And Russian [w] has a similar fate in Baltic Finnic, as
in mylo .- Karelian mujla 'soap'.
Attempts at using borrowing as a window to the psychological reality of
abstract phonology have not yielded valid universals. The above shows, however,
that in particular cases it may yield useful information. As other changes can
support abstract mechanisms, borrowing can be expected to do the same .
A B
Savoyard
English French Finnish Russian
animal pere hevonen losad'
vs.
In this way, borrowing knocks the feet from under the metaphor, and "immedi-
ate" fading is the result. Pronunciation borrowing, too, can affect primary
function only (e.g., the Lithuanian word for 'snow' is expected to be *snaigas).
Extensive Slavic contacts led to a replacement of the native ai with the Slavic
'e = ie (OCS snegu > Russian sn'eg, Polish snieg) yielding sniegas. In some
derivatives, however, the old ai remained: snaig(u)te 'snow flake ', snaigyti 'to
snow a little', and [snaiga 'snow hanging from trees'. Of course, the derivatives
based on sniegas have ie.
Borrowing is one of the main factors behind changes in lexicons. In English,
its effects are enormous. Borrowing has contributed to the many stylistic levels
and it has led to a considerable loss of motivated compounds and derivatives
(as we shall see). Metaphors and metonyms strive for novelty. With time they
fade, necessitating new metaphors, and so on. Loans are not very different; :
they carry considerable stylistic loading but are subject to fading like other
mechanisms, thereby increasing synchronic variation. And the stock can be
replenished by new loans, and so on.
life in general. Especially well represented are all kinds of terms connected with
trade items. In all these situations we know from history that the loanwords
reflect what went on quite accurately.
[8.10] Thus grouping of loanwords into semantic spheres can give valuable
support for historical inferences even when no other documentation is available.
(Of course cross-language comparison must first establish which words are
loans.) Germanic loans in Finnish cluster into roughly the same areas as the
French ones in English:
government and social order: airut 'messenger', hal!ita 'to govern',
joulu 'Yule', kihla 'security', kuningas 'king', kunnia 'honor', laina
'loan', murha 'murder', rikas 'rich', ruhtinas 'prince', sakko 'fine',
tuomita 'to judge', kartano 'estate', raha 'money', lunnas 'ransom',
t•akoilla 'to spy', vuokra 'rent', and so on.
religion: peijaiset 'funeral banquet', siunata 'bless', taika 'magic',
hurskas 'pious', and so on.
tools and skills: kaira 'auger', naula 'nail', mitta 'measure', saha
'saw', lukko 'lock', and so on.
housing and housekeeping: !a to 'barn', lattia 'floor', leipi:i 'bread',
saippua 'soap', patja 'mattress', kattila 'kettle', tupa 'living room',
and so on.
A justified inference from all this is that the Baltic Finnic speakers apparently
absorbed a Germanic-speaking upper class, although, of course, borrowing
can take place without absorption of peoples. At least the other available
evidence, mainly archaeological, does not speak against this hypothesis but
rather supports it, in that there seem to have been Germanic trading posts in
the Baltic. In this way, assessment of loanwords is an important tool for anthro-
pology and history, as such words record cultural contacts.
[8.11] One can also derive certain indications about the geographical
position of a language family in relation to other families by plotting correspond-
ing borrowings. Thus, for Finno-Ugric, we have the situation given in Figure 8-4.
When this type of a table is used in connection with the family tree (§ 15.1,
Figure 15-1), many inferences can be made. The importance for considering
the subgrouping and the particular phonological shape of the words comes
out in the first two rows. All the languages have loans from Old Iranian and
Slavic, but they must be assigned to different depths in the tree. Old Iranian
loans go into the protolanguage, whereas Slavic loans generally enter each
individual language separately. The splits in the tree are clearly reflected in the
table; for example, Hungarian does not have early Germanic loans and Baltic
Finnic does not have Turkic ones. Both contacts were obviously made in
different locations after the split. Sometimes the exact passage is not clear,
164 HIST O RICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
l=i
Loans
I In ..d
·a"'d
d
·;;: "'
"§
<!) d
~
(II
~
.;::!
....
(II
~
0. "C
.... .... -~ .,_.
>.
(II
.,_.
>.
bJ)
From 0.
(II 0
Q)
..d
....
>. 0
d
;:l
ii: .....:l ~ u N > 0"' :::c:
Slavic in general + + + + + + + +
Old Iranian + + + + + + + +
Proto-Baltic + (+) ? - - - - -
Proto-Germanic + (+) ? - - - - -
Old Norse + + - - - - - -
Scandinavian in general + + - - - - - -
Old Chuvash - - - - - - - +
Chuvash - - - + (+) + - -
Volga and Irtych Turkic - - + + - + + -
Other Turkic - - - - - - - +
FIGURE 8-4. The sources of loans for the Finno-Ugric languages. [Reprinted
from Robert Austerlitz," L'Ouralien," Le /angage, Encyclopedie de la Ph!iade,
25 (Paris, 1968 © Editions Gallimard).]
e.g., there is doubt whether the Baltic and Germanic loans entered Lapp directly
or came through Baltic Finnic (the latter is more likely, perhaps).
Sometimes one of the corresponding classical words is not cognate with the
whole set, or the English word is not, for example,
but it still takes part in the configurations and adds to the semantic fields of
English. When large-scale borrowing occurs between related languages, we
get regular sound correspondences between the items; note t- d- d (TOOTH, TEN,
FOOT, HUNDRED), th-t-t (TOOTH, THREE, FATHER, MOTHER), f-(p) - (p) (FOOT,
FATHER, FEATHER, FIRE, FISH), k - (g)-g (KNOW, QUEEN, YOKE, but see KIN with
affrication before e), b-f-(J) (BE, BROTHER), m-m-m (MOTHER), n-n-n (KIN),
r-r-r (FATHER, MOTHER), and so on (this of course clearly exemplifies Grimm's
law; §§ 4.9, 4. 10, 6.3). Such correspondences are curious in that, formally, they
are like morphophonemic alternation, but, psychologically, they need not be
connected at all, because there is no rule for the alternations. No clear environ-
ment for them can be specified (just try to write rules even for a simple case like
E-gyptfCopt-icJGyps-y!). Similar correspondences obtain in the Romance
languages through loans from Latin, for example, Spanish leche 'milk' vs.
lactar 'lactate ' , and noche 'night' vs. nocturno 'nocturnal' (ts-kt), or French
croire 'believe' vs. credible and loi 'law' vs. legal (wa-e +voiced stop). Such
situations originate easily when a language borrows from a literary norm of its
own past, for example, Old Church Slavic loans in Russian and Sanskrit loans
in Hindi. Scandinavian loans in English, Low German loans in Swedish and
High German, and Iranian loans in Armenian give the same configuration.
166 HISTORICAL LINGUISTICS: HOW DOES LANGUAGE CHANGE?
jalka 'leg, foot'), pize 'nest' (pesii), kandoms 'to carry' (kantaa), kargo 'crane'
(kurki). Russian loans have introduced initial voiced sounds, and these have
spread even beyond the original model, for example, hypercorrect bauk 'spider'
(Russian pauk). Often such rare foreign sounds have considerable descriptive
force, and they are preferred in onomatopoeic, descriptive, or suggestive forms,
or in intense forms and nonsense forms, whatever we call them. In Mordvin
the voiced sound has spread to native guffzems 'to growl' (kurnia) and guj
'snake' (kyy 'viper'), enhancing their emotional color. Similar spreads of rare
sounds that had entered originally through borrowing occur in all Baltic Finnic
and Lapp languages. The same phenomenon is well attested elsewhere also;
the Berlin dialect of German spreads its z, originally confined to loanwords
like genieren [z-] 'to embarrass, bother', to descriptive vocabulary, e.g., kuzeln
'to cuddle' and wuzelig 'unkempt'. Other languages could be cited as well;
note in this connection the sm- "reduplication" in American English, for
example, doctor smoctor and a few other words with s(m)-, schmalz, schlemiel,
mainly from Yiddish.
Earlier it was shown that loans helped to establish English /v, z/, and the
same is true of /dz, zf (§ 4.2). Foreign sounds thus typically come in as parts of
whole words, even if the words are imported precisely because their sounds
are felt to be expressive. This is true of bound morphemes as well. The hosts
of Latin/French words in -able, -ism, -ize, -tive, and so on, were the basis for
expanding these into new creations, for example, eatable, tokenism, macadamize,
and talkative.
It is customary to say that English has borrowed a great many Latin/Romance
phonological rules, as the essence of the Latin stress rule and the Romance
softening of k > s in certain environments (opaque/opacity). Such a statement
sounds as if the rules were borrowed without the actual items on which they
operated. In reality, the rules are a result of the importation of the units, the
individual words. Sound change had produced the French morphophonemic
alternations in clearly definable environments (§§ 4.4, 4.15). When both items,
such as opaque and opacity, had been borrowed into English, a synchronic
rule could be formulated. Because it had to cover the" same" facts in the" same"
environments, it turned out very much like the original French rule. Such rules
may well be different for different speakers, and it may take quite a while
before a child is exposed to all the necessary facts. Similarly, English has in
many cases borrowed both singular and plural formations of the Classical
languages, for example, phenomenon/phenomena and alumnus/alumni. The level
of bilingualism made this possible, and, also, English had the categories of
singular and plural to begin with. The pluralization rules were not borrowed
but rather the actual forms. The rules, whatever they are here, were just a by-
product.
The borrowing of intonation, which is frequent in convergence areas(§ 8.18),
would seem to come closer to straight rule borrowing if it carries a meaning
like 'question'. But note that intonational contours can as well be regarded as
"morphemes," and hence it is not primary rule borrowing. In American Finnish,
EXTERNAL CHANGE: BORROWING
the question intonation has been borrowed from English, and it sticks quickly,
even to Finns who speak normal Finnish. Finnish yes-no questions are marked
by the particle -ko with no characteristic intonation; this can be replaced by a
suprasegmental unit __j •
English loan, and then shifts into Russian. The French part is totally French,
so is the Russian, Russian. This is no mixed language, but two languages used
jointly. Similar sentences can be found in America and Russia among the
linguistic minorities, and, of course, elsewhere. On the other hand, sentences
like American Finnish Menin norttiin muuseja hunttaamaan 'I went north to
hunt moose' or Lude D'uod'i pivad, d'uod'i vinad, pajatet't'i, p/assit't'i 'One
drank beer, one drank liquor, one sang, one danced' are hardly mixed. In the
first, all but one word are loans from English, but the grammar is clearly
Finnish, and the same is true of the Lude sentence: the grammar is all Lude,
although, in vocabulary, only d'uodi is, the rest coming from Russian. The Gypsy
dialects are also often referred to as mixed, but, here also, much of the
vocabulary and phonetics is borrowed from the "host language." Consider the
following "pure" old English Gypsy sentence: Kom6va te jal adre mi Duve!esko
keri kana mer6va 'I'd like to come to my Lord's house when I die'. In modern
English Gypsy this goes I'd kom to jal adre mi Duvel's ker when mandi mers.
This is English in grammar except that it has an extra set of vocabulary, which
is characteristic of social jargons. Actually, classification by syntax or vocabu-
lary is an arbitrary decision; for example, Nissaya is considered basically
Burmese with Pali syntax and not Pali with Burmese vocabulary. Considerations
of continuity, and the social status or attitudes of the speakers, play a role in
such decisions.
The only certain conclusion that can be drawn is that all languages are mixed,
if mixing merely means borrowing. Some are, of course, more mixed than others.
Over half of English vocabulary is borrowed, which is also true of Swedish
(both about 75 per cent). But because Swedish has borrowed mainly from Low
German, it does not show so readily. Albanian, on the other hand, has borrowed
more than 90 per cent of its vocabulary, and even if Finnish has borrowed
only 20 per cent, it is still mixed in the same sense. The same mixing is true of
syntax, but there is no workable way of measuring that because of the universal
features prevalent in it.
In a puristic context, one speaks of "pure" and "hybrid" languages. Pure
language should avoid or even uproot loans from the outside, but encourage
them from the past of the same language (archaisms). Such conc~rn is school-
masterly and has value for genetic linguistics only insofar as it may provide a
sociolinguistic setting for certain innovations. The distinction between "pure"
and "hybrid" languages is untenable, of course.
[8.18] When adjoining and overlapping languages give and take, the
result can be what is called convergence (Sprachbund). In such convergence
areas, different languages may develop identical phonetics, similar phonological
systems, and very similar grammars, even if the lexical items and phonotactics
remained different. Famous convergence areas are, for example, India, the Bal-
kans, the Caucasus, and the Pacific Northwest. On the Marathi-Kannada
boundary one can find villages that have one basic grammar with two sets of
morphemes, one Marathi, the other Kannada. This leads to the startling position
EXTERNAL CHANGE: BORROWING 173
that it is the lexicon that does indeed determine genetic relationship, and not
grammar! The reason is, of course, that lexicon (including grammatical markers)
is heavily symbolic, whereas grammar is iconic and largely universal. Vocabulary
comes and goes, but when it is retained, it remains symbolic. Now, is modern
English Gypsy still Gypsy with English syntax?
Maybe there was a convergence area in the Baltic as well. We have seen a few
examples of far-reaching borrowing in Finnish from both Baltic and Germanic.
Although grammatical influence can at present only be guessed at, there seems
to have been a complete upheaval in the sound system. The two thousand or
so years preceding and following it have apparently been relatively stable in
comparison. Thus feeding Pre-Baltic Finnic into the convergence area and
taking it out, we get the configuration in Figure 8-5. Note that Germanic had
Verner's law alternation between/p x sand j3 o y z, which is a startling parallel
to the Baltic Finnic consonant gradation between p t k and j3 o y and s "' h
(viisas, (gen.) viisahan > Modern Finnish viisaan 'wise'). All the individual
phonological changes between Pre-Baltic Finnic and Late Proto-Baltic Finnic
are commonplace; but the cumulative effect makes the convergence hypothesis
impossible to ignore. If convergence areas can guide phonetic change like this,
it means that loss can be one side of borrowing. In this vein, it has been observed
that less prestigious languages avoid native forms that resemble obscene forms
in the prestige or upper language.
Such filtering does indeed seem to happen, and it is the reverse of adstratum
influence: one keeps one's language and borrows only the prestige phonetics.
It has been noted that Welsh has an" English" type of phonetics and phonology,
whereas Breton shows "French" features. Surely English and French, respec-
tively, are responsible for this.
[8.19] Convergence areas and adstrata are not the only situations referred
to as mixed languages in layman classification. Other candidates are various
trade jargons known as pidgins, which arise in superficial, often short-termed,
and limited cultural contact. They have sprung up since the European expansion
from 1500 onward, based on the languages of the maritime powers, that is,
English, French, Portuguese, and Spanish. A rarer phenomenon is a native-based
pidgin, as in the Chinook Jargon of the Northwest, Police Motu in Papua, and
Kituba in the Congo.
The belief that a pidgin is a mixture of the grammar of one language and the
vocabulary of another is oversimplified but very practical and fits quite well
the examples already mentioned. Pidgins are no more mixed than some other
languages, although they are drastically simplified in grammar and in the size
of their vocabularies. The aiding factors in the formation of a pidgin are the
very limited vocabulary, which is sufficient in such situations, and the univer-
sality of syntax. This vocabulary is largely directed toward universal human
experience (see Chapter 21), which makes syntax a side issue only. The concen-
tration on learning such key words can be verified in modern refugee or tourist
situations. Both sides try to let syntax take care of itself; for example, sentences
I74 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
Sprachbund filters
Pre-Baltic Finnic Baltic Germanic Late Proto-Baltic Finnic
p p p p(B--b)
t t t t(5--d)
k k k k(y--g)
b
d
g
6 B(b)
5' iS( d)
y y(g)
f
p
h(x) h
s s
s s(z) s s
s
z z
m m m m
n n(IJ) n(IJ) n(IJ)
n
IJ(Ij)?
1 1 1
I'
r r r r
v v v v
j j j j
c = ts c = ts
c=tS
c = t's
kt kt
ht ht
pt pt ft
mt mt
nt nt nt nt
In In
11 11
FIGURE 8-5. The filtering of the Baltic Finnic sound system through the
Baltic and Germanic ones. [Reprinted from Lauri Posti, "From Pre-Finnic
to late Proto-Finnic." Finnisch-Ugrische Forschungen, 31, 1-91 (1953).]
EXTERNAL CHANGE: BORROWING 175
such as me eat coconut (coconut me eat, and so on) or you catch train (train catch
you, and so on) cannot be misunderstood in a practical situation in which
philosophizing is incongruous. It is also normal that the pidgins developing
in slavery situations in the New World would resort a great deal to African
syntax, although it is often difficult to spot. This is a solution of least effort.
The masters would also imitate this, for various (scientifically false) reasons,
but the result is a system that does the work. Drastic simplification is charac-
teristic even in Kituba, even though the neighboring languages are quite similar
to its base, the dialect of Marianga. Pidgin formation implies elimination of
morphophonemic alternations and suppletion, reduction of grammatical classes,
and emphasis on single unaltered forms of lexical units. In short, practically all
inflection is stripped off; this mirrors an attempt to improve the efficiency of
the system, and it especially helps the person with a limited experience of the
dominant language. By concentrating on and hearing one form of any item,
both encoding and decoding are greatly simplified. One result is the "baby
talk" effect, which, incidentally, shows that speakers do, in fact, implicitly
know how to simplify. In other words, they are knowingly applying the principle
'one meaning, one form', and this is done so easily in these situations because
of the limited content the sign system has to handle. Note that here also, as in
similar "analogical" changes, there is a break in tradition. Against analogical
simplification is opposed the total community, which upholds linguistic con-
ventions; but, even so, analogical changes sneak in. In a pidgin situation,
however, there is no community or tradition, because the whole language is
largely a makeshift for economic profit (for at least one of the partners). In
more recent times it has become obvious that economic motives erode traditions
in other areas of culture as well. Note a certain parallelism to the lack of tradi-
tion in the Greek adoption of writing, which was very conducive to the rise of a
simple and efficient orthography. The social status of a pidgin is generally very
low indeed, below the less-valued dialects of the standard languages; and often
a gradual shift to one of the standard languages occurs. In areas where many
native languages are spoken, a pidgin may provide the most convenient lingua
franca, a common language for communication; in such a polyglot area Melane-
sian Pidgin English, for example, has remained vigorous. A pidgin is always a
lingua franca, but a lingua franca need not be a pidgin; for example, English
is a lingua franca in India, and, to an increasing degree, throughout the world.
The original lingua franca was pidginized Romance, mainly from the Riviera
area, and it served in the Mediterranean in various forms at various times during
and after the Crusades.
Artificial languages are pidgins coined in the calm of the study. The most
famous and successful one is Esperanto (contrived in 1887 by L. L. Zamenhof),
which can be classified as a kind of pidgin Romance or Neo-Latin, with loans
from other European languages; in other words, European Pidgin Romance.
A similar version of individually "pidginized" English is Basic English (con-
trived by Charles K. Ogden [1889- 1957]). The name is an acronym of British,
American, Scientific, International, and Commercial, and the "language" is
176 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
intended for international communication and as first steps into English, that
is, for typical pidgin purposes. It has not caught on very well (it is so modern
as to be copyrighted, which would seem to be inimical to the very idea of a
contact language).
We have seen that, generally, when words shorten, units multiply. English
and French have lost most of the Indo-European inflection, but this has been
compensated for by syntax (prepositional phrases, word order). The great
reduction of pidgin phonology, morphology, and vocabulary must likewise be
compensated for in syntax. Elaborate periphrasis must be used , that is, metaphors
and metonymy. Children are capable of the same, even with very limited vocab-
ulary; this is one side of the human capacity to analogize (Chapters 5 and 7).
Thus Melanesian Pidgin English has the metonymic (indexical) phrase woman
he brother belong me for 'my sister', he no got money for 'poor', and, for
example, metaphoric grass belong face 'beard ' and grass belong head 'hair'.
Basic English with its 850 words is more" wasteful" in that it includes the words
sister and poor, but not beard or hair. "Technical" terms like thyme must be
paraphrased as in goodly (note the regularity) smelling and tasting grass, smelling
grass tasting good, or the like. Pidginization is always directed toward greater
redundancy. Free pronouns rather than affixal forms result in longer phrases;
independent adverbial particles instead of tense morphemes, for example, do
the same. (Redundancy is another factor that facil itates communication and
that must be prominent in an efficient sign system.)
REFERENCES
General: Haugen 1950, Weinreich 1968, Deroy 1956, Martinet (ed.) 1968,
Petrovici 1969, Alatis (ed.) 1970; 8.2 Tabouret-Keller 1969; 8.3 Kalima 1915,
1936; 8.4 Aron 1930, Hoijer 1939, Yannay 1970; 8.5 Raun 1968, Maher 1969a,
1970b, Hyman 1970ab; 8.6 E. Itkonen 1966, Koivulehto 1970, Bernstejn 1961,
Halldorsson 1970; 8.8 Senn 1953, Patterson 1968; 8.9 Baugh 1951, Jespersen
1956, Hammerich 1954, H. Vogt 1954, Doerfer 1967, Kazazis 1972; 8.10
Hakulinen 1961; 8.11 Austerlitz 1968, Jacobsohn 1922; 8.13 Katicic 1966,
1970, Krahe 1970; 8.14 Ravila 1952, Specht 1952, Petrovici 1957, 1969; 8.15
Bach 1956, E. Itkonen 1966, Okell 1965, Burling 1970, Rohlfs 1922-1923;
8.16 Nielsen 1952, Szemerenyi 1964, Herman and Herman 1958; 8.17 Mueiier
(ed.) 1954, Fokos-Fuchs 1962, Havranek 1966, Petrovici 1969; 8.18 Becker
1948, Posti 1953, Emeneau 1956, C. F. Voegelin 1945, Weinreich 1958, Pisani
1966, Burling 1970, Martinet (ed.) 1968; 8.19 R. Hall 1966, Nida and Fehderau
1970, Welmers 1970, Ogden 1968; 8.20 Nida and Fehderau 1970, R. Hall 1958,
Martinet (ed.) 1968.
EXERCISES
2. Rewrite the same paragraphs in Basic English (for the rules, see, e.g.,
Webster's New World Dictionary of the American Language, College
Edition, p. 123, or Ogden 1968).
CHAPTER 9
(§ 8)
Borrowing
Sound change (§ 4)
Morphophonemic conditioning
(§ 6) Rule change
of sound change (§§ 4.27-30)
forms roll, give, and tell are used for the simple past. Homophony with the
present is avoided by the use of the auxiliary do, He does give •He gives' vs.
He gire •he gave'. The present has become the formally marked form, unless
one wants to derive the past by deleting does. Similar cases exist in many other
well-known languages.
The most famous case comes from Southwestern France in the history of the
Latin gallus •rooster' and cattus •cat'. In this area two sound changes worked
toward the merger of the two forms: (I) Latin -11- > -t, and (2) initial c- > g-.
Both •rooster' and •cat' should now end up as gat. But in a farming situation,
such a homophony is .. insidious" and detrimental to an efficient sign system.
And, indeed, one of the forms was replaced by another, namely, •rooster' by
bigey •village judge, deputy (vicar)', azii •pheasant', or put •young chick'. An
.. intolerable" homonymic clash resulted from the merger of Old English
lfttan • allow' and lettan •hinder'; one had to go, and it was the latter. It survives
only in the phrases without let or hindrance and a let ball (tennis term), which
has been folk-etymologized into net ball (though, in strict terminology, there
is still a difference between let and net balls). In a similar way, Old English cwen
•queen' and cwene ·wench' end up as /kwiyn/. Of these, queen has stood its
ground, whereas quean is strongly limited. It is also natural that the latter has
receded, as it has dozens of synonyms to replace it.
Even if the semantics are unrelated, homonymy is still avoided if it has
obscene overtones, because this is another impediment to communication.
Speakers react to embarrassment (either ridicule or annoyance) by avoiding
the homonyms to the tabu words. This has been observed in many languages, for
example, in American English, rooster and donkey are much more prevalent
than cock and ass. Finnish kuti • spawned' does not assibilate t > s as expected,
apparently because of kusi •pissed'.
Borrowing can be the mechanism to correct polysemy. In Finnish, kutsua
means •to invite' and •to call'. In the dialect spoken in the Province of Varmland,
Sweden, kuhtua is retained in the first meaning, and kal/oa has been borrowed
from Swedish for the second.
A B
1
'big pile of hay'
sheaf' •
Maastricht
'grain'
In the tabu cases, we saw that uncalled-for connotations and reactions may
lead to the elimination of a term, even when there is no danger of ambiguity.
A similar case can happen also in polysemic clashes. Again, in Belgium, in the
Brabant area, liiufig (Standard German form) means 'in heat' (of cows),
whereas in the East (Limburg-Rhine) it is applied to bitches. The two areas are
separated by a no-man's-land. The reason is not cognitive, but apparently
avoidance of ridicule (compare how funny it sounds when a foreigner says
184 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
[9.6 Holes in Patterns] From this internal cause we can move to another:
the symmetry or asymmetry of phonological systems. The hypothesis is that
systems strive toward perfect symmetry, that is, gaps tend to be leveled out.
Language is normally full of gaps of all kinds, and not only phonological ones
(e.g., in derivation, all possible combinations are never used), and this skewedness
is, of course, a prime target of analogy. Similarly, all possible combinations of
distinctive features need not be used in one language. If in the middle of an
otherwise perfect pattern an "expected" unit is missing, one speaks of a hole
in the pattern. In English, there is a gap in the combination of the forms good
and the comparative -er. When the "expected" combination is made the gap
is filled by goode;·(§ 5.14).
WHY DOES LANGUAGE CHANGE? 185
The vowel system of Middle High German was the following, after the umlaut
and before the lengthening of short vowels in open syllables:
i ii u i u fl
e 0 0 e 0 6
e ( )
ii a a ii
Here we have a twofold asymmetry, one between short front and back vowels
and one between short and long vowels. Balance can be achieved by eliminating
one front vowel or creating a new back vowel. In certain parts of Switzerland
e has indeed been eliminated by merging with e, in other localities by merger
with ii, with the result that the long and short systems correspond exactly. In
areas where four unrounded front vowels remained, there was a split:
~~~ · u~u
0~0 0~0
5 ;)
and the hole was filled. In the lengthening of short vowels, a is integrated in
three ways (in different areas):
fl
fl
6 /~(new)
a~ii a a
The last alternative obtains only in those areas where the short vowels have
four degrees of height.
A striking development toward symmetry is the first-syllable vocalism of
Proto-Baltic Finnic, in which the short vowels had one degree of height more
than the long ones, and contained the only rounded front vowel in the whole
system:
Finnish has filled every single gap and ended up with perfect symmetry. Another
asymmetry is known in the Serbo-Croatian system of velars:
g ( )
k X
186 HISTORICAL LINGUISTICS: HOW DOES LANGUAGE CHANGE?
[9.7 Articulatory Balance] Chain shifts would further show one aspect
of the principle of maximal differentiation. There seems to be a universal tendency
for phonological space, as defined by the articulatory possibilities (Figures 1-3
and 1-4), to be divided evenly among the units so that each has maximal elbow
room. Languages with one s-sound show [s]- or [s]-type phonetics for their
Js/ (e.g., conservative Finnish); languages with three vowels would normally
display i, a, and u (see§ 6.5); and so on. Further, a universal tendency in vowel
shifts is that tense vowels rise and lax ones get lowered. Hence push chain
shifts are in principle quite possible. Another regularity in the idea of maximal
differentiation is that languages with skewed vowel systems have more front
vowels than back vowels. This is a consequence of the shape of the mouth and
articulatory organs, which provide more room in the front area. Although this
factor is internal with respect to the human head, it should be classed as an
external one as far as the actual language is concerned. Then there is the problem
of "perceptual space," which may be relevant. Unfortunately, not much is
known about it as yet(§ 9.16).
Note how frequently linguists rely on the principle of articulatory balance
by using arguments based on holes in patterns. We saw above a case in which
German dil) was anomalous in its final IJ, because elsewhere it occurred before
velars only, dal)k. The hole could be filled by g, dil)g, which is then automatically
WHY DOES LANGUAGE CHANGE?
dropped by a phonetic realization rule (§ 6.15). Here the linguist tries to fill the
hole that sound change is trying to create (or has created, depending on one's
vantage point; see§ 19.6).
the nongrammatic area we have already seen forms like hussy < hiiswif, lord <
hliiford, lady < hliifdige; others could be added, for example, head< heafod.
(Lord and lady fall basically into the area of titles.) From the sailor's vocabulary,
one could name similar forms, for example, /bows'dn/ bosun ""' boatswain and
/st'dns'dl/ studdingsail. Developments like Eboracum > York occurred all over
Europe, where we have old records to prove the cases.
Another aspect of frequency in linguistic theory is the claim that sounds that
distinguish very few forms from others are more likely to disappear than those
z
with a higher functional yield. In English, the contrasts =1= is rather peripheral,
and its elimination (say, through merger into s, or z into dz would not disturb
many signs, whereas it would be quite a different matter with s =f z, or with any
of the other consonants, especially stops. The pairs k =f g, t =f d, and p =f b
differentiate among a host of words in English. In a way, a low functional load
is the same as a hole in the pattern, that is, a hole in the distribution of sounds
in the lexicon or morphology. The inbalance can be eliminated by increasing
the load of the precarious sound (e.g., by borrowing) or eliminating it. Here
again, scholars do not agree on the extent of the phenomenon, or even on its
existence.
There is one further area where frequency plays a role in change, namely,
resistance to pronunciation borrowing. Words with high local frequency tend
to be the last ones to be changed. Now if high local frequency acts as a barrier
to change from the outside, it is also supposed to initiate change from the inside.
The hypothesis goes that the most frequent sounds can get away with less precise
articulation. This variation goes one way, for example, toward less marking,
and sound change results. Thus one would say that the relative frequency of
the voiced stops d g in Indo-European launched Grimm's law. These were
marked with [+voice] in contrast top t k. As the most frequent stops they
tended to lose their marking and move toward t k, which in turn had to shift
out > (f) jJ x. We do know that variation is the basis of change, but the exact
role of frequency in this connection is unknown and unknowable. On the other
hand, speakers themselves seem to know statistics connected with speaking,
because in certain sociolinguistic situations the proportion of the cases in which
a rule applies is rather consistent considering all the possible cases in which it
could apply (Figure 3-1, §§ 3.3, 3.5, 9.3, 9.11).
without variation, one could not understand change, and without change, one
would not understand synchronic variation. Sound change shows the social
factors of change, in addition to the phonetic/physiological (determined by
the articulatory organs), phonological (determined by the pressures of the
system), and psychological factors (e.g., tabu).
The starting point of sound change is index formation. The articulatory basis
of pronunciation always ensures a certain amount of random variation. Culture
allows for the same, for example, in the area of clothing. Striking random
fluctuation (in pronunciation) has been quoted for example, from Papuan,
where the velar in voka 'coffee' varies (or varied) between kh, kx, g, and y,
or the French of the Ardennes, where l'eau 'water' can (or could) be [lot:J],
[lou], or [lao]. When a particular variant is given social interpretation, it becomes
an index of that group. In the same way, animal sounds are indexes of the
corresponding animals, and this index is often reproduced iconically in language
in naming the animal (onomatopoeia). Similarly in culture, variation can become
a social index, as in a particular clothing fashion. If a group which shows
social cohesion through pronunciation indexicality has wider appeal, its pro-
nunciation will be imitated; that is, others try to produce the index as iconically
as they can. This represents the combination 'pronunciation pronunciation'
which was missing in §§ 2.14, 2.15 (see end of§ 21.4). Again, the same happens
in culture, say the imitation of fashions if it is not proscribed (e.g., the imitation
of royal insignia, and so on). If on the other hand a group is stigmatized for
its pronunciation by a more prestigious group, it may ultimately give up its
index for the prestige model. Such indexes need not be absolute but can be
statistical parameters; that is, one throws in the index at the proper frequency
(see Figure 3-1, §§ 3.3, 3.5). This easily results in hypercorrection by those who
try to belong to the prestige group. They overdo the index. A similar reaction
in nonlinguistic culture (and scientific schools) is typified by the aphorism:
"The Irish are more Catholic than the Pope." This is also why upstarts, or
social climbers, are often easily spotted.
Much of this index identification is subconscious, but at times it reaches the
level of awareness. Even linguistically naive speakers can brand pronunciations
as vulgar, soft, hard, coarse, and so on. (N.B.: To do this they thus use indexes/
metonyms and icons/metaphors!) Indeed, misinterpretation of the index can
lead to serious misunderstanding among different groups. Young people may
interpret the pronunciation of older people as pompous and authoritarian, and
older people that of the young as provocative and irresponsible. An American
can interpret the British intonation as patronizing and insulting. Practically
every "major" sound change since the seventeenth century was explicitly
discussed by French orthoepists. This shows the struggle of variants and their
social implications, and, of course, reveals that the whole situation could be
observed. The first occurrence of English labialization wa > WJ is attested from
1640. In 1766 Buchanan used J in some words (ward, warn, want) and <e or a
in others (wabble, wad, wallop). In 1780 Sheridan connected the pronunciation
of quality with J with the meaning 'people of high social rank', whereas, in its
WHY DOES LANGUAGE CHANGE? 191
abstract meaning, the old ;e remained. In 1633 there was apparently no difference
in the vowels between good and blood, both with u·. Later in the seventeenth
century, when short u > A, the long u· was in the process of being shortened
(attested already in 1569 for good). Now this variation u'"'"' u· interfered with
the change (variation) u >A. Those words where the long variant was more
frequent were not affected, except by the later shortening (good), but those in
which the short u was dominant underwent u > A (blood). Thus the apparent
double outcome depends on the overlap of two sound changes, which went on
as synchronic variation in the grammars. In 1747 Johnson sometimes had
difficulty in deciding the proper pronunciation for his rhyming dictionary, for
example, whether great rhymed with state or seat. The best speaker in the
House of Lords decided without hesitation for (st)ate = (gr)eat, and the best
speaker in the House of Commons likewise for (s)eat = (gr)eat. These English
examples have shown three facts: (1) variation and change can spread from
word to word and need not be simultaneous throughout the vocabulary, (2)
statistics/frequency does play a role, and (3) identification with a social layer
exists.
In French, many grammarians have recorded the alternation we '"'"' wa (loi
'law', and so on). The variant wa occurred in monosyllabic words, and in
polysyllabic words before rand 1. The upper classes condemned wa and regarded
we the only acceptable pronunciation. Toward the end of the eighteenth century,
both pronunciations were almost on equal footing. When Louis XVIII, who
had fled in 1791, came back in 1814, and uttered: C'est moe /e roe (i.e., we),
he was quickly told that in his absence this pronunciation had become vulgar
and provincial. In other words, the complete social upheaval of the French
revolution reversed the social value. This is often true when new social classes
come to political power.
(to [a]) and 3 (to [~]), with two steps in between. Centralization indexes by age
groups are as follows:*
jAij jAuj
over 75 0.25 0.22
61-75 0.35 0.37
46-60 0.62 0.44
31-45 0.81 0.88
14-30 0.37 0.46
In 1933 the centralization for jail was about 0.86, for jaul only 0.06. For jauj,
centralization has been steadily rising. There is a clear difference in the geo-
graphical distribution as well:
jAij jAuj
Down-island 0.35 0.33
Up-island 0.61 0.66
or occupation
Fishermen 1.00 0.79
Farmers 0.32 0.22
Others 0.41 0.57
Labov gives such indexes for ethnic groups also, but the most revealing is the
correlation of centra:lization with the attitude of the speaker toward the island.
The indexes for four 15-year-old high school students are: down-island and
leaving 0.00-0.40 and 0.00-0.00, up-island and staying 0.90-1.00 and 1.13-1.19.
With those people who feel very strongly about the island, the index can go
even beyond 2.00. In general the attitude test gives the following centralization
averages:
NO. OF PERSONS jAij jAuj
40 Positive 0.63 0.62
19 Neutral 0.32 0.42
6 Negative 0.09 0.08
* The tabular material in§ 9.11 reproduced by permission from William Labov, "The social
motivation for sound change," Word, 19, 273-309 (1963).
WHY DOES LANGUAGE CHANGE? 193
[9.12 Change and the Social Setting] The social setting shows a common
mechanism of change. A feature is adopted by a group as a social index. Now,
understanding why a particular feature should be involved at a particular time
is beyond us, exactly like predicting fashions. If a group becomes a reference
group for some other, the latter will accept the index and exaggerate it. Hyper-
correction further spreads the feature, perhaps in combination with structural
symmetry. This establishes a new norm when the social situation stabilizes,
which again can serve as a model for other groups. Equally important is the
acknowledgment and realization of the heterogenousness of the transmission of
a change. It may be individual or occasional and it may first affect women or
young people, sailors or farmers, individual words or word classes (e.g., verbs
or certain semantic fields), and so on (see § 6.8). All this makes it possible for
many irregular forms to come into being "regularly," for example, the pronun-
ciation of good vs. blood. That is, grammatically irregular forms have a regular
explanation in terms of the relations in the speech community. Linguistic change
cannot be understood without data from the speech community, because lan-
guage exists for the community, is maintained by it, and refers to the culture
of the community. Sublinguistic (e.g., allophonic or individual) fluctuation is
given noncognitive social meaning, and this results in sound change. When one
variant wins out at the end, ultimate regularity is produced, if one excepts
certain irregularities here and there.
The particular cultural setting may involve a long history of writing. This
creates the additional possibilities of spelling pronunciation and archaisms,
which are loans from earlier stages of the same language. Semantic change, in
particular, emphasized that linguistic structure alone is not sufficient for the
unfolding of linguistic change. The causality of change resides in a complicated
texture of social, physiological, psychological, phonological (and other systemic)
factors. It is clearly wrong to seek only one factor which would explain every-
thing. One must acknowledge the psychological factor to be the strongest one-
that is, the general tendency toward simplicity and symmetry. It is also clear
that this chapter has been a general summary of the preceding ones, and that
we have continued to discuss more the 'how' than the 'why' of change. Nonethe-
less, the factors that have been delineated are all we have at the moment, and,
on the whole, it seems plausible that such factors combine to give causes of
change.
[9.14 Sound Change and Language Learning] It has been mentioned that
language is one manifestation of an innate ability to analogize(§ 7.8), and that
children, in particular, introduce analogical creations (§§ 3.3, 5.2, 5.4, 5.14).
The contribution by children to linguistic change may be quite considerable;
it is difficult to know its exact scope, however, because tradition usually prevails.
Linguists stated early that one of the chief factors of sound change is 'imperfect
learning' by children. The reason for this might well be biological/psychological;
that is, there may be an innate natural system of phonological processes,
manifested particularly in the postbabbling period. This innate system is gradu-
ally modified by linguistic experience so that the child comes closer and closer
to the standard; if he fails to any degree, change results. An innate system would
also explain the implicational laws or scales presented in § 6.5.
Such a 'natural phonology' (Stampe) is the first language of the child, a
kind of "innate speech defect," which automatically produces a substratum
effect in the acquisition of the pronunciation of the community. The child masters
the underlying phonological units earlier than he can produce all the contrasts.
Surface contrasts are eliminated through the application of the processes of
the innate system. Thus there is a "biological" pull toward the language-
innocent state of affairs and a social pull toward the language of the community.
WHY DOES LANGUAGE CHANGE? 195
Many linguists believe that the former is the more important factor and often
state it in the slogan: "Children simplify (restructure) grammars, adults com-
plicate them." The way adults "borrow" phonetic change (§§ 9.10, 9.11),
however, is no different in mechanism from language acquisition. In both cases
there is regular sound substitution (see § 8.4; Figure 8-1), although the inter-
ference comes from different sources. The regularity of substitution, of course,
results in the regularity of sound change. Such an innate sound system is also
compatible with the fact that sometimes change is very radical indeed.
When the child fails to suppress some innate process that does not apply in
the standard language, phonetic change occurs which looks like an addition
to the standard grammar. One of the natural rules appears to be that word-final
obstruents are devoiced, and English speakers must unlearn it (in contrast with
German speakers, for example). If they fail in this, the language gets an "addi-
tion," for example, d > t: [b~:·d] > [b~:·t] 'bed' (which, in fact, has become
standard in some Appalachian dialects). The word still remains distinct from
[b~:t] bet, because the lengthening of the vowel before voiced stops carries the
contrast jbe·tf vs. /bet/. The lengthening had been copied correctly (see§ 10.4),
which automatically "orders" it before the devoicing in the synchronic applica-
tion of rules. The innate application is unordered (strives toward perfect un-
marking), and thus synchronic order would result from the order in which
particular distinctions are copied from the standard language. (This is parallel
to the possible different chronology in dialect borrowing;§ 14.8.) An unordered
application of vowel lengthening and devoicing would produce [bat] for both
'back' and 'bad', but when vowel lengthening is applied first we get [bat]
'back' and [ba·t] 'bad', as above. Here again we see that our descriptive
mechanisms of change (or rather diachronic correspondences, § 6.22) look
totally different from the point of view of language acquisition; that is, they are
not historical explanations. The apparent addition, generalization, and unorder-
ing of processes arise in the child's failure to suppress, limit, or order processes
of the innate system, to the extent required by the standard language.
Such a natural phonology is the exact opposite of the empirist tabula rasa.
In spite of the "neatness" and the dramatic appeal of the hypothesis, it is highly
controversial; it serves, however, as a modern example of bold reasoning into
the causes of sound change. (We have not discussed those earlier attempts now
proved unfruitful-climate, geography, and so on.)
[9.15 Rules, Sound Change, and !conicity] Because all the different
mechanisms of change are so heavily diagrammatic (analogical), and because
change is, in general, connected with iconicity and indexicality, the traditional
interpretation of sound change as something very different would, if true, be
quite noteworthy. On the contrary, it is notable that sound change is indexical,
and iconic (especially in its spread). Since rules are largely iconic, and since
phonological rules handle the commutations, associations, and distributions of
distinctive features, they are iconic as well. They represent the relations in the
hierarchy of distinctive features and are thus diagrammatic (see §§ 1.13-1.16).
196 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
This inference is deductive. Now, induction is inference with the order of the
procedure reversed: we infer the rule from the case and the result. But the most
common type of reasoning is hypothetical inference, in which the rule and the
result are given and we infer the case. This is abduction, the everyday logic par
WHY DOES LANGUAGE CHANGE? 197
Universals
r------------------,
:I Grammar 1 II Grammar2
L--------- ----------~
Output2
Grammar 2 is inferred from Universals (the major premise) and Output 1 (the
minor premise) by abduction; if Grammar 2 is different from Grammar 1, we
may speak of abductive change. Output 2 is inferred or derived from Universals
and Grammar 2; if Output 2 is different from Output 1, we may speak of
deductive change. Note that deduction is always an experiment. In language
the test is whether Output 2 is acceptable to speakers who produce Output 1.
If it is, the general rule has been verified: the two grammars are the same for
practical needs, although they may be drastically different in structure.
Although the exact nature of the language universals is still unknown, their
existence is a certainty, and they are connected with man's innate capacity to
learn a language. We have at least seen some of the cornerstones of the uni-
versals, scattered throughout the preceding chapters. The most important one
is' one meaning, one form' (e.g.§ 9.3), which need not be repeated here. We have
seen various assimilation phenomena and other phonetic processes (Chapter 4;
see also§ 16.8) and marking conventions which somehow represent universals
(§§ 6.20, 6.21, 9.8, 9.9). Then, there is a universal phonetics (§§ 6.5, 9.14) and
articulatory balance between units(§ 9.7).
Many linguists have posited a particular 'language acquisition device'
(LAD) that would handle language learning alone. But there is no need for a
separate learning device or autonomous mechanism like that. Language ac-
quisition is a process of socialization and not verbalization alone. Language is
just one facet of the human capacity for analogizing (§§ 7.8, 9.14), in other
words, the human capacity for abduction. Significantly, diagrams for language
learning devices (to process the speech signals) tend to have compartments
corresponding to abduction, deduction, and induction. Perceptual differentiation
comes earliest and this leads to a discovery of perceptual values (abduction).
Then these values are given an articulatory optimization to find and produce the
favored contrasts; this is discovery of articulatory values ("creative behavior,"
deduction), aided by a device that handles articulatory balance (§§ 6.4, 9.3,
9.6, 9.7, 9.9). The principle of perceptual differentiation maximizes the degree
of perceptual contrast, and the principle of least effort minimizes articulatory
expenditure. The balance between these two factors keeps the phonology natural,
"easy-to-hear" and" easy-to-say." Finally there is interaction with the environ-
ment and adult speech in general that leads to normalization (induction).
This is quite correct, except that we must allow for the learning of culture at the
same time. Our diagram in fact does this; it is a general acquisition and learning
diagram. Note that the order of learning phonology is hierarchically ordered in
that first come sentence units, then words, unanalyzable syllables, and last,
distinctive features and segments. This exhibits a striking parallelism to the
development of the alphabet(§§ 2.7-2.11), or even language(§§ 1.27, 1.28).
Chapter 4 treated the mapping of sounds into other sounds, for example,
t > e, which looks as if Output 1 has been directly changed into Output 2.
But this was just a convenience of description. Similarly, the diachronic corre-
spondence rules of Chapter 6 took Grammar 1 directly to Grammar 2. Historical
change actually must go through abduction; this was more clearly seen in our
discussion of morphology and semantics (e.g., §§ 5.6, 5.16, 6.13-6.15, 6.19,
6.22, 6.23, 7.4), but the same is true of sounds also. Sounds cannot be shifted
directly on the articulatory scale (Chapter 4, and Figures 1-3, 1-4) in spite of
the convenience of such terminology. The child cannot learn all the articulatory
facts-many are simply not visible-but he has to abduce the sounds from his
perception. Here the child is well equipped since he can distinguish rather early
between features and things he cannot yet produce (primates and some other
animals have this ability as well). Articulatory space and ease do produce
random variation on which social forces feed (§ 9.11), but this happens through
abduction. Only in this way can we give a natural explanation for changes
like x > f (genoh > enough, hleahhan > laugh), f > x, e > f (§ 4.2 I); the child
reinterprets the acoustic signal. Thus, after the Pre-Greek *kw gave *kw > *kY
WHY DOES LANGUAGE CHANGE? 199
before i (§§ 6.16, 18.13), we had more or less the following acoustic scale with p,
t, and k: [t- kY(i) I k - kw(u) I kw(o) - p]. At some point somebody inferred
that [kY] could be taken as a variant of a dental, [kwu] as [ku], and [kwo] as
[po ]. This abduction was successful; we end up with p - t - k as indicated by
the vertical divisions above. Of course there is a fair amount of "articulatory
justification" in this, but as cases like x ;c f show, the auditory justification is
more powerful. This is why a reduction of an acoustic scale [t - p' I p] in a
Czech dialect into t - p is not so exotic after all (in the standard dialect p'
merged with p, which is more "normal" or natural). In general, near homoph-
ony can easily lead to a merger, also in terms of words. We have seen how the
Old English high vowel scale [i - y I u] was reduced to i - u by merging i and
y (§§ 4.5, 4.12, 4.13, 4.22). This is a reversal in the distinctive feature hierarchy.
As long as y was a clearly derived unit, rounding was the primary feature. At
some point frontness was inferred as basic, and this led to i quite naturally
through Universals (front vowels are characteristically unrounded). To take
another example, the voicing of intervocalic consonants is a natural assimilation
(articulatory ease, the speaker's "lazy tongue") that can easily catch on. Once
[voice] is interpreted as the primary feature, voiceless variants are likely to
disappear (compare§ 6.13).
Reversal of basic and derived features is particularly clear in the fading of
metaphors and so on(§§ 7.9, 7.15). The cultural situation allows for the learning
of base and derived forms separately or in the reverse order of the actual
history (irrigate [§ 5.5], counting one's beads [§§ 7.4, 7.15]). We see again that
culture change is parallel to linguistic change: there too boundaries shift
through abduction (§§ 7.3, 21.8). More generally, culture and human semiotic
systems show markedness reversal in marked contexts. As an example, consider
the distinction between formal and casual dress. In the everyday situation formal
wear is marked and casual clothes unmarked, but in the marked context of a
festive occasion, the values of formal and casual clothes are reversed (compare
other reversals in§ 21.13). Similarly, in the marked context of the underworld or
war killing can be unmarked activity. We have already seen how such facts
guide semantic change in terms of social jargons(§ 7.5). But the same phenom-
enon is operative in syntax also. In the marked subjunctive mood, the past vs.
present opposition (they knew vs. they know) is neutralized; the normally
marked past tense is used to the exclusion of the present (/ wish they knew).
The number opposition (they were vs. he was) is also neutralized here so that
the normally marked number is used to the exclusion of the unmarked number
(/wish he were).
Now we can profitably return to the German syllable-final devoicing. We
saw that devoicing was a sign of neutralization into the unmarked voiceless
member in Russian (§ 9.15). German manifests the features [tense] vs. [lax]
and hence the unmarked member should be [lax]. But note that this phenom-
enon occurs in syllable-final position. This position is marked in respect to
syllable-initial position. The order of learning syllable types goes (1) ev, (2)
eve, (3) ve, and (4) V. Voicing is learned earlier in syllable-initial position;
200 HISTORICAL LINGUISTICS: HOW DOES LANGUAGE CHANGE?
for example, when an English-speaking child can already produce [buk] 'book'
he goes on saying [pik] 'pig' (compare§ 9.14). We see now that in the German
case the usually marked member of the opposition takes up unmarked behavior
in the marked position of the syllable. Thus "devoicing" turns out to be
another misnomer which has to be taken as a traditional term.
Another universal factor in change is syllable structure. We have seen its
influence as a conditioning factor in terms of open and closed syllables or
various cluster rules. The assignment of syllable boundaries (and not only the
learning of syllable types) tends to be rather universal, although there are
language-specific differences (§§ 4.6, 4.7, 4.16, ll.l9). Syllable structure does
not only guide sound change, it can also be its target. The most unmarked
syllable structure is CV (open syllables only). Slavic seems to have had a "con-
spiracy" toward CV (drift, teleology) in its passage from Proto-Indo-European
to Proto-Slavic. Three changes are mainly responsible for producing open
syllables: (1) metathesis of liquids, CerCV > CreCV (§ 4.18), (2) monophthon-
gization CeyCV > CiCV, CewCV > CuCV, and CenCV > C~CV, and (3)
cluster simplification, for example, Pre-Slavic *supnos > sunu 'sleep, dream',
where both closed syllables drop the final consonant. The language ends up
with open syllables only (compare the Rotuman loss of one syllable or length;
§§ 4.6, 4.7, Figure 6-l: B). On the other hand, all the modern languages have
reverted to complicated clusters (by syncope), that is, a marked state of
affairs.
[9.17] Of course, it may take a long time before the change gets established
for good. Let us look at one more example of restructuring (§§ 6.13- 6.15,
6.19). Many English dialects have an underlying diphthong JyuJ after dentals:
ftyuwnf tune, Jnyuwf new, and so on. In some dialects the glide is considerably
weakened, so that it has been inferred as being an irrelevant accompaniment to
the dental. This leads to restructured forms like JnuJ without the glide: /nuw/
new. If this output is not acceptable to the community, it can be corrected into
the "proper" JnyUJ. Another possibility is to keep JnUJ and to add a patch-up
rule g ~ y after dentals to pacify the community. Now the outcome is again an
acceptable fnyuw f new. But the problem is that the rule applies only in certain
words which have to be specially learned (compare § 3.3), and thus hyper-
correction can creep in: Jnyuwnf for noon (which never had JyuJ). As time goes
by, such a patch-up rule is gradually eroded, because a person who was forced
to put it up has little or no reason to require it from his own children, and so on.
Somewhere on the line the patch-up rule can be omitted, as our predictions
would tell us.
Let us look at another typical readjustment. An English-speaking child
learns early a base form foot and a pluralization rule with -s (frequency and
basic meanings). He can now predict a plural foot-s(§ 6.21; the grammar gives
the -s rule, and the universals, 'one meaning, one form') . But this deduction is
not accepted by the community, and he has to redo it. Normally he does not
want to abandon his first attempt completely, but patches it up as feet-s
WHY DOES LANGUAGE CHANGE? 201
(compare § 5.4). Finally he learns that this is one of those cases that "defy
reason." This is a case without any danger of changing the underlying form.
Note that a reinterpretation of distinctive features is obligatory for an adult
borrower (§ 8.5). This shows that features are more important than segmental
units (which must be combinations of features).
[9.18] Folk etymology and semantic change show most clearly the short-
cuts abduction helps to make; even completely "unjustified" extrapolation can
become successful. The reversals in base and derived values we have seen are
by no means rare. If the historically derived form is the most frequent one or
otherwise represents basic morphological, syntactic, or semantic categories, it is
natural for the learner to take it as basic and derive the original form from it in
reverse (if alternation is not leveled out altogether). This is inverse derivation
(compare "inverse spelling" [§ 2.6] and "inverted reconstruction" [§ 18.14]),
the exact reverse of the original history, that is, history: a > b I -c, inverse
derivation: b ~a I- non-e (see the form of rules in §§ 6.7-6.10), which is a
synchronic process.
In the Yiddish revoicing phenomenon we saw how a base form (voiced stop)
asserted itself also on the surface (§ 4.27). Similarly, the German vowel length-
ening in the nominative singular proves that the derived long vowel in the
majority of the paradigm (one kind of frequency) had apparently been reanalysed
as basic, and it thus surfaced everywhere. Note also the role of frequency in
the reversal of markings in the pluralization of final-spirant nouns in English
(§ 6.21). When alternation is eliminated no inversion remains. But it is clearly
there in cases where the alternation is extended beyond the original items. The
preconsonantal lbet<JI better is more frequent than the prevocalic lbet<Jr/, and
was clearly taken as basic whereby the /r/ became an automatic transition
sound, and thus spread (§ 5.3). Of the various final -n deletion cases let us look
at English mine fmaynl (§ 4.24). Like lbet<J/, my was the original preconsonantal
derived form which came to be taken as basic, whereby the /n/ in mine was
reinterpreted as a derived predicative marker. And again, in some dialects it
has spread into other persons: his'n, her'n, our'n, and so on (the alternation is
made "purposeful use" of;§ 5.21). This is the same process that gave the final
stop in Estonian kuusk (§ 5.3). Some Estonian words even switch consonants
after d > 0 and g > 0 (§ 5.9). The derived form, 0, is taken as basic and we get
a reversal 0 ~ d and 0 ~g. We know this from the fact that sometimes the
wrong stop is reestablished: d > 0 ~ g, and vice versa.
Proto-Indo-European had a large number of roots of the shape CeRC
(§ 4.18), for example, Greek derk- 'see' (or English help), but relatively few
roots of the type CReC, for example, Greek trep- 'turn' (or English break;
see § 12.3). From such shapes another root variant was derived by the deletion
of the vowel, e ~ 0, whereby both CeRC and CReC result in Cf!.C. Thus in
Sanskrit, where e > a, we have the deletion as a~ 0: dars ~ drs 'see', kalp ~
kfp 'be adapted', myak$ ~ mik$ 'mix', and mrad ~ mrd 'crush', and so on.
These derived forms without a hold the majority in morphology, and came to be
202 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
taken as basic. The deletion rule was reversed in that now the vowel had to be
inserted in the grammatical environments where it had originally belonged. But
the vowel was inserted according to the majority pattern CeRC, and thus goes
always before the medial resonant: mik~ ~ *mayk~ > mek~, mrd ~ mard, and
so on. The old base form, if not ultimately lost, is often relexicalized into a
separate root not subject to vowel alternation at all, for example, myak~ 'join,
belong to' (compare§ 17.9).
One of two possible word orders is often called 'inverted'. This implies that
also linguists take the other one as basic. Reversal of word order easily occurs if
the frequency and the semantics are right. Even more peculiar are cases like the
Finnish head and attribute switches (§§ 5.17, 6.22, 7.13), although they can be
explained through abduction, the main force of syntactic change (§ 19.5). In
syntactic change a single case leads to a new syntactic pattern, where:as the
matter is quite different elsewhere in grammar, because a single direction may
be lacking. Linguists (and other scientists) have not been willing to acknowledge
abduction because of its unpredictability. Only when its results are regular have
they been happy to formalize the situation post facto, bypassing the actual
abductive link. But such regularity is "accidental" and does not represent the
essence of abduction (e.g.§§ 5.9, 5.10).
REFERENCES
9.2 Stern 1931, Vachek 1962; 9.3 Menner 1936, Jespersen 1941, Labov 1970,
Coates 1968; 9.4 Goossens 1969; 9.5 Dressler 1969b; 9.6 Martinet 1958, 1964,
Moulton 1960, 1961, 1970, Sieberer 1964, E. ltkonen 1966, P. Ivic (private
communication); 9.7 Labov 1971; 9.8 Zipf 1965, King 1967, Manczak 1968,
1969, Weijnen 1969, Onishi 1969, Greenberg 1969b; 9.9 J. Harris 1969, Cairns
1969, Vennemann 1971; 9.10 Joos 1952, Fonagy 1956-1957, 1967, Wang 1969,
Chen and Hsieh 1971; 9.11 Labov 1963, 1965; 9.12 Weinreich and Labov and
Herzog 1968, Labov 1970 and in Sebeok (ed.) 1963f. vol. 11, Anshen 1970,
Malkiel 1967; 9.13 Sapir 1921, Scur 1966, Greenberg 1969a; 9.14 Halle 1962,
204 HISTORICAL LINGUISTICS: How DOES LANGUAGE CHANGE?
Stampe 1969, Martinet (ed.) 1968, Graur (ed.) 3.127-200; 9.15 Andersen 1966,
1969, Shapiro 1969, 1970; 9.16 Peirce 1955, Knight 1965, Andersen 1969, 1972,
(lecture" Abductive and deductive change") 1971, RatHer Engell970, Lindblom
1971 , Coates 1968; 9.17 Andersen (lecture) 1971; 9.18 Anttila 1969b; 9.19
Knight 1965.
PART III
COMPARATIVE LINGUISTICS
(GENERAL NOTIONS AND
STRUCTURE):
HOW CAN CHANGE
BE REVERSED?
CHAPTER 10
PRELIMINARIES TO THE
HISTORICAL METHODS
PHONEMIC ANALYSIS
of work has been directed toward the principles and procedures for arriving
at this level of phonetics. These principles are generally known as phonemic
analysis, and the functional phonetic surface units as phonemes. Contrary to
general belief, the phoneme was intended as a practical (rather than theoretical)
help in writing down and analyzing the utterances of a language, with a minimum
of symbols and without sacrificing the relevant distinctions, because a key
permits an unambiguous mapping of the phonemes into the actual sounds and
vice versa. A phonemic notation is one that does not write any phonetic detail
that can be predicted from this notation. In other words, phonemic analysis is
a handy storing procedure when we do not want or need to carry around minute
phonetic detail in our operations. The key converts the phonemes back to sound
when necessary. This technique is similar to the dehydration of food, also for
practical purposes: water can be put back when needed. That is, these procedures
do not replace phonetics or water, they just do without them for some
purposes.
of them. The same is true of Finnish, in which all eight vowels can be defined
with a minimal octuple:
[10.3 Postediting and the Role of the Linguist] One frequent indeter-
minacy in the area of segmental sounds is 'one or two phonemes' (unit or
cluster)? In English, [tsm] chin contrasts with [dzm] gin, ([thm] tin, [dm] din,
al)d [sm] shin). Because of patterning and other criteria, many linguists phonemi-
cize these as /cin/ and /Jin/, although /c/ never contrasts with the sequence ftsf.
The /c/-notation is still quite unambiguous: initially it is preceded by stress
onset, whereas /t/ and /s/ are separated by one (why choose vs. white shoes).
Here, then, many linguists are happy to remain below the level of strict phone-
mics, even when this stress onset can be written as a juncture or boundary;
2Io CoMPARATIVE LINGUISTics: How CAN CHANGE BE REVERSED?
that is, I+ cl vs. It + sl could as well and even better be I+ tsl vs. It + sl.
Often these junctures require grammatical information, and generally they are
used freely in editing the results of the method. In German, there are a few cases
where the palatal and velar spirants contrast, for example, [ku·9~n] Kuhchen
'little cow' and [ku·x~n] Kuchen 'cake, pastry'. The occurrence of the variants
can be predicted grammatically, and one generally writes the morpheme bound-
ary in Kuh + chen, that is, lku· + x~nl. Such pairs are peripheral in German,
it is true, and they always involve the diminutive, for example, Tauchen 'little
rope' vs. tau chen 'dive', Schlauchen 'little sly one' vs. schlauchen 'to use a hose
for filling a barrel', or almost minimal pairs in Pfauchen 'little peacock' vs.
fauchen 'hiss' and Frauchen 'little woman' vs. rauchen 'smoke'. It is also rather
difficult to find a native speaker who would acknowledge all these forms.
Linguists differ greatly in their use of such boundary markers or junctures.
Many factors play a role here, for example, the tradition of the language, or the
linguistic school in which the linguist was trained, or simply the linguist's prefer-
ences for theory or even for some kind of implicit elegance. Thus a junctura!
analysis may be rejected for one language (e.g., English above) and preferred
for another (e.g., German above), even by the same linguist.
The stops in spill, still, and skill are complementary to both lp, t, kl and
lb, d, g/. After [s] in initial position, only a lax unaspirated voiceless stop occurs.
If that's tough and that stuff can be [oretsthAf] and [oretstAf] without indication
of stress onset, as they, in fact, sometimes are, we would seem to have a contrast
in the dentals. Such pronunciations, however, are extremely peripheral, and this
is why one relies on the position of the stress onset, which makes the dentals
complementary: forets't~f/ vs. foret' st~f/; stress onset correlates well with mor-
pheme boundaries that's + tough and that + stuff, because some boundaries
can always be phonetically implemented. In some varieties of English, the pro-
nunciation of stuff is auditorily (at least) identified with duff /d~f/, if the [s]
is somehow eliminated (e.g., through tape erasure). English /d/ is a lax stop in
many dialects, and voice is apparently not the distinctive feature in it. Thus, in
this variety of English, still should be written /sdil/. Very few linguists do it
-apparently for orthographical and practical cross-dialectal reasons.
adze-adds, Polly-Piili, and so on; compare§ 9.14.) The flap itself is complemen-
tary to both ftf and /d/, and its assignment to either one is impossible; thus,
by the strict principles of phonemic analysis, it should be a phoneme by itself,
/D/; compare still, which should be fsdil/ in some dialects, but which, contrary
to such an identification is universally written as jstilf. The importance of the
method is that it pinpoints the differences in [rayo~r] and [ra·yo;)r], for example.
It is at this point that the linguists usually leave the method and rephonemicize
shortness +flap as /t/ and length +flap as /d/, for reasons dictated by the rest
of the grammar, for example, the relationship with jrayt/ vs. fraydf. Then, of
course, many generative phonologists would say that one should not use such
a phonemic method in the first place. Be that as it may, in actual practice they
implicitly use something very similar.
history produces an output of sound and meaning, and the linguist has to work
the data backward toward the deeper origins. Reconstruction is a predictive
inference; that is, we "generate" part of the past, as it were, and we do this
initially with the analytic item-and-arrangement method, which is the compara-
tive method (Chapter 11).
As a final reminder, it is very useful to remember that a coin has two sides,
that is, that there are different complementary aspects of methods and their
domains in linguistic investigation in general. These two aspects with their
characteristic constituents can be listed in two columns (and these should,
again, be taken only as helpful cardinal points that enable the beginner to
organize his thinking about various aspects of linguistics):
analysis synthesis
item-and-arrangement process
units rules
comparative linguistics historical linguistics
hearer/linguist speaker
induction (see§ 9.16f.) deduction
Some linguists believe that a native hearer analyzes by synthesis only, but the
notion of "analysis-by-synthesis" is too one-sided to be of so much value. It
does, of course, occur to a great degree because hearers are also speakers and
can use contextual information for interpretation. A person who has to start
with pure analysis is a nonnative hearer, a linguist.
This discussion of phonemic analysis is extremely sketchy; its purpose is just
to recall the basic principles of the method. Many linguists, for theoretical
reasons, emphatically reject the kind of analytic phoneme described here, but
we must repeat once more the two main practical reasons why the concept and
the method must be known even to genetic linguists: access to past publications
(written in that framework) and access to the other methods. This will become
clear in the subsequent sections. The better one knows phonemic analysis, the
easier it will be to handle the other analytic methods.
MORPHOPHONEMIC ANALYSIS
2
IT] drrnv•
dr i ven
chrnld
ch i ldren
wrnde
w i dth
divrn ne
div i nity
3
[;] sillne
s a nity
opillque
op a city
dill me
d a msel
Sp~n
Sp a niard
br~d
~ r~d (present) br~the v
r ea d (past) br e d br ea th N
2
IT] brn,,
b i t
hrnd,
h i d
etc.
PRELIMINARIES TO THE HISTORICAL METHODS 2I5
The r- and d-sets are constant, and thus the vowel sets do differentiate meanings
and, consequently, contrast. This becomes clearer once we write the vocalic
sets with single symbols (letters) rid vs. rEd (compare a German minimal pair,
216 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
§ 10.10). This is quite parallel to, say, /pin/ pin vs. /pen/ pen. Thus the past-tense
forms rld-d and rEd-d assume the vocalism ridd and redd, and, finally, a geminate
is simplified, giving /rid/ and fred/ read. (Historically, this analysis of ride/rid
is wrong, because the paradigm derives from the type drivefdrovefdriven.) In
other words, the vowels tell us that there is a morphophonemic dental suffix in
these forms. This suffix has no other manifestation.
These same alternations do seem to occur under grammatical conditioning in,
for example -teenften, life/live V /liv/ (the adjective /layv/ live has the same
vocalism as the noun; the difference in spirants will be treated below), but we
cannot pursue them here, because our purpose is to show the method, not how
far the method can take us.
singular
plural
r;;l
~
t~th g~se
t l::J th g l::J se
~ tul~
~ m 1 ce
l,o_u,se
I 1 ce
These cases are so few that one would not necessarily set up special morpho-
phonemic vowels for them (see § 4.5), but more compelling evidence can be
found for the following consonant alternations:
singular houls]e
plural hou~es
These alternating spirants contrast with the invariant ones in chief/chiefs,
faith/faiths, and so on, so that there has been reason to posit two morpho-
phonemic classes of spirants. Here again one does run into indeterminacy,
but a completely regular pattern is found in
share an /s/, but the "upper" members of the sets make them contrast and
consequently we need four different labels for the sets: ldl, ltl, lkl, and lsi.
When one finds other sets that share members with these, one again tries to
see whether they are variants of the ones already established or independent,
for example,
....."'
.5
en
ss::
S::
~ 8
] ·:;
,E:e
- ·-
0 ()
<:ti
.__,s::
0
~
0.
FIGURE 10-1. The parallelism between the phoneme and the morphopho-
neme.
218 COMPARATIVE LINGUISTICS; HOW CAN CHANGE BE REVERSED?
[10.9] In many cases there remain peripheral alternations that are unique
or extremely limited in number. In such cases it does not pay to establish
special morphophonemic symbols, and linguists usually treat them as exceptions
of some kind, for example, something related to suppletion. Many English
speakers do not associate bleak and bleach in their synchronic knowledge of the
language, in spite of the close similarity in meaning and phonetic shape of the
two words. There are other examples of this /k "' ts/ alternation, for example,
break/breach (where the second member is a derived noun), stink/stench (where
the second member is a derived causative), and duke/duchess (where the second
member is a derived feminine). Parallel to these are stick/stitch and drink/drench,
but these are now synchronically separate although historically the same. Of
course, there are also many types of alternation in the vocalism of these forms
but we shall ignore them here. Thus, in a synchronic description, one has to
posit two different forms like lstikl and lstitSI. This has clearly happened, that
is, it has become obligatory, for book and beech, which once were closely
connected in the same way-not to speak of other more opaque cases like
wrinkle 'clever trick' and wrench 'false interpretation' (both from a root
originally meaning 'twist'). Such "exceptions" are a nuisance in a synchronic
grammar, although they show a fair amount of regularity; but they are often
the main source for historical hints in connection with internal reconstruction,
the historical side of morphophonemic analysis (Chapters 12, 18). Thus the
method should always be carried as far as possible, because it turns up handy
stepping-stones into history. For the synchronic grammar, the debris from the
use of this aspect of the method creates an area of considerable reshuffling,
which surpasses that of phonemic analysis. Here, too, one eventually reaches
a level where the services of the method become redundant or useless in the
synchronic context (§ 9.19).
[dl
schnei en sie [dl
en Jdl [tl rei [tl en schel ftl en
geschni ~en geso ~en = l!J vs. ~ = geri ~en geschol ~en
'cut' 'boil' 'ride' 'scold'
Whatever the ultimate assignment of these sets, they contrast and belong to
PRELIMINARIES TO THE HISTORICAL METHODS 2I9
'hand'
'trade'
han ill
han d el(n)
vs. han m ieren
han t el
'handle'
'hand bar'
pe t i jo d i
pe t i-ii jo d i-a
pe t i-n jo d i-n
pe t i-t jo d i-t
pe t e-i-llii jo d e-i-Ila
6 7
'bed' 'iodine'
FIGURE 10-2. A selection of Finnish front vowel and dental sets (limited to
five forms out of twenty-eight).
and so on. For practical purposes these words have been quoted in the nomina-
tive singular only, because more of the paradigm is supplied in Figure 10-2
through other items. The sets are labeled numerically because the method has
not yet shown which ones, if any, have to be grouped together. Now, because
of such minimal pairs, these three Finnish front vowel sets clearly contrast,
exactly like the dentals in German bunt vs. Bund. Within sets 2 and 3 the alterna-
tion is, of course, environmentally conditioned: in 3 fef occurs only before i,
and /i/ occurs elsewhere; in 2 the vowel is syncopated before the plural i and
the partitive morph -tii, and is replaced by i in final position (nom. sg.); and in I
e is unalternating all the way. Sets 1 and 2 contrast only in stem-final position;
elsewhere we have sets like 1, for example, the first vowels in vesi, neiti, and
peti. This situation reminds us of the English flap problem encountered in
phonemic analysis (§ 10.4), that is, there is a contrast only in one particular
environment, and indeterminacy in the assignment of initial vowels (to either
1 or 2). We shall return to this, but for now let us go by the method, which
gives us three morphophonemes, say, I = Je 1 J, 2 = Je 2 J, 3 = JiJ.
Turning now to the dental sets we can immediately see that 6 and 7 contrast:
6 is unalternating ftj, and 7 unalternating /d/. Set 5 shares both ftf and /d/ and
thus clearly contrasts with both 6 and 7. The alternation is phonetically con-
ditioned: jtf occurs in open syllables, /d/ in closed ones. Thus, from the end,
the sets can be labeled, for example, 7, Jd J, 6, Jt 1 J, and 5, Jt 2 J. There remains
set 4, which shares ftf and /d/ with Jt 2 J (5), but otherwise shows an extra dimen-
sion in having /s/ in addition. Because of this /s/, it has been taken as another
independent set contrasting with Jt 2 J, say Jt 3 J, but a closer look reveals that the
PRELIMINARIES TO THE HISTORICAL METHODS 221
nom. sg.
gen. sg.
joilli
jo 0 en
;ill a
i 0 an
su~u
su v un
ky~y
ky v yn
'river' 'age' 'family' 'ability'
10 10 11 12
sylilli
syl j en
jar ill;
jar j en
muilli
mu kin
lii~a
Iii g an
kaftlo
vs. ka~t o or
katltlo
ka~on ka t on katl!J on
'loss' 'roof'
222 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
sufkla sukfkla
vs. sur;]k a or
su~an su k an sukl!Jan
'brush' 'sock'
The important fact is that the method does confront the minimal pairs here in
either case. The second analysis definitely seems to be better for sukka, because
it does not require any new machinery (i.e., set 8 includes this environment,
too), and for katto we would get a new variant (13) of the morphophoneme lt2l
(5). All this agrees with the analysis that length in Finnish is taken as a double
occurrence of the single unit.
So far we have been treating Finnish morphophonemic alternations that are
phonetically/phonologically conditioned, just like most of the English alterna-
tions above. But Finnish also has cases parallel to English toothfteethfteethe,
wreath/wreathe, and so on:
tah [tl
o-mme N 'our wish' malkl-is-Hi adj. 'hiiiy' (part.)
tah ~o-m me V 'we wish' (§ 4.23) ma l_!j
-i-sta N 'from the hills'
The surface environments are exactly the same for both alternants; that is,
in terms of the sounds that actually occur, the alternation is induced in clearly
defined grammatical categories. In short, the possessive suffixes (-mme, and so
on) do not induce consonant gradation, and neither does the adjective suffix
-ise- (whose final vowel is le 2 1 and thus drops in the partitive).
will recur below. Small capitals represent half-voiced stops. Now we are ready
to start the analysis.
Set 16 contrasts with 18, and 17 with 19; the first pair has double stops, the
second has overlong stops in the nominative. The genitives are the same for all
and thus cannot be used for predicting the nominatives. The length difference
in the nominative (16 and 18 vs. 17 and 19) is conditioned by the preceding
Dentals:
14 15 16
inf. ko'OO•D nom. sg. niej~a kielhttla
1 sg. kol d am gen. sg. nie oa kie o
a
'dig at random' 'young lady' 'hand'
17 18 19
kulhftla
kii o a
koiMia
ko o a
o~a
o oa
'six' 'spawn' 'new'
Velars:
20 20 21 22
nom. sg.
gen. sg. j~lh·t tkkla ~~~kl• juol~e
JO yy a so yy a a IJ e juol g e
'river' 'family' 'age' 'leg'
22 22
tol~e t!ol~a
tSol g a
tol g e
'feather' 'saliva'
Corresponding longs:
23 23 24 24
inf. ka lhttl ·eD nom. sg. nuol hitle solhkkla alhkkle
1 sg. ka htt am gen. sg. nuo htt e so hkk a a hkk e
'cover' 'seine' 'sock' 'old woman'
All consonants: 25 26
(other than stops) kiefll] a
n:Jmmla
ne m a kiel!J a
'name' 'tongue'
FIGURE 10-3. A selection of East Lapp consonant alternations.
224 COMPARATIVE LINGUISTICS: HOW CAN CHANGE BE REVERSED?
vowel. If the vowel is short (17, 19), the intervocalic stop is overlong; if long
or a diphthong (16, 18), the stop is double. Indeed, this recurs in the rest of
the material as well (20 vs. 21, 25 vs. 26). (Further evidence would show that
such overlong stops occur only in nouns, not verbs, which is another interesting
grammatical conditioning.) Now we can turn to sets 15 (genitive o as in 16-19)
and 14 (i5 as in 15). It is easy to see that 14 and 15 are complementary to 16-17;
[htt] occurs between vowels and [5] after resonants (paii5eD/piildam 'be afraid/
I am afraid', kufoajkuyda (or kuyDa) 'hay patch'), which are also long in the
nominative. These same environments determine the occurrence of [o] and [d]
in the genitive. The consonants turn out to be always long in the nominative,
and, in addition, a voiceless stop is preaspirated. All these dentals can be pre-
dicted from two units, which the method gives us, the morphophoneme It!
(14-17), and ldl (18-19). The same principles hold for the velar sets 20-22 as
well: they are all environmentally conditioned and thus variants of one morpho-
phoneme lkl . The longs are also here best taken as clusters of the single stops,
!ttl (23) and jkk l (24). Also lm l (25) and III (26) are unambiguous starting
points for the nominatives nel'nma (overlong because of the preceding short
vowel, which is long in the genitive only) and kiella (double because of the
preceding diphthong).
Morphophonemic analysis is a method of establishing invariant sound units
in the framework of paradigms or the total machinery of the grammar. We have
seen how easily it eliminated consonant gradation from both Finnish and Lapp.
The (mainly) phonological conditioning in Finnish and the grammatical con-
ditioning in Lapp made no difference to the method. The Lapp situation would
make one seriously suspect that the consonant alternations were originally
determined by phonetic conditioning (compare wreath/wreathe, and so on,
§§ 4.2, 4.5). And, indeed, if monosyllabic words had been included, they would
have shown the genitive marker -n:Ki 'who', Reii-n 'whose' (compare Estonian
maa-n-tee, the only relic of the genitive -n in that language; §§ 4.24, 5.15). Thus
there is a section of vocabulary that retains the original phonological condition-
ing. This section confirms our suspicions. But the point is that morphophonemic
analysis operates without such confirmation quite well; but such historical
guesses need not be included in a pure synchronic context.
(I) Drop of je21 after t before partitive -tii and plural-i-: vettii, vetillii
(2) jt 2j-+ /d/ in closed syllable: veden, neidin
(3) !e2!-+ /i/ before #: veti
(4) lt2l-+ /s/ before /i/: vesi, vesillii
(or the like, see§ 6.6)
PRELIMINARIES TO THE HISTORICAL METHODS 225
and in Lapp,
(I) It!-* [D] in nom. sg. after resonants (and finally): nief.va, kolvev
(2) It! in gen. sg. and 1st sg.--* [d] after resonants, [o] intervocalically: koldam,
kieoa, kuoa
(3) Lengthening of consonants and clusters in nom. sg.: kietta, niefDa, koooa,
nuoite, nemma, oooa
(4) Extra lengthening in nouns with a short vowel preceding: nemma, oiioa,
kuita
(5) Lengthening of short vowels in the genitive and 1st sg.: kuoa, nema, koldam
(6) Preaspiration of voiceless stops: kiehtta, nuohitefnuohtte, kuhita [or the
like]
Note that the rules always turn out to rely heavily on the environments that
made the analysis possible in the first place (§ 10.6). As we see, this synthetic
part mirrors the framework of historical linguistics, because sound changes
occur in a strict historical sequence (§§ 6.1-6.3). The difference is that in the
synchronic framework one can take the shortest way even when it violates
our knowledge of the actual history(§§ 6.5-6.6). Often, however, such ordering
of synchronic mapping rules mirror the actual historical changes to some
degree ; for example, there is a fair amount of relative chronology in them. In
the Finnish case, we must first have the raising of e > ito get the right environ-
ment for the assibilation ti > si and so on (§ 6.6).
It has already been mentioned that when the method encounters rare or
unique alternations, for example, English manfmen, one is hardly willing to
construct special vowels for such cases. The alternation in manfmen, however,
is not completely random, since there are other vowel alternations in English.
The case of goose/geese and tooth/teeth is more systematic, because there are
at least two instances of it, but certainly not systematic enough to warrant a
special vowel to carry the alternation (one might rather use an exception feature
like [+umlaut] or the like; see§ 10.16). These words will have to be somehow
listed or marked as" regular" exceptions. The postediting problems of morpho-
phonemic analysis surpass those of phonemic analysis; one hardly ever tries
to apply the method up to its logical conclusion and to adopt its results without
any questions asked in the establishment of morphophonemes. Postediting is
the final touch in which every factor is drawn in and inductive methodology is
mellowed by deductive expectations from universal theory. The results of the
method were not adopted in the framework of the item-and-arrangement model
of listing allomorphs, which in any case did not involve morphophonemes, and
the result was that all items were exceptions in this sense. The logic of the gram-
mar must naturally prevail over the logic of the method (§§ 5.14, 9.19).
nursery and loanwords, and other such descriptive material (notable exceptions
are itse 'self' and ko!me 'three'). The two vowels merge, if the words with
le 1 1 are marked for a special feature, say, [+affective]. This feature indicates
just all those semantic areas mentioned and it blocks vowel alternation. Or,
second, we may erase le 2 1 from the underlying forms altogether and establish a
boundary marker (juncture) of some kind. At this boundary we can now write
the automatic union vowels i and e where needed. We notice further that
consonant gradation fails to occur in the same categories listed above. By
adopting two features [native] and [foreign], we might have one dental only,
ltl, which would be unalternating /t/ in connection with [-native] and unalter-
nating /d/ with [+foreign]. The same is true of the velars as well, for example,
REALIZED AS
nalle [-native] nal!ejna!!en
vete [+native] vesifveden
peti [-native] petifpetin
joti [+foreign] jodifjodin
liika [+foreign] liigaj!iigan
liika [+native] liikaj!iian
muki [-native] mukifmukin
etc.
Note now that [-native] in this scheme is not the same as [+foreign]. In the
same way, one has used these features to indicate the differences in English
spirants, for example,
PLURALS
chief [-native] chiefs [f]
faith [-native] faiths [El]
knife [+native] knives [v]} .
sheath [+native] sheaths [o) VOICe
Now, in the environment [-native] a spirant fails to voice. One should note
that such features would not necessarily reflect historical borrowing; a histori-
calJy native word like oath has to be listed as [-native] in those varieties of
English in which the plural is jowEls/ (rather than jowozj), to ensure the voiceless
spirant. In a way it creates a situation that is the reverse of the acclimatization
of loans, taking away the official citizenship rights. The feature [+foreign] fails
in those Finnish words with more than one stop that are different with respect
to voicing, for example, toga and Golgata 'Golgotha'. Hence one seems to
need ldl and lgl after alJ, but the voiceless stops can be combined with the use
of features as indicated. All this is quite parallel to what we did to Finnish
tykin, for example, by extracting a feature F(rontness) and writing the word as
Ftukin. Such features as [±foreign] are rather vague as a device for blocking
the operation of certain rules, and they were presented here only as historical
curiosities of a procedure that enjoyed popularity during the mid-1960s. To
PRELIMINARIES TO THE HISTORICAL METHODS 227
block a rule one can use the nonoperation of the rule as the most concrete
feature, for example,
peti [-consonant gradation]
faith [-voicing rule] etc.
That is, rather than create a special feature that ultimately negates a rule we
directly use the negation of that rule. This makes our phonology as concrete
as possible. In the above case, this same desirability prescribes Finnish morpho-
phonemes ldl and jgj. These are now always /d/ and /g/ and no special conversion
rules are required.
Diacritic feature notations like [±foreign] show the extent linguists like to
theorize for their own delight irrespective of "innocent" speakers whose speech
they supposedly describe and explain. A child learns his language without
knowing that borrowing exists; he has no use for history in any form. For him
read/read /riyd-red/ and breed/bred are on the same footing as bite/bit and
ride/rid (original ablaut), and in fact hide/hid is now conjugated exactly the same
as ride (analogy; past passive participle hidden). The child is always right, as
long as his output is accepted, whereas the historical linguist is wrong when he
does similar things in (the [hopefully] initial stages of) his analysis (§§ 9.16,
9.19, 10.7, 18.16, 18.17). It is important to realize that both the language
learner and the linguist must use abduction(§ 9.16), and are thus bound to make
"mistakes" here and there. This aspect is emphasized throughout the book.
REFERENCES
General: Bloch 1948 and the references in§ 1.1; 10.4 Schane 1971; 10.6 Sigurd
1970, Schane 1971; 10.9 Wartburg 1931 , Malmberg 1969 ; 10.10 Vennemann
1968; 10.12-10.14 Anttila 1968; 10.14 E. Jtkonen 1941 , 1946, T. I. Itkonen 1958;
10.15-10.16 Anttila 1969a; 10.17 Anttila in Sebeok (ed.) 1963f. vol. 11.
EXERCISE
On the basis of the following additional forms, complete the analysis of Finnish
consonant gradation. Only two characteristic environments are given, the
nominative and genitive singular.
NOM. GEN. NOM. GEN.
kulta ku!!an ' gold' pel)kki pel) kin 'bench'
pelto pe11on 'field' pupu pupun 'bunny'
part a parran 'beard' Kaapo Kaapon 'Gabriel'
sorto sorron 'oppression' }obi Jobin 'job'
rant a rannan 'strand' papu pavun 'bean'
pinta pinnan 'surface' rupi ruren 'scab'
pinna pinnan ' peg' napa naran 'navel'
ruffa ru!!an 'roll' ha!pa halvan 'cheap'
hyrrii hyrran 'top' kifpi kih;en 'shield'
kortti kortin 'card' turpa tun·an 'snout'
tontti font in 'lot' tun•a tunan 'protection'
Riku Rikun 'Dick' korpi korven 'dark forest'
grogi grogin 'drink' kumpu kummun 'small hill'
ku!ku ku!un 'going' rumpu rummun 'drum'
pa!ko palon 'pod' Iampi !am men 'pond'
parku parun ' weeping ' sut:i suven 'summer'
t'irka riran 'office' savu savun 'smoke'
puku purun 'dress/suit' nappi napin 'button'
palo palon 'fire' sappi sapen 'gall'
hol)ka hOI)I)an 'pine' kuppi kupin 'cup'
rul)ko YUIJI)On ' stem' norppa norpan 'marble seal'
ho!kki ho!kin ' cigarette lamppu lampun 'lamp'
holder' tumma tumman 'dark'
For other selections, see Cowan 1971.
CHAPTER 11
a~ ~
a n
a nn
0 il
s o n
s oh n
we ~
h o. m e
h e1 m
il a~ il u~
h ai r
h aa r
h ou s e
h au s
~ u~
m ou s e
m au s
In spite of the semantic identity and the formal similarity of the boxed-in
sounds in the first two items, we do not have regular sets of correspondences;
these are unique. No other items where these recur can be found between these
languages. Thus the requirement of regularity diminishes the possibility of
being misled by chance similarity. Actually, it eliminates chance similarity
altogether, but it is no help in another kind of complication. We said that it is
impossible to find another pair of words where English d corresponds to Latin
d. But note Latin dens 'tooth' and English dentist, where the semantic side is
certainly acceptable and where we have another such set (see §§ 8. 12, 8.13). Or
consider
[11.3] Let us now go back to the six items from the three Germanic
languages. Here we clearly have basic core vocabulary, which seems to guarantee
inheritance. Excluding stops, the Germanic consonant correspondences are
very straightforward, as the cartouches show: each language reflects the same
sound. To apply the comparative method we treat these sets of correspondences
like phones in phonemic analysis, that is, we combine all noncontrasting sets
into one unit; all such sets can be written (labeled) with one and the same
symbol. Thus the m-set and the h-set clearly contrast, because we have the mini-
mal pairs house and mouse. One would not combine m's and h's anyway,
because there is no phonetic justification for it. Phonetically there is more reason
to try to see whether them-set contrasts with then-set (both are nasals). These
sets do occur in the same environments, for example, in final position in son
and home, hence contrast and must be labeled differently. For mnemonic
purposes and reasons of common sense we label the h-sets with *h, the m-sets
with *m, and the n-sets with *n. The asterisk is the traditional indicator for the
fact that the following symbol(s) (letter[s]) is (are) shorthand for a whole car-
touche, a set of correspondences; more generally, * means 'not directly docu-
mented'. In this sense, *h means that we have h in Swedish, h in English, and
h in German, and so on, in other words, what we can read in a set (cartouche).
These sets recur through a great amount of additional material, for example,
in the numbered items of Figure 11-1 (see § 11.4), which supplies evidence for
stops: *m (5), *n (11, 12, 15, 24, 36), and *h (7, 16, 17). Further similar simple
unalternating sets are *r (14, 16, 22, 24, 25, 26, 31, 33) and*/ (2, 5, 6, 8, 13, 21,
34, 36), where the choice of symbol is again obvious, because these contrast with
232 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
1 2 4
~0~~
~ iWa aa~p
i p e
ei f e
pp e~
I e
5
a pf e I
6 7
liJu~
8
~ a~a ~ !r!lem
~~
o[;Jma
I oo m
I u m e
10
~ i~a
i v e
e b en
11
ave
a ben ~owe~
~
a~t en
a t er
a ss er
* * *
12 15
~m~.~e 6~r:Jm
l;J l!JeiQ ~a~~
16 17 18 19
ja~ ~a
~ ~
e~t
~
and
ea
e r
r t
z
ea t a
i tz e
ooth
ahn
u~
out
au s
22 23
25
ITl ~~~
~~ei
26
~ ~~g
1ng
i ng
28
~ a~e~
~ o~e~
rn
a th e r o th e r
a~-ar
a y -s
a t e r u e r
tt a g -e
30 31
~~ o~e~
rother
r u d e r
* * *
FIGURE 11-1. Swedish, English, and German consonants lined up for the
comparative method.
THE CoMPARATIVE METHOD 233
32 33 34 35
o~a
rn
k
00
o chen
[
~ a~
a Y ]
o gg en
'rye'
36 38 39
~:~em ~a~e
~
a~en
a w 'saying'
l;Ja~eW ~~Wen a g e
* * *
40 41 42 43
44 45 46 47
bin~a
bind
hanrnia
han d le
bor~a
bur d en
mor~
mur d er
bin d en han d eln biir d e mor d (es)
FIGURE 11-1. [Continued]
the units already established (the nasals and *h). So far it has been easy to see
how these sets contrast, because their phonetic makeup has been so obvious
and constant within each set. Now we have to look beyond standard orthography
and remember that the German s in sohn (and 3, 35, 37) is [z] against an [s]
in haus and maus, and an [s] in stein and stuhl (12, 13). In other words, instead
of the orthographic single set s-s-s we have actually three s-sets:
1' 2' 3'
Swedish
English ~s ~s ~s
German s z s
Closer inspection reveals that these sets are complementary, occurring in
mutually exclusive environments: 2' occurs in syllable-initial position before
vowels, 3' before a t-set, and 1' in syllable-final position. Hence they are allosets
of one and the same unit which we can label *s. So it turns out that German
orthography would have given us a shortcut in this matter in any case. For the
application of the method it does not matter whether one applies it to ortho-
graphic units or to phonetic units. It matters only for the correctness of the
results, because sometimes orthography obscures facts (distinctions) that must
234 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
be known in order to secure correct results. This phonetic check on the s-sets
proves that at this point the orthography does not obscure relevant phonetic
detail. Orthography is mentioned here in anticipation of those cases where
orthography is all we have.
[11.5 Labials) Looking first at the labials we see that Swedish hasp bf v,
English p b f v w, and German pf b f v, but we have six sets of phonemic corre-
spondence (as opposed to the relationships in the traditional orthography) in
Figure 11-2. The numbers indicate the proper environment for the sets, which
can be looked up in Figure 11-1. To facilitate talking about these sets, let us
call the first two p-sets, the next two b-sets, and the last two spirant sets. Once
again, we are going to treat the sets like phones to find out whether they contrast
or not.
In phonemic analysis, phonetic similarity makes two phones potential
candidates for membership in a single phoneme, and the method then checks out
Swedish p p b v f v
English
German
1, 2 1, 3 4, 5 6, 7 8, 9, 10, 11
25
FIGURE 11-2. Labial sets taken from Figure 11-1 with numbers indicating the
original item.
THE COMPARATIVE METHOD 235
such cases. As we have already seen with the s-sets, the same is true of the regular
sets of correspondences in connection with the comparative method. Thus with
the labial sets the best strategy is to see first whether the two p-sets contrast.
After all, they are the same in Swedish and English and differ only in German.
Exactly the same situation obtained in the s-sets. We have only to 1efer to the
actual environments given in the table (Figure 11-1), and it is clear that the set
with a German affricate (pf) occurs initially, and medially after short vowels;
the set with a German f, after diphthongs (and long vowels). Of course, the
evidence reproduced here is rather meager, but those who know German will
immediately see that this is indeed true. Hence the two p-sets are complementary,
conditioned variants of one and the same unit, and they can both be represented
with one symbol, say *p. The structure of the b-sets is the reverse of the p-sets,
because here German is the same for both, and Swedish and English differentiate
between stop and continuant articulation. But they are complementary also,
because the first one occurs in initial position and the second medially; they
can thus be written with, say, *b. Since *p and *b occur in the same environments
they also contrast. Another labial set can be found which has stop articulation
in German:
Swedish
English lorn dorn
in lea f dea f
German lau b tau b (devoicing; see§§ 2.4, 6.15)
This set occurs in (syllable-)final position, and thus contrasts clearly with *p
(lopp-leap-lauf, upp-up-auf). The b-sets, however, never occur in this environ-
ment but only in syllable-initial position, and hence this new set is a variant of
*b. Note that English does not supply any information for syllable position,
as both live and leaf are monosyllables, but the other languages do: livafleben
vs. 16vflaub. Swedish and German thus provide the information that explains
the contrast v =If in English live vs. leaf (see § 4.2). This is a typical situation
in the comparative method. Obviously, the method works only when different
languages have undergone different independent changes. English has lost final
vowels to a greater degree than Swedish and German, but it does not matter
here, because the comparative method uses the combined evidence from all the
languages to which it is being applied. Each language can potentially supply
useful information. The combined evidence has been emphasized by boxing in
the sets of correspondences for enhancing the notion of unity derived from
diverse sources (languages). Again, as in the case of the s-sets, we see that German
orthography, for example, laub for [p], is efficient.
As for the spirant sets, they contrast with each other, and with both stop sets.
It is easy to choose a label for the constant f-f-f, namely, *J, but the voiced
one, v-w-v, would seem to suggest either *v or *w. It is perhaps easier to derive
*w > v than *v > w, and hence *w seems to be preferable, both because of
articulatory simplification (bilabial more marked than labiodental) and for
mnemonic reasons (since the German orthography uses w).
236 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
We have now treated seven labial sets of correspondences, which were all
regular in the sense that they occurred more than once in this material (which is
just a small selection). These seven sets were made up of four Swedish and
German units, and five English units. This is a general characteristic: the number
of sets of correspondences can be bigger than the highest number of units in
any of the languages. Also, the number of reconstructed units is independent of
the number of sets with which we start out, although it obviously cannot exceed
the number of the sets. For the Germanic labials we got four, *p, *b, *J, *w, a
number very close to the number of units in all three languages. Notice, however,
that the distribution of these reconstructed units is completely different from
those of the units in the source languages.
[11.6 Dentals) The situation with regard to the dental stops and spirants
is more complicated than the one with regard to the labials. The s-sets (where
each language had spirants only) have already been treated. This leaves us-two
units from Swedish (t, d), four from English (t, d, B, o) and three from German
(t, ts, d), but these yield seven sets of correspondences as in Figure 11-3, where
Swe dish t t t t d d d
English
German
12, 13, 16, 17, 10, 19, 22, 23, 14, 27, 25, 26 30, 31
14, 15 18 20, 21 24 28,29
FIGURE 11-3. Dental sets taken from Figure 11-1 with numbers indicating the
original item.
the sets have been organized according to voicing and spirantness. These are
features that we expect to be relevant, because they were important with regard
to the 1abials. The best place to begin is with the three voiceless sets, which are
differentiated only in German (exactly like the threes-sets). The first one (t-t-t)
occurs only after sets with a spirant (at least in German); the second (t-t-ts)
initially and after short vowels (at least in German) or after *r (16); and the
third (t-t-s) after long vowels (at least in Swedish). Thus they are conditioned
variants of the same unit, which we can label *t. The conditioning of the variants
is generally the same as for the labials; both groupings are thereby mutually
supported. In the remaining four sets, both Swedish and German agree for
those sets where German has voiceless stops (t). These two sets are different
THE COMPARATIVE METHOD 237
o
because of the contrast d =f in English. As for their distribution, the set with
the English spirant (d- o-t) occurs only before a vowel followed by the r-set,
that is, *r. Complementary distribution based on so few examples may of course
always be wrong, but here it is all we can use. With more evidence, for example,
with the information that there are English dialects with [d] infather, we would
be inclined to combine these two sets under *d. Now, the two remaining sets
also have spirants in English. Because they both occur in initial position, they
contrast with each other as well as with *t and *d, which both also occur in the
same environment. The symbols we need to label these contrasts have to be
different from the established stops, and since spirantness is present in English,
spirants should be used. Thus we can label the set t-e-d with *p (thorn rather
than e is appropriate in Germanic, for historical reasons) and d-o-d with *o.
It turns out that the groupings come closest to the English units.
One further dental set involving an English spirant is
Swedish
klii.w-
English in oaewth clo th
German ei d klei d (devoicing; see§§ 2.4, 6.15)
Swedish
klii.Wer
English clo th es
German klei d er
This is identical to *6, and must be assigned to it by the method. The result
is that we have reconstructed singulars * Vp, *k/Vp and plurals * Vo- V, *kiVa- V
(where V is a cover symbol for the vocalic nucleus, which will not be further
treated). Such a situation is by no means typologically impossible; in fact, it is
exactly what English morphophonemics shows. But note that we made a selec-
tion from English, since the plural oaths can also occur with [9]. If we had
chosen this form and the plural cloths, we would have gotten a completely
different result- *p for all forms. The sections on morphophonemics showed
that the voiced spirants are regular in old inherited material, and thus we
238 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
probably chose the right variant. This shows, however, that dangers lurk at
every stage in our work. Synchronic variation may impose far-reaching effects
on our reconstructions.
So far we have treated the nucleus of the evidence for dental stops. With
further material and similar collation, other sets emerge. Let us look at items
40- 47, where the words have been given in such a form as to avoid German
syllable-final devoicing. This would just complicate the application of the
method, but as we have already seen, it would not block it. We get the following
two additional sets :
A B
Swedish
English
German ill40, 41 rn 42-47
Neither of them occurs in word-initial position, and hence they are potential
candidates for membership in those units that do, that is, *t, *d, *p, *a-in
other words, all the units we have so far established. Both sets A and B contrast
with each other since they both occur after*/ (i.e., set l-1- 1). Hence they contrast
also with *d, which also follows */ (English scold-German schelten). Set B,
being voiced all the way, would now seem to belong to *o, and this leaves
set A as a candidate for *p, which does not otherwise occur word-medially
after*/. This is the direction in which the method points; the evidence, however,
is too scanty for a definite answer.
(11.7 Velars] Finally, let us turn to the material that shows the velar
sets of correspondences, not counting the h-sets, which have already been
dealt with. Swedish supplies k g y (saga 35), and even t; English, k g y w :
(length), and German, k g x. Again the number of sets of correspondences
exceeds the number of the units in any one language, as shown in Figure 11-4.
The k-sets, one of which occurred also in the item cloth(es), are obviously
complementary: k- k- k occurs initially and k-k-x medially and finally. Hence
they can both be labeled with *k. The g-sets provide a more complicated situa-
tion. One, g- g-g, occurs initially only and is thus complementary to the rest,
misunderstanding that the comparative linguist works chiefly with old written
records or "dead" languages. Historical linguistics suffers from the same kind
of misunderstanding, since obviously its entire framework is very historical
indeed. This has led to a further misconception-that the comparative method is
valid only for the older Indo-European languages with written records (see §§
1.21, 1.22), with the implication that all "unwritten" languages need another
method. One could easily imagine a situation in which the European languages
were unwritten and, say, the American Indian languages had a long tradition
of writing. Writing is a historical accident depending on the socioeconomic
factors of the culture and not on the particular shape of the language. The three
European languages in this sample were presented in this spirit; this is how
they would appear to a Martian linguist sampling them. Traditional orthography
was retained, but that was ancillary and was not the basis of the actual treatment
(and, of course, such orthographies might also be available to the Martian
linguist).
But more important, the choice of Modern Germanic languages enables
us to check on the results of the comparative method: since the Germanic
languages were already written down a thousand years ago, we can go at least
that far toward the ultimate cutoff point and see how our groupings hold. {This
will be most valuable as a reminder for those situations where documentation
is not available.) Any reconstruction is valid only for those languages that have
been used '(i.e., included in the buildup of the sets); thus the units we have
tentatively reconstructed cannot be called Proto-Germanic, because all the
Germanic languages were not used. They can only be labeled Proto-Swedish-
English-German, or the like. Neither is there any certainty of such an exact
historical unity; it means only that we have a base from which we can derive
all the languages used in the establishment of the proto-units. This always carries
some historical truth; how much is another matter.
Turning now to the oldest records of English, we find that Old English father
and mother did have a [d], j<l!der and modor, and that daughter and night dis-
played a fricative [x], as reconstructed, which was written with the same symbol
as [h]: doh tor and niht. In the first instance, the method hinted at the right
direction , in the second, its result is confirmed. Further, have and live come
from Old English habban and libban, which in fact show b's, although long
ones. Those items that contained sets with German g's and where English was
so diverse all turn out to have orthographic ( g) 's at least, seemingly representing
[y] in back vowel environments and [y] in front environments, for example,
saw < sagu, bow < boga, nail < n<l!gel, rye < ryge, fowl < fugol, and lay <
lecgan, say < secgan (where cg is probably [ddz], that is, an affricate derivable
from g). In other words, Old English almost matches Modern German and
settles the issue in favor of one medial *g. The method by itself was able only
to suggest this.
As for the dentals, Old Swedish also shows voiced interdental spirants. Old
English, on the other hand, displays only a tendency toward complementary
distribution of (p) and ( 5) , the latter occurring only more frequently in voiced
THE COMPARATIVE METHOD 241
1 2 3 4 5
Check through
Old English A further
and Old High Adjusted check through
Items German forms reconstruction Gothic
thing ping (OHG) no change no change
thorn porn no change no change
brother brooor no change p
burden byr5en no change p
murder mor<Sor ~
no change p
cloth clap 0 6 p
clothes cliiGas no change p
oath iip 6 p
oaths a<Sas
1-------1
no change p
gold gold no change p
bind bintan d no change
mild milti d no change
handle hantalon d no change
environments than not. Depending on the documents, both [o] and [9] (which
are in complementary distribution) can be written with either jJ or o. Old High
German bears witness to a spirant in thing, that is, OHG thing> NG ding
and thiob > dieb 'thief' as ther, dher > der 'the'(§ 11.9). It also "replaces"
some of the Modern German d's with t's. In Figure 11-5 we can now tabulate
our reconstructed spirants, the additional older evidence, and the consequent
correction of the results of the comparative method. Now *p occurs only in
initial position where it contrasts with *o, which, however, occurs in this
environment only in pronouns. As for the set t-d-d (field and shield), we find
older Swedish spellings fiild- and faldh- (sixteenth century). It becomes clear
that there is also a word skold which means 'shield', whereas skylt is a technical
word meaning 'sign'. But also fait is a technical term occurring mainly in
military contexts. OHG scilt shows that shield contains a *d and OHG feld
speaks for a *o in field. Swedish fiilt and skylt still do not fit in these, and the
evidence is in all ways characteristic of a borrowing situation (compare tooth-
dentist). The technical meanings of these two words, together with historical
evidence. pointing to strong contacts between Sweden and the Hanseatic states,
make the borrowing hypothesis attractive; and, as it happens, the Low German
forms are velt and schilt, with exactly the right dentals. These two items are
typical of loans which are not spotted right away but undergo the method and
may lead to chimerical reconstructions.
24z. CoMPARATIVE LINGUISTics: How CAN CHANGE BE REVERSED?
[11.9] We have seen that our method was unable to penetrate completely
all the changes that accumulated during a thousand years. In English, the diffi-
culties arose partly from the changes a > d and d > a, which are very compli-
cated indeed, but more significantly from the fate of medial velars. German
has undergone a tricky change t > d, which is not phonetically regular (i.e.,
OHG skeltan > schelten, but milti > milde). The comparative method rests on
the assumption of regular phonetic change. Irregular change (various types of
analogy and borrowing) immediately blocks its historically correct course. All
three types of change do occur, and often we have no way of sorting them out.
Hence the result of the comparative method is always highly tentative. Further
evidence requires a revision as often as it confirms the earlier results. The
Germanic case shows that in the majority of items the method did give correct
results, which were confirmed by the earlier stages of the languages used. On
the other hand, in the reconstruction of interdental spirants, the older evidence
necessitated a revision.
If we turn now to Gothic, the "oldest" substantially recorded Germanic
language, we would have to make further changes, namely, to collapse *p and
*a into *p (column 5 in Figure 11-5), because now we have, say, the following
two sets:
~
orne and
~
Swedish et broder ed
English orn at brother oath
German orn as bruder eid
Gothic orn ata bropar aips
Gothic supplies just one p where the other languages required two, *p and *o.
English has been written phonemically, in terms of Old English, as p. The
first set occurs word-initially and the second word-medially and -finally, but
also word-initially in pronouns (e.g., du-thou-du-jJu). This grammatical condi-
tioning of the set should be noted, because it is not normal in the comparative
method. Many scholars would rather interpret it as conditioned by weak stress,
or the like, to get a phonetic environment for it. But the change is still completely
regular and thus enables the comparative method to penetrate it. When Gothic
is brought into the picture, all major branches of Germanic are represented and
the result could now be called Proto-Germanic. But note that Gothic is not a
check on our initial use of the method; since a new language is involved, we
have a new reconstruction altogether. We do not have a check on the method
but on the total Germanic evidence. At the same time, the results of the first
application stand up quite well indeed.
We have now seen the method, the reliability of its results, and why one
should always use the earliest evidence available, because then one can eliminate
many cases of unexplainable change. When we do not have early documents,
we apply the method to any other material available. But we always have to
allow for a certain range of confidence in the results. They need never be perfectly
correct historically. They are only correct as a starting point from which we can
THE COMPARATIVE METHOD 243
derive unambiguously all the languages used in the method; to some degree
this is also historically true. The method is very powerful and very useful, but
not omnipotent.
(11.10 The Number of Units and the Number of Correspondences] We
have seen one important aspect of the comparative method: the number of
reconstructed proto-units is independent (I) of the number of units in any of
the languages being used and (2) of the number of sets of correspondences
among the languages being used. The decisive factor is the number of contrasts
within the sets; that is, all noncontrasting sets are grouped together into one
proto-unit. Theoretically, two units from two languages can form four sets of
correspondences. This we have already seen, as is clear from the following sets,
showing Swedish and German t and d:
a b c d
Swedish
German
Now, neither the two units for each language nor the four sets by themselves
determine the number of proto-units. As already mentioned, only contrasts
count. Sets b, c, and d all occur in initial position, and thus contrast. Set a,
which occurs only after spirants, is thus complementary to the three others.
We know that there is a further set t-ts in initial position, and since this is
voiceless all the way, we would be inclined to assign set a to it. Thus all four
sets, being built up by only two units from each language, do in fact contrast,
and we need four proto-units to account for them. The choice of symbols for
them would be easy for a (and t-ts) = *t and d = *d, but rather tricky forb
and c. A provisional answer might be *t2 and *d2 forb and c, or vice versa, or
else *p and *o (either way). To choose appropriate labels, the linguist would
have to use his sense of patterning and his experience. What is important is
that the method lays out the contrasts, the relevant distinctions needed to derive
the outcomes for both languages; it unveils the relations between the units, but
it does not choose the labels for the units. This has to be done by the linguist.
Thus, although the method is rather simple and mechanistic, the linguist is
needed for the postediting of its product, as has already been noted many
times.
(11.11] We have seen a situation that is very frequent indeed: two units
from two languages building three contrasting sets. Let us look at six items from
European and Syrian Romany (Gypsy), both varieties having an sand an s,
but which form three sets of correspondences:
m·Pap mamek
European Romany
Syrian Romany rna s
'snake' 'month'
m~ and
m
This small selection makes it look as if the set s-s occurred only next to a set
e-a. In that case it would be complementary to both s-s and s-s, and we would
look for further items to see whether the hypothesis holds or not. But further
evidence shows that these sets are not complementarily distributed; rather, all
occur in the same environments and thus contrast. Now s-s can be obviously
labeled *s, and s-s as *s. The middle set s-s needs something different, and the
best strategy is to choose a phonetic symbol that is between the two, namely,
a fronto-palatal *s. The important thing, again, is that the three contrasting
sets be labeled differently, that is, by three symbols (at the same time, one tries
to stick as close as possible to the phonetics actually manifested in the material).
In the case of Romany once more we can check the reconstruction, because it
is an offshoot of the ancient language of northern India, a related form of which,
Sanskrit, was recorded quite early. The corresponding Sanskrit words are $0{
'6 ', O${ha ' lip' ($ = s), satam ' 100 ', dasa' 10 ', sarpa 'snake', and miisa 'month'.
These forms go well beyond Proto-Romany, because they point to sounds that
were not preserved in it at all, for example, retroflex f, but they confirm the
existence of three different sibilants. The conclusions of the method were basi-
cally right. But once all of Romany had, for example, replaced t by s, t could
not be recaptured on the basis of Romany alone. This is a limitation of the
comparative method. We need different independent changes in different lan-
guages to be able to reconstruct earlier stages.
Exactly the same configurations of s-sets can be found in correspondences
between South and East Slavic vs. West Slavic, Hebrew vs. Arabic, Georgian
vs. the other Kartvelian languages, and so on. The method requires three
proto-units, and it can (or should) be right, as was actually demonstrable in the
Romany case.
[11.12) Similar to the above cases is the question of the velars in Indo-
European. The most famous isogloss in the world is the one that divides these
languages in two groups, the centum and satem (see Figure 15-2:1). Centum
languages have plain velars and labiovelars (i.e., two units) corresponding to the
satem sibilants and affricates ([s, s, z, dz], for which the cover term 'coronal'
will be used in the diagram) and plain velars (i.e., again two units); but these
form three sets of correspondences:
THE COMPARATIVE METHOD 245
A B c
Centum velar labiovelar
Satem coronal velar
These sets are now abstract cover symbols whereby voice and "murmur" have
been ignored; that is, they actually embody distinctive features abstracted from
segments in which voice, voicelessness, or murmur co-occur. These three sets
are frequently labeled with three different symbols, for example (using voiceless
stops), by a palatal*" (A; see item TEN§ 11.13/Figure 11-6), a plain velar *k
(B), and a labiovelar *kw (C). Sets A and C occur in the same environments
and a contrast is consequently clear. Set B, however, occurs only after *u and
*s and before *rand *a, that is, u B r (one environment of which must be
s a
present), whereas set A does not show up in these environments. In other words,
sets A and B are environmentally conditioned variants of the same unit, and
can thus be written with one symbol, *k, again using voiceless stops. The
complementary distribution is not quite watertight, however, which is hardly
surprising in a time span of some two thousand years (the reason being apparently
dialect borrowing); and that is why many Indo-Europeanists go on writing the
allophonic variants*" and *k-exactly as we could have done with our Swedish-
English-German *h-and *-x- (§ 11.7). In the selection of words for Indo-Euro-
pean vowels (Figure 11-6), we shall see the three voiced velars, *g (FIELD,
DRIVE), *g (YOKE), and *gw (COME), where, to be sure, *g occurs before *r
against the rule, but *g after *u according to it.
This section shows that the method can be applied to abstracted features and
not to segments only. But such features have to be handled in connection with
meanings. They also complicate the bundling operations for getting back to
the segments. It is much simpler to apply the method first to segments and do
the componential analysis on the reconstructed protosegments; at the same time,
one can look back at the attested phonetics of the daughter forms. The reason
is that one also needs a knowledge of subgrouping for phonetic reconstruction
(§ 18.8f.).
1 2 1 3 7 5 c 8 6 14
7 E 9 4 9 E 10 B 4 10 A 12
v~· dMva ~:
e0 i theos
vi dua
~~u :~~:m e~~u 1:_;~:s
i u g u r u b-er
st w·
sth a t o s
st a t u s
~~i ~ w~ ~
p a t e r
w i duw6 j u k 0 0 OE r u d-u st a p 0 s fa d a r
'widow' 'yoke' 'red' 'stand-stood' 'father'
11 B 12 11 A 12 G 13 14 23
(dft)dh~(mi)
bhr~
phr a t ~0
e r
t
fr a t e r
m~ ~0
m a
m a
t e r
t
t e r
(ti)th
f
e (mi)
e (ci)
d~n~m
d 6 ron
d6num
br 6 p a r m 6 d a r (ga)d e (ps) OCS[d a r ii 0]
'brother' 'mother' 'do-deed' 'gift'
G 15 23 16 17 18 6 19
dh~m~s
th u m 6 s
fum us
m~•- ~kane
m us
m us no
~jMgeoks~
d~i~
de k a
dec em
OCS[d y m ii 0] OHGm us ns kan t e h un
'smoke-mind' 'mouse' 'one' 'increase' 'ten'
(eke)
20 D 21 22 A
g~tis
b a sis
-v en ti6
-q urn ps
~cccc m~tis
-m a tos
men t-
-m un ds
'come-go' 'un-' 'mind'
FIGURE 11-6. Indo-European sets of correspondences lined up for the
comparative method.
THE COMPARATIVE METHOD 247
Some of the consonants in Figure 11-6 are as straightforward as in the
Germanic cases above, namely, the sets for *s (Is, STAND, MOUSE), *r (FIELD,
CARRY, RED, FATHER, and SO on), *m (SMOKE, MOUSE, MIND, MOTHER),
*w (SHEEP, WIDOW). Let us now extract the three sets that predominantly
display t's, sets A (FATHER), B (BROTHER, STOOD), and C (IS):
A B c
Sanskrit
Greek
Latin
Germanic
Set C is the constant with which Germanic (in A and B) disagrees. These are
then the areas that need scrutiny. Are the three sets (A-C) complementary or
not? Set C occurs after s, set A in noninitial position before an accented vowel,
and set B after an accented vowel or in initial position. The accent is given by
Greek and Sanskrit. However, set B does occur before the accent in STOOD;
here it is part of a voiceless final cluster in Gothic. (Elsewhere a d does, in fact,
occur, as we have seen in English stood; compare further (home)stead and Swed-
ish stad 'town'.) In MOTHER, Greek and Sanskrit disagree on the place of the
accent, but we have to rely on Sanskrit alone, which matches the Germanic
forms in every case. All three sets are now environmentally conditioned (by
other such sets) and we can label all of them with one symbol, conveniently *t.
Set D is different from B only for Greek, which has s instead of t. But the s is
conditioned by the following i, and hence set D is also a variant of *t (and we
can reconstruct the starting point for Grimm's and Verner's laws,§§ 4.9, 4.10,
6.3).
Or, take set E from WIDOW, F from RED, and G from DO and SMOKE,
which give us sets that are different only because of Latin:
E F G
~
Sanskrit h
Greek th
Germanic d
Latin b
We notice easily (when we survey the total evidence in these languages) that
set G occurs initially, set F medially next to *u or *r, and set E medially else-
where. Once more the method does its service and we can label all the sets
with "one" symbol, say, *dh (we anticipate here, because there are other sets
that require the symbol *d, that is, the initial sets in GIFT and TEN). This then
is how one continues all through the phonology until one reaches units that
can no longer be combined with each other.
248 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
[11.14] The vowel sets are numbered with Arabic numerals. The uniform
sets 1, 7, and 9 clearly contrast, and just for mnemonic purposes these are thus
best labeled *a, *i, and *u. Set 3 contrasts with *a in having an o in two lan-
guages, and an o occurs also in sets 2 and 4, which are complementary to set 3
in that they occur in final syllables. The difference between 2 and 4 is confined
to Latin 0 vs. u, of which 0 occurs further without a following s, and only
after r (compare RED, and add sacer00 'sacred', vir00 'man '). Further, this r
shows up as er in this environment. But now we see that 3 does not contrast
with 2/4; their differences are conditioned by the position in the word, as has
already been noted. Hence the sets 2/3/4 do not contrast; they are all complemen-
tary, their distribution being conditioned by their phonetic environments (other
such sets). The most sensible, unambiguous, and mnemonically good symbol
for these sets is *o. In addition, set 23 is also a noncontrasting variant of this
same *o, although it brings in extra variety through Old Church Slavic. Now,
sets 5 and 6 contrast with the already established units in that at least half of
the languages show an e, but they are different within Germanic, which has e
vs. i. The two sets do not contrast between themselves, however, because set 5
is conditioned by set 8 (and other sets not included here), but 5/6 do contrast
with the other sets. For these reasons, then, we label them separately, with *e.
As for set 8, it is a conditioned variant of *i (7), although here we do need
grammatical conditioning in Latin (a final *i drops after stops in verbal endings).
We have now arrived at five short vowels:
(7, 8) i u (9)
(5, 6) e 0 (2, 3, 4)
a
(1)
But we still have not accommodated set 10 (FATHER and STOOD), which can
be paraphrased as 'i in Sanskrit, a elsewhere'. Although the majority of the
languages show a, set 10 clearly contrasts with *a, because of Sanskrit, and it
also contrasts with *i, because of the languages other than Sanskrit. By the same
token it contrasts with all the other vowels as well and thus requires a symbol of
its own. Traditionally, the symbol *a (schwa) has been used, which is between
i and a (and the other vowels as well), and which fits in the middle of the vowel
triangle.
Set 11 has long vowels only, and it thus contrasts with all the short vowels
established, as it does also with sets 13 and 14. The best labels for these are
obviously *a (11), *e (13), and *o (14). Set 12 contains an e in Greek, and e in
Latin, and an a in Gothic, but since half of the set agrees with *e, one must look
closer for a possible ultimate identity. Indeed, 13 occurs in root syllables (i.e.,
we do not count prefixes) and 12 in final syllables before *r. Hence they can
indeed be combined into one unit. Sets 15 and 16 do not contrast between
themselves, because Slavic always has y against u elsewhere, but they obviously
contrast with all the other vowels we have reconstructed, hence *u.
THE COMPARATIVE METHOD 249
It is now clear that one continues this procedure by taking more and more
material under scrutiny. Here we must leave the systematic handling of the
Indo-European evidence, with a brief reference to what we would encounter
if we continued. We would have to reconstruct an *i for which no material is
included here, as well as i- and u-diphthongs. For the latter, just two items are
given, set 17 for *oi and set 18 for *au. In some cases single vocalic segments
in Sanskrit (and Greek) correspond to a sequence of vowel plus a nasal or
liquid in the other languages, as in sets 19-22. Such sets obviously contrast
with the others already established; but is 19 different from 20, or only its
variant, or vice versa? Sanskrit and Greek agree; Latin and Germanic reverse
m's and n's. But a word-final m does not occur in oldest Germanic, nor an m
before t in Latin. Hence the sets do not contrast and must share one label.
Here Indo-Europeanists combine the Sanskrit and Greek evidence for a single
syllabic segment with that of Latin and Germanic, which point to a nasal, and
write a vocalic *m. Sets 21 and 22 show an irregularity in Latin, wh~re the i
occurs only in this negative prefix. Otherwise, both clearly contrast with *m
by showing an n in Latin and Germanic. Thus the label *tz. By the same token,
one would encounter evidence requiring us to reconstruct *r and */, and, as
in most other vowels, also length: *l]i, *8, *f, and *[. Here we shall completely
ignore diphthongs with a long initial part, for example, *oi.
One reconstructs nonobstruents for Proto-Indo-European as shown in
Figure 11-7 (the boxed-in units have no evidence at all in our selection of data).
Most Indo-European handbooks write these units as given entities. Interestingly
..
c:J
i u w
y
1 2 3 4
A Finnish kii~i tie~ iiii nei~i liih[]eii
Lapp kie htt a tie htt eo nieJ fi a liex t aD
'hand' 'to know' 'young 'to leave'
lady'
5 6 7 8
ku~u ku~un liih~ en nuo~a
ko oo a ko 5 a lieu t am nuo hh e
'spawn' 'spawn' 'I leave' 'seine'
(nom.) (gen.)
1 2 3 4
B Finnish
Lapp
jo[JJ i
jo hkk a
jal~a
juoi oe s~[;]un
so yy a
~~JO [;Jen
yy a
'river' 'leg' 'family' 'river'
(gen.) (gen.)
5 6 7 8
Finnish Lapp
dental velar dental velar
l katto sukka nuohite sohkka sohkka
II kat on kato sukan suka nuohtte kiehtta sohkka
III kadon suan kie<Sa soyya
'roof' 'loss' 'sock' 'brush' 'seine' 'hand' 'sock' 'family'
this situation is parallel to the information provided by the accent in Greek and
Sanskrit that supplied environments for grouping consonant sets together. That
is, we have the three-way contrasts as indicated by the rows of Figure 11-9. In
other words, the rows indicate phonemic identity within each language and
articulatory set (i.e., dental and velar in this material), whereas the columns
indicate morphophonemic identity within themselves but morphophonemic
difference with other columns. These columns contain the sets we found to
contrast with each other in the section on morphophonemics, but here they have
been written so that the phonetic/phonemic identity of their members match in
the same rows. The phonemic contrasts between the rows are clear, because
minimal pairs immediately establish them: kattofkato, katonfkadon, tahtommef
tahdomme 'our wish/we wish'. The table, then, is a summary and a visual
comparison between morphophonemics and phonemics and their overlapping.
Using these phonemic contrasts we must group the sets under three contrasting
units within each articulatory set: *t (Al-4), *d (AS-6), *tt (AS), *k (Bl-2),
*g (B3-7), and *kk (B8). It must be remembered that the three units (within
each articulatory group) arrived at here depend on the material used, and in no
way mirror the number of units one would have to posit with fuller evidence.
The method can never go beyond the material to which it is being applied. Only
if the selection happens to represent the total population adequately will the
result of the method have wider validity. It is always helpful, therefore, to
expand the selection until it includes all the evidence.
What is of interest in this particular Finnish-Lapp selection is that the
comparative method does assign consonant gradation to most of the words
used, as we see by writing the reconstructed units in their proper environments,
for example, (nom.) *kat V, (gen.) *kiiden 'hand', and (nom.) *suku, (gen.)
*sugun 'family', and so on. However, in one item, where both languages have
alternation, the comparative method eliminates it, (nom.) *kudu, (gen.) *kudun
'spawn'. This would be true of all those consonants that alternate in Lapp,
252 COMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
but not in Finnish (e.g., m, n, !, and so on). We should expect this here, however
(i.e., Finnish has no alternation but provides the phonetic environments for
the Lapp alternation)-unlike the case where both languages participate.
Sets 13 and 14 are complementary, and sets 11 and 14 both occur in the same
environment (medially) and thus contrast, the method now requiring two
units, *e (II) and *i (14), for example, (nom.) *kiiti, (gen.) *kiiden 'hand'.
The vowel alternation turns out to be the same as in Finnish. From these
reconstructed forms one can unambiguously derive the Lapp vowels, but there
is no guarantee whatsoever that Lapp ever had the alternation also. From the
time that Finnish (all of Baltic Finnic) shows the change of final *-e > -i, the
method assigns the result (i "' e alternation) also to the protolanguage.
Thus in both cases (consonant gradation and i "' e alternation) the comparative
method gives us alternation, if we use surface phonology (phonemes or phones),
but no guarantee of exact historical truth. This contrasts sharply with the
results yielded by the Indo-European material. But even in cases where the
method yields uncertain historical facts, it still gives us a possible starting point
from which the languages used can be unambiguously derived. And of course
such starting points will always be historically true in some sense.
show an e-stern, like kiisi/kiiden. The old paradigm was thus of the type *neisi/
*neiden, and such a nominative lurks in the Karelian compounds neiz-akka,
neiz-akku 'old maid' (with akka 'old woman'; compare English compounds
husband and shepherd, which contain old shapes, § 5.15). The i-stems are late,
and found in Proto-Baltic Finnic only; our application of the method, therefore,
had to rely on Finnish. The lateness of the Finnish paradigm also explains
the lack of assibilation in the i-stems, that is, difference in chronology (§ 4.29).
But note that in spite of the Lapp anomaly, the reconstruction of consonantism
was basically right.
Forms like liihteiifliihden 'to leave/! leave' did not have consonant gradation
in Early Baltic Finnic in the cluster *kt: *liikte-oiikf*liikte-m. After the change
*k > h gradation was analogically extended to the word. In Lapp all words
are subject to gradation in one way or another, and the Proto-Lapp paradigm
went *lekta-oekf*leyiii-m. This is a perfect example of independent drift in
both languages, triggered by the common kernel of at least phonetic consonant
gradation (see§ 9.13). In such cases the comparative method is helpless.
where the period indicates syllable boundaries, and the hyphen, morpheme
boundaries. The genitives are as expected because the syllables .det. and .vut.
are closed ( < .tel., (u).kut.), but in the nominative .de. and .vu. are anomalous,
because here one would expect .te. and .ku., respectively. But since the genitives
show that the single t of the -ton/ton '-less' must be derived from the underlying
double tt of the suffix 1-ttoma-1, one must assume that a double It was still
present in the nominative at the time when single stops gradated. The order
of gradation is thus (ignoring vowel harmony):
NOMINATIVE GENITIVE
(I) underlying lkate-ttomal Ikate-ttoma-n I
(2) gradation of single stop kadettoma kadettoman
(3) drop of final vowel kadettom kadettoman
(4) gradation of double stop kadetom kadettoman
And then at some point, final m > n. Thus Finnish still shows a clear hierarchy
THE COMPARATIVE METHOD 255
in the gradation of single and double stops. Finno-Ugric scholars use this
information correctly in assuming that the double gradation is later than *t - *d
and *k "' *g, and they generally write a phonetic difference, for example, *tt -
*lt, *kk "' *kk. The method could not, of course, go beyond the material it was
applied to. Because it had to rely on Finnish (and Lapp) disyllabic stems only,
its result was too much like Finnish (all evidence not considered here). The
moral of this is that everything in a language must be subjected to different
methods for solid reconstruction, and even so, many indeterminacies will
remain. But our treatment can be only so detailed, as our main purpose is to
get a feeling of what is involved in applying the methods and how reliable they
are.
the starting point of Grimm's and Verner's laws, and that it also automatically
provides the rules for them, that is,
REFERENCES
General: Meillet 1967a, 1926, Hoenigswald 1950, 1960a, Pike 1957, Thieme
1964, Anttila 1968, Devoto 1969, Katicic 1970, Voyles 1971, Allen 1953, Ellis
1966; 11.8 Moulton 1954; 11.12 Krahe 1970, Voyles 1971; 11.13-11.14 Krahe
1970, Szemerenyi 1970; 11.18 E. Itkonen (private communication); 11.19 Leppik
1968, Anttila 1969a.
EXERCISES
For practicing the comparative method the instructor can devise artificial
material of any complexity necessary. Such exercises are provided here as models.
Another procedure, which allows for comparison with the actual "history,"
is to split the class in two and to hand each part a word list. Depending on the
number of languages needed each group is further divided, letting students in
each section introduce changes one after the other. When sufficient change has
accumulated the two groups can switch lists. The original starting point and
the relative chronology are now documented and one can check afterward
how close the reconstruction comes. For selections from real languages, see
Cowan 1971.
THE CoMPARATIVE METHOD 257
LANGUAGE A LANGUAGE B
top 1 dip 3 tob 1 des 5
tuk 2 dop 4 tuk 2 duo 6
tes 5 diu 7 tip 3 diu 7
tuo 6 doo 8 top 4 doo 8
2. Artificial. Reconstruct the protoforms for the following forms from two
languages, A and B. Assume that any item in the list could be matched by
many others.
A B A B
1. dim dim 2. eo io
3. oen oin 4. te ti
5. dim din 6. onep onit
7. med med 8. nae nai
9. IJO IJO 10. UlJa UlJa
11. utu utu 12. oap oap
13. pep pit 14. keo keu
15. akot akut 16. IJoka IJuka
3. Artificial. Reconstruct the protoforms for the following items from two
languages. The meanings are again given with numbers only. Arrows indicate
tones, each language having two, a falling and a rising one. The tone is written
after the form, although its domain is the whole word.
A B A B
1. lJUip t lJOep t 2. pau!J t paon-l-
3. tOl) t ton t 4. tim t dent
5. kopt gopt 6. keum-l- keon-l-
7. pant ban t 8. kolJ t gont
9. mip-l- mep-l- 10. neb-l- nebt
11. tud t dod-l- 12. poig-l- poeg-l-
13. lJUm t lJOn t 14. nuom-l- noon+
15. tag t tagt 16. pe!J t pent
!7. pott bott 18. tik t tek t
19. nak-l- nakt 20. mont mont
21. kott gott 22. kaumt gaon-l-
23. toug-l- toog t 24. pumt bon-l-
25. koq kott
4. Artificial. Reconstruct the protoforms for the following items from two
languages. Note that now you have to pass semantic judgments and weed out
258 COMPARATIVE LINGUISTICS; HOW CAN CHANGE BE REVERSED?
A B
1. kiup 'sweetheart' hipu 'witch'
2. taul 'mother' palu
3. tomp 'horse' pomfi 'foal'
4. sapp 'bread' sappi
5. kruuv 'raw' hruuvi
6. riusp 'peg' riusfi 'pole'
7. ras 'glass' rasi
8. proupp 'eye' froppu 'window'
9. top 'wound' top
10. krukr 'crane' hruhri
11. vipp 'loan' vippi 'theft'
12. ramiur 'walk' ramiru
13. soik 'high' soiho
14. siok 'small' sioha
15. tau! 'painting' paulu
16. kour 'gold' koru
17. pou 'boy' pau 'table'
18. rasp 'three' rasfi
19. suuk 'lip' suhu
20. kauk 'five' kouk 'grandmother'
21. kop 'cigar' kop
22. tiut 'window' pipu 'door'
23. sik 'taxes' sik 'burden'
24. mafu 'peasant' mafu 'farmer'
25. vautt 'berry' vattu
26. nius 'rye' nisu
27. soik 'round' soihe 'oval'
28. ras 'box' rase 'chest'
5. Artificial. Reconstruct the protoforms for the following items from three
languages. Assume that the material does not contain borrowings, and so on.
Numbers stand for meanings.
A B c
1. gukura kuxra kiigra
2. Iipar lifri Iiabri
3. gaukaura kauxra kaugra
4. rutsu rupsu rutsu
5. rapina rafna raibna
6. gipina kifna kibna
7. de teke teke
8. gies kixsu kieksu
THE COMPARATIVE METHOD 259
INTERNAL RECONSTRUCTION
lJ~
pres.
pret. sg. b ai t k~s~
k au s
b~ndan
b a nd
pret. pl. b i tum k u sum b u ndum
p.p.p. b i tans k u sans b u ndans
The first two sets are of the type in divine/divinity vs. sanefsanity, or of course
like bite/bitten; that is, a diphthong alternates with a short vowel. They obviously
contrast; but then we notice that the i and u, respectively, run through the whole
set, suggesting a different segmentation:
b~· itan
b a it
k~· usan
k a us
b~· ndan
b a nd
b 0 itum k 0 usum b u ndum
b 0 itans k 0 usans b u ndans
Now we see that the first two diphthongs contrast on account of i and u and
not the alternating vowel, which now resembles the vowel set in bind. We had
264
INTERNAL RECONSTRUCTION 265
exactly the same situation with Finnish katto and sukka (§ 10.13). We further
note that the set i-a-u-u as in bind, occurs only before nasals and therefore, is
conditioned by the set n-n-n-n; it is thus a variant of i-a-0-0, however we
would label it. What happened in the above case was that the vowel alternation
tended to fuse with its environment, thus creating more possibilities for seg-
mentation. Rather than carry out internal reconstruction in this familiar way
to establish all the contrasting alternations, we shall concentrate on the environ-
ments of the variants of this one unit, that is, i, a, and nothing. In this way our
internal reconstruction has a slightly different emphasis from that presented in
Chapter 10. Such a treatment of vowels and environments means the study of
canonical forms (favorite sound shapes of morphemes, e.g., eve, evee,
evev, and so on), and we shall see that the essence of the laryngeal theory is
also a fusion of the alternating vowel and its environment. Thus this chapter
will not be superfluous after the previous one, since it will show a new side to the
flexibility of internal reconstruction.
To facilitate this new aim we simply turn around the above arrangement by
ninety degrees, which gives us
I.
II.
b~itan
k i usan bmit
k a us
bitum
kusum
bitans
kusans
III. b i ndan b a nd bluJndum b lu]
ndans
The set s- s contrasts with s-r, because both occur medially, and they require
different labels, *sand *z, or so (see§ 4.10). Internal reconstruction in Gothic
gave no hint of this. We have seen the internally reconstructed Pre-Gothic
ablaut pattern CiRC""' CaRC""' CRC; the comparative method provides a
different version:
CiiC CiC
CinC Cue
CeRC CaRC CuRC
where R means the resonants not already specified above it. Thus for inter-
consonantal n m r I, the vocalicity segment u existed already in Proto-Germanic;
and the present vowel was generally e, not i, which occurred only before another
i and n. Only the past tense forms CaRC match for both methods. If we now
apply internal reconstruction once more to the result of the comparative
method, we can easily eliminate the "aberrant" i-forms, since they occur
only with i and n. And the "extra" u in CuRC can be stripped off again. The
result is a pattern CeRC ""' CaRC ""' CRC, a combination of both comparative
and internal evidence. The principal contribution of the comparative evidence
was the present-tense vocalism e. Gothic also has e (written ai), but only before
rand h (wairpan 'to throw', saihwan 'to see', and, in the previous chapter, we
saw hairan 'carry' and taihun 'ten'); i never occurs before r and h, so e is only
a variant of i (actually there is one exception, hiri! 'come here!').
This is ablaut. Essentially there is nothing more to it (with the exception of the
environments for the variants, which we shall discuss shortly); it is only this
simple alternation in terms of paradigms or words belonging together in deriva-
tion of some kind. It should be remembered that the few examples given here
are just a suggestion of the actual material to which the comparative method was
· applied; but the method gave us the units that we handle and organize here.
Obviously the total evidence cannot be reproduced in this book. When possible,
we let Greek stand in for Proto-Indo-European vocalism. (The previous chapter
showed that Greek agreed almost to the dot with our reconstructed units.)
This is only an expedient, since Greek of course is Greek and not Proto-Indo-
European.
When this alternating unit e ,..., o ,..., 0 occurs next to a resonant we have the
same situation as in the Gothic case:
2. ei ,..., oi ,..., i (Greek pres. leipo, perf. teloipa, a or. elipon 'leave')
eu ,..., ou "'U (fut. eletlsomai, perf. ei!d!outha, aor. d!uthon 'come')
er ,..., or "'! (pres. derkomai, perf. dedorka, aor. edrakon 'see')
el ,..., of "'I (pres. ste/16, noun st6/os, perf. estalmai 'send')
em ,..., om "'lfl (pres. nemo, noun nome 'divide'; Gothic niman ,..., nam ,...,
numans 'take')
en ,..., on ~~ (minos • mind'' perf. memona, p. p. p. -matos 'think')
268 CoMPARATIVE LINGUISTICS; How CAN CHANGE BE REVERSED?
The difference from pattern 1 is that when the vowel disappears, the resonant
carries the syllable by becoming vocalic. But note, this happens only when the
resonant is interconsonantal, as in the cases here. Otherwise, the distribution of
allophonics (using again Rasa cover symbol for all six resonants) is
An example is Greek ace. patera, -patora, gen. patr6s (with *r), but dat. pl.
patnisi (with *r, because it is flanked by consonants) 'father'. We see that
pattern 2 contains exactly the same vowel alternation as 1, namely, e,..., o ,..., @.
Only the phonetic environments are different, mainly because of the allophonics
of the resonants. The resonant may also precede the alternation e ,..., o ,..., @,
for example, Re "' Ro "' F./R (Greek pres. tdp6, noun tropos, aor. etrapon
'turn'), and the resonant takes on the syllabicity when the vowel is gone,
exactly as above.
This is the overwhelming regularity against which divergencies have to be
coordinated and evaluated.
There is a pattern to the distribution of the vowels, otherwise it would not
be morphophonemic at all. Characteristically-to name only a few categories-
e occurs in presents and s-stem nouns, o in prefects and o-f a- stem nouns, and no
vowel at all in past passive participles, ti-stems, and certain aorists. Now,
Proto-Indo-European has long vowels too; and when we group the long vowels
according to these same categories, we get pattern 3:
I. e 0 (i}
2. eR oR F.
3. e 6 a (Greek pres. tithemi; noun thomas 'place ' ; Latinfaci6 'do')
a 6 a (pres. phiimi; noun phon~; ph/isis ' speak')
6 6 a (pres. dfdomi; noun do ron; Latin datus 'give')
In other words, except for a few isolated cases like *path 'father', the vowel *a
alternates with long vowels. We remember also from the reconstructed inventory
that *a is the only vowel without a corresponding long (Figure 11-7). And here
it occurs in formations where only resonants show up and the other vowels do
not. Such striking complementary distribution always points to ultimate same-
ness, and Saussure was the first linguist to juxtapose patterns 2 and 3. The
INTERNAL RECONSTRUCTION
first row of pattern 3 shows the same vowels as 2, and when we strip them away
we are left with the frames
2. R R ~ R R
3. - a length length a
That is, length occurs in those categories where the resonants of pattern 2 are
consonantal, and *a, where the resonants are vocalic. Equally important is the
fact that the long syllabic resonants occur in morphemes that elsewhere have
*a's,
2. eR oR ~
3. e 0 a
{Sanskrit noun bhavitar, p.p.p. bhuta 'to be',
4. eRa oRa jJ. janitar, p.p.p.jata 'be born'
3. - a
4. Ra Ra ~
But since the term R is a constant throughout line 4, we can ignore it for the
time being; eliminating it from the table, we get a startling configuration
3. - a length length a
4. a a a a length
which clearly shows that *a and length are strictly complementary. The only
correct conclusion is that length and *a represent one and the same thing. It is
further clear from the collocations above that this unit patterns exactly like the
resonants, and Saussure therefore posited a consonant *H (we do not use his
orthography), which gives a *a in interconsonantal position (parallel to ~) and
length in postvocalic position, that is,
H= length a 0 0
[12.4] Thus the anomaly of the vowel *a and length could be eliminated
by evaluating it against the regular pattern 2:
2. eR oR R
3'.
oH
oH
4'. eRH oRH RH
This analysis is further supported by the fact that length does not occur with the
consonants either (on the morphophonemic level) in Proto-Indo-European.
The current tendency in Indo-European linguistics is to remain with this
analysis. But there is one irregularity in the otherwise perfect pattern: the e-
column has one *a and one *o, in categories where hundreds of *e-vowels
occur. They can be eliminated only by assigning the vowel color to the *H's.
That is, instead of three vowels and one *H, we have one vowel and three
*H's: *H1o *H 2 , and *H 3 , of which *H 2 assimilates an adjacent *e into an *a,
and *H 3 assimilates an *e into an *o before lengthening operates. Mnemoni-
cally, the best way of writing these *H's is *E, *A, and *0. Now the pattern is
perfect in terms of the vowels:
3". eE oE E
eA oA A
eO oO 0
Only the vowel *e is affected by *A and *0; in all other environments these
units lengthen the preceding vowel, whatever it is, for example, also the vocalic
resonant *f!..
Proto-Indo-European root structure happily accommodates our three units
*E, *A, and *0. The majority of Indo-European roots begin with a consonant
and end in a consonant, for example, *bher- 'carry', *gwem- 'come', and
*sed- 'sit'. Pattern 3' reinterprets the long vowel roots like dhe- as this same
type: dheH- = dheE-, that is, CeC-. On the other hand, there is only a handful
of roots beginning with a vowel, and here also we frequently encounter the
anomalous vowels *a and *o; for example, *es- 'be', *ag- 'drive', *aug- 'in-
crease', and *okw- 'eye' (Latin oculus). Of these initial vowels, the *a and *o
are remarkable in that they ablaut exactly like our *eA and *eO; that is, there
is alternation *a ,...., *o, but *o as in *okw- is always *o. Here we do not need the
lengthening power of the *H's, but since they lengthen only postvocalically,
we can safely put them into the initial position, where they assimilate the
following vowel before dropping out: *Ees-, *Aeg-, *Aeug- and *Oekw-. Thus
even though the H is not physically there in the attested forms its imprint is
left on the next vowel. Now we can derive the comparatively reconstructed
vowels from the internally reconstructed sequences in three steps:
INTERNAL RECONSTRUCTION 271
small selection was taken for treatment here-and the method is the best there
is: internal reconstruction, alias morphophonemic analysis. Of course, different
scholars bring the method to different depths. As has already been mentioned,
many remain happy with one *H = *h; postulating the three *H's (*E, *A, *0)
still represents a central position, as it were; and others go on splitting the
*H's further (even to as many as ten!). But it seems that we need three and
certainly at least one. In our present discussion this particular hypothesis has
explained four' anomalies': (1) the curious status of *a, (2) length, (3) unexpected
*a and *o in certain positions, and (4) roots with initial vowels. We have to
stop here, but it can be at least mentioned that there are further difficulties that
are resolved by the laryngeals, for example, (5) the Proto-Indo-European
voiceless aspirate stops ( = *C + H), (6) the so-called Brugmann's law in
Sanskrit, and (7) the Proto-Indo-European prevocalic vocalic resonants. (That
is, contrary to the usual distribution of allophonics, one reconstructed sequences
like *C.(?.-V. But these typically occur only where the evidence points to the
preconsonantal shape *CJJ,.H-C. Thus the aberrant sequence is actually *C.(?.H-V,
in which a prevocalic, or an intervocalic,* H drops out.) Certainly an explanation
that takes care of so many different phenomena cannot be doubted, but these
last points must be left to Indo-European linguistics for a full treatment. We
have seen the principle, and that will do here.
Thus further comparative evidence, which was not available at the time of the
internal reconstruction, confirms a substantial part of the results. There is at
least one h, and it occurs exactly in the positions assumed. Problems do exist,
however : namely, that *E has no reflex beyond the fact that *e remains as e
(but compare Germanic, where the Indo-European accent leaves no evidence
beyond the alternation of d and p). Thus Hittite apparently confirms the exis-
tence of two laryngeals at least (possibly even three). Hittite orthography has a
rather poor fit with the sounds of the language, which adds difficulties. But we
can leave all this to Hittitologists and Indo-Europeanists. The important point
INTERNAL RECONSTRUCTION 273
REFERENCES
General: Hoenigswald 1946, 1960b, Marchand 1956, Chafe 1959, Kurylowicz
1964 and in Sebeok (ed.) 1963f. vol. 11, Anttila 1968 and in Sebeok (ed.) 1963f.
vol. 11; 12.1 Vennemann 1968; 12.3 Saussure 1879, Hjelmslev 1970, Winter ( ed.)
1965, Lindeman 1970, Szemen!nyi 1970; 12.4 Keiler 1970.
CHAPTER 13
[13.2) We saw a clear case of the different results of the methods when
reconstructing the Gothic and Germanic verbal ablaut (§ 12.2). We started
from Gothic forms (R = i u m n r I; exceptions are mentioned first; that is,
when r is listed above R, it is excluded from the cover symbol):
h i
Cere cue
A. CiRC CaRC CuRC
Internal reconstruction on Gothic gave us Pre-Gothic:
B. CiRC CaRC CRC
The result of the comparative method on all the Germanic languages ended in
Proto-Germanic:
i i
CinC cue
C. CeRC CaRC CuRC
276 COMPARATIVE LINGUISTICS: HOW CAN CHANGE BE REVERSED?
We see how the comparative method yields surface forms (i.e., A and C are
very similar in structure). Internal reconstruction again on Proto-Germanic
gave Pre-Germanic:
D. CeRC CaRC CRC
CRC
Pre-Gothic
i
CM CeRC [CuC]
CaRC
i
[CinC] CuRC Proto-Germanic
IR CiRC
[13.3] Schematically we can also compare and visualize the results of the
two methods in terms of the Finnish and Lapp dentals used above (§§ 10.12,
10.14, 11.15, 11.16). Figure 13-2 roughly indicates the boundaries of the units
(internally reconstructed ones are written with small capitals, and alternations
are shown by curved lines). The diagram is drawn for the disyllabic stems used
in the reconstruction and ignores the adjustment of§ 11.19, which, however, is
indicated in parentheses. Boundaries labeled a, b, and c coincide in the columns
as indicated, but others do not. It is also interesting to note that a and b set off
vocabulary items involving morphophonemic longs. Both methods happen to
give the same number of units, but only if we include a Finnish D = Jdl in
column 2, based on one word only, (nom.) sydiinf(gen.) sydiimen 'heart' (Lapp
tsiioejtsaJoam), because clear loans like jodifjodin 'iodine' must be discarded,
although synchronically they are also n-words. If we now accept sydiin as a
genuine relic of an unalternating *d, we immediately uncover an abundance of
CoNCLUSION TO THE METHODS 277
1 2 3 4 5
Finnish IR CM IR Lapp
tt tt hit
a J
r TI' r TI' I r a
J Jl
(h) II htt / /
I
b [ ]
ltl-
I
I
I
- b
t D
r---
r T r c
J I J,;
t~ ~s 6/
I
......____ I
I
I
J
r T
1/ d r
d J
D 66
d
I
r
J
-------
D 66
pre- proto- pre-
analogical cases in Finnish. That is, this relic supports the reconstruction of
. *kuduj*kudun 'spawn' without alternation and shows that the Finnish outcome
kutufkudun has shifted to the type vesifveden and neitifneidin, and so on. Such a
(single) relic often gives important information; here it enables us to distinguish
between analogic change and sound change. But this revelation is possible only
by combining internal and comparative evidence; that is, the comparative
evidence (Finnish syllable structure) showed that the Lapp length alternation
was secondary in, for example, nemmajnema 'name', kiellajkiela 'tongue',
and koooajkooa 'spawn', and that (because of the lack of Lapp stop/continuant
alternation) the Finnish gradation in kutujkudun was a Finnish development.
The one relic sydiin strongly supports this interpretation. This situation is
characteristic, and linguists would always like to see such matching in the re-
sults of the two methods.
278 COMPARATIVE LINGUISTICS: HOW CAN CHANGE BE REVERSED?
[13.5 The Structure and Domain of the Methods] Although the methods
are independent of each other as far as application is concerned, we have clearly
seen that their mechanisms are identical. All of them-phonemic analysis, the
comparative method, and morphophonemic analysis/internal reconstruction-
handle some kind of sound units in connection with meaning, whether it is
lexical (as mainly in the comparative method), grammatical (morphophonemic
analysis/internal reconstruction), or both (phonemic analysis, but also the
comparative method). Also, some kind of conditioning is stated, either phonetic
(phonemic analysis, the comparative method) or phonetic/grammatical (mor-
phophonemic analysis/internal reconstruction), either within one language
(phonemic analysis, morphophonemic analysis/internal reconstruction) or with
conditioning shifting from language to language (the comparative method).
This last fact is an automatic consequence of the makeup of the sound units of
the comparative method, namely, sets compiled from different languages.
Because the structures of the methods are equivalent, their respective outputs
are also equivalent. All methods give ultimate units from which there is a one-
way mapping relation to the lower units (or to units in different languages)-
in short, to the units from which one started. This mapping relation is generally
expressed as a given rule component of the grammar. For example, the result of
the comparative method, the protophoneme, can be mapped (rewritten) into
the attested sounds depending (first) on the particular language, and (second)
on the particular environment in the language. The morphophoneme or the
prephoneme (the result of internal reconstruction) can be mapped into phone-
(me)s depending on the particular grammatical and/or phonological environ-
ment. The more severe (strictly phonetic) constraints of phonemic analysis give
units (phonemes) with a bidirectional (one-to-one) mapping relation with lower
units (phones), that is, the situation known as the biuniqueness relation. Be-
cause the structure of the methods is the same, it will be illustrative to feed
different units and conditioning environments into it and see where the differ-
CONCLUSION TO THE METHODS 279
Input - Output
History
Number of Phonetic Condition- Name of Name of vs. des-
languages units ing procedure units cription
Pre-
IR H
Single Gramma- e phoneme
------ t- - - - - - -
tical ·aIn"' t------
Morpho-
Sets of .: morpho-
a corre- f CJ phonemic phoneme D
sponding ~ analysis
b.O
sound
-... ------ ------ ------
.5 proto-
-
units Vl CM phoneme H
In
Many
=Q Pandia-
d e Phonetic u !ectal dia- D
phoneme
c analysis
Single Phones
Phonemic phoneme D
analysis
a b
ences originate. Figure 13-3 summarizes our discussion of the three procedures;
their common mechanism is called 'contrasting mechanism'. The diagram is
not an algorithm that would automatically produce a certain output, given the
input, but rather a convenient approximation of what linguists do. Since
the mechanism is the same for all the procedures, all differences originate in the
input, which has three compartments (the columns), each allowing for two
possibilities. Now, supplied with information on the number of languages, the
makeup of phonetic units (sets or phones), and the nature of the conditioning
allowed, the 'machine' finds out whether these units contrast or not, that is,
whether these units differentiate meanings or not. All units that do not contrast
in this way are grouped together by the machine, and the linguist can label such
groups with one symbol. Characteristically, the machine handles the following
combinations:
1
· : ~ j = internal reconstruction, morphophonemic analysis
2. a e c = phonetically conditioned morphophonemics
3. de f = comparative method on morphophonemes
4. de c = comparative method, pandialectal analysis
5. a b c = phonemic analysis
280 CoMPARATIVE LINGUISTICS: How CAN CHANGE BE REVERSED?
The graphic comparison in Figure 13-3 clearly brings out the interlocking and
overlapping of the procedures. Box a is listed twice so that the diagram can show
the three most usual combinations in as many rows. The units given by the
output will then have a one-way mapping relation back to box b, and the output
units can be fed again into the machine, for example, by putting the phonemes
back into boxes e and b. Thus one usually applies the comparative method (4)
to phonemes (result of 5). The exact linguistic status of these units is not so
essential, and for Lapp we made no appeal to such questions. The essence
of Finnish consonant gradation and most English alternations, for example, is
phonetically conditioned morphophonemics (2). Internal reconstruction (I)
was applied to selections from attested languages, but it can also be applied to
a reconstructed protolanguage (result of 4), as is shown by the laryngeal theory.
Or if we applied it to the reconstructed column 3 of Figure 13-2, we would get
the same result as column 4, the Lapp state of affairs (in other words, Pre-Lapp
clearly penetrates Early Baltic Finnic). If we apply the comparative method
(4) to the morphophonemes (result of 1) of the languages used, the result (3) is
similar to internal reconstruction. This is again made quite clear by Figure
13-2. Applying it to the material underlying columns 2 and 4 we would again
end up with column 4, the Lapp state of affairs. Here then, the order of applica-
tion does not make any difference; the comparative method on columns 1 and
5 gives column 3, and internal reconstruction on it gives column 4; or, internal
reconstruction on columns 1 and 5 gives columns 2 and 4, and comparative
method on these gives column 4. The reason for this homogeneity is of course
the grammatical conditioning which always operates somewhere in the sequence
of applications of the two methods. We can even keep the usual comparative
method but allow for grammatical conditioning of the sets. Let us use three
varieties of Lapp and Baltic Finnic that have dropped the final-n of the genitive,
which had induced phonologically conditioned consonant gradation. After the
drop of -n, gradation is grammatically conditioned in all the languages:
The two velar sets do not contrast, if categorial information is used; one occurs
in connection with the nominath·e, the other with the genitive, and we can
reconstruct one velar, say *k, for both sets. If we were using phonetic condition-
ing, the method would automatically assign grammatically conditioned con-
sonant gradation to the protolanguage as soon as the last language dropped the
final -n. This would be an instance of multiple merger (exactly the same change
in all the language here), which is not recoverable by the comparative method;
the well-defined grammatical environment, however, might make us suspect
what actually happened-compare English wreath/wreathe (/8 ,...., of). The
point is that once the characteristic grammatical conditioning of internal
CONCLUSION TO THE METHODS
underlying units, that is, how to decide between a synchronic and diachronic
interpretation; such criteria must be developed for the units and not only for the
rules. Even in the absence of an algorithm here, at least there is agreement that
these notions must be kept strictly separate, because they serve different pur-
poses; this is why we label the units differently, although the names might not
be the best possible ones. Thorough philological screening is important for
the recovery of real history, and, interestingly enough, in etymological work
we often have the same dichotomy as in units vs. rules, namely, the origin
versus the history of an item. Both are important for their own sakes(§ 17.6).
Both historical methods, then, give us ultimate units as starting points for
various mapping rules. There has been a choice available in the presentation of
material: for example, some scholars tend to favor the units, reconstructed
protophonemes (independent sound correspondences); others take the units
as given and use a set of ordered rules, as many of the scholars favoring the
proto-units have done in practice.
The first one presents no difficulties, but the second one does, because it is
11-IJ.
highly characterized; the speaker has to learn something new. He has to learn
every single morpheme where he must replace his [n] with an [IJ]. There is no
simple mapping relation (rule) from his dialect to the other one, as there is in
the reverse case: a speaker with [IJ]'S can always replace them with [n]'s, once
he has learned this correspondence. It is easy to learn that all verbal forms in
-in(g) are affected. But the speaker who has to travel the one-to-many scale
tends to overdo the "difficult" (characterized) part in cases that he has not yet
thoroughly learned. Results are then "mistakes" like coffing and chicking.
Such mistakes show that sound correspondences are clearly (one would like
to say consciously) handled. The mistake is not due to method but to its appli-
cation; the speaker makes an unjustified shortcut and secures an outcome that
would look more like the target dialect. In other words, he lets some kind of
proportional analogy enter his handling of the sound correspondences(§§ 5.1-
5.3). The use of such analogy is quite characteristic of other language learning
situations also, as we have seen. Such interference is expected, because as a
synchronic mechanism the comparative method is only part of the machinery of
the speaker. In comparative linguistics, however, the method is used, as it were,
in "sterile" laboratory conditions and it can be kept from being contaminated
by other factors, at least to a much greater degree.
Thus English speakers who communicate across the American and British
dialect boundary acquire the knowledge of the following (and many other)
correspondences (we have to assume that these speakers know nothing about
writing):
A B
). re· re· mass, cad, etc.
2. re· a· dance, half, etc.
3. a· a· father, balm, etc.
4. a·.~ a· barn, harm 'foamy yeast', etc.
5. a ;') pot, hot, etc.
6. D t writer, latter, etc}
(§ 10.4)
7. D d rider, ladder, etc.
Structurally the situation is exactly the same as between mutually unintelligible
but related languages. One cannot map one way from A into B because of sets
I and 2, and 6 and 7-and not from B to A either, because of sets 2 and 4. In a
pandialectal grammar one would then, sometimes, have to rely on the distinc-
tions in American English and sometimes on those in British English. We can
now set up pandialectal units that differ from both outcomes, but from which
there is of course an unambiguous one-way mapping relation to both A(merican)
and B(ritish), for example, (I) ce, (2) ce·, (3) a· (these first three sets cover two
units in each language; compare§§ 11.11, 11.12), (4)a ·r, (5)a( = ;,), (6)!,(7)
d. This setup is now a dialect cohesion (§ 13.8) that connects the two varieties
of English, exactly as the reconstructed units connect the outcomes of two
(or more) related languages.
284 COMPARATIVE LINGUISTICS: HOW CAN CHANGE BE REVERSED?
[13.8] The result of this pandialectal analysis can be called dialect cohe-
sion, and this is the synchronic counterpart of a reconstructed phonological
system. The units of dialect cohesion have been called diaphonemes in Figure
13-3. Thus we must add another characteristic combination for the method of
treating sound units; a e c = pandialectal analysis. Structurally it contrasts
with the comparative method in that its sets of correspondences have a different
source; they are different dialects rather than different languages or different
grammatical environments in one and the same language or dialect (as in
internal reconstruction/morphophonemic analysis). It is a truism that one
cannot draw a boundary between two dialects of the same language and two
closely related but different languages. Thus it is also reasonable that the method
is the same in both areas. It is the linguist's task to try to determine when such a
dialect cohesion has no more synchronic validity but must be given a historical
explanation in terms of reconstructed protoforms. This partly depends on the
cultural situation and the alertness of the speakers, and of course we must
exclude professional linguists altogether (see§ 18.17). Children speaking, say,
Swedish and German (not to speak of more closely related dialectal varieties
like Bulgarian and Macedonian or the Scandinavian "languages") may quite
well develop a fair number of synchronic correspondences between the two,
without being able to give any historical explanation to them (and in most
cases the correct explanation is actually borrowing rather than inheritance).
The most startling case is reported for Lapp and Finnish. Although we have
seen very clear correspondences between the two languages, the languages are
definitely not mutually intelligible. They are as much apart as English and
German in this respect. Still, Lapp children can map Lapp words into Finnish
and vice versa, exactly as between dialects. Obviously they master the sets of
correspondences, though here again they cannot give them a historical inter-
pretation. Thus once more, this time in connection with the comparative
method, we get an indication of unity within linguistics, in terms of the methods
used in diachrony and synchrony.
Since we have also seen that regular sound change depends on dialect inter-
action, and that sets of correspondences depend on regular sound change, we
see even more reason for the structural parallelism between dialect cohesion
and the comparative method. Such dialect cohesion guides regular sound change
(Chapters 3, 9). The comparative method, on the other hand, starts at the other
end of the material and works backward in time. It is natural that both aspects
have the same basic structure.
The acceptance of the comparative method in descriptive linguistics has been
hindered by a fallacious concentration on uniform idiolectal grammars. Histori-
cal linguistics shows, however, that diversity is a must if we want to understand
change. And it is also a logical necessity, since these fictional unified speakers at
least understand other varieties of the language (see Figure 1-7).
A final remark on Figure 13-3. Phonemic analysis is the only method that
remains unpaired with a historical aspect, and we have indeed seen that changes
generally do not occur by phonemes (com pare § 4.12). When phonemic changes
CONCLUSION TO THE METHODS z8s
occur, they are most often the results of other changes. But phonemic analysis
is the school where the basic mechanism of the other methods can best be
learned; it is here that the structure of all these methods is displayed with
maximal clarity and simplicity.
REFERENCES
General: Hermann 1907, R. Hall 1950, Anttila 1968; 13.4 Hjelmslev 1970;
13.6 Katicic 1966, 1970; 13.7 Dyen 1963, Bailey 1969, Anttila 1969a; 13.8 Dyen
1963, 1969, Anttila 1969a.
PART IV
LINGUISTIC RECONSTRUCTION:
A SYNTHESIS OF VARIOUS
LINGUISTIC AND CULTURAL
NOTIONS
CHAPTER 14
DIALECT GEOGRAPHY
(14.2 The Rise and Impact of Dialect Geography] In 1876 August Leskien
published his famous slogan that sound laws have no exceptions, J. Winteler
published the first rigorous dialect monograph, and Sophus Bugge deciphered
the runes. The last half of the 1870s was the beginning of modern linguistics in
many respects. Also in 1876, Georg Wenker wanted to prove Leskien's claim by
checking it with the High German consonant shift (e.g., Dorpf Dorf 'village',
datfdas 'that', and makenfmachen 'make'; the first item in each pair is Low
German, agreeing with English thorp, that, and make). Proof was supposed to
come from the "pure" folk dialects, because the standard and the city dialects
were often obviously mixtures of the surrounding dialects. Wenker expected to
find a sharp boundary, on one side of which High German sounds prevailed,
on the other Low German; in other words, he hoped to find sharp dialect
boundaries in general. He failed in this, but found the isoglosses instead. An
isogloss is a line that separates an item a from not-a whether this is a word, a
phonetic or syntactic peculiarity, or whatever. An isogloss therefore allows us to
reconstruct the spread of any linguistic feature, and dialect boundaries are
defined by isoglosses, either one or more. The boundaries for the above items
show the widest separation along the river Rhine. The isogloss between datfdas
crosses the river south of Coblenz, that of Dorpf Dorf south of Bonn, that of
makenfmachen between Dusseldorf and Cologne, and the isogfoss for ik/ich
'I' reaches the farthest north, crossing the river at Urdingen. This is the famous
Rhenish Fan (an obligatory example in any book on the topic, although we
shall break the rule in not reproducing the map). The lines form a fan-like
picture, since they merge into a common stem farther east; the situation
shows that every word has its own spread. Wenker had discovered dialect
geography.
Also in 1876, two Frenchmen, independently of Wenker, wanted to find the
boundary line for Provencal vs. French. With a questionnaire of twenty words
they walked zigzag along the expected boundary; like Wenker, they found iso-
glosses, or dialect geography, rather than a neat dialect boundary. Ferdinand
Wrede extended the German study into every German-speaking village in
Europe (44,000 points), and Jules Gillieron organized what became the French
dialect atlas. Wrede had to rely on questionnaires filled out by school teachers,
whereas in France the field work was carried out (in some 640 localities) by
Edmond Edmont, a student of Abbe Rousselot, the most famous phonetician
of the day. The German survey could at least give reliable information on
syntax, which the French survey did not include at all. Other projects have
DIALECT GEOGRAPHY 291
built on the French and German (and other) experiences; for example, the New
England field workers were instructed by Jakob Jud and Paul Scheuermeier,
who had worked with Karl Jaberg on the Italian atlas. Hans Kurath, the director
of the linguistic atlas of the United States and Canada, prepared a questionnaire
of I ,000 questions. A historian picked the towns so that connections with
England could be studied as well, and twenty hours were spent with one infor-
mant, four to a locality, thus permitting investigation of age differences, and so
on. The survey began with nine field workers, but it eventually became clear
that one is better than nine, and subsequent areas were investigated by G. S.
Lowman, and after his death, by Raven McDavid. After the publication of
the New England dialect maps the survey was published in list form to cut
costs.
The results have been the same in principle everywhere: that there is always
variation, which has been described in Part II, and that there need not be
"clear" dialect boundaries (i.e., tight bundles of isoglosses). Every feature can
have its own spread. This seemed to be a complete denial of the regularity of
sound change, and the maxim sound laws admit of no exceptions (Leskien) was
replaced by every word has its own history (Gillieron, Schuchardt), especially in
the Romance field. Both positions of course have truth in them; it is too opti-
mistic to expect that only one principle would contain the whole truth about such
a complicated phenomenon as language change. The first maxim relies on a
successful result of sound change, the second stresses the spread of change,
which takes place in terms of individual words, grammatical categories, social
layers, and so on (see § 9.10, 9.11). But most important of all, the maxim
specifies words, which easily come and go, and do indeed have unique histories
(Chapter 17). The same is true of morphological elements in general, although
it is most obviously true of words. The spread of a change depends on social
forces, which can shift before a particular change has attained regularity. The
isoglosses also clearly show the role of communication and social interaction in
change. Isogloss bundles tend to cluster along barriers that impede communica-
tion, for example, mountains, swamps, lakes, and political or religious bound-
aries. Many of the German dialect boundaries follow medieval diocese limits,
splitting even single towns, and we have seen that social dialects are different
within the same geographical spot (Chapter 3). In other words, different beliefs
act "like mountains between people." From a different point of vit:w, isogloss
bundles can be looked at as lines of weakness in the network of oral communica-
tion. Although dialect geography at the outset was atomistic in orientation
(i.e., it studied single items), it led to valuable results, especially when coupled
with the study of the distribution of cultural artifacts. Cultural diffusion is
another aspect of communication. The combination of linguistic and ethno-
graphic data is known as the WiMer und Sachen technique (words and things).
The value of this method is particularly great in studying semantic change;
often we must know the exact cultural context to understand it (see Chapter 7).
The Worter-Sachen method is one aspect of philology and etymology; it is a
292 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
prerequisite for solid reconstruction (see Chapter 17), and must be employed
whenever possible.
For the realization of the first element of the diphthongs one needs mainly
the two rules: (1) a~af- vowel+ voiceless consonant, and (2) a~a?f-u.
This is the common rule pool in this sample: Charleston needs only (I) and New
Bern only (2); Winchester applies first (1), which of course gives the same lower
row as in Charleston, and then (2), which can now apply only in the upper
right-hand corner, making it the same as in New Bern; and the Roanoke
vicinity undergoes first (2) and then (1), that is, it has the same rules as Winches-
ter, but in reverse order. It has been possible to characterize four dialects with
only two rules and a specification of their order of application (of course, other
rules would be necessary for the facts not yet taken care of). The German Ding
case (§ 6.15) and the Finnish teeq case (§ 6.17) showed exactly the same situation
for dialect characterization with different rules or different rule order. If all the
rules in the overall rule pool can be hierarchically ordered in a scale from
top to bottom, dialects can be specified on the basis of where they appear on the
scale, as shown in Figure 14-1. When no reordering of rules occurs, one can
Dialect 1
Dialect 2
Dialect 3 ~-----------------1
Dialect 4
~--------------1
Rules 1 2 3 4 5 6 7 8 9 10 11 12
(Top) (Bottom)
FIGURE 14-1. Four dialects characterized by a pool of twelve rules.
simply state that dialect 3 takes all the rules from R 1 on (but skipping two),
dialect 2 from R 2 to R 10, dialect 4 would apply R 5 before R 4, and so on.
There seems to be more order in this than in defining a dialect with a bundle of
isoglosses (e.g., D 1 [I 1, I 3, I 7]), but of course it is basically the same concept.
It has even been suggested that the name dialect be replaced by 'climacolect ',
to agree with such rule scales (C.-J. Bailey). Note that these rules apply to the
underlying common units, diaphonemes or whatever one wants to call them.
We have seen many times already that such units and rules are complementary
concepts. It is not enough that two dialects share most of their rules; they have
294 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
to have common units as well (e.g., words, morphemes, and so on). The phono-
logical rules for two different languages in a convergence area may be about the
same, but no dialect cohesion can be constructed by the speakers if there are no
regular sets of correspondences (diaphonemes) (see§§ 13.7, 13.8).
r---------------------------------------------------,
earlier
y,arents
Sea
II':~
Finland\\
:
:
~
~
: U.S.S.R
md
\ ...·•
·:
[14.7 Areal Linguistics] In other words, we have again run into the
notion of correspondences and mapping relations, which are central to (genetic)
linguistics. These were able to establish direction of borrowing (§§ 8.6, 8.13).
If there is unambiguous two-way mapping within the members of the corre-
spondences, no priority can be stated. But a one-way mapping relation establishes
direction, as above. And when one cannot map either way, one reconstructs a
diasystem from which there is a one-way mapping relation to all the systems used
in the reconstruction. And whenever correspondences and mapping are involved,
some kind of comparing/contrasting is being carried out. The French dialect
maps, of which Figure 14-2 is a sample, not only led to results in dialectology in
general, but also to a whole school of linguistics known as areal linguistics,
practiced mainly in Italy, in which the units of comparison are languages,
dialects, forms, or sounds from them. With three areal norms one can in many
cases establish the relative ch~onology of the two sides of an isogloss line.
1. The earlier form is preserved in the more isolated area, for example, the
mountainous Auvergne region in Figure 14-2.
2. Lateral areas preserve the older forms, which also was obvious in Figure
14-2.
3. The larger area shows the original form, except when the minor area is the
more isolated one, or when the minor area represents the sum of lateral areas.
For example, by comparing the dental spirant of English mother against the t
in Latin miiter, Greek mdter, Sanskrit matd, and Russian mat' (see Figure 11-6),
one would conclude that t was the original sound, if English cannot be taken as
an isolated or lateral area. English is geographically lateral or isolated, but
there is no further evidence for the others' having innovated together. These
norms are handy rules of thumb, but like other methods of analysis they can
lead to incorrect interpretations; especially when coupled with the principle of
linguistic seriation in its many aspects, reliable hypotheses can often be put
forward. (For example, in the example of the English dental spirant just given,
LINGUISTIC RECONSTRUCTION: A SYNTHESIS
any hypothesis that t is earlier than () would receive strong support from the
seriation principle, because t > () is considerably more likely than () > t.)
Note that seriation itself is not a rigid principle all the way through various
subsections of grammar. In a borrowing situation one could map from the
language in which the form was motivated (iconic) into the one where it was not
(§ 8.6). But if the same situation, anomaly versus regularity, obtains across an
isogloss, it is the anomalous form that is generally older. If two dialects have
plurals like kine vs. coH·s, the former is the older one. Within the same grammar,
plurals like men and mice are likewise older than boys and horses. We have here
variations of the principle of !ectio difficilior (§ 2.17), a principle that helps in
establishing chronology. Linguistic seriation is another manifestation of it.
Areal linguistics claims that the traditional comparative method or comparison
between languages was just its forerunner, because the latter also projected
into the past forms and sounds (e.g., *tin mother) which exist synchronically in
(at least some corner of) the investigated area. Only when complex corre-
spondence relations are used (i.e., in establishing diaforms that do not exist
anywhere as such [*seft V]), is comparative linguistics different from areal
linguistics. Whether one wants to build a separate doctrine of areal linguistics
in addition to dialect geography or not, the importance of spatial distributions
for genetic linguistics cannot be questioned. We shall also see how dialect
geography does further service for comparative linguistics in helping to establish
relationship models (Chapter 15).
REFERENCES
Yurak Samoyed
Samoyed Tavgi
Uralic
(c. 4000 B.C.) Selqup
[Kamassian]
[etc.]
Finno- Ugric
Ugric
(c. 3000 B.C.)
Permian
Votyak
Zyrien
Finnic
(c. 1500 B.C.) Mordvin
Cheremis
Finnish Livian
South
Karelian Veps Votian Estonian
FIGURE 15-1. A family tree of the Uralic languages. [Based on Robert
Austerlitz's lecture notes, reproduced by permission.]
302 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
middle point. 'Indo-European', on the other hand, is based on the spread from
Europe to India, that is, it is named by its end points. The tree is also a handy
measure of relative closeness. All these languages of course have dialects,
so that one could go on adding finer and finer branches, which is one advantage
of the tree diagram. Note that the Baltic Finnic node has two nuclei. The problem
is the exact position of Lapp, but since it is so closely related to Baltic Finnic,
it seems plausible that it branched off from the upper nucleus, called Early
Baltic Finnic. Now, Lapp is usually referred to as Lapp, although the diversity
within it is far greater than within Baltic Finnic. This is why the three subgroups
are written out. On the other hand, the Baltic Finnic languages are generally
enumerated separately even though they are best regarded as dialects of the
same language. Some obscure old social reasons are seemingly responsible for
this radically different treatment of Lapp and Baltic Finnic; compare the general
American attitude (which is only now receding) that the Negro dialects are not
language at all.
[15.2 How to Draw Trees: Subgrouping) The family tree given here did
not draw itself. It is the result of painstaking scrutiny of the actual linguistic
facts, and moreover linguists disagree among themselves here and there,
because "facts" can be interpreted in different ways. It is relatively easy to
establish a family of languages, for example, all the Uralic languages in the
diagram. This is given by regular sets of correspondences. But correspondences
put all the units of each language on equal footing, and can do no more than
group the languages into a brush formation (one node), not a tree. Similarly,
it is relatively easy to see which (e.g., Classical) manuscripts represent the same
story (see§§ 2. I 7, 2.18), but determining the sequence of copying is much trickier
and often indeterminate in detail. When a number of manuscripts have the
same unusual mistake, it is clear (or rather, it is the simplest hypothesis) that the
mistake was copied from the original. The probability of repeating the mistake
independently so many times is practically nonexistent. All the extant mistakes
are thus continuations of the original mistake. In dialect geography we saw that
the same logic is used for the interpretation of lateral areas. The same form in
discontinuous localities does not speak for an independent origin for each
locality, but argues that they continue an original unity (when there was one
area only). Now, this is the principle in drawing family trees for languages. If
two or more languages share a feature which is unlikely to have occurred
spontaneously in each of them, this feature must have arisen once only, when
these languages were one and the same. The more features pile up in this way
for a particular group of languages, the firmer the conclusion is that these
languages represent an original unity, which is represented by a node in the
family tree. But note now that the establishment of a language family uses the
same principle, in terms of the linguistic sign: the colligation between meaning
and form is arbitrary, and if any two or more languages have compatible forms
(ascertained on the basis of sound correspondences) linked with compatible
meanings, and if borrowing is not likely, it is simplest to hypothesize that we
have relationship, that is, a language family, which will be represented by a node
ALTERNATIVE RELATIONSHIP MODELS 303
in the tree. This is an important criterion for subgrouping, in spite of the danger
of interference from borrowing. The more morphological signs (grammatical
markers) are involved, the better guarantee we have against borrowing.
Let us now refer back to the Uralic family tree. It was mentioned that the
Baltic Finnic languages are so close to each other that they can be taken as
dialects of one language; their variety is roughly on the order of the Romance or
Slavic languages. The total grammar of the Baltic Finnic languages is essentially
identical. Hundreds of basic vocabulary items are shared as well. There is no
possibility that all this could be accidental, and thus we assume that the gram-
mars and the vocabulary arose once only, and that the slight differences are due
to later change in each form. Let us mention one specific Baltic Finnic innova-
tion, a change that breaks away from tradition (or a "common mistake"): all
the languages have undergone the change of final -e# > -i (§§ 4.29, 6.6, 10.12,
11.17). All these languages and Lapp (except for Veps, Livian, and a variety of
Southern Lapp) have consonant gradation, and this is one of the features
defining Early Baltic Finnic, although it is not really certain in what form, if any,
it existed in the protolanguage (§§ 11.15-11.19).
The same situation as in Baltic Finnic obtains in Germanic. Here again the
grammars are obviously very similar, as well as the vocabulary (§§ 11.2-11.8).
Even the linguistically uninitiated notice the necessity for a common node in
the tree, however they word it. Consider, for example, the Germanic strong
verb system (§§ 12.1, 12.2), essentially the same everywhere. It is utterly impos-
sible that each language would have independently carried out the following
innovations: shift of the Proto-Indo-European perfect into (1) a preterite, or
(2) a few auxiliary "presents" (e.g., can, may, shall), and (3) the formation of
the weak conjugation (with the preterite formed with dental stops). Or take the
Germanic innovation of replacing the Proto-Indo-European syllabic resonant
by a back vowel -plus the corresponding consonantal resonant, *lJ. > uR (in
certain environments we have subsequent umlauts u > o). Again it is more
reasonable to assume that this happened only once, in the protolanguage,
rather than many times. As these common innovations keep accumulating, our
analysis and our reasons for positing a common node become vindicated.
was always bifurcations, that is, binary splits. This is how August Fick drew the
Indo-European family tree, and this was made popular by August Schleicher.
Schleicher first split Proto-Indo-European into Aryan-Greco-ltalo-Celtic vs.
Slavo-Germanic. The former was then divided into Indo-Iranian vs. Greco-Italo-
Celtic, the latter into Germanic vs. Balto-Siavic. Greco-Italo-Celtic gave Italo-
Celtic and Greek. Albanian somehow sprang from the stem of Greek, and as
for the remaining two-part names, their final splits are obvious. A binary
split is, of course, possible, and in fact the Uralic tree has many of them (some
linguists posit even more than indicated in Figure 15-1). This is also why the
Finnic node has two nuclei. It can be split in two by lowering Volga Finnic
down the stem. But to limit oneself to binary splits only is too stiff a requirement;
this became quite clear when dealing with closer relationships, where adjacent
dialects share features (e.g., the Baltic Finnic languages or the Germanic
languages). It is impossible, without being arbitrary, to give priority to some
features, and the "tree" remains a brush-like formation with one node only.
This is also how one has to draw the Indo-European family tree for the following
subgroups: Indo-Iranian, Armenian, Greek, Italic, Celtic, Germanic, Balto-
Slavic, Albanian, Anatolian, and Tocharian.
The drawing of trees is of course based on actual linguistic differences-in
fact, isoglosses, as in dialect geography. Instead of indicating the isogloss line
between two items, say, a I b, one concentrates on the actual derivation of the
items, a 1\ b, because this can show history and the aim of the tree is to show
derivational history, that is, the actual splits. Now, such splits need not be
absolute, because borrowing is always possible; in a convergence situation
borrowing can be quite considerable in every part o.f the grammar. Rather than
draw splits often based on arbitrary selections of features and of dubious chro-
nology, Johannes Schmidt proposed (1872) that changes be represented like the
mesh of chain mail, or like circular ripples on water, which spread outward
from the point of origin of the change. He called this way of handling the
relationship the wave theory. A few years later, when dialect geography was well
established, it became apparent that it was not mere theory, because this was the
way an isogloss map behaved. Of course the choice of isoglosses for this new
purpose can be as arbitrary as the selection of innovations in drawing trees.
Trees can represent splits, waves the actual spread of features. Both ways of
representing relationship can of course be as bad or good as the linguists make
them, within the limits imposed by the models themselves.
Tocharian
24 §
G~~j
·aQJ
E
1-<
()
7_t__:£!QJ -~ <
u
-
til
..... 4
Greek
, ____ -
I
23 9 23 13 14
12 16 10 8 14
25 1 4
FIGURE 15-2. A dialect map of the Indo-European languages.
[15.5 Synthesis of the Models] The problem with the family-tree model
was that it did not spell out the actual linguistic items defining the nodes. It did
not allow for diffusion either, and this then led to the wave-theory model. And
now we see that the isogloss map has a problem in handling chronology. It
spells out the isoglosses, but they are all on the same flat surface without chrono-
logical depth. Seriation can be used, but only for one isogloss at a time, atom-
istically and without relation to other isoglosses. In short, this leads us back
to family trees, which show time depth. This is an indication of the comple-
mentarity of the two models, a view championed by Leskien. Both diagrams
are visual aids which show, in a single picture, interrelationships within the
whole family or its subgroups. Models are, of course, icons of these relations,
but since models represent certain hypotheses, they are icons of hypotheses.
Overliteral interpretation is the greatest danger, since a picture (diagram) of a
hypothesis cannot be more accurate than the hypothesis itself. It should also be
no wonder that pictures of the same object, taken at different angles, look
different, although they have a substantial common core. And both pictures
tend to be fuzzy as well, because the object is vague as a consequence of many
historical gaps.
How would one now lift up a tree out of the flat map of Figure 15-2? One
must give priority to some isoglosses over others. Schleicher's tree (§ 15.3)
would seem to take isogloss II as the first basic division, then numbers 17 and
4, and so on. It is not claimed that Schleicher used these very isoglosses for his
tree, but the isoglosses provided in Figure 15-2 would enable one to set up
Schleicher's tree. On the other hand, if isogloss I is given highest priority, we get
the most famous subgrouping into centum and satem languages, west and east,
and this is again a two-way split, as always in Schleicher's tree. Isoglosses 22 and
24 gave rise to another two-way classification: Hittite (or Anatolian) versus the
rest of Indo-European. To stress the very deep-cutting nature of this cleavage
(22, 24, and others not given here) a name Indo-Hittite has been proposed for
the family. The term has aroused almost unbelievable rage in some linguists,
although the facts have not changed at all. The dialect map remains exactly as
before; the new name just emphasizes certain isoglosses. In this sense the Indo-
Hittite hypothesis is based on internal analysis. Of course new names often
stress old facts from a new angle; for example, rule manipulation supersedes
analogy in this way. There is traditional justification in keeping old names. It
308 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
*XYZ
0
X y z
--
0
could be rather difficult (not to say pointless) to ban the word horse and replace
it with, say, hay burner, just to stress the fact that horses eat hay. The term
horse can and does cover that as well, and similarly every Indo-Europeanist
knows that he can go on using the name Indo-European irrespective of the actual
isoglosses that split the family.
If we choose isogloss 23 as the earliest division, we get a two-way split between
Tocharian, Hittite, and Greek versus the rest. One could now postulate that after
Hittite had left the scene, Tocharian and Greek vocalized *yH (Greek yajia,
Tocharian yii), and this would give another two-way split. The fact is, however,
that the only solid two-way tree remains the Indo-Hittite one, that is, the one
based on isogloss 24. If one does not like the name or the tree, the fact still
remains that Hittite is the most aberrant dialect of Indo-European.
There is one unfortunate gap in the procedure of converting a map into a
tree. There is no single way of deciding which isogloss is basic. This is why so
much controversy arises. Linguists go basically by their feelings or intuitions,
even to the point of trying to justify the correctness of linguistic theories. Such
intuitions often brand the work of other linguists as "wrong" or "uninterest-
ing." No wonder the Indo-Hittite hypothesis has not been "settled."
F. C. Southworth has ~uggested a notation that would combine the family
tree and wave diagrams. In Figure 15-3:1 we have four isoglosses (a, b, c, and
d) defining three languages (X, Y, and Z) without time depth. This map can be
converted into a family tree (Figure 15-3: 2) if we assume that dis earlier than a
or b. This tree can be made more detailed by indicating successive stages with
bumps in the branches and by stretching the isoglosses through the time axis:
the higher an isogloss reaches, the earlier it pitched in (Figure 15-3: 3). But the
tree is of course different if isogloss a is the earlier one (Figure 15-3: 4). Two
different trees now represent two different histories for the isogloss situation
(Figure 15-3:1 ). Southworth also suggests converting the isoglosses into an
envelope around the family tree (Figure 15-3: 5-8). If an early split is not
bridged over by later isoglosses, the envelope will indicate the split accordingly
(Figure 15-3: 5). But if the early split is connected by later changes, the notch in
the envelope is shallow (Figure 15-3: 6). This now makes the distance between
the highest node and the second branching significant (in Figure 15-3, compare
combinations 5 and 6). In a brush-like tree (Figure 15-3:7, 8) the branching
distance remains the same, but the different depths of the envelope notches
indicate the absence or presence of isogloss overlap.
REFERENCES
CLASSIFICATION OF LANGUAGES
ISOLATING INFLECTING
roots Chinese Arabic
stems Samoan Modern Greek
groups Georgian
but for Subiya and Turkish, syntactic criteria were used, with concentration
on nouns for the former and verbs for the latter.
In addition to the above classification according to the makeup of the words
in relation to form and meaning, Max Mi.iller introduced the terms 'analytic'
and 'synthetic' to refer to the segmentability of units. W. D. Whitney added the
term 'polysynthetic' for cases when segmentation was especially difficult.
These terms have been used concurrently with 'isolating' or 'agglutinative',
'flectional', and 'incorporating,' respectively, because such types allow for
corresponding segmentation.
312 LINGUISTIC RECONSTRUCTION! A SYNTHESIS
When these types are combined with different techniques and syntheses we get a
three-parameter classificatory scheme as in Figure 16-1. Chinese is now a simple
1. Fundamental
type 2. Technique 3. Synthesis
A Isolating (relating words together) Analytic
B Agglutinative} Synthetic
c Fusional within words Polysynthetic
D Symbolic
or 'mildly', and so on, and compounds like 'symbolic-fusional' are used. But
the system is still unable to represent exact details; Sapir of course was aware
of it and warned that the strong craving for a simple formula has been the
undoing of linguists, and that languages cannot be pigeonholed (stated thus
already by Humboldt). But Sapir's multiple-parameter typology was a definite
step forward, and it has been extended by Joseph Greenberg to yield exact
numerical values.
,J:i
..... "'
~
C1>
Typological indices
·c:
~
"'=
1:::
JJ.I
"d
1:::
.~
"'
1-<
,J:i
.~
Oil
.....;:I
~
~
-:a
~
"'C1>
E
~
1:::
~
0
E
~ C1> 1::: ~ 1:::
til 0 p.. JJ.I >- til <r:: "'
IJ..l
Synthesis M/W 2.59 2.12 1.52 1.68 2.17 2.55 1.06 3.72
Agglutination A/J .09 .11 .34 .30 .51 .67 - .03
Compounding B/W 1.13 1.00 1.03 1.00 1.02 1.00 1.07 1.00
Derivation D/W .62 .20 .10 .15 .35 .07 .00 1.25
Gross
inflection I/W .84 .90 .39 .53 .82 .80 .00 1.75
Prefixing P/W .16 .06 .01 .04 .00 1.16 .00 .00
Suffixing S/W 1.18 1.03 .49 .64 1.15 .41 .00 2.72
Isolation 0/N .16 .15 .52 .75 .29 .40 1.00 .02
Pure
inflection Pi/N .46 .47 .29 .14 .59 .19 .00 .46
Concord Co/N .38 .38 .19 .11 .12 .41 .00 .38
i u i i u
e o e :;, o
a re a ~
e a
Traditional
Latin American English Turkish
(Triangle) (Square) (Cube)
FIGURE 16-3. Characterization of vowel systems through geometric arrange-
ments.
Fisch hat sich geniihrt 'the fish has nourished itself'); and wanftsyaks 'it-to
itself-fish-eat-present' = 'the fish eats itself' ='the fish gets eaten' (compare
Spanish ellibro se vende 'the book is sold').
Syncrisis is a generic aspect of the study of variation (Figure 1-7), and the
same tools have to be used as in dialectology: diagrams (see Figure 3-1) and
numerical parameters(§ 9.11).
Can this
Can this result Evidence
Agreement in result easily Can this or proof
different parts accident- from be in- of rela-
of language ally? borrowing? herited? tionship?
Agreement in the principles
of syntax, morphology, and Yes Yes Yes No
sound system
Agreement in descriptive
and onomatopoeic Yes Yes Yes No
vocabulary
Agreement in easily
No Yes Yes No
borrowable vocabulary
Multiple agreement in the
basic and rather un-
No No Yes Yes
borrowable vocabulary
with sound correspondences
Considerable and frequent
agreement in gram-
matical formants (endings, No No Yes Yes
prefixes, auxiliaries) and
sound correspondences
tion. Any language can be derived from a universal deep structure with omnip-
otent transformations. Thus the whole notion of syntax is rather limited if not
useless for classification. On the other hand, there is no doubt that syntax can
also be inherited. But it has not been possible to establish genetic relationship
purely on syntactic criteria. Whenever syntactic facts have been retrieved,
sound correspondences have also played a part (§§ 19.8-19.11). The fullest
treatment of the role of syntax in the question of language relationship (by
D. R . Fokos-Fuchs) comes to the conclusion that syntax alone cannot be used,
but neither should it be neglected. This is true, of course. The problem is that
syntax can be described so many ways, and for no language family do we have
commensurate syntaxes as a basis for further study. Take Indo-European as an
example. To prove genetic relationship by syntax alone one would have to
write a grammar for each language on exactly the same principles and show that
a different subgrouping from the use of other criteria would result. Then one
would have to show why this grouping is more valid than the others based on
different criteria. No such program has been carried out. Rather, all facts of
320 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
Hamito-
Nostratic Altaic Uralic Dravidian PIE Kartvelian Semitic
*t t- -t- t- -tt- t- -t(t)- t t t (t[-p])
1: •]
REFERENCES
General: Humboldt 1970 [1836], Steinthal and Misteli 1893, Finck 1910, Horne
1966, Greenberg (ed.) 1963, 1966b, and in Sebeok (ed.) 1963f. vol. 11, Uspenskij
322 LINGUISTIC RECONSTRUCTION; A SYNTHESIS
1965, Martinet (ed.) 1968, Graur (ed.) 3.493-682; 16.2 Robins in Sebeok (ed.)
1963f. vol. 11; 16.3 Hodge 1970; 16.4 Sapir 1921; 16.5 Greenberg 1954, Kroeber
1960, Householder 1960, Voegelin and Ramanujan and Voegelin 1960; 16.6
Lounsbury (personal communication); 16.7 Nida 1969, Nida and Taber 1969,
Alatis (ed.) 1968, Kazazis 1967, Uspenskij 1965; 16.8 Jakobson 1958, K.
Schmidt 1966, Skalicka 1967, Kuipers 1968, Greenberg 1969a, Hoenigswald in
Greenberg (ed.) 1963, Hodge 1970; 16.9 Brugmann 1884, Kroeber 1913,
Hymes 1959, Hjelmslev 1970, Haas 1966, Dolgopol'skij 1967, Martinet (ed.)
1968, Katicic 1970; 16.10 Fokos-Fuchs 1962, Teeter 1964; 16.11 Swadesh 1963,
Dolgopol'skij 1967, Illic-Svityc 1967, 1968, Martin 1966, R. Miller 1967; 16.12
The World: Fraenkel 1967, Martinet (ed.) 1968, Meillet and Cohen (eds.) 1952;
North America: Powell 1966, Boas 1929, Hoijer 1946, Pinnow 1964, Haas 1966
[1969]; South America: Loukotka 1968; Africa: Greenberg 1966a; Oceania:
Dyen 1965b; Europe: Lewy 1964, Bastian 1964; Soviet Union: Vinogradov (ed.)
1966f.-Current Trends in Linguistics (Sebeok ed. 1963f.) assigns the following
volumes to the following areas: 1 (Russia and East Europe), 2 (East Asia, South
East Asia), 4 (lbero-America, Caribbean), 5 (South Asia), 6 (South West Asia,
North Africa), 7 (Sub-Saharan Africa), 8 (Oceania), 9 (Western Europe), and
10 (North America).
CHAPTER 17
This is why restatements like those in Chapter 6 have enjoyed such an enthusiastic
reception; they look different, at least. At the opposite pole, the controversy led
to a historical linguistics that acknowledges only the human mind with its in-
dividual intuitions as the driving force of change. Linguistic change is just history
of the expressions of the mind, that is, art history in its widest sense. Grammar
is but part of literary history, which itself belongs integrally to culture history.
Aesthetics is the sole ruler of philology. This position is known as the idealist
school of linguistics. The appeal of the idealist position is reflected clearly in
the German term of abuse for the sterile mechanistic (Neogrammarian in the
negative sense) approach to language, Lautschieber, that is, 'sound shifter ' ,
known even among laymen. It has already been mentioned that this mechanical
approach is merely being continued as rule manipulation.
Polarization into opposite camps is quite common in the sciences. Two general
principles are mainly responsible for this, and A. Kaplan has called them (1) the
drunkard's search and (2) the law of the instrument. The first principle says that
it is easier to look for a lost key under a street lamp because "it is lighter there"
than where it was actually lost. And the second can be exemplified as follows:
give a small boy a hammer, and he will find that everything he encounters needs
pounding. It is indeed often good strategy to start the investigation "where it
is lighter"; and it is no wonder that the scientist formulates problems in such a
way that what is needed for their solution are those very techniques at which he
is most skilled. There is often considerable pressure from the scholarly community
or school as well. If the linguist is aesthetically inclined, he might see all of
language in that light; if he has been trained to observe sound shifts, he finds
this the most significant line to follow. On the other hand, if his background
encompasses rule writing, he finds mere rules everywhere and denies other
parts of language. We have all heard stories of doctors who always prescribe
the same medicine no matter what the ailment is. The beneficial effect of the
law of the instrument is that scholars can ride their ideas to the utmost, until
others pull them back. The danger is that one does not listen to others, but
brands them as unscientific (or something more colorful); the price for being
trained one way is trained incapacity to do things otherwise. It is clear that in
their behavior both the philologists and the formalists are identical. They have
different lamps and different hammers, but they certainly find that the same
objects need their pounding. Actually the philologists are rarely worsted in this
battle; the real stalemate is reached among different schools of structuralism
(including the generative-transformational approach) where the small-boy
behavior is strongest: the cry "My indoctrination can beat your indoctrination"
stands for "My father can beat your father," even when the hammers of both
parties chip away about as inefficiently.
Structuralism has not only pervaded linguistics, but also other areas connected
somehow with philology (ethnology, sociology, philosophy, and so on). Partic-
ularly influenced has been literary criticism, which has always been closely
allied with philology. The tendency to consider a literary work independently
of the total culture has been strong. Structuralism has now tended to become
PHILOLOGY AND ETYMOLOGY
is back where it belongs. Similarly, mathematics for the sake of mere mathematics
is of limited appeal in the humanities or other sciences, but as an aid it is rather
central. This does not deny the possibility of a mathematical study of language
or its usefulness, and many would indeed like to define linguistics this way.
But for genetic linguistics it would have little to offer, because language changes
when it is used by people.
the kingdom; the original meaning of the form survives in Finnish Ruotsi
'Sweden' and in the place-names containing Roslags on the coastal area of
Uppland. This stretch north of Stockholm was a pol itical unit named Roslagen,
home of the expeditions to the East. All are derived from Old Norse r6psmenn
or r6jJskarlar 'rowers, seamen'. In a similar way Normandy in France is from
norpmenn 'northerners', a name that reflects another Scandinavian invasion
and settlement (A.D. 911). Similarly, France is named after the Franks, a tribe
of Germanic invaders, and Lombardy in Italy is a modern form of Langobardia,
named after the Lango bards ('long-beards'), a Germanic tribe that invaded
Italy and settled in the area (A.D. 568). Finland and Finn refer originally to
Lapps (compare OE finnas, Norwegian finner 'Lapps'; in addition, the word
Lapp is perhaps of Finnish origin). At times the transferred meanings are startling,
as in Modern Greek Romioi 'Greeks', derived from Roma 'Constantinople',
which became the capital of the Roman empire in 330.
The Finnish adjective santillinen 'punctual' would seem to be connected with
saanto 'rule'. There are many other pairs that differ in the length of the rad ical
vowel, for example, riilea 'cool',..., t'i!u 'cold, chill', riippua 'hang' (intr.),...,
ripustaa 'hang' (trans.) and kaappa,..., kapala 'paw'. The connection with
saanto becomes less certain with the revelation that dialects and older Finnish
also have santillinen, with a back vowel. This itself does not disqualify the ety-
mology, since there are pairs with both back and front vowels, for example,
raisu,..., raisy 'quick, boisterous', loka,..., loka 'dirt ', and tollo,..., tol!o 'simple-
ton'. But the base of the adjective seems to be an adverb santilleen/siintilleen
'at the proper time', and the correct explanation is obvious in terms of Catholic
Finland in the Middle Ages, when the calendar was known by the names of
the saints assigned to the days . The base form santti is a borrowing from Swedish
sankt, itself ultimately from Latin sanctus 'saint', and it survives in dialects in
folkloristic contexts. To do something santilleen meant thus to do it on the right
saint's day, at the right time. The base santtifsantti has in turn borrowed the
meaning of the adverb and adjective and means' a punctual or particular person'
(compare fast, Figure 7-4: C, § 7.9). Change of religion (the Reformation)
accounts for the loss of the base word in its primary functions; the connection
of santillinen and saanto has no historical foundation, although for many
speakers it might have some folk-etymological meaning (similar form, similar
meaning).
The European words for 'moustache' can be traced back to Italian mostaccio
in the West and to Greek moustaki (ou = [u]) in the Balkans. The word mustaks
[miistaks], apparently 'upper lip', occurs once in Greek (Plutarch, A.D. 100),
and once it was suggested that this might be the etymon of moustaki and the
Italian word; scholars have repeated this suggestion as though it were obvious
and self-validating, despite serious problems. Most importantly, there is no
way of relating the vowels of the first syllables; Greek mustaks, if borrowed
around the end of the first century, would have given Italian *mistacchi and
Greek *mistaki. When a scholar (Maher) was finally sufficiently bothered by
the difficulties and improbabilities of the standard etymology, and reinvestigated
PHILOLOGY AND ETYMOLOGY 329
the words, a much more plausible account was formulated. The Italian word
comes from mustum 'new wine', in a derivative mustiiceus 'wine-doused', a
name of a cookie as well as metaphorically of the moustache (attested in the
seventh century). The original meaning survives in the diminutive mostacciolo,
a spiced cake, and the Italian displays the formal scheme mosto-mostaccio-
mostacciolo. Greek moustaki is a borrowing from Latin and has originally
nothing to do with Doric nnlstaks 'upper lip', although the two words were
mixed up in the learned circles. Italian mostaccio survives only in a metaphorical
meaning, 'snout' (vulgar for 'face'). For the original meaning the Greek shape
was borrowed back as mostacchio or mustacchio (see § 8.8). This account fits
quite well with what we know about naming the moustache: culinary terms are
quite common, for example, soupstrainers. And of course borrowing back and
forth can occur; compare English sport, originally from French, which has
borrowed it back. That scholars make and keep false connections like the
mustaks etymology shows that they are human, and sometimes folk-etymologize
in the fashion of naive speakers (as in siiiint6/siintillinen above).
Apparently all languages have cases of the above kind (§§ 5.5, 7.8). Words
that derive from different sources can become psychologically connected, and
words that ultimately come from the same source can be completely separated
in the speaker's conciousness. A good case is dough "'jigurejjiction "' (para)dise
(§ 8.12). The original root *dheigh- meant 'mold, give shape'. In Germanic, it
gave dough (food preparation), in Iranian and Latin, 'to mold clay' (Iranian
*pari-daiza- 'molded around', that is, 'walled garden'). When these two words
finally reached English their meanings had already shifted, Persian through
Greek into a narrower, and Latin into a more abstract and general one. Knowl-
edge of earlier building practices is necessary to see that German Wand 'wall'
is likewise connected with winden 'to wind, twist', because mud walls had a
wicker frame armature (wattle) made out of saplings, willow, or the like.
It has become clear that evidence from the material culture, archaeology, and
history may be crucial in linguistic explanations (and vice versa, of course).
[17.6 The Adjustment of Origin and History to Each Other] One of the
objections of the classical philologists to genetic linguistics, which was becoming
independent at the beginning of the nineteenth century, was the neglect of syntax
by the linguists. Syntax was the central area for the Classicists, and language
was always studied in the context of full texts well integrated with the total
culture. Here they were completely right, as was shown by semantic change,
which depends heavily on both the cultural and the syntactic context. Let us
review another case where syntax and semantics meet. In general, the Latin
accusative is the case that is continued by the Romance nouns, or more precisely,
the oblique stem that had melted together with the accusative. In some instances,
however, the Latin nominative unmistakably survives, as in Italian moglie
'wife', uomo 'man', andre 'king'; and in religious meanings suora 'sister',
frate 'brother', and prete 'priest'. These words denote persons who occur
frequently as subjects in sentences, and as titles/vocatives; for both functions
330 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
Latin used the nominative in these words. Latin (nom.) serpensjserpentem (ace.)
'snake' survives in both cases in ltalian: serpefserpente. The first (the old
nominative) apparently derives from the biblical context of paradise. The snake
is admittedly not a person, but it is personal, since it speaks. Only the nominative
serpens occurs in the Vulgate. A few other words like lampa 'lamp', tempesta
'storm', and some (often foreign) birds like struzzo 'ostrich' also continue the
nominative, because the nominative survives as various types of subject. The
area where the Latin nominative is chosen for survival (see § 22.5) is the inter-
section of the syntactic notion of 'subject/agent' and the semantic notion of
'animacy'. To understand when personification like serpens > serpe and the
other cases take place, we need the total culture as the background : here, (I) a
religious literary legacy, (2) geographic distance from Africa and knowledge of
the distribution of fauna in the world, and (3) that natural phenomena occupy
subject position in Indo-European (wind, rain, lightning, and so on). All these
cases are examples of syntactic petrification, and such change can never be
predicted totally, nor is it ever perfectly regular (see § 7.13). Note also similar
splits from English, for example, the shade/shadow types (§ 5.8) and an adjective
like glad which occurs only in predicative position: I am glad, but not *a glad
boy. Thus it is not surprising that the fringes of a feature' animacy' (i.e., subject
position) would attract a few stray forms like lampa, tempesta, and struzzo.
Lampa goes with the natural phenomena and characteristically occupies the
subject position like tempesta and is thus different from other pieces of furniture,
and struzzo would occur more likely in sentences like 'The ostrich is a big bird
that lives in Africa', and so on.
Many scholars would derive serpe from an accusative serpem, which would
presuppose a Latin nominative serps or serpis. The object here is to push the
difficulties of historical derivation to the starting point (without caring what
happens there). Here the principle of reconstruction that says that protoforms
must be reconstructed in such a way as to allow a simple derivation of the
occurring forms is pushed too far. In this approach one relies blindly on the
sound correspondences, which makes the starting point (or protolanguage) a
repository of all the difficulties. These difficulties are expressed in a multitude
of coexisting forms. Whenever there is a formal problem of derivation, one
manipulates the starting point. Linguistic literature and even handbooks are
full of this shortsighted procedure, which is actually based on the implicit
assumption that there is only one kind of change, regular sound change in
phonetic environments. Germanic verbs cognate to can have a vowel between the
velar and the nasal against the 'regular' velar-nasal-vowel(§ 8.12). The problem
is within Germanic, which has both forms (typified by knowfcan ; compare
serpefserpente). The most widely encountered solution in handbooks recon-
structs a Proto-Indo-European shape *gona- (attested in Germanic) next to
*gno- (attested everywhere, Germanic included). Now the Germanic problem
is solved, but Proto-Indo-European has acquired a monstrosity ; linguists get
away with it, because the speakers cannot protest any longer. One has to use
typological expectations and synchronize both the origin and the history for a
PHILOLOGY AND ETYMOLOGY 331
maximal fit. As it happens, can is plainly an analogical formation; serps is not
only not attested in Classical Latin, it is extremely unlikely from the Indo-
European side as well. Now, the fact that Venantius Fortunatus, a fifth-century
author from Northern Italy, actually once writes serps cannot be taken at face
value. Given the literary prestige of Latin and the gap between it and the spoken
vernacular, both amply attested, it is very likely that Fortunatus is the first
recorded scholar' to make sense' out of serpe by creating the expected nominative
serps. This is a case of hypercorrection for him, since the mapping relations
between the two varieties were known to him. A sociolinguistic explanation
based on the attested tug-of-war between a prestige norm and a vernacular
colloquial form is superior to mere speculation about unlikely Latin words.
[17.9 The Service of Etymology for Comparison) What has been shown
above is that etymology (and philology) form an integral part of genetic lin-
guistics, both in historical and comparative aspects. Etymological screening is
an obligatory prerequisite for reconstruction, to keep the protolanguage from
becoming burdened by the debris of our ignorance and high-handedness. The
elimination of English can and Welsh blif (and dozens of other similar forms)
as inheritances has a far-reaching consequence for Proto-Indo-European
morphophonemics. The corresponding roots are *gn6- and *gwela-, NOT *gona-
and *gwle-. A doubtful mechanism of alternation in the place of vowel within
the root, CeRC "' CReC, can be eliminated altogether. By adjusting the
starting points and the derivatory histories for a maximal fit, we get precision
for reconstructions (see§ 9.18).
Although experimentation is impossible in history, later independent finds
often confirm an earlier analysis (see§ 1.24). These take on the function of the
experiment in other sciences. In etymology, principle 4 serves this end. The fact
that the reconstructed Proto-Indo-European *negwhro- 'kidney' was found after
all, in a mangled form, in Latin ren reconfirms our faith in the predictive power
of reconstruction (see § 18.17).
[17.10 The Blending of Philology into Other Disciplines] One of the areas
of philology is textual criticism, which is also concerned with origin and deriva-
tion (history), but based on the text as a unit, not on the word(§§ 2.17, 2.18).
As a parallel from folklore we have the historic-geographic method of folklore
investigation, which studies folk tales (legends, games, riddles, ballads, and so
334 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
on). A large number of variants is necessary with many component parts. The
distribution of the components on the map and the internal seriation between
variants can give hints toward establishing an approximate original, its age and
place of origin, and the vicissitudes of the story. The similarity to dialect geog-
raphy is obvious-even the principle that each tale or item has its own history.
There is even a parallel in naming: areal linguistics is known as the Italian school
and the historic-geographic method as the Finnish method of folklore investi-
gation.
This short note is a final reminder that philology blends into nonlinguistic
aspects of culture. We must, however, content ourselves with this introduction
to the subject, having come full circle back to where we started the chapter
(see § 21.2).
REFERENCES
General: Breal 1893, Gamillscheg 1927, Hockett 1948a, Guiraud 1964, Ross
1965, Malkiel1968, Bakell968, Schulze 1966; 17.1 Kaplan 1964, Starosta 1969,
Garvin 1970, Vossler 1904, R. Hall 1963, Barthes 1967, Piaget 1968, Auzias
1967, Schiwy 1969, Ducrot 1968, Lane (ed.) 1970, Ehrmann (ed.) 1970; 17.2
Reid 1956, Hymes 1968a; 17.4 Szemerenyi 1962; 17.5 Malkiel 1968, Maher
1970c, Nirvi 1969; 17.6 Szemerenyi 1962, Maher 1969c, 1969d, R. Hall 1969,
Romeo 1969; 17.7 Breal1893, Szemerenyi 1962; 17.8 Malkiell968; 17.9 Anttila
1969b; 17.10 K. Krohn 1926.
CHAPTER 18
RECONSTRUCTING PHONOLOGY
There is always a danger of positing too many zeros for the matchings, and
this is why one can start with bigger sequences, for example, matching -ains
with -an in I. Here, however, ai matches a also in II, and n-n recurs many times
among the numerals so that a matching s-0 is rather obvious. But in III it is
better to match ija-eo, rather than i-eo + j-0 + a- 0 , or i-e + j-0 + a- o, and
so on. It is too early to find where such zeros fit in best. We can now extract the
following matchings:
I. 1. ai-a
2. n-n See pairs in VII, IX, and X
3. s-0
II. 4. t- t See pairs in IV, VI, and VIII
5. w-w
RECONSTRUCTING PHONOLOGY 337
Ill. 6. p-p VIII for jJ
7. r-r IV
8. ija-eo
IV. 9. f-f V, VII, IX
10. 6-e
11. au-ea [No matching for -id- and -eo-]
12. g-g
13. a-a
14. d- d
15. iu-eo kiusan-ceosan 'choose', triu-treo 'tree'
16. k-c
17. s-s VII and IX
v. [No matching for the medial. Closest material is nasal
followed by voiceless spirant]
18. VN-V munps-mup 'mouth' (see§ 4.16)
19. m- m lamb-lamb
20. u-u fulls-full, un--un-
VI. 21. ai-io
22. hs-x wahsjan-weaxan 'grow' and VIII
23. e-1£ merjan-miiran 'preach', swes-swiis 'one's own'
24. j-0
25. a-ea
VII. 26. U-0
27. b-b bileiban-belifan 'remain'
28. b-f
29. i-e lisan-lesan 'gather'
30. ei-i -leipan-lipan 'go' and IX for u-o
31. i-eo
32. 1-1
VIII. 33. h-h
34. au-a
35. ai-e
36. au-o saurga-sorg 'sorrow' and VIII
37. a-0
IX. 38. i-i [OE -g- cannot be matched with Gothic 0]
X. [Gothic -aihu- and OE -ie- remain unmatched]
Four pairs of elements could thus not be matched at all:
IV. -id- -eo-
v. -im- -i-
IX. -0- -g-
X. -aihu- -ie-
In spite of the obvious close relationship between Gothic and Old English four
sequences remain without a match. But, otherwise, the situation is clear: there
338 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
must be a genetic relationship underlying the matchings, and now one would
carry on with the comparative method to see which matchings can be combined
into one proto-unit. The great diversity of vowel matchings especially is due to
the various Old English umlaut phenomena, and it would eventually come out
that these matchings are indeed environmentally conditioned correspondences.
When the linguist starts to talk about correspondences he is already making defi-
nite historical claims (e.g., borrowings have been weeded out, clusters versus
single units have been decided, and so on).
[18.4] The ease with which one can establish a closed circle of matchings
within some kind of basic vocabulary is a quick practical measure both of the
degree of genetic relationship, and of the possibility of additional reconstruction.
Of course the significance of matchings depends on (I) the length of the words,
(2) the phonemic inventories, and (3) the number of words. If we have just a
few matchings in a few short forms or suffixes in languages with "poor" inven-
tories, we do not have a good case for relationship, and further comparative
work looks unpromising.
RUSSIAN- GERMAN
I. t-d tam-dart 'there', brat-Bruder 'brother', ty-du 'thou', tri-drei
'three', togda-dann 'then', (e)tot-derfdies- 'that/this'
2. t-t stojat'-stehen 'stand', mat'-Mutter 'mother'
3. d-t den'-Tag 'day', segodnja-heute 'today'
4. d-ts sidet'-sitzen 'sit', desjat'-zehn 'ten ', serdce-Herz 'heart'
5. d-s voda- Wasser 'water', edim-wir essen 'we eat'
6. s-s syn-Sohn 'son', est' -essen 'eat', sidet'-sitzen 'sit', sest'-setzen
'set'
7. s-h segodnja-heute 'today', serdce-Herz 'heart'
8. i-g moino-mog!ich 'possible', leiat'-liegen 'lie'
RECONSTRUCTING PHONOLOGY 339
9. ts-xt moc'-Macht 'power (might)', noc'-Nacht 'night'
10. p-f pjat'-funf 'five', pro-fur 'for', polnyi-mll 'full'
11. b-b brat-Bruder 'brother', ljubit'-lieben 'love', byt'-iclz bin 'be'
12. k-L' kto-1rer 'who', kotoryj-ll'elch 'which', kogda-1\'ann 'when'
13. r-v roda-Wasser 'water', dra-zll'ei 'two'
14. r-0 drer'-Tur/Tor 'door', twj-dein 'your'
15. l-l ljubit' -lieben 'Jove', lezat'-liegen 'lie', polnyj-mll 'full'
16. m-m moj-mein 'my', moc'-Macht 'power', mat'-Mutter 'mother'
17. n-n ne-nein/nicht 'not', nicto-nichts 'nothing', syn-Sohn 'son', nos-
Nose 'nose', nu-nun 'now'
18. r-r tri-drei 'three', pro-fur 'for', brat-Bruder 'brother', serdce-Herz
'heart'
19. e-e 6
20. a-u I, 2
21. o-a 9, 17
22. 0-0 I 0 and solnce-Sonne 'sun'
23. e-ai 17 and cel}j-heil 'whole', xleb-Laib 'bread'
24. e-a 3 and vera-wahr 'true'
25. u-i 11 and sjuda-hier 'here'
Not only do matchings occur among lexical items, but also in pronouns and
the like (which are short forms):
m- m- '1st sg.'
t-Jtt.:- d- '2nd sg.'
n- n- 'negation'
to- da-jde- '(th-) deictic'
k(o)- wa-Jwe- '(wh-) question'
-es' -(e)st '2nd sg. pres.'
-et -(e)t '3rd sg. pres.'
-em -en '1st pl. pres.'
-ete -(e)t '2nd pl. pres.'
[18.6] These matchings show that Russian and German are plausibly
relatable on the basis of contemporary evidence and that Schleicher was too
cautious. Indeed, a similar test between Russian and French also yields a
positive result, although a meager one. No actual reconstruction is readily
possible in either case, but the evidence for distant relationship is substantial.
Thus the matchings between Russian and German and Russian and French
give tangible reality to the possibility of ascertaining distant relationship even
in cases of languages with no recorded history.
Two ways of using basic vocabulary as the starting point of comparative
work have been presented. A third practical beginning is the glottochronological
list (the Swadesh Jist), whose items were originally selected so as to minimize
the likelihood of borrowings (§§ 22.13, 22.14). It gives the linguist two hundred
items among which matchings should show up, if they occur at all.
Saek ki~'
Siamese ki n 1 Iii~'
li n 4
vi~'
fo n 5
re~'
he n 5
vee~'
khwu: n 5
etc.
Saek
pe~' rU~'
Siamese he n 5
hi n
va~'
fa n 5 1 Y"~' ruu~'
kh;n n 4
rJJ n 4
etc.
[18.9) In phonemic analysis one starts from phonetics to get to the pho-
nemes (Chapter 10). In reconstruction one often makes phonetic inferences only
342 LINGUISTIC RECONSTRUCTION; A SYNTHESIS
[18.10] One should note that phonetics has great heuristic value in the
actual analysis, that is, in the application of the comparative method. Sets
that contain the same or similar sounds are likely candidates for inclusion into
the same protophoneme. We saw this in the Germanic reconstruction (§§ 11.5-
11.7), where sets were arranged according to the phonetics contained in them.
We saw also that this phonetics helped determine the choice of the symbols
for the reconstructed units. This is the general approach in reconstruction.
Phonetics does enter into the initial stages in an implicit manner, and as a final
touch to the reconstruction one returns to it explicitly. In phonetic reconstruc-
tion, as in any other kind of reconstruction, one tries to minimize the steps in
derivation. For derivation one needs the subgrouping (family tree), and this is
why phonetic reconstruction must be done as a final touch. One simply distributes
the actual phonetics of the sets of correspondence to the proper branches of the
tree. Figure 18-1 shows a family tree for four languages. A set of correspondences
A-B-C-D has been aligned with the proper branches. But before we can proceed
we have to paraphrase briefly two basic principles of the algebra of classes,
namely, Boolean algebra.
Here we need only the basic notions of addition and multiplication of classes.
If we define a class that contains all women, and another that encompasses all
Americans, the addition of the two yields a class that contains either women or
Americans. If we let A represent the class of all women and B that of all Ameri-
cans (Figure 18-1 ), we can present the sum of them as in stage 2, A + B. There
is an overlap in the middle-obviously those members that belong to both
classes, who are both American and women. This overlap is the product or
intersection of the two classes. Addition is an either-or relation, multiplication
both-and. The peanut-shaped shaded figure A + B represents the sum of the
two classes A and B. The checkered elliptical leaf shape in the middle is the
intersection. The two facing crescents represent non-American women (hori-
zontal lines) and American men (vertical lines), but the crescents are part of the
sum. A sum need not contain an intersection; for example, the sum of a class
of all stones and all songs is a class with either stones or songs, without any
RECONSTRUCTING PHONOLOGY 343
(A+ B)(C +D)
3. Intersections
(both-and)
2. Sums (either-or)
intersection. It seems that all students are implicitly aware of these Boolean
notions of addition and multiplication; the former gives disjunctive definitions,
the latter conjunctive. The phoneme as a family of sounds, say English /k/ as a
class of [k, K, kh, and so on] (see§ 10.2, Figure 10-1), is disjunctively defined:
fk/ is either [k] or [k'] and so on, that is, a sum. Conjunctive definition gives
the bundle of distinctive features; that is, /k/ is an intersection of [voiceless],
[velar], and [closure] (both-and). Striving after conjunctive definitions is central
to linguistics (see§ 20.7).
Now we are ready to go back to stage 1 in Figure 18-1 and the set A-B-C-D.
Let us assign the value p-f-p-b to it, which is simple enough for exemplification.
When one ascends the tree one takes the sum of what occurs below for each
node. Thus for stage 2 we get A + B and C + D. The shaded areas represent
now: horizontal-line crescent = [closure], checkered leaf= [labial] + [voice-
less], vertical-line crescent = [friction], rising-line crescent = [voiceless], dia-
mond-checkered leaf= [labial] + [closure], and falling-line crescent = [voice].
But we need not dissect this far because we need just the sums. Since the tree
continues higher up, we carry both sums to the next node (for better legibility
we round off the peanut shapes into circles). When we reach the top node we
take the intersection of A + B and C + D, the shaded leaf shape, which means
[labial] + [closure] + [voiceless], that is, *p. The right-facing crescent ([fric-
tion]) and the left-facing one ([voice]) were discarded. The result is the obvious
one, obtainable through intuition, but Boolean algebra makes it explicit. Because
the method handles one protosound at a time, it does not give the answer all
by itself. One still has to consider each intersection in relation to other intersec-
tions; that is, the contributing factors are (1) subgrouping, (2) intersections,
344 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
and (3) relations between intersections. The last point allows a considerable
share in the decision making to typology and language universals, so mere
Boolean algebra can be overweighed (compare the Saek-Siamese case).
[18.11] Let us assume we have a case of three languages that show a set
g-r-h. Let us consider three possible trees:
1/A
g r h
In case I it would come to an intersection right away. If r is assumed to have
been uvular [R] at an earlier point, it could be derived from [y]. Continuancy
wins (2 to I) over closure; the best candidates are *y or *g (voicing also wins
out 2 to 1). Now we would have to look at other intersections. If there is one
with voicing, closure, and velarity, that one has a better claim for *g, and we
pick *y for this one. Each language requires now a single step, Verscharfung (g),
'rhotacism' (y > r), and devoicing (h). In case II continuancy occurs over the
deepest split which makes *y more likely at the outset, and in III velarity has
the same position making *g a good candidate, since spirantization would
occur only once in the subgroup r-h. Here we see clearly the importance of
comparing the intersections. In every case if there exists another set with (velar)
stops throughout, *y must be chosen for g-r-h.
First stage p t k
Second stage P
A B
tl\d(h)) (\
k y( g(h))
Third stage
/A
p m v t d
I r 0 1 n r j
/~
v k 0
FIGURE 18-2. Derivation trees for Finnish consonant gradation.
v j lJ
LINGUISTIC RECONSTRUCTION: A SYNTHESIS
of checking our inferences (see §§ 11.8, 11.9). If possible, one should always
consult any earlier evidence, and earliest Finnish records show that d was a
spirant, written dh or d, and that instead of a v for the velar *y there was a
written g!z or g. In fact, one Western dialect still retains [o] for d; some other
Finnish dialects do not have d at all; instead various "continuants" occur
(j, v, I, flap f) or else 0. The variant d of Standard F innish is thus a late spelling
pronunciation (on the Swedish model). We have been able to posit a unified
phonetic stage for the weak grade and get the relative chronology given in
Figure 18-2. (The outcomes of the second-stage spirants do not necessarily
represent direct continuations as seemingly implied by the lines. Some are no
doubt transition sounds filling the gaps left by the loss of the spirants. But such
transition sounds generally observe "natural phonetics" from the environment.)
~
Do matchings or sound KNogenetic
correspondences emerge? elationship
Yes l
K
Unit reconstruction
Apply the mechanism of Fig.l3-3.
Is it possible to reach units in a Distant
0 straightforward way (clear
relationship
environments, etc.)?
~----------------------
Next simple unit solution
Yes~
--
Rule Formulation No
Can all the daughter forms be ~
Yes l
Devise the simplest set of rules
0 and incorporate them with a
corresponding family tree
~ Additions
Any universal/typological Yes
constraints (areal, ~ Filling gaps
physiological, etc.)
Yes I !No
Subtractions
Protoforms (units), rules and relative
chronology, and the family tree
Output
FIGURE 18-3. Flow chart for the basic steps in the reconstruction procedure.
LINGUISTIC RECONSTRUCTION: A SYNTHESIS
between steps 3 and 4; step 6 also participates in this extensively. The most
eloquent example of this has been the etymological adjustment of origin and
history of an item (§ 17.6).
We have also seen how the "aberrant" members could be unified under one
"expected" invariant unit. If the" straightened-out" unit clashes with a previous
straight one (e.g., English f-v as *f clashes with f-J, which should be an *f by
better rights; the same is true of Finnish p-v as *p, or Greek p-t-k as either
*p, *t, or *k), one has to choose other labels, for example, English *j2 , Finnish
*p2 , and Greek *kw, where the numbers can then be interpreted with different
chronology, and so on. The above tabulation shows the same relational setup
as analogy, which also attacks forms that get out of line, for example, English
fowozf oaths is being replaced by fow9s/ (analogy, of course, does not provide
subscripts). This proportion in analysis is of course one side of the proportion
in change (see§ 9.2). The difference is that the linguist gets a historical pre-form,
whereas analogy gives a new synchronic form. Analogy is future-oriented,
internal reconstruction past-oriented. But both are based on clashes between
morphemic and semantic structure (see§ 9.19).
REFERENCES
RECONSTRUCTING GRAMMAR
MORPHOLOGY
'BE' 'CARRY'
1st sg. esmi bhero
3rd sg. esti bhereti
provides at once a third singular *-ti (see also Sanskrit aniti, § 12.5) and a first
singular *-mi. In this partial paradigm of 'carry' the subtraction of *-ti leaves
a stem *bhere-, but additional evidence would show that the root here is just
*bher-, so that *bher-e- is the present stem. Now, whenever this *-e- occurs,
the first singular is *-o and not *-mi; *o is a morphophonemic alternant of
*-e- (as *e "' *o ablaut), which leaves length as '1st sg.': *bher-o-H (*-o- also
occurs in 1st pl. *bher-o-mes and 3rd pl. *bher-o-nti). Because of the a-a corre-
spondence in the perfect, for example, Greek oida: Sanskrit veda 'I know', one
351
352 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
can further identify the *Has probably or possibly *A (see § 16.3). Chapters 10
and 12 have already shown how morphophonemic analysis, alias internal
reconstruction, combines morphs into invariant morphemes, or even into
canonical forms such as by rewriting *estt as *Eesti (with an initial consonant;
§ 12.4).
[19.2] This normal analysis now yields morpheme boundaries and the
sequences of morpheme slots, for example, *es-ti, *bher-e-tiwith ROOT + [e + ]ti.
We get systematic inferences about morpheme order, that is, word internal
syntax. There is always the danger than any particular combination of morphemes
never occurred in the protolanguage, even if it is present in the daughter lan-
guages. But it still means that our sequenltial formula for combination is valid,
since it must be responsible for the independent combinations in these daughters.
The problem here is that reconstruction is always positive; we have no sure way of
reconstructing the absence of something, ~:xcept for trivialities like being certain
that there was no word for airplane in Proto-Indo-European.
This is how one continues the reconstruction of morphology. It is simply
internal reconstruction and normal synchronic linguistic analysis, once the
comparative method has provided enough substance to be handled. The differ-
ence from synchrony is that the results are far less certain, but in the previous
chapter it was pointed out that even synchronic analysis often violates native
psychological reality. Again we have reached a juncture with direct access to
descriptive linguistics; those who want to brush up their knowledge oflinguistic
analysis have to refer to the references given (for§ 1.1).
Types A-C are no longer productive; type D is, with no formal difference
RECONSTRUCTING GRAMMAR 353
between noun and verb. Similarly, nominal derivatives like song from sing and
bond from bind are unproductive as opposed to nouns like a run, a slip, and so
on. Suffixes like gift from give, birth from bear, and filth from foul, are also
limited in their range of combination. Among causatives an unproductive type
E contrasts with an analytic one F:
E. drink drench F. laugh make laugh
sit set run (make) run
lie lay X make X
fall fell
In fact, the synchronic connection between the columns in E has broken down
and is purely historical. The analytical type is unlimited (e.g., make sit, and
make lie). In English, then, generally, the more productive a type is, the less
variation or fusion it has. The plural formations repeat this principle, for
example, ox-en (A), mice (B), calve-s (C), and faith-s (D). Calves has the analytic/
agglutinative plural marker, but the variation f"' v and the unproductivity of
it makes the type older thanfaith-s, for example. Note that by the same token
warmth should be later than filth, for example, because it has less variation than
the latter (-th normally carries umlaut). It should be one of the last forms
created before the suffix lost its productivity altogether.
We have, of course, seen (§§ 4.2, 4.5) how the umlaut and other alternation
phenomena are almost always due to the loss of a suffix (B, C). Now we can
also match type D a fish-to fish with a plural type a sheep-many sheep. There
is no alternation and no suffix. Is sheep then a recent type also? No, because it
is not productive. It is peculiar to a few animal names, particularly in hunting
or catching situations; that is, they are now virtually mass nouns like water.
This is a semantic area that correlates well with earlier "cultural" practices.
[19.4] In contrast to English, Finnish does not have a scale that equates
alternation with non productivity:
A. NOUN VERB
sy!ke- 'saliva' sy!ke- 'spit'(§ 10.13)
tuule- 'wind' tuule- 'blow'
kuse- ' urine' kuse- 'urinate' (§§ 8.3, 9.3)
tuke- 'support' tuke- 'support'
su!a- 'liquid, unfrozen' su!a- 'melt'
lohko- 'portion, section' lohko- 'split, partition'
tahto- 'want' tahto- 'want'(§§ 4.23, 10.13)
B. STEM 1 STEM 2
sure- 'to mourn' sur-u- 'sorrow'
pure- 'to bite' pur-u- 'chewed pulp'
sana- 'word' sano- 'to say'
liittii- 'to join' liitto- 'alliance'
354 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
C. NO CONSONANT GRADATION
peti 'bed' pupu 'bunny'
muki 'mug' /aku 'licorice '
auto 'car' kapu ' captain'
Finnish has a rich derivational apparatus, a few instances of which are given
in B, without indication of the direction of derivation. In general, nouns and
verbs, verbs and causatives have different stems. ln opposition to this type we
have A, which is a curious anomaly indeed; there is no difference between the
nominal and the verbal stem. We have already seen, however, that the verbs
show consonant gradation in some cases where the nouns do not (§ 10.13).
Otherwise consonant gradation prevails in A and B, but is absent in C, where
the meanings represent objects of material culture (typical of loans), slang words,
and hypocoristic words (second column). Type D is characterized by much
more technical semantics (plus some slang: jobi) correlated with the phonetic
peculiarity of voiced stops and complicated consonant clusters. Such stops
belong only to innovative dialects and are not used by all social or regional
dialects. This type is thus a clear newcomer. And so is E, which displays the
most complicated kind of alternation in which consonant length is coupled
with vowel length (except that voiced stops do not alternate). The extremely
technical meanings point to recent date, and the type is productive for such
Greek terminology.
The Finnish situation thus shows relative growth of suffixing (A to B) and
lessening of consonant gradation (A and B to C), as well as formal features
which are integrally tied to the functioning of the language (D and E-a kind of
derivational consonant gradation with gratuitous vocalic complications as well).
Although much in the evaluation of both English and Finnish alternation
RECONSTRUCTING GRAMMAR 355
matches, there are some characteristic differences. First of all, the existence of
alternations or derivatory suffixes alone does not guarantee the chronology.
One also needs information on productivity and semantics. English started
with stem variation and ended up without it (a fish-to fish, D), Finnish is the
reverse (A is clearly a relic, the list here including practically all the existing
stems of this type). In the plural, however, English showed an old type without
variation (one fish- two fish). Finnish consonant gradation is alternation belonging
to the oldest types, and then again a recent peripheral pattern (E) has similar
alternation. An inference that one would be likely to draw from the English
and Finnish configurations is that English has lost flectional apparatus and that
Finnish has acquired it. Basically this is quite correct, with the difference that
for English the time span here is about a thousand years, for Finnish perhaps
four thousand. We shall not present the comparative and documentary evidence
that confirms our analysis, because the purpose was to show what kind of
inferences can be made from internal evidence. Moreover, much of the English
material has, in fact, been mentioned earlier.
Finally, it should be noted that these inferences about relative chronology
concern morphological types or patterns only and not the individual members
or items, because any word can analogically shift types (e.g., English cow from
B to D or oath from C to D). Such shifts occur usually toward the more pro-
ductive (newer) types. These rather systematic inferences concerning the relative
chronology of morpheme types are quite parallel to the reconstruction of the
sequences of morpheme slots. In the latter case we also faced the uncertainty of
not knowing whether any particular morpheme had in fact been combined
with some other morpheme in the protolanguage (to which the slots themselves
can be assigned).
SYNTAX
[19.5 Syntactic Change] It has been generally agreed that the unit of
speech is the sentence (only now is this conception changing in favor of a whole
text, a discourse, as the basic unit; see§§ 19.8-19.11, 20.1-20.6). So far relatively
little has been said about sentences and syntactic change in general. First of
all, there is the practical problem of space: to characterize syntactic change one
would first have to present a great deal of background information. It has
always been easier to present change in terms of the various units that build up a
sentence, because without units there would be no sentences either. This is a
simple and obvious point that is easily forgotten. The arbitrariness of the lin-
guistic sign resides in these building blocks of the sentence and not in the purely
syntactic rules, which are heavily iconic. Because syntactic change is analogical
it interferes heavily with reconstruction. This creates the following paradox:
since syntax is iconic, it is an important factor in change (which is largely
iconic) but a peripheral aspect in reconstruction (which requires symbolic
factors). Both aspects have already been mentioned. The first has been implicit
LINGUISTIC RECONSTRUCTION: A SYNTHESIS
The plural and the singular match perfectly except for the Z (menee). We have
often seen that such irregularities are prime targets for analysis, exactly like
analogy (§ 18.16). We have immediate success here, since archaic and poetic
Finnish goes poika menevi in the singular. A v shows up where expected, and
Z can be replaced by X", in which only the final vowel deviates from the pattern.
Now it would be very attractive to be able to combine -va and -vi, and the sim-
plest way is to posit grammatical conditioning of sound change: a final -a in
predicative position yields -i. And now X" can finally join the pattern as X'.
Predicative position is rather prone to provide change (see § 4.24), and it is
quite common for a nominal sentence to give a verbal tense. The participle
became first a verb syntactically, and then (with sound changes) also semanti-
cally. Earlier we saw that the accusative singular of the participle produced
an uninflected form(§ 5.17).
Thus this evidence, and that in the Finnish type A, support each other:
both point to a lack of differentiation between noun and verb in Pre-Finnish
(§ 19.4). And indeed comparative evidence makes it likely that this was true of
Proto-U ralic.
Latin have compound verbs with more or less idiomatic meanings, for example,
Greek en-ktesasthai 'to acquire possessions in a foreign country' (to buy in)
and hupo-thesthai 'to take a mortgage (to place under), Latin oc-cidere 'to kill'
(to strike against), and con-surgere 'to ris1~ up together'; these can be repeated
in the discourse without the preverb, but the meaning undergoes no modification.
We have here a kind of syntactic ellipsis or haplology (AB 'X' ... AB 'X' >
AB 'X' ... B 'X'). Now Latin could perhaps have borrowed this rule from
Greek; but it is very unlikely, because it occurs in the earliest legal language of
both languages. According to the three-witness requirement, we could not
assign this deletion rule to Proto-Indo-European. But when we see that Hittite
has the same rule (e.g., [appa] pai 'gives back' and [ser] sarnikmi 'I make
restitution') we have the third witness, and the rule can be assumed for Indo-
European. Counting the number of witnesses is not a reliable criterion, however;
a deletion of the above kind would seem to be quite natural in any language
having similar verbal compounds. In syntax one needs the total situation far
more urgently than in other areas of grammar, and below we shall see that
even a single witness can be crucial.
The formal similarity of the last two columns to the pronoun *so-f*to- is obvious.
There must be a historical connection. The best inference is that the Indo-
European pronoun is a fusion of the earlier sentence connective plus the enclitic
pronoun (isogloss 24 in Figure 15-2). It is now possible to see a natural reason
for the reinterpretation of *so as a nominative (no change in subject), and the
less frequent combinations *t(o) + os and *s(o)-om were lost. (But Old Latin
has sum 'him', sam 'her', and sos 'them'!) The principles of linguistic change
tell us how much easier it is to merge independent units rather than split them:
for example, English not could not have split into nan wiht, so the development
must have been the reverse (§ 9.8). The Hittite facts have to be interpreted as
the lectio difficilior, and we see how syntax has crystallized into morphology.
Earlier we saw how particular sentences can freeze into adverbs, that is, words
(willy-nilly ).
LINGUISTIC RECONSTRUCTION: A SYNTHESIS
The relation between these endings is best known through the difference between
a present *bhere-ti 'he carries' and an imperfect *e-hhere-t 'he was carrying'.
Sanskrit has also a verbal form called 'injunctive' which does not indicate
either tense or mood, for example, (md) bharat 'he (may not) carry' (*me
bhere-t). When one lines up all these forms one gets a clear segmentation of
forms:
SANSKRIT PROTO-INDO-EUROPEAN
bhara-t-i bhere-t-i
a-bhara-t e-bhere-t
bhara-t-u (imperative 3rd sg.) bhere-t-u
md bhara-t me bhere-t
Generally, prefixes and suffixes exclude each other, and it is quite obvious that
the -i of the primary endings is a temporal suffix added to the person marker.
Against this complementarity of suffixes and prefixes we have a type with both,
as in Sanskrit pra-hharat-i 'he carries forward' (*pr6-bhere-t-i). In the earliest
attested Indo-European a pre-verb like *pro is still a free word, whereas the -i
is not; in many languages it is obligatory. It looks as if the verb had incorporated
the particle -i not unlike the Indo-European enclitic pronouns incorporated
the sentence connectives. Old Irish does indeed bear this out; all prefixes require
secondary endings (known as conjunct), for example, from 'carry' > 'take'
(ni-heir 'he does not take', do-heir 'he gives'):
In a similar way Romance developed single tenses (e.g., the future) out of
analytic phrases, and also infixed pronouns. Portuguese retains the original
Latin state of affairs most clearly, as in fabulare habeo > fa/ar hei > falarei 'I
will speak'. This is directly parallel to French parlerai 'I will speak', which is
an unbreakable single word. In Portuguese, however, pronouns can still occur
between the infinitival part and the present forms of haver: lemhrar-me-ei 'I
will remember' and chamti-lo-ei 'I will call him' (the orthography is conven-
tional, compare Brazilian chamal-o-(h)ei). Portuguese further has a compound
future with the above components in a different order: hei-de /ho dizer or
hei-de dizer-/ho 'I will say to him'. Such a situation is very similar to the Irish
development (compare Oneida in §16.3). The Proto-Indo-European imperfect
prefix *e is presumably another adverb, as is *m~ 'don't'. Once the tense or
mood is fully specified, the discourse continues with injunctives, which indicate
person and number only. And these injunctives are linked to the previous sen-
tences by sentence connectives.
The analysis of these two interrelated syntactic problems from Indo-European
-pronouns and sentence connectives, and sentence connectives and injunctives
-was based on the seriation method. This means that internal criteria were
used to single out apparent relics, and these relics were combined for Proto-
Indo-European. The crucial evidence came from Hittite and Old Irish for the
connectives and from Sanskrit and Old Irish for the injunctive, although the
complexities of the Irish evidence for connectives were not elaborated upon (e.g.,
the so-called 'dummy' prefixes like no- < *nu) (see isogloss 15 in Figure 15-2).
is reduced to
V + P + {;;} . .. V + P,
if the tense or mood remains the same. For Proto-Indo-European one would
be tempted to write such a rule if semantics were taken as syntax-dependent.
This shows how easy it is to devise all kinds of rules which probably do not
reflect history, for the injunctive is clearly the original verbal form, ousted by
those verbs that incorporated temporal and modal particles. History shows
accretion of forms, but typical syntactic rules would treat this as deletion.
Of course history can go both ways. Languages may develop complicated
congruence phenomena or lose them (often with much of inflection in general).
Germanic developed a complicated adjective inflection, which has now been
lost in English. The Finnish congruence between adjective and noun in case and
number is also an innovation. One cannot predict the direction of change in
advance without evaluating the total evidence at hand.
§§ 6.5, 18.7, 19.10, 21.14, 21.20). Very important are implicational universals
of the type that the reconstruction of a phenomenon a makes the reconstruction
of the phenomenon b probable (compare§ 16.8).
By necessity the directionality in syntactic reconstruction is thus from sounds
to forms to case syntax, and so on (§ 20.2; compare §§ 22.1, 22.2). Note that
ultimately this procedure can indeed reach discourse analysis as was exemplified
by the Indo-European sentence connectives(§§ 19.8-19.10). The reconstruction
of morphology is also often impossible without syntax(§§ 19.5, 19.6, 19.9, 19.10),
although morphology need not always be fossilized syntax.
REFERENCES
General: Havers 1931, Fokos-Fuchs 1962, Katicic 1970; 19.1 Krahe 1970,
Szemerenyi 1970; 19.2 Hymes 1956, Dressler 1969a; 19.5 Breal 1964, Porzig
1924, Havers 1931, Allen 1951, Watkins 1965, Skrelina 1968, R. Hall 1968,
Szemerenyi 1968, Traugott 1965, 1969, 1971; 19.6 Anttila in Sebeok (ed.)
1963f. vol. 11; 19.7 Watkins 1966; 19.8-19.9 Sturtevant 1939, Szemerenvi )970;
19.10 Meid 1968ab, Kiparsky 1968b; 19.12 Dressler 1971.
CHAPTER 20
RECONSTRUCTING SEMOLOGY/SEMANTICS
The morphological units spelled with small capitals represent forms that are
connected with sound correspondences; that is, a Latin genitive corresponds
in this sense to the Greek genitive, although it does not mean that the two cases
have the same function in both languages. On the contrary, the case corre-
spondences have a different range for the: three Latin cases and the two Greek
ones (see§ 7.2). The glosses are a translation shorthand for the syntactic-seman-
tic environments. The sets represent "pure" cases without prepositions. One
can now note that syntax and semantics play a role in the environments of the
morphological sets (which themselves are reconstructed on the basis of sound
correspondences). The configuration suggests immediately that at least the 'of'
set and the' forjto' set contrast; then one should see whether all those sets having
364
RECONSTRUCTING SEMOLOGY /SEMANTICS
Latin ABL belong together, or whether the Greek with only GEN and DAT allows
us to predict the Latin. Since the 'with' and 'in' sets are formally the same these
are the most likely candidates for being the same. First of all, the 'in' set is
rather limited in both languages, which normally use the preposition in + the
'in' set for the meaning 'in'. Seriation would hold that a productive form
marked twice analytically is the younger one. But then both languages have also
a few forms that are not the same as the Latin ablative or the Greek dative,
respectively (e.g., Latin "ablative" media urbe 'in the center of the city', or
Carthiigine 'in Carthage' vs. the "genitive" Romae 'in Rome', or an "-i" case
in domi 'at home'; ruri 'in the countryside' and Greek "dative" Delphois 'in
Delphoi'; and kukloi 'in a circle' [both normally with the additional en 'in']
vs. oikoi 'at home' or ekei 'there'). These formal differences, which are clearly
relics, are now added to the relic peculiarity of the lack of in 'in'. The indication
thus is that the 'in' set does not point toward unity with the 'with' set at all,
and they cannot be combined.
[20.2] In fact, all the sets contrast with each other because they carry
the different meanings as indicated. We see that our method is quite different
from that in phonology, after all. Here we have to know meanings to predict
the forms, because meanings are the environments. In phonology we had to
predict sounds in the environments of other sounds. We used the meanings to
give us the forms for analysis, but in the actual analysis such meanings could be
pushed into the background. Here we are clearly coming back to meaning. To
reconstruct the actual case morphemes, we would use vague meanings such as 'a
grammatical suffix with local meaning'. This would be enough background for
the comparative method, which would let the sound correspondences ferret
out protoforms, if possible. Here we have come back to the link between
meaning and form, the basic makeup of the linguistic sign. This same buildup
provided the forms for the comparative method, and we see once more that
different parts of language are indissociable and complementary in a natural
situation. The linguist's vivisection does not always do justice to the joints.
The small selection of Latin and Greek above is extremely simplified because the
cases can be formally quite different (in the different declensions). This is why
the labels GEN, ABL, and so on, do not represent form alone, but also syntactic-
semantic equivalences (slots as it were). This is why the members of these case
correspondences (G EN, D AT, and so on) also represent higher levels of grammati-
cal hierarchy (sounds~ forms~ case syntax). Thus our task is not the recon-
struction of single case forms for each set, after all, as would seem from the
table; that would not be possible with these languages. The reconstruction
results in a series of semantic cases, that is,' of', 'from', 'with', 'in' and 'for/to'.
(The actual forms are a different matter.) It is expedient to assume that the
protolanguage had such categories, since the phonology and morphology
establish clear protoforms. Although reconstruction is based on both form and
meaning, we once more get an indication that both sides obey the laws of inde-
pendent structures.
LINGUISTIC RECONSTRUCTION: A SYNTHESIS
[20.3] This small selection from cas'~ syntax is typical: it takes so much
exposition compared with the final yield. It was also chosen so that a further
comparative check would be possible from Indo-European languages not used
in this analysis. Having made the inference that five semantic cases can be
assumed for the protolanguage on the basis of five case correspondences in five
semantic-syntactic environments, it is also likely that there was a more uniform
formal representation of the five semantic cases (in particular the 'in' case had
relic forms clearly pointing to extra forms). Otherwise it would be difficult to
understand the almost haphazard formal case representation between Latin and
Greek. If, on the other hand, we assume that there was a richer case system as a
starting point, it is easy to see how the number of cases would be reduced with
the expansion of prepositions; the history of English is another good case in
point. Although only the 'in' case was mentioned as co-occurring with a prep-
osition in (infen), this is true of others as well, for example, with (cumfsun =
syn) and from (defap6). Cases without prepositions look older (restricted
relics).
The evidence from other languages confirms our guesses quite well. Sanskrit
has five local cases: gentive 'of', ablative 'from', instrumental 'with', locative
'in', and dative 'for/to', although an ablative separate from the genitive exists
in one declension only. In fact, this formal matching with the assumed semantic
cases, which could be adduced from other languages as well, proves the issue.
We started out from apparent morphology, which turned out to be case syntax
and ultimately semantics, that is, structural semantics independent of form.
Semantic reconstruction is also heavily applied historical morphology and
syntax.
*oi(No)- *S(E)M-
Greek ofne 'one' (on dice) heis < *sems 'one' (masc.)
oios 'alone'
Latin iinus 'one' sin(gulus) 'single'
sem(el) 'once'
San- eka 'one and only' sa(krt) 'once'
skrit era 'all alone, only'
(Avestan 'one')
OE iin, NE (al)one, on(ly)
ocs inu 'one, another'
Sanskrit represents here Indo-Iranian and the selection is limited to forms clearly
meaning ' one' (except perhaps for Latin singulus). Since exact synonyms are
rather rare in a language, the next step is to see whether the formal difference
RECONSTRUCTING SEMOLOGY /SEMANTICS
represents a semantic difference as well. Are there other forms available with
compatible semantics that would show more about the semantic range of the
forms? Indo-Iranian has compounds with sa-, for example, Sanskrit sa-jo:}a
'having one will' or 'having the same will, whose will is the same'. The meaning
'one and the same' is suggested by other occurrences: the full grade sam means
'together', as in san-dhi 'putting together'. Greek agrees quite well with Indo-
Iranian, for example, ha-ma 'together' and hom-os 'common' (sometimes
without the h-, as in a-lokhos 'having the same bed', i.e., 'wife'). Glosses like
'one, same, together' have already hinted that Germanic same also contains the
*som- reflected in Greek 'common'. Sanskrit, in fact, displays the exact counter-
part to Germanic same and Greek homos: sama 'same', and an enclitic sama
'any, every'. The form recurs in Germanic 'together', German zu-sam-men,
and Swedish till-sam-mans; compare further German sam-meln and Swedish
sam-/a 'to gather'. Latin sim-ul 'together', sim-i/is 'like', Russian sam-yj
'same', and the prepositions/adverbs German/Swedish samt 'with' and Balto-
Siavic (e.g., OCS su) 'with' further expand the reflexes for this semantic area.
But clearest of all is the Scandinavian preverb sam- 'together', as in Swedish
sam-tal 'conversation' and sam-arbete 'joint work, collaboration'.
But even for the numeral, Germanic has reflexes of *s(e)m-. Old English has
a pronominal adjective sum 'a certain one, one, a; some'. In Modern English,
it can mean unspecified (greater) number or quantity, as in some twenty people,
but as a suffix it means a specified (smallish) number, as in threesome, that is,
'three together [as one]'. The meanings of OE iin 'one, a certain one; the same;
only, alone' partially overlap with those of sum, but this is hardly surprising,
since we are dealing with a sphere of 'unity'. Sum occurs in what is practically
paradigmatic alternation with a derivative of iin (&n-ig > any), as in Do you
have some bread? No, I don't have any.
Other languages and additional forms could be adduced, but this is enough
for retrieving one interesting point. Proto-Indo-European apparently had two
notions of unity, 'one-alone' and 'one-together'; the importance of the distinc-
tion is borne out by the different roots to express it. Of course many languages do
express these aspects by derivatives of 'one', for example, Finnish yksi 'one',
yksi-n 'alone', and yhde-ssii 'together' ('in one'). In Indo-European such
derivations also occur: Latin iinitiis 'unity', a together meaning from the alone
word, and singulus 'single', an alone meaning from the together word. The
meaning 'same' fits in quite naturally with 'together', since it means 'something
must be taken together with what was known or mentioned before'; it is another
side of a known unity. That the words like iin > one or Russian odin also mean
'same' (They lire in one house, compare alokhos above) is a natural phenomenon,
but Old Church Slavic shows a meaning 'another' for inu, which is 'not the
same', that is, clear contrast.
[20.5] Now it is clear that even a numeral like 'one' can be semantically
very complex, because it overlaps the pronominal sphere. The question is
whether the singulative and the collective unity have other formal expressions
368 LINGUISTIC RECONSTRUCTION: A SYNTHESIS
in Indo-European. Indeed they do. Consid1~r, for example, the singulative suffix
*-(i)on and the collective *-(i)ii-, in a Gre:ek name like Plat-on 'the flat-faced
one', hegem-on 'leader', Latin Cat-o(nem) 'the sly one', and Greek sophia
'wisdom ', Latin miseria 'misery'. Greek has clear syntactic reflexes in pantes hoi
anthr6poi 'all the people , and autos ho hippos 'the horse itself' (singulative) vs.
hoi pantes anthropoi 'the people all together ·· and ho autos hippos ' the same horse'
(collective). Here word order (Pronoun-Article-Noun vs. A- P- N) differentiates
the semantic units we have delineated through different words and derivatory
suffixes. It is quite well known that the feminine singular and neuter plural are
formally identical in Indo-European (*-[i]a-), and these can be taken as repre-
senting an original collective, which implies plural ity as unity. Thus Greek !zenia
'reins' occurs both as nominative plural neuter and as nominative singular
feminine, and Rigvedic tanii as nominative plural 'descendants' and as fem inine
singular ' offspring'. This collective -ii contrasts with the singulative -s (nom . sg.
masc.), for example, OCS nog-a 'foot '/Lith. nag-a' hoof' vs. Greek 6nuk-s 'claw',
Greek phord 'produce, fruit' vs. pharo-s 'tribute, payment', pharo-s 'bearing
(favorable) wind', and Sanskrit hfma 'winter' vs. hinu:i-s 'cold, frost'. Support
for this collect ive/plural comes also from the fact that a neuter plural subject is
combined with a singular predicate in Greek ([pl.] ta zoia [sg.] trekhei ' living
things move'), earliest Avestan, and sometimes in Vedic Sanskrit as well. In
Baltic the third plural verb was lost altogether and this 'collective singular'
usage was generalized.
[20.6] The purpose of this demonstration was to show that one can
continue combining morphological and syntactic evidence for semantic recon-
struction. Gaps and mistakes would not be identifiable at this point. As a final
conjecture Jet us look at another possibility in this semantic sphere of unity (we
shall take the following forms at face value for the moment). Swedish has two
compounds for 'the present time' : nu-tid and sam-tid, that is, with 'now' and
'one-together'. German has corresponding Jetzt-zeit, also 'now-time' and
Gegen-ll'art 'presence', in which the first part is the same as in English again.
The meanings of again include 'once more' and 'back into a former position',
and an obsolete 'back'. Such meanings come into the area of 'same', that is,
repetitions of earlier things, even if againstfgegen normally mean togetherness
of a hostile nature. Is it really possible to combine meanings 'one, now, again,
back' into one semantic unit? We would not take these as variants of one se-
mantic unit in Modern Germanic, but is it possible for earlier times? It would
seem so; Oneida has such a unity sememe with exactly the variants mentioned,
expressed, for example, by the s-in Skanyataliyo (§ 16.7), although in certain
environments another morpheme is used.
But German Gegenwart acquired temporal significance first in the eighteenth
century ; and Jetztzeit was coined in 1807, Swedish nutid being formed after the
latter. Thus the possibilities explored above diminish substantially. The moral
of this is that in semantic reconstruction it is easy to go too far, even though each
step can be perfectly substantiated or paralleled from another language.
RECONSTRUCTING SEMOLOGY / SEMANTICS
Let us refer back to the buildup of anaphoric pronouns from sentence con-
nectives(§§ 19.8, 19.9). The notion of' sameness' is inherent in *nu, which means
'carry on, add this sentence to the preceding ones, the same discourse continues'.
Since it appears that *to 'and' was used with a change of subject, the attractive
hypothesis emerges that the mystery *-sm- in local cases of the anaphoric
pronoun is actually the zero grade of *sem-, that is, *to-sm-6i 'and to this same
one, to the aforementioned '-the *-sm- in effect canceling the change of subject
usually associated with *to, and contributing to the simplification of what pre-
sumably was a very complex set of paradigms. Once more we come back to
syntax. Reconstructing structural semantics is very complex indeed. We cannot
expect easy results, since there are great difficulties in reaching agreement on
semantic structure even in synchronic situations. Reconstruction is many more
times difficult.
the air', which is obviously absurd, and the words must thus be regarded as
homophones only. Independent evidence tells us that this is the correct inter-
pretation. English board (1) 'flat (long) piece of wood', (2) 'food' (room and
board), and (3) 'committee, body of people', presents a formidable problem on
purely internal evidence, but if we know the groaning board or that Swedish
bord means 'table' (smorgasbord§ 8.6), the metonymic chain is obvious. We
see the intersecting indexical joints' board',,' a board used as a piece of furniture',
'what is put on the board', and 'those who sit around the board for some pur-
pose'. Internal seriation shows that one meaning could be taken as original,
the others as conditioned variants. The intersection of the meanings of French
fille is 'woman' (§ 7.12). This would give a starting point for jeune fille 'girl' at
least, although not so easily for (ma) fille 'daughter', both of which are syn-
tacticaJJy marked. Semantic derivation would at least suggest that' prostitute' is
not original, and the pronominal combination of ma fil/e seems to be hierarchi-
cally deeper than a compound jeune fille (see §§ 20.16, 21.17). The intersection
'woman' is not historically accurate, but seriation might have pointed toward
the correct origin, Latinfi/ia 'daughter'.
[20.9) Translation into the linguists' own language can obscure the issue
and strengthen apparent homophony. Thus Greek trephO is usually listed under
two meanings, (I) 'rear, bring up, cause to grow' and (2) 'curdle, congeal'.
Close observation of the Greek passages shows that these are variants of one
meaning: 'to favor the tendencies inhere:nt in the matter, to allow growth'.
When you Jet children have their biological way they will grow up, when you let
milk take its course, it will curdle, and so on. Here the intersection of the two
meanings gives the simplest solution, and the example shows that the danger of
missing legitimate intersections of meanings is about as great as trying to create
illegitimate ones. Semantics is notoriously an area where common sense cannot
be replaced by mechanical reconstruction steps.
[20.10] But the principles of Figure I 8-1 can serve as guidelines. English
sheep and Sanskrit chiiga 'goat' share the protoform *sK.egwo-. In addition, both
languages have reflexes of a word *owi- 'sheep' (ewe; Figure I I -6). Because of
the latter, the meaning' *sheep' for *sK.eg'"o- is unlikely, but could it have been
a straight intersection 'small cattle' covering both sheep and goats? We do
not know, because some kind of a goat is also possible. Note that we do the
same thing here as in phonetic reconstruction: we compare intersections.
Since *owi- 'sheep' has a better claim for 'sheep' (the evidence goes well
beyond Sanskrit and English), we would have to find something nearby (see
§§ 18.1 0, 18.11 ). All over the world we have similar problems with domesticated
animals. Shifts occur between them very easily, which means that all we need is
the context and that the context need not be elerated into a protomeaning (see
fille above). If in American Indian languages the same form can mean either
'cow', 'goat', or 'dog', it does not justify a reconstruction '*domestic animal'
in itself. In northern Eurasia the same form can mean either 'dog' or 'reindeer',
RECONSTRUCTING SEMOLOGY /SEMANTICS 371
but a reconstruction '*sledge-pulling animal' would be wrong, because this
could have been no more than the characteristic context for metonymic shifts.
speakers (e.g., bake/batch; § 18.17). But when the unknowns of history are
combined with the present unknowns of semantics, we have to be content with
the few methodological approaches and gains we have established so far. One
should note particularly the paraiielism between phonetic and semantic features
(Figures 1-3-1-5) in reconstruction (Figure 18-1), as wen as the similarities in
their respective relative difficulties.
readily interpreted. The Hupa call Mt. Shasta (in northern California) nm-ms-'an
lak-gai 'white mountain', whereas the Yana have a single unanalyzable term
for it, wa<gafu·. The inference is that the Athabaskan-speaking Hupa are new-
comers in California compared to the Yana.
Most languages yield readily to such inferences. The fact that Seneca (an
Iroquois language) has unanalyzable roots with the meanings 'to be in water'
(-o-; compare Oneida kahy-6-/oras, § 16.2), 'to fetch water' (-jc-), and 'depth
of water' (*-!mot-) seems to show prolonged association with water. Similarly,
the unanalyzable roots for the snowsnake (*-/mas-) and hoop and javelin games
(*-ket-) point to the antiquity of these games among the Seneca. The Iroquois
religious system has two components, (northern) shamanistic rituals pertaining
to a hunting environment and (southern) agricultural ceremonies. Both sections
have been regarded as the earlier one. Now, the linguistic evidence unambiguously
supports the greater age of the shamanistic complex, because its terminology is
comprised mainly of unanalyzable roots, whereas the agricultural area makes
use of descriptive compounds.
Among the dangers and uncertainties in this method, only positive evidence
counts. That is, the fact that a meaning is represented by regular morphology
(cowfcows § 19.4) or an iconic compound (top cow § 7.6 and mere-hengest
§ 7.8) does not mean that the item cannot be old. Then, borrowings give un-
analyzable forms that can be very recent (squaw and wigwam). And further, a
form may change meanings, so that the semantic inference we are making belongs
to another notion altogether (for example:, meat would appear to belong to
'flesh', but actually it once represented 'food', see § 7.13). All this shows once
again that for ultimate solutions various principles must be synchronized and
fitted with care, and with sensitivity for such puzzles.
[20.17 Conclusion] Chapter 18 is notably different from 19 and 20,
because in phonological reconstruction it is easiest to separate units for analysis.
Chapters 19 and 20, on the other hand, form a tight-knit unit in which mor-
phology, syntax, and semantics are inextricably interrelated. Because gram-
matical and syntactic reconstruction requires such an enormous amount of
background information on the languages used, as well as long practice, only brief
hints could be supplied in this introduction. Even so, the essential guidelines
have been described, and if the student wants to continue in this particular area,
he has to develop the necessary ability for" scientifically trained fantasy" himself,
without which syntactic and semantic reconstruction is a mere game on paper.
REFERENCES
General: Benveniste 1954; 20.1 Hoenigswald 1960a; 20.5 Lehmann 1958, Maher
1965, 1969b; 20.8-20.11 Benveniste 1954; :W.13 Maher 1970e; 20.14 Schrader
1883, Sapir 1912, Krahe 1970, Thieme 1953 and in Sebeok (ed.) 1963f. vol. 11,
Dressler 1965, Scherer (ed.) 1968, R. Schmitt 1967, Benveniste 1969, Cardona
and Hoenigswald and Senn (eds.) 1970, lFriedrich 1970; 20.15 Tovar 1954,
Swadesh 1959; 20.16 Sapir 1916, Chafe 1964.
PART V
CONCLUSION:
LINGUISTICS AS PART
OF ANTHROPOLOGY
CHAPTER 21
(21.2 Linguistics and the Study of Man] Both psychology and anthro-
pology have recognized linguistics as a methodological model and as the most
precise of these particular sciences. The sociologists acknowledge the" cruel truth"
that awareness of language can do more for them than sociology can do for
377
378 CONCLUSION: LINGUISTICS As PART OF ANTHROPOLOGY
linguistic studies. The issue is not that simple, however, since it is exactly
sociological information that is crucial in understanding linguistic change
(Chapter 9). Language is necessary for sociologists and society for linguists; the
truth of course is that the best results for all social sciences are secured by well-
planned cooperation. Thus one must pay equal attention both to the autonomy
and to the integration of the social sciences. Linguistics has secured a central
position in these interdisciplinary studies, largely because of its unusually
regular self-contained patterning and because of its central role in the totality
of culture (although the former has led to the dangerous tendency of looking
only into language-internal facts in studying change). Language is a constituent
of culture and functions as a substructure. Therefore it is easier to abstract
linguistics from the remainder of culture and define it separately than vice
versa. One does not usually realize that even mathematics is firmly rooted in
ordinary language, since it is verbal activity presupposing language. Money and
economics can also be viewed as a language:, because the circulation of money is
one kind of sending messages. This is part of wider communication (for example,
social intercourse), and in all these areas linguistic models are useful. Because
social life is not conceivable without the existence of communicative signs,
semiotics is part of sociology. Linguistics is the study of verbal messages, which
makes linguistics a small hierarchical part of sociology after all. And of course
the wider end of semiotics leads into biology and neural systems (Chapter 22).
CHANGE
[21.5 Study, Traces, and Processes of Change] The cultural and social
evolutionists believe that they have been successful in establishing unilinear
schemes of social evolution, parallel to the development of man from infant to
adult. A scale of evolution from primitive to sophisticated with three basic
categories (savagery, barbarism, and civilization) was very popular in the
nineteenth century. Every society was supposed to have risen through the same
380 CONCLUSION: LINGUISTICS As PART OF ANTHROPOLOGY
hierarchy (unless it was still standing on the lowest rung). Such typological
schemes are untenable today as are so many other earlier ones (Chapter I 6,
§ 21.11). But the essence of such notions is salvaged by neo-evolutionism, which
makes use of a typology better suited to the enormously complex phenomena.
[21.8] When the fit between cultural norms or suppositions and actual
behavior (i .e., the deep structure and the surface structure) becomes rather loose,
one can expect restructuring toward the actual behavior, for example," bringing
laws up to date." This is of course another vague parallel, but it is instructive in
that here also the change creeps into the underlying structure from the surface
(compare§ 9.16).
not lend itself to the precise segmentation and comparison necessary to establish
genetic relations, linguistics as the oldest cultural science can help. No cultural
aspect can be studied without reference to the linguistic signs used for it. Exam-
ples show also that linguistic change tends to be accelerated when the culture of
the speakers undergoes rapid changes; for example, the Norman conquest of
England brought in loanwords, new spelling, and it was at this time that loss
of grammatical endings was most rapid (in the literary tradition at least). All
this is summed up as the transition from Old to Middle English. When the
culture of a people is relatively static or slow to change, linguistic change tends
to slow down also. An often cited example of this is Lithuanian, whose speakers
led the same kind of life for millennia. Certainly such correlation is not random,
even if individual cases might not seem to comply.
[21.10] It has been observed that the nomadic way of life often corre-
sponds with lack of great dialectal variation, even when the area occupied by
such speakers is large. Thus, roughly, the European societies have undergone
change through various forms of states, organized religions, and the industrial
revolution; the Indo-European languages do seem to have changed more rapidly
than, say, the Turkic ones. Dialectal stratification is the basis of internal change,
and correlates with social evolution as well. The Paleolithic hunting and gather-
ing societies had low population density, no fixed abodes, and autonomy of the
social units. There was no specialization of labor and no stratification in society,
and consequently no social or prestige dialects, but equality of single ones.
There were different dialects, of course, but they were all taken as equal, as is
still true in the Paleolithic way of life. Father, mother, children, aunts, uncles
can all speak different dialects. People branch into different tribes and get
together occasionally for such matters as the arrangement of new marriages.
With the Neolithic economy of planned food production, population increased;
the social unit remained about the same size, but it was still rather mobile
because of the slash-and-burn method of agriculture. Surplus people branched
off to found new settlements. The family-tree model is appropriate here. But
branching off often meant marriage in another community, and thus the wave
theory is also operative. The Neolithic social unit remained self-sufficient and
without prestige dialects until urban revolution produced a stratified society
and prestige dialects, and apparently much faster change in language. Note that
prestige is one factor in addition to the Paleo- and Neolithic variations along the
scales of sex and age. Ecological influence finds its way into the grammar; for
example, hunters occasionally have to stay away from their families, which
encourages development of hunting tabus and other sex-linked linguistic
features.
It seems thus quite likely that socioeconomic factors correlate with change (see
Muller, § 16.3). In the Soviet Union, N.Y. Marr took this hypothesis too far
when he correlated language with the Marxist doctrine of social evolution through
the following stages: primitive horde, slave-holding, feudal, capitalist, socialist,
and communist (see § 21.5). Of course Russian did not change with the revolu-
tion, except for new vocabulary.
RECONSTRUCTION
(universal) function? And the study of individual cases leads to the reconstruc-
tion of the long-term cultural history of a society. This is the so-called historical
method, which concentrates on well-defined small geographical areas. The
method implies a detailed study of customs in relation to the total culture and
their relation to, and distribution among, the neighboring tribes. The method is
thus culture-internal with glimpses into neighboring geography. It is dialect
geography with cultural items. The comparative method studies the results of
historical growth, the historical method the processes of growth. The two are
complementary and essential for understanding societies and their institutions.
understand others, and to understand others, one must look into oneself" is
Schiller's variant of the principle ofsyncritical comparison in ethnology. We study
history in order to make the present understandable through the past, but the
past would be inaccessible to us, if we could not use our own experience to inter-
pret the traces of past life and societies. In other words, synchrony, diachrony,
and syncrisis are intimately connected, as in linguistics (see Figure 1-7).
[21.21] Thus, using the living paradigms offered by the Eskimos or the
Papuans, anthropologists can supply rather realistic detail for the life of the Ice
Age mammoth hunters of northern Europe, or of the Neolithic peasants
penetrating into the European virgin forests. The folklore of modern gathering
communities may give clues to earlier rituals. The way such communities make
and use tools brings life to archaeological finds. As in linguistics, all theories
must be checked against "living paradigms." And without such present-day
examples much would remain incomprehensible. The psychic unity of mankind
serves as a typological check on reconstruction, even if alone it does not carry
historical meaning other than species specification in relation to other primates.
(21.24] Of course history can play tricks on us, and this is why one has to
remember that the method yields probabilities only. But these are very valuable
for ethnology when used together with other nonlinguistic evidence. Modern
mobility and immigration interferes with the method, but in such cases we
generally have documentation to go by. When one finds the Russian colony in
California, one would presumably guess that it came from Russia, or from
Russian Alaska. But simplest derivations are not necessarily always right
historically. The colony is known to have moved first to China, from there to
Brazil, and then to California. Unless loanwords record such routes (see Figure
8-4), mere migration theory is powerless. Every historical method has its weak-
nesses, of course.
REFERENCES
General: Kroeber 1935, 1944, (ed.) 1953, 1963ac, Rice (ed.) 1931, Carroll1953,
Beals and Hoijer 1965, Pike 1967, Jakobson 1969, Hallowell 1956, Henle (ed.)
1965, Hoijer (ed.) 1954, Current Trends in Linguistics vol. 12 (Sebeok ed.), and
see also the selections in the Bobbs-Merrill Reprint Series in Social Sciences;
388 CONCLUSION: LINGUISTICS As PART OF ANTHROPOLOGY
21.2 Voegelin and Harris 1947, White (ed.) 1956, E. Hall 1959, Fast 1970,
Birdwhistell 1970, Bouissac 1970; 21.3 Lenneberg 1960, L. A. White 1959,
Steiner 1969, Wescott 1969, Fox 1970, Woolfson 1970; 21.4 Kroeber 1919, L.A.
White 1945; 21.5 L. A. White 1945, Eisenstadt 1968, W. Moore 1968, Service
1968, E. Vogt 1968; 21.6 E. Hall1959; 21.7-21.8 Oliver 1964; 21.9 Hoijer 1948,
Holmberg 1954; 21.10 W. Miller 1967, Maher 1970d; 21.12 Whorf 1956,
Hormann 1970, Shands 1970; 21.14-21.15 Irving 1949, Eggan 1954, Leach 1968;
21.16 Paor 1967, Deetz 1967, Sapir 1916, Romney 1957; 21.17 Sapir 1916;
21.18 Ehrich and Henderson 1968; 21.19 W. Schmidt 1939, Greenberg 1957;
21.20-21.21 Childe 1951; 21.22-21.23 Sapir 1916, Dyen 1956, Romney 1957,
Diebold 1960.
CHAPTER 22
[22.1 The Genetic Code and Language] One of the recent striking findings
of molecular biology is the structure of the genetic code, which revolves around
compounds known as DNA. These are nucleic acids whose immense molecules
are arranged in double helices. When the strings separate, each helix attracts
the missing part of the original double helix, thus providing exact copies of the
original. A DNA molecule must contain all the information for all the details of
the future plant or animal. How this staggering requirement is met is not known.
DNA is built out of four nucleotides (A, G, T, C), and this gives a four-letter
alphabet of life, combined according to the "grammar of life," as yet undis-
covered. (Curiously, this matches the number of mythical cardinal humors:
blood, phlegm, choler, and melancholy.) The order of the letters is significant,
as well as certain bracketing, that is, commas or "spaces." A DNA code word,
a codon, consists of three letters. The message in DNA is taken up by another
nucleic acid molecule, RNA, and this single-stranded messenger takes it into
the protoplasm, where the transfer RNA each bring their own amino acids,
which form proteins according to the message in the first RNA. In spite of the
attractive chemistry, all this is pure witchcraft, that is, beyond our understanding.
How does DNA regulate growth and specialization of cells of the same genetic
inheritance? How does it determine a bird's song? We do not know. But one
parallel to language is obvious: we have a deep structure, DNA, and rules, RNA,
that take the message to an ultimate surface structure. And, indeed, geneticists
have freely used linguistic terminology; S. K. Saumjan in Russia uses the
biological terms genotype and phenotype as the counterparts of deep and surface
structure. DNA is subject to mutations, the equivalent of deletion and metathesis.
An alphabetic unit can also produce a different result in different positions of the
genetic message, that is, contextual meaning. Now it seems that there is one
crucial difference compared with language: the message goes one way only,
from DNA to RNA, which would be parallel to a grammar for a speaker only.
Such grammars have indeed been popular in linguistics around the late 1960s.
But it now seems certain that certain cancer-producing viruses, consisting of
only RNA and a protein sheath, may make their own DNA once they invade a
host cell. This message from the RNA stays permanently in the host cell, and
cancer is passed on during cell division. This evidence makes the DNA-RNA
389
390 CONCLUSION: LINGUISTICS As PART OF ANTHROPOLOGY
relations more similar to language, where the machinery is habitually used both
ways (Figure 1-1). Note also the curious facts of regeneration. When the poste-
rior half of a planariandevelops a new front part, it will include the brains as well.
The rear end develops its own future control mechanism(§ 5.7).
[22.2] The parallels are startling, though not perfect. One could find
more; for example, is a grammar of rules only, with no units, a "virus" or a
"cancerous grammar"? A more important question is this: Does genetics
support the view that deep structure is after all the place for determining genetic
relationship of languages? (§ 16.10). The answer is apparently no, because
note that the alphabetic compartments are reversed in language and the genetic
code. The chemical alphabet occurs in the deep structure, the acoustic one in
language in the surface structure. In genetics the "arbitrariness" lies in the fact
that a particular combination of the nucleotides is connected with a particular
meaning (e.g., man, cat, worm, and so on). Similarity in DNA cannot be ran-
dom; it means closer historical connection. Since all forms of life are related,
the matter is one of subgrouping, and it works out as expected; for example,
the primates have closely related DNA as well as similar physical shape, and
so on down the hierarchy. It would seem that DNA parallels support the
position presented in the preceding chapters exactly. Of course such parallels
do not necessarily have any validity whatsoever. But they are so striking that
one can raise the question whether the isomorphism between genetic and verbal
codes results from a mere convergence of similar needs, or whether the verbal
code was imposed directly on the molecular structural principles (see § 21.3).
Language is in part at least a molecular endowment of man, and both the genetic
code and language are anticipatory models of the future (see§ 1.16).
CHANGE
[22.4 Direction and Mechanism of Change] According to the second law
of thermodynamics, all natural (chemical and physical/__.. processes tend to
GENETIC LINGUISTICS AND BIOLOGICAL GENETICS 391
proceed toward increased entropy, that is, greater chaos. In other words, they
tend to run down. Organic matter (life) is sharply the reverse, because in it
order tends to increase. The general trend of biological evolution is toward
organisms of greater efficiency as a whole, and this is achieved by more and
more complex synorganization and interdependence of parts. Cells can either
be independent units or collaborate to complex ends. This is in a way a parallel
to language, where various components and units can be studied in isolation,
but where ultimately everything is combined for reaching one end, that of
providing a flexible means for communication. Language and biology share
complicated synorganization in their systems. Life, like language, is complex,
self-reproducing, and self-varying matter, and both inherently reside in the
same structure and imply each other. Mutation is an accidental affair; it takes
place in all directions, but it provides the raw material for evolution. Natural
selection converts accident into apparent design and determines its direction.
Those mutations that improve the fit of the organism to its environment usually
make the organism more fecund and favored (selected) for survival; others will
be ultimately rejected. New and old forms can exist side by side for a long time,
if the mutation is not fatal outright; or, with isolation, both can survive in
different environments. Evolution has also been vexed by the problem of whether
change comes from within or from without. Gene mutations do come from
within, but the future of such changes is mediated by natural selection. Adapta-
tion to the environment is the main causative agent of organic evolution. In
this sense, evolutionary changes come from the environment, that is, interaction
between inheritance and environment. No organism is without a genotype
(deep structure), and no genotype can exist outside an environment, that is, a
continuum of space and time.
[22.5] The parallels to the above in linguistic change are obvious. In the
previous chapters we have seen that linguistic change is a complex phenomenon
in which both internal and external causes must be recognized. In environmental
factors the social ones are the most important, and social factors are of course
further anchored in a particular society, a particular age, and a particular
historical event. Social factors play a role in human evolution as well, because
various incest tabus and marriage systems either prohibit or encourage exchange
of genes between particular individuals. A linguistic innovation is either accep-
ted or rejected by the speakers, or both can be preserved in different environ-
ments, for example, once/one's, mead/meadow, my/mine, metal/mettle etc.
(§§ 5.7-5.9, 7.9). Also linguistic change is subject to choice, a kind of Darwinian
selection. The speakers can choose to change or not to change, and this is
clearest in situations with social motivation. One can sometimes choose one's
group membership and linguistic solidarity. The universality of change (§ 9.1)
makes no change very baffling indeed, and language does not stay put as long
as biology can (for example the ant has remained practically the same for some
fifty million years). Choice is still a force that governs both change and no
change, and in language of course there may be no change in grammar even
392 CoNCLUSION: LINGUISTICS As PART OF ANTHROPOLOGY
if there is in phonology. But over rather short stretches (some one hundred
years or so) language always changes somehow. However, in language there
does not seem to be any improvement or progress in organization; that is,
language retains its basic structure (Figures 1-1, 1-2). There is an efficiency
factor, however, in that areas vital for the community will be matched by
language. The language of the Aymara is adapted to the importance of the
potato, that of the Arabs to the camel, and that of the Eskimos to snow. Thus a
language will increase its efficiency for any particular cultural situation by
developing new vocabulary or syntactic patterns, without violating the universal
laws of language design. The implications of this are explicit in the cultural-
relativistic slogan that every language is as good as any other for the society that
uses it; that is, the Aymara have a rich vocabulary for potatoes, and the Western
languages for airplane screws and nuts. The same seems to be true of organisms
as well: they all are well fitted for the environments they live in. Of course,
languages can borrow from each other to meet specific needs; organisms can-
not: a bird cannot borrow the burrowing and eating habits of a worm. Only
man has been successful in this; he has borrowed underwater travel from the
fish (submarines), armor from the rhinoceros (tanks), flying from the birds
(airplanes), and so on. Ultimately, all this is possible only through language.
[22.6 Teleology, Mentalism, and Function] Even if the genetic code and
the mechanism of natural selection explain one side of biological evolution, it
is not enough for understanding life, or the mind. No notion of awareness of
existence can be extended from physics or chemistry into biology; a non-
mechanistic factor must be involved, and most biologists agree on this point.
Parapsychology, the study of extrasensory perception and psychokinesis, is an
established fact, although still a disquieting one for many scientists (who believe
that they have open minds). Its statistical proofs are impeccable. On the other
hand, statistics shows unequivocally that man cannot exist. Mere evolutionary
accidents would never have produced man. Biologists have had to return to
Aristotle's final causes, which are goal-directed, purposive, and self-regulating.
It is not enough to ask how or for what reason; rather, it is legitimate to ask
for what purpose(§ 9.1). It is the nature of the organism to be oriented toward
the change that occurs. The earlier monistic views based either on matter or on
mind alone are inadequate. This has changed into a both-and attitude and the
beginnings of psychosomatic biology.
Language is also a teleological or goal-directed system. It maintains a certain
dynamic equilibrium for functioning through time, providing a vehicle for
anticipation, initiative, and foresight(§ 21.3). This is quite typical of functional
systems. It can be briefly mentioned here that the notion offunction is applicable
where the following conditions are fulfilled: (l) the object of study can be taken
as a system forming a unitary whole (which is not to say a monolithic lump);
(2) the unitary whole can be ordered as a differentiated complex so that it is
possible to talk about part-whole relationships; and (3) the parts are elements
that contribute to fulfilling the purpose for which the ordered whole has been
GENETIC LINGUISTICS AND BIOLOGICAL GENETICS 393
set up, that is, the parts must help maintain the whole in a persisting or enduring
state. Phonology alone can satisfy these criteria, since its parts, distinctive
features (or units) and rules, ensure speech synthesis (see§ 9.15); the same is true
of other parts of language, or of culture in general, although such a notion of
function was originally borrowed from biology. Keeping the necessary homeo-
stasis, that is, functioning, the language has to change to stay the same, to
continue to fill its purpose. One of the factors here is the psychological-mental
one of one meaning-one form, shortcuts in particular environments, whether
grammatical or cultural. And then of course one cannot predict which mecha-
nism will be chosen for new nomination when it is needed, or even without a
need (see§ 9.5). As it turns out, the psychological factors of biological evolution
are completely beyond our grasp, at least so far, whereas they are on the whole
understandable where linguistic change is concerned. No doubt much more will
be found out in the future.
RECONSTRUCTION
subgrouping coincides with typology. Syncrisis thus sifts the shared characteris-
tics, seriation orders them, and the "confusion" of descriptive facts is turned
into history. Schleicher phrased the same principle for language, in that one
need only convert the side-by-side arrangement of systems (e.g., man and fish)
into the successive stages of evolution (fish first). We have seen how seriation
works in making inferences about relative chronology in linguistics (§ 14.7).
There is also a biological equivalent of the comparative method, although it does
not come out as explicitly as the corresponding one in linguistics: there is
certain form (e.g., a bone structure in mammals) which carries a function (i.e.,
meaning, say, locomotion); but the form may be environmentally conditioned
(e.g., the leg, the wing [of a bat], and the flipper [of a dolphin]-these three are
conditioned by land, air, and water, respectively, and thus they are variants of
one mammalian protoform).
[22.10] Physical appearance that is out of line from the expected seriation
reveals "loans" geographically. Sometimes the distribution of certain genes or
gene combinations shows distinct spatial orderliness in the form of statistical
averages that produce a gradient over the map. These gradient steps are called
clines. For example, skin color, dependent on the presence of a complex organic
molecule called melanin, has a general distribution of darker at the equator and
lighter in the north, or away from the equator. This is similar to the "cline"
in French dialect geography with k-ts-s from south to north, that is, decrease in
obstruction (see § 14.6). The reason apparently is that heavy pigmentation
protects against radiation and thus has survival value. Where there is less sun to
cope with, pigmentation decreased accordingly, because all evidence points to
dark as the original color. If these hypotheses are true, the distribution of skin
color clines on the map automatically reveal certain movements(" borrowings").
Thus it is apparent that the New World was populated later than the Old
GENETIC LINGUISTICS AND BIOLOGICAL GENETICS 395
(although darker pigmentation has developed around the equator in the New
World also), that the majority of the Indonesian population are latecomers,
that the Eskimo have been in the north a shorter time than the Scandinavians,
and so on. In short, this locates disjoint seriation. Often there is other evidence
to support such reasoning.
(22.11) This leads to the question of using diversity for relative chronology
as in migration theory (§§ 21.22, 21.23). The area of greatest diversity is the one of
longer settlement. Thus the racial diversity of man in the Old World points to
oldest settlements there, because this way we need to posit fewer large-scale
migrations. Mutations take time to pile up, and where they have exerted their
influence most, the most time has flown by. In language we could use similar
seriation in that diversity represents an older state of affairs. For example, a dialect
with cow/kine is "older" than one with cow/cows(§§ 14.7, 19.3, 19.4); but note
the case of Finnish type E, which is the latest arrival(§ 19.4). And indeed, the
same danger lurks in biology as well, in the form of adaptive radiation. A famous
example is Darwin's ground finches, from the Galapagos Islands, songbirds
derived from some New World species. The islands provided a perfect evolu-
tionary setting, without competitors or predators. The normal seed-feeding
mode of life was of limited use on the islands, and those variants that could take
advantage of un-finch-like sources of food would flourish. The group now
consists of seed-eaters, omnivorous ground-eaters, insect-eaters, leaf- and bud-
eaters, and a woodpecker type-four genera and fourteen separate species.
Although the birds are still very similar, they have developed beaks that are
adapted to these different modes of food gathering. On the mainland there were
no such low-pressure niches available, and no such differentiation has occurred.
Reconstruction is a very delicate puzzle game indeed. Note that the filling of
ecological niches by various forms of life has a certain similarity to filling the
holes in phonological patterns (§ 9.6). When genetic drift alters a single species
in the same environment, there is often the difficulty in deciding when it becomes
a new species. The same problem recurs in linguistics-for example, the bound-
ary between Old and Middle English (see§ 21.9).
(22.13] Genetic linguistics has its own decay dating, known as glotto
chronology. One of its underlying assumptions is that some items of the vocab-
ulary are better preserved than others: lower numerals, pronouns, body parts,
natural objects, basic actions, and so on-items referred to as the basic core
vocabulary. Words that are intimately tied with cultural items suffer loss with
the items themselves; for example, the vocabulary of falconry went with the
practice (any dead metaphors surviving today are synchronically unanalyzable).
Another assumption is that the rate of attrition in the core vocabulary is con-
stant, so that about 81 per cent of the original two-hundred-item core vocabulary
set would be preserved after a thousand years. The time of separation, t, is now
equal to the logarithm of the percentage of cognates, c, divided by twice the
logarithm of the percentage of cognates retained after a thousand years of
separation, r:
t = log c
2log r
The cognates are decided through strict observance of meaning; for example,
'animal' between English and German comes out as animal/Tier, that is, no
cognation, even though Tier can be etymologically aligned with English deer
(§ 7.12). Generally, inspection and matchings are used, because inaccuracies
can be expected to cancel each other out.
Glottochronology has often been equated with lexicostatistics, but it is
advisable to take the latter as a wider field of statistics in the service of historical
vocabulary studies, including stylostatistics for determining disputed authorship.
Stylostatistics can also be used for establishing relative chronology within the
output of one author. But the prime function oflexicostatistics is the determina-
tion of the degree of lexical relationship among related languages. This can be
expressed in dips calculated, for example, by the formula
d = 14 log c
2log r
or by other similar means. When used for lexicostatistical subgrouping such a
method tells relative closeness only, and its potential value is great for language
families not yet worked out in detail by the conventional methods. For many
languages only word lists are available, so a lexicostatistical subgrouping gives
a useful interim result and may give valuable hints for the other methods as well.
There is a certain parallel to biological subgrouping ascertained by counting the
agreements between DNA of the various forms of life. Although the method
GENETIC LINGUISTICS AND BIOLOGICAL GENETICS 397
yields numerical indices (compare Figure 16-2) between languages, these need
not be interpreted to the letter, if nonnumerical evidence points to the contrary.
As always in historical investigation, one-sided rigidity must be avoided. Note
that low numerical values mean diversity, and such figures can be used for
migration theory. If low figures cluster in one geographic area, it means older
settlement (see §21.22).
proposed universals in culture can claim (see§ 21.14). Of course, there are cases
where a strong literary tradition, contact, or strong tabu effects have distorted
results of the method. The method is not without value, but neither is it omnip-
otent. Although claims about chronology are weak, further inquiry is justified
by the results so far. Lexicostatistics must contribute to and draw from the
general theory of culture change, and the numerical expressions of linguistic,
cultural, and biological distance should be correlated. We have returned to an
area where more has to be found out about language typology (how to pick out
the basic vocabulary) and the influence of society, for example, how it affects
the rate of change (see §§ 21.9-21.11), or the formation of tabus. No area of
genetic linguistics is settled for good. The student meets new challenges every-
where.
REFERENCES
General: Hockett 1948b, Kroeber 1963b, Beals and Hoijer 1965, Zipf 1965,
Lenneberg 1967, Spuhler (ed.) 1967, Jakobson 1969; 22.1 Crick 1962, 1966,
Nirenberg 1963, Yanofsky 1967, Beadle and Beadle 1967, Baltimore and Temin
and Mizutani 1970; 22.2 Chao 1967; 22.3 Darlington 1947, 1955, Brosnahan
1960, 1961, Hogben 1956; 22.5 Greenberg 1959, Hymes 1961; 22.6 Emmet 1958,
Segerstrale 1968; 22.7 Hoenigswald 1963, Maher 1966, Neumann 1967; 22.10
Huxley 1953, Brosnahan 1961, Brace and Montagu l965; 22.12 Paor 1967,
Deetz 1967; 22.13-22.14 Kroeber and Chretien 1937, Swadesh 1951, 1952, 1955,
Lees 1953, Gudschinsky 1956, Arndt 1959, Hymes 1960, Dyen 1962, 1964,
1965ab, Dyen and James and Cole 1967, Martinet (ed.) 1968, Sankoff 1970,
Lees 1953, Chretien 1962, Bergs1and and Vogt 1962, Teeter 1963, Fodor 1965,
H. Vogt 1965, L. Campbell 1971, Hymes and Sherzer, Rea, and Hattori in
Sebeok (ed.) 1963f. vol. 11.
APPENDIX I
399
APPENDIX II
400
BIBLIOGRAPHY*
ALA TIS, JAMES E. (ed.). 1968. Contrastive linguistics and its pedagogical implications.
Monograph Series on Languages and Linguistics No. 21 (19th Round Table
Meeting). Washington, D.C., Georgetown University Press.
- - - (ed.). 1969. Linguistics and the teaching of Standard English to speakers of
other languages or dialects. Monograph Series on Languages and Linguistics
No. 22 (20th Round Table Meeting). Washington, D.C., Georgetown University
Press.
- - - (ed.). 1970. Bilingualism and language contact: anthropological, linguistic,
psychological, and sociological aspects. Monograph Series on Languages and
Linguistics No. 23 (21st Round Table Meeting). Washington, D.C., Georgetown
University Press.
ALLEN, W. S. 1951. "Phonetics and comparative linguistics," Archivum Linguisticum,
3, 126-136.
- - - . 1953. "Relationship in comparative linguistics," Transactions ofthe Philological
Society, 52-108.
ANDERSEN, HENNING. 1966. Tenues and mediae in the Slavic languages: a historical
investigation. Cambridge, Mass., Harvard University dissertation.
- - - . 1969. "A study in diachronic morphophonemics: the Ukrainian prefixes,"
Language, 45, 807-830.
- - . 1972. "Diphthongization," Language, 48, 11-50.
ANSHEN, FRANK. 1970. "A sociolinguistic analysis of a sound change," Language
Sciences, 9, 20-21 (February).
ANTTILA, RAIMO. 1968. "The relation between internal reconstruction and the
comparative method," Ural-Altaische Jahrbiicher, 40, 159-173.
- - - . 1969a. Uusimman iiiinnehistorian suunnasta ja luonteesta. Publications of the
Phonetics Department of the University of Turku, 5 (August).
- - - . 1969b. Proto-Indo-European schwebeablaut. University of California Publica-
tions in Linguistics, 58. Berkeley and Los Angeles.
ARENS, HANS. 1969. Sprachwissenschaft. Der Gang ihrer Entwicklung von der Antike
bis zur Gegenwart. 2 Freiburg and Munich, Karl Alber.
ARNDT, WALTER W. 1959. "The performance of glottochronology in Germanic,"
Language,35, 180-192.
ARON, ALBERT W. 1930. "The gender of English loan-words in colloquial American
German," Language Monographs, 7, 11-28 (December). Baltimore, Linguistic
Society of America.
AusTERLITZ, ROBERT. 1968. "L'Ouralien," Le langage. Encyclopedie de la Ph~iade,
25, 1331-1387. Paris, Gallimard.
AuziAS, JEAN-MARIE.-1967. Clefs pour le structuralisme. 2 Paris, Seghers.
AxMANOVA, OL'GA S. 1966. Slovar' lingvisticeskix terminov. Moscow, Sovetskaja
Enciklopedija.
* Superscript numbers indicate editions.
401
402 BIBLIOGRAPHY
BACH, ADOLF. 1934. Deutsche Mundartforschung. Ihre Wege, Ergebnisse und Aufgaben.
Heidelberg, Carl Winter.
- - . 1956. Geschichte der deutschen Sprache. 6 Heidelberg, Queiie & Meyer.
BAILEY, CHARLES-JAMES N. 1969. "The integration of linguistic theory: internal
reconstruction and the comparative method in descriptive linguistics," Working
Papers in Linguistics, 2, 85- 122 (March). Department of Linguistics, University of
Hawaii, Honolulu. [Also in Stockwell and Macaulay (eds.).]
BAKEL, JAN vAN. 1968. "Transformational etymology," Orbis, 17, 435-458.
BALTIMORE, DAVID, HOWARD M. TEMIN, and SATOSHI MIZUTANI. 1970.
"Viral RNA-dependent DNA polymerase," Nature, 226, 1209-1213 (June 27).
BARTHES, RoLAND. 1967. Elements of semiology (transl. by Annette Lavers and
Colin Smith). Cape Editions 4. London, Cape.
BAR TO LI, MATTEO. 1925. lntroduzione alia neolinguistica. Geneva, Olschki.
BASTIAN, OTTO. 1964. Die europiiischen Sprachen. Dalp-Taschenblicher. Bern and
Munich, Francke.
BAUGH, ALBERT C. 1951. A history of the English language. London, Routledge &
Kegan Paul.
BEADLE, GEORGE, and MURIEL BEADLE. 1967. The language of life. Anchor Books
A579. Garden City, N.Y., Doubleday.
BEALS, RALPH L., and HARRY HOIJER. 1965. An introduction to anthropology. 3
New York, Macmillan.
BECHERT, JOHANNES, DANIELE CLEMENT, WoLF THUMMEL, and KARL HEINZ
WAGNER. 1970. Einfiihrung in die generative Transformationsgrammatik. Lin-
guistische Reihe 2. Munich, Hueber.
BECKER, HENRIK. 1948. Der Sprachbund. Leipzig, Humboldt-Blicherei.
BENDIX, EDWARD H. 1966. Componential analysis ofgeneral vocabulary: the semantic
structure of a set of verbs in English, Hindi, Japanese. IJAL Publication 41 [40].
Bloomington, Ind., Indiana University.
BENEDIKTSSON, HREINN. 1970. "Aspects of historical phonology," The Nordic
languages and modern linguistics. Proceedings of the International Conference of
Nordic and General Linguistics, Univ. of Iceland, Reykjavik, July 6-11, 1969
(Hreinn Benediktsson, ed.), 87-142. Reykjavik, Visindafelag fslendinga.
BENNETT, HoBART. 1969. "Manifestations of i-umlaut in Old English," Linguistics,
50, 5-26.
BENNETT, WILLIAM H. 1968. "The operation and relative chronology of Verner's
Jaw," Language, 44, 219-223.
BENVENISTE, EMILE. 1939. "La nature du signe linguistique," Acta Linguistica, 1,
23-29. [Also in Readings in linguistics (E. Hamp, F. Householder, and R. Auster-
litz, eds.), 2, 104-108 (1966). Chicago, University of Chicago Press.]
- - - . 1954. "Probh!mes semantiques de la reconstruction," Word, 10, 251-264.
- - - . 1969. Le vocabulaire des institutions indo-europeennes. 2 vols. Paris, Minuit.
BERGSLAND, KNUT, and HANS VOGT. 1962. "On the validity of glottochronology,"
Current Anthropology, 3, 115-153.
BERNSTEJN, S. B. 1961. Ocerk sravnitel'noj grammatiki slavjanskix jazykov. Moscow,
Nauka.
BEVERIDGE, W. I. B. 1950. The art of scientific investigation. Vintage Books V-129.
New York, Vintage Books.
BHAT, D. N. SHANKARA. 1968. "Is sound change gradual?"Linguistics, 42,5-18
(August).
BIBLIOGRAPHY 403
BIGGS, BRUCE. 1965, "Direct and indirect inheritance in Rotuman," Lingua, 14,
383-445.
BIRDWHISTELL, RAY L. 1970. Kinesics and context: essays on body motion communi-
cation. Philadelphia, University of Pennsylvania Press.
BIRNBAUM, HENRIK, and JAAN PUHVEL (eds.). 1966. Ancient Indo-European
dialects. Berkeley and Los Angeles, University of California Press.
BLANSITT, EDWARD L., JR. (ed.). 1967. Tagmemic theory, current research in tag-
memic description, grammatical analysis. Monograph Series on Languages and
Linguistics No. 20 (18th Round Table Meeting). Washington, D.C., Georgetown
University Press.
BLOCH, BERNARD. 1948. "A set of postulates for phonemic analysis," Language,
24, 3-46. [Bobbs-Merrill Language-5.]
BLOOMFIELD, LEONARD. 1933. Language. New York, Holt.
BoAs, FRANZ. 1929. "Classification of American Indian languages," Language, 5, 1-7.
- - - . 1966. Race, language, and culture. New York, Macmillan (Free Press).
BocHENSKI, I. M. 1965. The methods of contemporary thought (trans!. by Peter Caws).
Harper Torchbooks TB/1377. New York, Harper & Row.
BOLINGER, DWIGHT L. 1949. "The sign is not arbitrary," Boletin del Instituto Caro
y Cuervo, 5, 52-62. Bogota.
- - - . 1968. Aspects of language. New York, Harcourt.
BoNFANTE, GIULIANO. 1945. "On reconstruction and linguistic method," Word, l,
83-94, 132-161.
- - - . 1946. "Additional notes on reconstruction," Word, 2, 155-156.
BOTTIGLIONI, GINO. 1954. "Linguistic geography: achievements, methods and
orientations," Word, 10, 375-387.
BouiSSAC, PAUL A. R. 1970. "The circus as a multimedia language," Language
Sciences, 11, 1-7 (August).
BRACE, C. LORING, and M. F. ASHLEY MONTAGU. 1965. Man's evolution, an intro-
duction to physical anthropology. New York, Macmillan.
BREAL, MICHEL. 1893. "On the canons of etymological investigation," Transactions
of the American Philological Association, 24, 17-28.
- - - . 1964 [1897]. Semantics: studies in the science of meaning (trans!. by Nina Cust).
T1228. New York, Dover.
BREMER, OTTO. 1894. "Relative Sprachchronologie," Indogermanische Forschungen,
4, 8-31.
BREND, RuTH M. 1970. "Tagmemic theory: an annotated bibliography," Journal of
English Linguistics, 4, 7-45.
BRIGHT, WILLIAM (ed.). 1966. Sociolinguistics. Proceedings of the UCLA sociolin-
guistics conference, 1964. Janua Linguarum, Series Maior 20. The Hague, Mouton.
BRIGHT, WILLIAM, and A. K. RAMANUJAN. 1964. "Sociolinguistic variation and
language change," Proceedings of the ninth international congress of linguists
(Horace Lunt, ed.) 1107-1113. The Hague, Mouton.
BROSNAHAN, L. F. 1960. "Language and evolution," Lingua, 9, 225-236.
- - - . 1961. The sounds of language. Cambridge, Heffer & Sons.
BRUGMANN, KARL. 1884. "Zur Frage nach den VerwandtschaftsverhiHtnissen der
indogermanischen Sprachen," Internationale Zeitschrift fiir allgemeine Sprach-
wissenschaft, 1, 226-256.
BuRLING, ROBBINS. 1970. Man's many voices. Language in its cultural setting. New
York, Holt.
404 BIBLIOGRAPHY
EBELING, C. L. 1962. Linguistic units. Janua Linguarum, Series Minor 12. The
Hague, Mouton.
EGG AN, FRED. 1954. "Social anthropology and the method of controlled compari-
son," American Anthropologist, 56, 743- 763. [Hobbs-Merrill A-57.]
EHRICH, ROBERT W., and GERALD M. HENDERSON. 1968. "Culture area," Inter-
national encyclopedia of the social sciences, 3, 563-568. New York, Macmillan
(Free Press).
EHRMANN, JACQUES (ed.). 1970. Structuralism. Anchor Books A 719. Garden City,
N.Y., Doubleday.
EISENSTADT, SHMUEL N. 1968. "Social evolution," International encyclopedia of
the social sciences, 5, 228-234. New York, Macmillan (Free Press).
ELLIS, JEFFREY. 1966. Towards a general comparative linguistics. Janua Linguarum,
Series Minor 52. The Hague, Mouton.
EMENEAU, MuRRAY B. 1956. "India as a linguistic area," Language, 32,3-16.
EMMET, DoROTHY. 1958. Function, purpose, and powers. New York, Macmillan.
EsPER, ERWIN A. 1966. "Social transmission of an artificial language," Language,
42, 575-580.
HAAS, MARY R. 1966. "Historical linguistics and the genetic relationship oflanguages,"
Current Trends in Linguistics (Thomas A. Sebeok, ed.), 3, 112-153. The Hague,
Mouton. [Expanded version The prehistory of languages, Janua Linguarum,
Series Minor 57, 1969.]
BIBLIOGRAPHY 409
HAKULINEN, LA URI. 1961. The structure and development of the Finnish language
(transl. by John Atkinson). Bloomington, Ind., Indiana University Press.
HALL, EDWARD T. 1959. The silent language. T402. Greenwich, Conn., Fawcett
Publications.
HALL, RoBERT A., JR. 1950. "The reconstruction of Proto-Romance," Language,
26, 6-27. [Also in Readings in linguistics (Martin Joos, ed.), 1, 303-314 (1958).
University of Chicago Press.]
- - - . 1958. "Creolized languages and 'genetic relationships'," Word, 14, 367-37:1
- - - . 1960. "On realism in reconstruction," Language, 36, 203-206.
- - - . 1963. Idealism in Romance linguistics. Ithaca, N.Y., Cornell University
press.
- - - . 1964. Introductory linguistics. Philadelphia, Chilton Books.
- - - . 1966. Pidgin and creole languages. Ithaca, N.Y., Cornell University Press.
- - - . 1968. "Comparative reconstruction in Romance syntax," Acta Linguistica
Hafniensia, 11, 81-88.
- - - . 1969. "Italian serpe, serpente again," Language Sciences, 8, 27-28 (December).
HALLDORSSON, HALLDOR. 1970. "Determining the lending language," The Nordic
languages and modern linguistics, 365-378. See Benediktsson.
HALLE, MoRRIS. 1962. "Phonology and generative grammar," Word, 18, 54-72.
HALLOWELL, A. IRVING. 1956. "The structural and functional dimensions of a
human existence," The Quarterly Review of Biology, 31, 88-101. [Bobbs-Merrill
A-105.]
HAMMARSTROM, GoRAN. 1966. Linguistische Einheiten im Rahmen der modernen
Sprachwissenschaft. Kommunikation und Kybernetik in Einzeldarstellungen 5.
New York, Springer.
HAMMERICH, Louis L. 1954. "The Russian stratum in Alaskan Eskimo," Word, 10,
401-428.
RAMP, ERIC P. 1963. A glossary of American technical linguistic usage, I925-I950.
Utrecht, Spectrum.
HARRIS, JAMES W. 1969. "Sound change in Spanish and the theory of markedness,"
Language, 45, 538-552.
HARRIS, ZELLIG S. 1951. Structural linguistics. Phoenix Books P52. Chicago,
University of Chicago Press.
HAss, WILBUR A. 1969. "The psychological reality of different types of phonological
change," Papers from the fifth regional meeting of the Chicago Linguistic Society,
343-347. Chicago, Department of Linguistics, University of Chicago.
HAUGEN, EINAR. 1950. "The analysis of linguistic borrowing," Language, 26, 210-
231. [Also in Lass (ed.), 1969:58-81.]
HAVERS, WILHELM. 1931. Handbuch der erkliirenden Syntax. Ein Versuch zur
Erforschung der Bedingungen und Triebkriifte in Syntax und Stilistik. Heidelberg,
Carl Winter.
HAVRANEK, BoHUSLAV. 1966. "Zur Problematik der Sprachmischung," Trauaux
linguistiques de Prague, 2, 81-98.
HENLE, PAUL (ed.). 1965. Language, thought and culture Ann Arbor Paperbacks 97.
Ann Arbor, Mich., University of Michigan Press.
HERMAN, L., and M.S. HERMAN. 1958. Foreign dialects; a manual for actors, directors,
and writers. New York, Theater Arts Books.
HERMANN, EDUARD. 1907. "Uber das Rekonstruieren," Zeitschriftfur Vergleichende
Sprachforschung, 41, 1-64.
410 BIBLIOGRAPHY
- - - . 1931. Lautgesetz und Analogie. Abhandlungen der Gesellschaft der Wissen-
schaften zu Gottingen, Phil.-Hist. Klasse, Neue Folge, Band 23,3.
HILL, ARCHIBALD A. 1936. "Phonetic and phonemic change," Language, 12, 15-22.
- - (ed.). 1969. Linguistics today. New York, Basic Books.
HJELMSLEV, Lours. 1963. Prolegomena to a theory of language (trans!. by Francis
J. Whitfield). Madison, Wis., University of Wisconsin Press.
- - - . 1970. Language: an introduction (trans!. by Francis J. Whitfield). Madison,
Wis., University of Wisconsin Press.
HocKETT, CHARLES F. 1948a. "Implications of Bloomfield's Algonkian studies,"
Language, 24, 117-131. [Also in Readings in linguistics (Martin Joos, ed.), 1,
281-289 (1958). Chicago, University of Chicago Press, and Hymes (ed.) 1964:
599-611.]
- - - . 1948b. "Biophysics, linguistics, and the unity of science," American Scientist,
36, 558-572. [Bobbs-Merrill A-115.]
- - - . 1958. A course in modern linguistics. New York, Macmillan.
- - - . 1960. "The origin of speech," Scientific American, 3-11 (September).
[Scientific American Reprints 603.]
- - - . 1961. "Linguistic elements and their relations," Language, 37, 29-53.
- - . 1965. "Sound change," Language, 41, 185-204.
- - - . 1967. "Where the tongue slips, there slip 1," To honor Roman Jakobson, 2,
910-936. Janua Linguarum, Series Maior 21. The Hague, Mouton.
HOCKETT, CHARLES F., and ROBERT AsCHER. 1964. "The human revolution,"
Current Anthropology, 5, 135-168. [Bobbs-Merrill A-306.]
HODGE, CARLETON T. 1970. "The linguistic cycle," Language Sciences, 13, 1-7
(December).
HOENIGSWALD, HENRY M.1946. "Soundchangeandlinguisticstructure," Language,
22, 138-143. [Bobbs-Merrill Language-48.]
- - - . 1950. "The principal step in comparative grammar," Language, 26, 357-364.
[Bobbs-Merrill Language-49, and Readings in linguistics (Martin Joos, ed.), 1,
298-302 (1958). Chicago, University of Chicago Press.]
- - - . 1955. "Change, analogical and semantic," Indian Linguistics, 16, 233-236.
- - - . 1960a. Language change and linguistic reconstruction. Phoenix Books P178.
Chicago, University of Chicago Press.
- - - . 1960b. "Phonetic similarity in internal reconstruction," Language, 36, 191-192.
- -- . 1963. "On the history of the comparative method," Anthropological Linguistics,
5, 1-11.
- - - . 1964a. "Allophones, allomorphs, and conditioned change," Proceedings of
the ninth international congress of linguists (Horace Lunt, ed.), 645-649. The Hague,
Mouton.
- - - . 1964b. "Graduality, sporadicity, and the minor sound change processes,"
Phonetica, 11, 202-215.
- - - . 1965. "Phonetic reconstruction," Proceedings of the fifth international congress
of phonetic sciences (E. Zwirner, ed.), 25-42. Basel, Karger.
- - - . 1966. "Criteria for the subgrouping of languages," in Birnbaum and Puhvel
(eds.), 1-12.
H 6 FLER, OTTO. 1958. Die zweite Lautverschiebung bei Ostgermanen und Westgermanen.
Ttibingen, Niemeyer.
HoFMANN, J. B., and H. RuBENBAUER. 1963. Worterbuch der grammatischen und
metrischen Terminologie. Heidelberg, Carl Winter.
BIBLIOGRAPHY 411
HOG BEN, L. 1956. "Human biology and human speech," British Journal ofPreventive
Medicine, 10, 63-74.
HOIJER, HARRY. 1939. "Chiricahua loan-words from Spanish," Language, 15,
110-115.
- - - . 1946. "Introduction" to Linguistic structures of native America. Viking Fund
Publications in Anthropology 6, 9-29. [Bobbs-Merrill A-118.]
- -- . 1948. "Linguistic and cultural change," Language, 24, 335-345. [Also in
Hymes (ed.), 1964:455-466.]
- - - (ed.). 1954. Language in culture. Chicago, Chicago University Press.
HoLMBERG, ALLAN R. 1954. "Adventures in culture change," Method and perspective
in anthropology (Robert F. Spencer, ed.), 103-113. Minneapolis, University of
Minnesota Press. [Bobbs-Merrill A-119.]
Ho:REJsf, VLADIMiR. 1964. "Chronologie relative et linguistique structurale,"
Travaux linguistiques de Prague, 1, 107-110.
HoRMANN, HANS. 1970. Psycho/inguistics. An introduction to research and theory
(transl. from rev. ed. by H. H. Stern). New York, Springer.
HoRNE, KIBBEY M. 1966. Language typology: 19th and 20th century views. School
of Languages and Linguistics Publications. Washington, D.C., Georgetown
University Press.
HousEHOLDER, FRED. W. 1960. "First thoughts on syntactic indices," International
Journal of American Linguistics, 26, 195-197.
HousTON, SusAN. 1969. "A sociolinguistic consideration of the Black English
children in northern Florida," Language, 45, 599-607.
- - - . 1970. "Competence and performance in Child Black English," Language
Sciences, 12, 9-14 (October).
HUMBOLDT, WILHELM VON. 1970 [1836]. Linguistic variability and intellectual
development (transl. by George C. Buck and Frithjof A. Raven). Miami Linguistics
Series 9. Coral Gables, Fla., University of Miami Press.
HuNTLEY, D. G. 1968. "Two cases of analogical feature substitution in Slavic,"
Language, 44, 501-506.
HuxLEY, JULIAN. 1953. Evolution in action. London, Chatto & Windus.
HYMAN, LARRY M. 1970a. "How concrete is phonology?," Language, 46,
58-76.
- - - . 1970b. "The role of borrowing in the justification of phonological grammars,"
Studies in African Linguistics (UCLA), 1, 1-48.
HYMES, DELL H. 1956. "Na-Dene and positional analysis of categories," American
Anthropologist, 58, 624-638.
- - - . 1959. "Genetic classification: retrospect and prospect," Anthropological
Linguistics, 1, 50-66. [Bobbs-Merrill A-125.]
- - - . 1960. "Lexicostatistics so far," Current Anthropology, 1, 3-44. [Bobl:s-Merrill
A-126.]
- - - . 1961. "Functions of speech: an evolutionary approach," Anthropology and
education (Frederick C. Gruber, ed.), 55-83. Philadelphia, University of Pennsyl-
vania Press. [Bobbs-Merrill A-124.]
- -- (ed.). 1964. Language in culture and society. A reader in linguistics and anthropol-
ogy. New York, Harper & Row.
- - - . 1968a. "Linguistics: the field," International encyclopedia of the social sciences,
9, 351-371. New York, Macmillan (Free Press).
- - - . 1968b. "Sociolinguistics," Language Sciences, 1, 23-26 (May).
412 BIBLIOGRAPHY
ILLIC-SYITYC, V. M. 1967. "Materialy k sravnitel'nomu slovarju nostraticeskix
jazykov," Etimologija 1965, 321-373. Moscow, Nauka.
- - -. 1968. "Sootvestvija smycnyx v nostraticeskix jazykax," Etimologija 1966,
304-355, 401-410. Moscow, Nauka.
IR YING, JOHN A. 1949. "The comparative method and the nature of human nature,"
Philosophy and Phenomenological Research, 9, 545-557.
ITKONEN, ERKKI. 1941. "Uber den Charakter des ostlappischen Stufenwechsel-
systems," Finnisch-Ugrisclre Forsc/wngen, 27, 137-167.
- - - . 1946. Struktur und Entwicklung der ostlappischen Quantitiitssysteme. Memoires
de Ia Societe Finno-Ougrienne 88. Helsinki, Suomalais-ugrilainen Seura.
- - - . 1966. Kieli ja sen tutkimus. Helsinki and Porvoo, Soderstrom.
ITKONEN, TERHO. 1970. "Ovatko aanteenmuutokset vahittaisia vai harppauksel-
Jisia ?," Virittiijii, 74:4, 411-438.
ITKONEN, T. I. 1958. Kolton- ja kuolanlapin sanakirja (Worterbuch des Kolta- und
Kolalappischen). 2 vols. Lexica Societatis Fenno-Ugricae 15. Helsinki, Suomalais-
ugrilainen Seura.
IYIC, MILKA. 1965. Trends in Linguistics. Janua Linguarum, Series Minor 42. The
Hague, Mouton.
J ACOBSOHN, HERMANN. 1922. Arier und Ugrofinnen. Gottingen, Vandenhoeck &
Ruprecht.
J AKOBSON, RoMAN. 1949. "The phonemic and grammatical aspects of language in
their interrelation," Proceedings of the sixth international congress of linguists
(Michel Lejeune, ed.), 5-18. Paris, Klincksieck.
- - - . 1958. "Typological studies and their contribution to historical linguistics,"
Proceedings of the eighth international congress of linguists (Eva Sivertsen, ed.),
17-35. Oslo, University Press.
- - - . 1965. "Quest for the essence of language," Diogenes, 51, 21-37.
- - - . 1968. Child language, aphasia and phonological universals (trans!. by Allan R.
Keiler). Janua Linguarum, Series Minor 72. The Hague, Mouton.
- - -. 1969. "Linguistics in its relation to other sciences," Proceedings of the tenth
international congress of linguists, 1, 75-122. See Graur (ed.).
JANERT, KLAUS LuowiG. 1961. "Studien zu den Asoka-Inschriften. III," Nach-
ric/rten der Akademie der Wissenschaften zu Gottingen.l. Phil.-Hist. Klasse, No. 1,
1-25.
JENSEN, HANS. 1969. Die Schrift. 3 Berlin, VEB deutscher Verlage der Wissenschaften.
JESPERSEN, OTTO. 1941. Efficiency in linguistic change. Det Kg!. Danske Viden-
skabernes Selskab. Hist.-fil. Meddelelser 27, 4.
- - - . 1956. Growth and structure of the English language. Oxford, Blackwell.
- - - . 1964. Language, its nature, development and origin. The Norton Library
N229. New York, Norton.
Joos, MARTIN. 1952. "The Medieval sibilants," Language, 28, 222-231. [Also in
Readings in linguistics (Martin Joos, ed.), 1, 377-378 (1958). Chicago, University
of Chicago Press.]
- - - . 1962. The five clocks. IJAL Publication 22. Bloomington, Indiana University.
JucQUOIS, GuY. 1970. "La theorie de Ia racine en indo-europeen," La Linguistique,
6, 69-102.
KALIMA, JALO. 1915. Die ostseefinnischen Lehnworter im Russischen. Helsinki,
Suomalaisen Kirjallisuuden Seura.
BIBLIOGRAPHY 413
- - - . 1936. Itiimerensuomalaisten kielten balttilaiset lainasanat. Helsinki, Suomalaisen
Kirjallisuuden Seura.
KApLAN, ABRAHAM. 1964. The conduct of inquiry. Methodology for behavioral science.
San Francisco, Chandler Publishing Co.
KATICIC, RADOS LAY. 1966. "Der Entsprechungsbegritf in der vergleichenden Laut-
und Formenlehre," Indogermanische Forschungen, 71, 203-220.
- - - . 1970. A contribution to the general theory of comparative linguistics. Janua
Linguarum, Series Minor 83. The Hague, Mouton.
KAZAZIS, KosTAS. 1967. "On a generative grammar of the Balkan languages,"
Foundations of Language, 3, 117-123.
- - - . 1968. "Sunday Greek," Papers from the fourth regional meeting of the Chicago
Linguistic Society, 130-140. Chicago, Department of Linguistics, University of
Chicago.
- - - . 1970. "The relative importance of parents and peers in first-language learning,"
General Linguistics, 10, 111-120.
- - - . 1972. "The status of Turk isms in the present-day Balkan languages," Aspects
of Balkans: continuity and change (H. Birnbaum and S. Vryonis, eds.), 87-116.
The Hague, Mouton.
KElLER, ALLAN R. 1970. A phonological study of the Indo-European laryngeals.
Janua Linguarum, Series Practica 76. The Hague, Mouton.
KENT, ROLAND G. 1936. "Assimilation and dissimilation," Language, 12, 245-258.
KETTUNEN, LAURI. 1962. Eestin kielen iiiinnehistoria. 3 Suomalaisen Kirjallisuuden
Seuran Toimituksia 156. Helsinki.
KEYSER, SAMUEL JAY. 1963. "Review of The pronunciation of English in the Atlantic
states, by Hans Kurath and Raven I. McDavid," Language, 39, 303-316.
KING, ROBERT D. 1967. "Functional load and sound change," Language, 43,831-852.
- - - . 1969a. "Push chains and drag chains," Glossa, 3, 3-21.
- - - . 1969b. Historical linguistics and generative grammar. Englewood Cliffs, N.J.,
Prentice-Hall.
KIPARSK Y, PAuL. 1965. Phonological change. Cambridge, Mass., M.I.T. dissertation.
- - - . 1968a. "Linguistic universals and linguistic change," Universals in linguistic
theory (Emmon Bach and Robert T. Harms, eds.), 170-202. New York, Holt.
- - - . 1968b. "Tense and mood in Indo-European syntax," Foundations of Language,
4, 30-57.
KIRN, PAuL. 1963. Einfiihrung in die Geschichtswissenschaft. Sammlung Goschen
270. Berlin, de Gruyter.
KNIGHT, THOMAS S. 1965. Charles Peirce. W 885. New York, Washington Square
Press.
KNOBLOCH, JOHANN. 1961f. Sprachwissenschaft/iches Worterbuclz. Heidelberg, Carl
Winter.
KocH, WALTER A. 1970. Zur Theorie des Lautwandels. Hildesheim/New York,
Olms.
KoHLER, KLAUS J. 1967. "Structural dialectology," Zeitschriftfiir Mundartforschung,
34,40-44.
KoiVULEHTO, JORMA. 1970. "Suomen laiva-sanasta," Virittiijii, 74:2, 178-183.
KouTSOUDAS, ANDREAS. 1966. Writing transformational grammars: an introduction.
New York, McGraw-Hill.
KRAFT, V. 1955. "Geschichtsforschung als exakte Wissenschaft," Anzeiger der
Osterreichischen Akademie der Wissenschaften, 19, 239-251.
414 BIBLIOGRAPHY
KRAHE, HANS. 1970. Einleitung in das vergleichende Sprachstudium (ed. by Wolfgang
Meid). Innsbruck, Sprachwissenschaftliches lnstitut der Universitiit.
KRAMSKY, JiiU. 1969. The word as a linguistic unit. Janua Linguarum, Series Minor
75. The Hague, Mouton.
KROEBER, A. L. 1913. "The determination of linguistic relationship," Anthropos
8, 389--401.
- - - . 1919. "On the principle of order in civilization as exemplified by changes of
fashion," American Anthropologist, 21, 235-263. [Bobbs-Merrill A-137.]
- - - . 1935. "History and science in anthropology," American Anthropologist, 37,
539-569. [Bobbs-Merrill A-136.]
- - - . 1944. Configuration of culture growth. Berkeley and Los Angeles, University of
California Press.
- - - (ed.). 1953. Anthropology today. An encyclopedic inventory. Chicago, University
of Chicago Press.
- - - . 1960. "On typological indices 1: ranking of languages," International Journal
of American Linguistics, 26, 171-177.
- - - . 1963a. Anthropology: culture patterns and processes. A Harbinger Book H018.
New York, Harcourt.
- - - . 1963b. Anthropology: biology and race. A Harbinger Book H017. New York,
Harcourt.
- - . 1963c. An anthropologist looks at history. California Paperbacks Cal 87.
Berkeley and Los Angeles, University of California Press.
KROEBER, A. L., and C. D. CHRETIEN. 1937. "Quantitative classification of Indo-
European languages," Language, 13, 83-103.
KROHN, KAARLE. 1926. Die folkloristische Arbeitsmethode. Cambridge, Mass.,
Harvard University Press.
KROHN, ROBERT K. 1969. English vowels. Ann Arbor, University of Michigan
dissertation; University Microfilms 70-4123.
KRONASSER, HEINZ. 1952. Handbuch der Semasiologie. Heidelberg, Carl Winter.
KUIPERS, AERT H. 1968. "Unique types and typological universals," Pratidtinam
(Indian, Iranian, and Indo-European studies presented to F. B. J. Kuiper on his
60th birthday, ed. by J. C. Heesterman, et alii), 68- 88. The Hague, Mouton.
KuRATH, HANS, and RAVEN I. McDAVID, JR. 1961. The pronunciation of English
in the Atlantic states. Ann Arbor, Mich., University of Michigan Press.
KuRYLOWicz, JERZY. 1945- 1949. "La nature des proces dits 'analogiques'," Acta
Linguistica, 5, 15- 37. [Also in Readings in Linguistics (E. Hamp, F. Householder,
and R. Austerlitz, eds.), 2, 158-174 (1966). Chicago, University of Chicago
Press.]
- - - . 1964. "On the methods of internal reconstruction," Proceedings of the ninth
international congress of linguists (Horace Lunt, ed.), 9-36. The Hague, Mouton.
- - - . 1968. "A remark on Lachmann's law," Harvard Studies in Classical Philology,
72, 295-299.
MAAS, PAUL. 1958. Textual criticism (trans!. by Barbara Flower). New York, Oxford
University Press.
MAHER, J. PETER. 1965. Indo-European origins of some Slavic grammatical categories.
Bloomington, Ind., Indiana University dissertation; University Microfilms
66-1474.
- - - . 1966. "More on the history of the comparative method: the tradition of
Darwinism in August Schleicher's work," Anthropological Linguistics, 8, 1- 12.
- - - . 1969a. "English-speakers' awareness of the distinctive features," Language
Sciences, 5, 14 (April).
- - - . 1969b. "Phonetic law and suppletion in the history of the paradigm of the
Slavic ethnica in -jane," General Linguistics, 9, 167-172.
- - - . 1969c. "The paradox of creation and tradition in grammar: sound pattern
of a palimpsest," Language Sciences, 7, 15-24 (October).
- - - . 1969d. "Italian serpe: why not *serpente?" Language Sciences, 6, 5-8 (August),
8, 30-35 (December).
- - - . 1970a. "A note on the English passive with reference to some general questions
of theory," Language Sciences, 13, 30-31 (December).
- - - . 1970b. "Distinctive-feature rhyme in German folk versification," Language
Sciences, 19, 19-20 (February 1972).
- - - . 1970c. "Italian mostaccio," Zeitschrift fur romanische Philologie, 87, 320-333
(1971).
- - - . 1970d. "Sociological reality in linguistic theories" (to appear).
- - - . 1970e. "Etymology and generative phonology in traditiona11exicon: a study
of Latin aqua 'water', aquila 'eagle', aquilus 'dark', aquila 'north wind'," General
Linguistics, 11, 71-98 (1971).
BIBLIOGRAPHY 417
M ALKIEL, Y AKOV. 1967. "Multiple versus simple causation in linguistic change,"
To honor Roman Jakobson, 2, 1228-1246. Janua Linguarum, Series Maior 21.
The Hague, Mouton.
- - - . 1968. Essays on linguistic themes. Berkeley and Los Angeles, University of
California Press.
MALMBERG, BERTIL. 1963a. Phonetics. Tl024. New York, Dover.
- - - . 1963b. Structural linguistics and human communication. Kommunikation und
Kybernetik in Einzeldarstellungen 2. New York, Academic Press (Berlin, Springer).
- - - . 1964. New trends in linguistics (trans!. by Edward Carney). Bibliotheca
Linguistica 1. Stockholm and Lund, Naturmetodens Spn1kinstitut.
- - - . 1969. "Synchronie et diachronie," Proceedings of the tenth international
congress of linguists, 1, 13-36. See Graur (ed.).
MANCZAK, WITOLD. 1958. "Tendences generales des changements analogiques,"
Lingua, 7, 298-325, 387-420.
- -- . 1968. "Le developpement phonetique irregulier dfl a Ia frequence en russe,"
Lingua, 21, 287-293.
- - - . 1969. Le developpement phonetique des langues romanes et lafrequence. Krakow,
Nakladen Uniwersytetu Jagiellonskiego.
MARCHAND, JAMES W. 1956. "Internal reconstruction of phonemic split," Language,
32, 245-253.
MAROUZEAU, J. 1961. Lexique de Ia terminologie linguistique (fran~ais, allemand,
anglais, italien). 3 Paris, Geuthner.
MARTIN, SAMUEL £.1966. "Lexical evidence relating Korean to Japanese," Lan-
guage, 42, 185-251.
MARTINET, ANDRE. 1958. "Function, structure and sound change," Word, 8, 1-32.
- - -. 1964. Economie des changements phonetiques. 2 Bern, Francke.
- - - (ed.). 1968. Le langage. Encyclopedie de Ia Pleiade, 25. Paris, Gallimard.
MEETHAM, A. R., and R. A. HUDSON (eds.). 1969. Encyclopedia of linguistics,
information and control. Elmsford, N.Y., Pergamon Press.
MElD, WoLFGANG. 1968a. Indogermanisch und Keltisch. Innsbrucker Beitriige zur
Kulturwissenschaft, Sonderheft 25. Innsbruck, Sprachwissenschaftliches Institut
der UniversiHit.
- - - . 1968b. "Indo-European and Celtic," Scottish Studies, 12, 45-56.
MEILLET, ANTOINE. 1926. Linguistique historique et linguistique generale. 2 Paris,
Champion.
-. - - . 1967a. The comparative method in historical linguistics (trans!. by Gordon B.
Ford, Jr.). Paris, Champion.
- - - . 1967b. The Indo-European dialects (trans!. by Samuel N. Rosenberg). Alabama
Linguistic and Philological Series 15. University, Ala., University of Alabama Press.
MEILLET, ANTOINE, and MARCEL CoHEN (eds.). 1952. Les langues du monde.
New ed. Paris, Champion.
MEISINGER, OTHMAR. 1932. Vergleichende Wortkunde. Beitriige zur Bedeutungslehre.
Munich, Beck'sche Verlagsbuchhandlung.
MENNER, Robert J. 1936. "The conflict of homonyms in English," Language, 12,
229-244.
- - - . 1945. "Multiple meaning and change of meaning in English," Language, 21,
59-76.
MILLER, Rov ANDREW. 1967. "Old Japanese phonology and the Korean-Japanese
relationship," Language, 43, 278-302.
4I8 BIBLIOGRAPHY
PAOR, LIAM DE. 1967. Archaeology, an illustrated introduction. Pelican Books A859.
Baltimore, Penguin.
PATTERSON, WILLIAM T. 1968. "On the genealogical structure of the Spanish
vocabulary," Word, 24, 309- 339.
PAUL, HERMANN. 1920. Prinzipien der Sprachgeschichte. 5 Halle, Niemeyer.
PEr, MARIO. 1966. Glossary of linguistic terminology. New York, Columbia University
Press.
PEIRCE, CHARLES SANDERS. 1955. Philosophical writings of Peirce (ed. by Justus
Buchler). T217. New York, Dover.
PENZL, HERBERT. 1957. "The evidence for phonemic changes," Studies presented to
Joshua Whatmough (Ernst Pulgram, ed.), 193-208. The Hague, Mouton. [Also
in Lass (ed.), 1969:10-24.]
PETROVICI, EMIL. 1957. Kann das Phonemsystem einer Sprache durchfremden Einf/uss
umgestaltet werden? Janua Linguarum, Series Minor 3. The Hague, Mouton.
- - -. 1969. "Interpenetration des systemes linguistiques," Proceedings of the tenth
international congress of linguists, 1, 37-73. See Graur (ed.).
PrAGET, JEAN. 1968. Le structuralisme. 3 (Que sais-je?), No. 1311. Paris, Presses
Universitaires de France. [The English version, Structuralism, trans!. and ed. by
Chaninah Maschler. Harper Torchbooks TB/1610. New York, Harper & Row,
1971.]
PIKE, KENNETH L. 1947. "Grammatical prerequisites to phonemic analysis," Word,
3, 155-172. [Bobbs-Merrill Language-72.]
- - - . 1957. Axioms and procedures for reconstruction in comparative linguistics:-an
experimental syllabus. Rev. ed. Santa Ana, Summer Institute of Linguistics.
- - - . 1966. "A guide to publications related to tagmemic theory," Current trends
in linguistics, 3, 365-394. See Sebeok (ed.).
- -- . 1967. Language in relation to a unified theory of the structure ofhuman behavior. 2
Janua Linguarum, Series Maior 24. The Hague, Mouton.
PrLATI, LEONA LAMPI. 1969. "The Fox dialect: the influence of Finnish on a local
American English dialect," Neuphilologische Mitteilungen, 70, 145-158.
PINNOW, HEINZ-JURGEN. 1964. Die nordamerikanischen Indianersprachen: ein
Oberblick ilber ihren Bau und ihre Besonderheiten. Wiesbaden, Harrassowitz.
420 BIBLIOGRAPHY
PISANI, VITTORE. 1966. "Entstehung von Einzelsprachen aus Sprachbiinden,"
Kratylos, 11, 125-141.
PoRziG, WALTER. 1924. "Aufgaben der indogermanischen Syntax," Stand und
Aufgaben der Sprachwissenschaft: Festschrift fiir Wilhelm Streitberg (J. Friedrich
et alii, eds.), 126-151. Heidelberg, Carl Winter.
- - - . 1954. Die Gliederung des indogermanischen Sprachgebiets. Heidelberg, Carl
Winter.
PosTAL, PAUL M. 1968. Aspects of phonological theory. New York, Harper & Row.
Posn, LA URI. 1942. Grundziige der livischen Lautgeschichte. Memoires de la Societe
Finno-Ougrienne 85. Helsinki, Suomalais-ugrilainen Seura.
- - - . 1948. "Till fragan om brytning och omljud," Sprakvetenskapliga siillskapets i
Uppsala forhandlingar 1946-1948. Uppsala Universitets Arsskrifter, 13, 39-59.
- - - . 1953. "From Pre-Finnic to late Proto-Finnic," Finnisch-Ugrische Forschungen,
31, 1-91.
PowELL, J. W. 1966. Indian linguistic families (Preston Holder, ed.). Lincoln, Nebr.,
University of Nebraska Press.
PULGRAM, ERNST. 1951. "Phoneme and grapheme: a parallel," Word, 7, 15-20,
[Bobbs-Merrill Language-76.]
- - - . 1953. "Family tree, wave theory and dialectology," Orbis, 2, 67-72.
- - - . 1959. "Proto-Indo-European reality and reconstruction," Language, 35, 421-
426.
- - - . 1961. "The nature and use of proto-languages," Lingua, 10, 18-37.
- - - . 1964. "Proto-languages as proto-diasystems: Proto-Romance," Word, 20,
373-383.
- - - . 1965. "Graphic and phonic systems: figurae and signs," Word, 21, 208-224.
RAFFLER ENGEL, WALBURG A VON. 1970. "The LAD, our underlying unconscious,
and more on 'felt sets'," Language Sciences, 13, 15-18 (December).
RAMANUJAN, A. K. 1968. "The structure of variation: a study in caste dialect,"
Structure and change in Indian society (Milton Singer and Bernard Cohn, eds.),
461-474. Viking Fund Publications in Anthropology 47. Chicago, Aldine.
RAUN, ALo. 1968. "Native speaker and distinctive features," Lingua, 21, 364-350.
RA VILA, PAAVO. 1952. "Onomatopoieettisten ja deskriptiivisten sanojen asema
kielen iHinnesysteemissa," Virittiijii, 56:4, 262-274.
REDDY, MICHAEL J. 1969. "A semantic approach to metaphor," Papers from the
fifth regional meeting of the Chicago Linguistic Society, 240-251. Chicago, Depart-
ment of Linguistics, University of Chicago.
REICHSTEIN, RuTH. 1960. "Etude des variations sociales et geographiques des faits
linguistiques," Word, 16, 55-99.
REID, T. B. W. 1956. "Linguistics, structuralism and philology," Archivum Linguis-
ticum, 8, 28-37.
RICE, STUART A. (ed.). 1931. Methods in social science. A case book. Chicago, Univer-
sity of Chicago Press.
RISCH, ERNST. 1966. "Historisch-vergleichende Sprachbetrachtung und Dialekt-
geographie," Kratylos, 11, 142-155.
ROBINS, R. H. 1964. Genera! linguistics, an introductory survey. Longmans' Linguistics
Library. London, Longmans (Bloomington, Ind., Indiana University Press).
- - - . 1967. A short history of linguistics. Bloomington, Ind., Indiana University
Press.
BIBLIOGRAPHY 421
RoHLFS, GERHARD. 1922-1923. "Apul. ku, kalabr. mu und der Verlust des Infinitivs
in Unteritalien," Zeitschrift fiir romanische Philologie, 42, 211-223.
RoMEO, LUIGI. 1969 "Italian serpente: why also serpe," Language Sciences, 8, 28-30
(December).
RoMNEY, A. KIMBALL. 1957. "The genetic model and Uto-Aztecan time perspec-
tive," Davidson Journal of Anthropology, 3, 35-41. [Bobbs-Merrill A-192.]
Ross, A. S. C. 1965. Etymology. University Paperbacks 110. London, Methuen.
RUWET, NICOLAS. 1967. Introduction a Ia grammaire generative. Paris, Pion.
Boolean algebra, 342-46, 369-72 comparative linguistics, 3-4, 19, 20, 21, 43-45,
borrowing: 104, 144, 154-77, 180, 230-31 , 241, 202, 212-13, 327
316, 328-29, 332, 382, 392, 394; direction of, comparative method: 50, 229-56, 266, 278-82,
158-60; as evidence of cultural contacts, 298, 331, 385; and alternations, 250-52; in
162- 63; as evidence of earlier stages, 160- anthropology, 383; in biology, 394; and
61; and geographical position, 163-64; in- grammar, 364-66; limitations of, 244, 252,
ternal evidence of, 164-67; and levels of 254, 280; and segments or features, 245; in
grammar, 167-71; pronunciation borrowing, synchrony, 47-50, 282-84, 383-84
156, 157, 162; resistance to, 169, 188; and comparative-method machines, 203
structure of lexicon, 161-62; and stylistic competence, 107, 123, 124, 129
levels, 162; from within, 141, 145, 328 complementarity: of methods, 278; of methods
bow-wow theory, 28 of reasoning, 196-98, 202; of scliools of
Braille, 40 thought, 323-25, 341
brain order, 113 complementary distribution, 208, 215, 220-21,
break in traditions, 175 224, 237, 239, 268-70
Breton, 173 components, 6-11
British English, 42, 90-91, 201, 283 compounds, 38, 144
broad transcription, 207, 212, 222 concreteness, 10
buffer zone, 182-83 conditioning, 62, 229
Bulgarian, 284 confirmation, 240- 42
Burmese, 170, 172 conjunctive definition, 343, 369
consonant gradation, 97, 122, 221-22, 250-55,
345-46
call systems, 27 contamination, 42, 71, 76, 91, 95, 299
calque, 140 context: and protomeaning, 370; and semantic
capacity for language, 20, 27, 176, 198 reconstruction, 372
case syntax, 364-66 contiguity, 142
Catalan, 298 contrast, 188, 208-12, 214, 231, 233, 235, 247,
cause and effect, 13, 127-29, 194 248, 251
causes of change, 179-81, 193, 380 contrastive linguistics (analysis), 21, 316
Celtic, 171, 304-305 convergence, 172-74, 394
center of gravity (diversity), 387, 395, 397 coordination, 151
chance: exclusion of, 230, 231, 385 Coptic, 311
change: 13, 318; all-pervasiveness of, 179; bio- core pattern, 292
logical, 390-93; and description, 12; in eval- correspondence: 11, 27, 28, 159, 165-66, 170,
uation, 148-49; irreversibility of, 394; mech- 335-41 passim; among dialects, 48; form-
anisms of, 179-81, 195; as persistence and meaning, 100-101; language-writing, 34;
existence, 380, 393; primary, 124, 130; and sound-letter, 43
sign types, 14, 19, 133-34; and social regis- creation, 106, 125, 127, 153
ter, 128; and usage, 128; and variation, 84- creole, 176
85 cross-cultural comparison, 383
checking the reconstruction, 224, 239-42, 253- cultural association, 384
55, 318, 346, 366, 371, 386 cultural context, 153
Cheremis, 74, 164, 170, 299, 301 cultural diffusion, 291, 382
Cheyenne, 387 cultural expansion, 139
children's speech, 89, 92, 93, 101, 105, 113, 126, cultural reconstruction, 383-87
127, 176, 194-95, 200-201 cultural seriation, 384
Chinese, 38, 45, 79, 231, 311, 313, 317, 320 cultural situation, 146
Chinook Jargon, 173, 176 cultural universals, 383
Chiricahua, 157- 58 cultural variation, 190
choice: of labels, 243, 244; selection for survi- culture: 19-20, 310; and biology, 378-79;
val, 330, 391 change in, 199, 379-83
classification: areal, 321- 22; as explanation, cumulative evidence, 173, 385
179-80; genealogical, 317, 318-22; of lan- Czech, 75, 78, 199
guages, 11, 310-16; of mixed languages, 176-
77; of rule change, 123-25; of semantic Danish, 89
change, 148- 49; of signs, 12- 18; of sound dead languages, 23
change, 57; typological, 310-18 deadlocks on methods, 210-11, 220, 239, 242,
clines, 394 252-53, 340-41
clothing, 199, 379 decay dating, 395-96
Cockney, 77 deduction, 93, 103, 129, 151, 196-98, 200, 202,
code switching, 171 213, 225, 362
coinage, 139 deep structure, 123, 389-90
collateral relationship, 19 definition: of consonants, 7-9; of language,
color terms, 10- 11 12-13, 27, 31; of vowels, 7
combined evidence, 235, 252, 277 dephonologization, 70
comparative anatomy, 393 description, 129, 179-80, 198
INDEX 431
descriptive linguistics, 20 epigraphy, 43
descriptive order, 113 Eskimo, 10, 162, 311, 314-15, 392
descriptive prerequisite, 3 Eskimo-Alent, 387
determinatives, 33 Esperanto, 175, 176
Devanagari, 40 Estonian, 79, 80, 82, 91, 97, 98, 100, 102, 116,
devoicing, 117, 195, 196, 199-200 120, 139, 201, 211, 224, 280, 301
diachronic correspondence, 129, 153, 195, 198 etalon language, 316, 317
diachronic linguistics, 20 Etruscan, 23
diachrony, 21, 202-203 etymology, 326-33
diacritic features, 211, 226-27 euphemism, 139-40, 145-46
diagrammatic representation of lexical vs. European Pidgin Romance, 175
grammatical meanings, 17, 196 evidence: 112; of change and variation, 35-37;
diagrams: 16-17, 50, 170; between different for syntactic rules, 357-58
languages, 255; showing distinctive-feature evolution: 20, 153; and linguistics, 393
hierarchies, 195-96 exception features, 225
dialect, 289 exceptions, 64, 86, 127, 218, 225
dialect boundary, 290 excessive shortness, 184
dialect cohesion, 283-84, 292 excrescence, 67-69, 70, 72, 80
dialect correspondence, 47-50, 58 experimentation, 24, 333, 383
dialect geography: 182-83, 289-99, 304-309; explanation: 24, 84, 126, 131, 197, 383; of
in the service of comparative linguistics, 306 change, 179-81
dialect variation, 255, 282-84 extension, 91, 97, 98, 104, 107, 125, 127, 130,
dialectology, 47 148-49
diaphoneme, 283, 284, 292, 293-94 external causation, 180
diasystem, 292, 298-99, 317 extrapolation, 131, 201
diatopicality, 21
differentiation, 107, 143-45, 170, 382 factors of change, 179-81, 193, 380
diffusion, 155 facts, 18, 23-24, 112, 113
diglossia, 52 faculty for analogy, 105
directionality, 166, 362-63, 389-90 fading: 145-56, 148, 149-50, 153, 162, 199,
discovery procedures, 202-203, 362 356; degrees of, 151
disjunctive definition, 343, 369, 370 family trees, 44, 300-304, 307-309, 342-44,
dissimilation, 71, 74-75 381, 386
distant relationship, 320-21, 338, 397 fashion, 190
distinctive features: 6-9, 115, 158, 195-96, 245; feature phoneme, 211
shown diagrammatically, 195-96 feeding order, 110
distribution of phonemes, 167 field work, 21
division of linguistics, 20-22 figures of speech, 141, 148
Dravidian, 321 filling gaps, 120, 184-86, 395
drift, 194, 200, 252, 254, 352 Finnic, 300-301, 304
drum and whistle speech, 39 Finnish, 14, 72, 73, 74, 76, 79, 80, 83-84, 92,
drunkard's search, 324 103-104, 113, 114, 116, 119-20, 121-23,
Dutch, 57 124, 126, 129, 139, 140, 149-52 passim,
155-64 passim, 167-68, 169-72 passim, 182,
ease of articulation, 189, 198 185,186,203,209,211,219-22,224,225-27,
ecological influence, 381 228, 230, 231, 250-55, 274, 276-77, 280,
effect, 131 284, 290, 301, 320, 328, 344, 345-46, 348,
efficiency, 392 353-55,357,362,367,369
Egyptian (writing), 38, 39, 45 Finno-Ugric, 149, 163-64, 169, 300-301
elaboration, 143-44 flectionallanguages, 311-12
ellipsis, 138, 142, 143-44 flective languages, 310-11
empirical science, 24 Flemish, 183
English, 9-11, 14, 17, 26, 33-37, 40, 41, 42, 51, flip-flop, 112
57-59, 64-80 passim, 84-97 passim, 101, 102, focal area, 294
104, 108, 109, 110, 114-15, 119, 125, 126, folk etymology, 92, 142, 144, 160, 182, 201,
127, 136, 137, 140-49 passim, 151, 155, 158- 203, 369
62 passim, 164-65, 169, 172, 176, 177, 184, folklore investigation, 333-34
187, 188, 190-91, 196, 201, 203, 208, 209, form: 142; of rules, 114-17
210, 214-18, 219, 226-27, 229-42, 247, 249, Fox dialect, 48, 50, 53, 171
267, 297, 311-14 passim, 318, 320, 327, 329, Franco-Provencal, 298
333, 335, 352-53, 355, 356, 366-67, 368-73 Franklin, 171
passim, 387, 394, 396 French, 14, 42, 59, 69-74, passim, 80, 84, 85,
entelechy, 193 93, 101,109, 110, 126, 134-36,138,140, 142,
entropy, 391 143, 145, 148, 149, 154, 157, 160-61, 162, 165,
environment: 61, 231, 233, 235, 247, 248, 256; 167, 168, 171, 172, 173, 176, 182, 184, 187,
in evolution, 391; in syntax-semantics, 364- 190, 191, 230-31, 234, 294, 298, 317, 333,
65, 366 335, 356, 361, 370
432 INDEX