On Linguistic Method: Selected Papers [Reprint 2020 ed.] 9783112317679, 9783112306505

Citation preview

ON LINGUISTIC METHOD

JANUA LINGUARUM STUDIA MEMORIAE NICOLAI VAN WIJK D E D I C A T A edenda curai

C O R N E L I S H. VAN S C H O O N E V E L D STANFORD UNIVERSITY

SERIES MINOR NR. XXX

1964 M O U T O N & CO • THE H A G U E

ON LINGUISTIC METHOD Selected

Papers

by

PAUL L. GARVIN

1964

M O U T O N & CO • THE H A G U E

© Copyright 1964 by Mouton & Co., Publishers, The Hague. The Netherlands. No part of this book may be translated or reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publishers.

Printed in The Netherlands

TABLE OF

CONTENTS

INTRODUCTION

7

FUSED UNITS IN PHONEMICS ( JOINTLY WITH MADELEINE MATHIOT)

12

O N THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

22

EVALUATION PROCEDURE IN LINGUISTIC ANALYSIS

36

.

.

OPERATIONS IN SYNTACTIC ANALYSIS

44

SYNTACTIC UNITS AND OPERATIONS

56

A STUDY OF INDUCTIVE METHOD IN SYNTAX .

.

.

.

63

AUTOMATIC LINGUISTIC ANALYSIS. A HEURISTIC PROBLEM .

78

A DESCRIPTIVE TECHNIQUE FOR THE TREATMENT OF MEANING

98

O N STRUCTURALIST METHOD

144

STRUCTURALISM BEYOND LINGUISTICS

148

THE STANDARD LANGUAGE PROBLEM: CONCEPTS AND METHODS

153

INTRODUCTION1

The papers assembled in this volume are concerned with descriptive method in linguistics and in some aspects of language-and-culture research. The procedures proposed for linguistic analysis are based on the assumption that linguistic data are in essence amenable to empirical treatment, and that the techniques necessary for this can be inferred from a common-sense consideration of some of the consistently observable characteristics of natural languages. The aim of such an approach is to provide a frame for the acquisition of reliable knowledge about languages, rather than a systematization or "explanation" of things already known or assumed to be known. Several models have recently been proposed in linguistic which involve an "as if" mode of thinking. An "as if" model in effect suggests: "Since we cannot deal with our object of cognition effectively or directly, let's look at it as if it were something else with which we can deal, and which is sufficiently similar to it so that we can generalize back." Thus, let's look at the grammar of a language as if it were a probability matrix or a sentence-generating machine. A variant of the "as if" model is the reductionist model, in which certain properties of the object of cognition are eliminated because they are considered intractable. It is thought that this reduces the complexity of the object. In linguistics, a common application of this reductionism is the attempt to eliminate considerations of 1

The views presented in this Introduction were previously stated in my "From Model to Procedure," in: Proceedings of the National Symposium on Machine Translation, Los Angeles, 1960, H. P. Edmundson, ed. (Prentice-Hall Inc., Englewood Cliffs, N.J., 1961), pp. 367-70. For a more detailed discussion, see my "The Definitional Model of Language," in: Natural Language and the Computer, Paul L. Garvin, ed. (The McGraw-Hill Book Co., Inc., New York, N.Y., 1963), pp. 3-22.

s

INTRODUCTION

meaning from the study of language which is then defined as being form only, or as being describable in terms of form only. What is studied, then, is not the object itself but the particular model. Such an approach is necessary if the object of cognition indeed does not lend itself to direct investigation. I contend that this is not so in the case of language. It is perfectly possible to look upon language the way we have observed it to be in our past experience and to systematize this observation by abstracting from it a set of assumptions about the characteristic properties of our object of cognition, such that it is possible to study the object in terms of these properties. If we formulate these properties, we obtain, not a model in the previous sense, but a definition in the classical sense of the word. Such a definition will, however, serve as a perfectly good frame for the acquisition of knowledge, and why not, therefore, consider it a model in its own right and call it a definitional model. To be useful, the properties set forth in a definitional model must be general enough not to prejudge the results, and specific enough not to be trivial. They must allow the differentiation of one's object of cognition from other similar or related objects, and they must provide suitable points of departure for a procedure. They must allow the deduction of the equivalence principle and relevance criterion proper to one's own field of study. I stated the essentials of such a definition of language as far back as 1948 in a paper before the 29th International Congress of Americanists;2 the analytic work which has led me to this definition and has followed from it has not invalidated it. As all classical definitions, my definition contains a genus and a differentia. It places language into the larger class of phenomena to which it belongs by stating that it is a system of signs. It differentiates it from other systems of signs by a particular set of structural properties. These properties can be formulated in terms of three sets of 2

"Structure and Variation in Language and Culture," in: Sol Tax, ed., Indian Tribes of Aboriginal America, Selected Papers of the XXIX Intern. Congress of Americanists 3.216-221 (Chicago, 1952).

INTRODUCTION

9

levels: two levels of structuring, the phonemic and morpheme respectively; two levels of organization, namely selection and arrangement; and several levels of integration, along which the scale of units of increasing complexity is arranged. Note that my definition of language as a system of signs does not include the term "vocal". The reason is that the vocal character of spoken language is an irrelevant substantive property, since it is a mode of manifestation rather than an essential attribute of the system. To limit language to this one mode of manifestation would violate the requirement of generality for the definition. Not also that in stating levels of integration I have not specified a particular number of levels, but merely that there be several levels, in order not to impose an unnecessary specificity of structure on any particular language. The three sets of levels are intended to differentiate the language which is the linguist's object of cognition from other systems, often called languages, which are not. Thus, when logicians talk about a simplified language as defined, for instance, by the vocabulary "a", "b", "c", and the syntax " + " , " - " , " = " , this is not a language in our sense, since it lacks, of the three sets of levels, both the required two levels of structuring, and the levels of integration. Similarly, the language of the bees as described by von Frisch is not a linguist's language, since it, too, is limited to levels or organization only. From the properties stated in the definition can be deduced a set of methodological principles on which procedures can be based. The generic part of the definition, namely that language is a system of signs, allows me to posit the association of form and meaning through the sign function. From this association follow the equivalence principle and relevance criterion proper to linguistics. The linguistic equivalence principle will allow us to differentiate between what is same and what is not same in linguistics: same is what is functionally equivalent, and not necessarily what is substantively identical. Allomorphs, for example, have different forms but the same meaning - they are substantively not identical, but

10

INTRODUCTION

functionally equivalent. Conversely, homonyms are substantively identical, but functionally not equivalent. Thus, only if form and meaning are considered together can sameness and difference be established. A consequence of the linguistic equivalence principle is the linguistic relevance criterion: that which affects functional equivalence is relevant, that which does not affect functional equivalence, although it may affect substantive identity, is not relevant. The phonemic and morphemic levels of structuring differ in the type of the association of form and meaning: phonemes are meaning differentiators, morphemes are meaning carriers. Consequently, in morphemics the association of form and meaning is in the nature of a co-variance, in phonemics it is not. This is operationally of the greatest significance: where there is co-variance, one of the covariants can be made into the independent variable, whereby the other co-variant becomes the dependent variable. This is what happens in the elicitation of paradigms which is one of the standard techniques for handling morphemic data. In phonemics, on the other hand, since there is no co-variance, paradigms can not be elicited but can only be compiled ex post facto. The two levels of organization, which are stated in terms of the relevance of arrangement in addition to selection, allow not only the manipulation of paradigms as obtained above to establish the membership of each class, but also the manipulation of sequential order which leads to the differentiation of classes from one another. The levels of integration imply the existence of not only minimum units, namely phonemes and morphemes, but also of fused units, that is, non-minimal units which exhibit characteristic overall properties. For each fused unit can be posited an internal structure and an external functioning as separate attributes. The separateness of these attributes allows the application of the two basic procedures of dropping and substitution to fused units in morphemics. In dropping, portions of sequences are omitted from the whole, leading to the establishment of relations of dependence, which are one mode of external functioning. Classes of fused units can be defined in terms of these relations. In substitution, units of different internal structure are substituted for each other in controlled frames, such

INTRODUCTION

11

as the ones resulting from the above relations, within which they have the same external functioning. Inventories of classes can be compiled in terms of this substitutability. In actual analytic practice, procedures stemming from different properties of the structure are used in order of their applicability to the problems encountered, rather than in the order in which the model was presented. They do, however, fall into two broad classes: those drawing upon the co-variance of form and meaning - formmeaning procedures, and those drawing upon the regularity of recurrence of linguistic units - distributional procedures. The papers in this volume are arranged in the order of the observed tractability of linguistic data: phonemics, morphology, syntax, meaning, language and culture. The work of assembling and revising these papers was done under the sponsorship of the AF Office of Scientific Research of the Office of Aerospace Research, under Contract No. AF 49 (638)-1128. Thompson Ramo Wooldridge Inc. Canoga Park, California November 1962.

PAUL L . GARVIN

FUSED UNITS IN PHONEMICS* (jointly with Madeleine Mathiot)

0. The thesis of this paper is that an integrational approach to phonemic problems will yield both increased analytic efficiency and greater interpretive power as over a primarily linear treatment. By a linear approach we mean an investigation of phonemic patterns based on the assumption that segmental phonemes constitute sequences, with certain special phonemes - called junctural - forming anchorages of the sequences, and with other special phonemes (such as stress and pitch) - called suprasegmental - accompanying the linear segmental phonemes in additional, coterminous sequences. Hockett's discussion of "microsegments" and "macrosegments" has been a significant move in another direction of analysis.1 Pike's chapter on "Higher-layered units of the manifestation mode of the uttereme" 2 is an attempt to develop the theory of the integrational approach. In spite of his cumbersome terminology (e.g. "hyperphonemes" for phonemic fused units), we find ourselves in basic agreement with his point of view. In the present paper, we want to take an operational rather than theoretical approach and present a phonemic analysis based on the integrative rather than the linear assumption. * Originally published in Word, 14.178-86 (1958). The title under which this paper was published in Word, "Fused Units in Prosodic Analysis," was suggested by the editor of the journal. We are here going back to the title we used when we originally wrote the paper. 1 Charles F. Hockett, "Peiping Phonology," Journal of the American Oriental Society 67.253-67 (1947); also id., A Manual of Phonology (Baltimore, 1955), p. 43: "Ultimate phonologic constituents do not occur in an utterance as the individual bricks occur in a row of bricks. Rather, they occur in clusterings, these occur in still larger clusterings, and so on, up to the level of the whole utterance." 2 Kenneth L. Pike, Language in Relation to a Unified Theory of Human Behavior, Part II, Prelim. Ed. (Glendale, Calif., 1955), pp. 41-70.

FUSED UNITS IN PHONEMICS

13

The integrational approach posits that segmental phonemes do not simply constitute sequences, but that they fuse into units of higher orders, just as morphemes fuse into words or other higherorder units in morphemics. These higher-order phonemic units are assumed to have, as overall characteristics, (a) the prosodic features which in the linear approach are said to constitute the suprasegmental phonemes, and (b) the boundary features which in the linear approach are said to constitute the juncture phonemes. Finally, both the prosodic and the boundary features are assumed to be systematically related to each other, in terms to be established by the analysis.3 On the basis of the above, emphasis in phonemic procedure will be shifted to the following two operations: (1) the establishment of fused-unit boundaries to obtain proper frames for distributional statements; (2) the computation, within the spans of fused units of different orders, of the maximum predictability of phonetic detail from a minimum of features assumed to be phonemic. This is particularly relevant in the determination of prosodic features. Thus, the units and features serving as the best minimum base for maximum predictability will be said to enter into the phonemic pattern. The description which we are presenting is that of Chamorro, a language of the Philippine branch of the Malayo-Polynesian family, spoken on the island of Guam. 4 The data, consisting of isolated examples and narrative monologue, were recorded on tape and analyzed on the basis of consistent auditory impressions of phonetic oppositions.5 For a discussion of phonemic predictability, see Noam Chomsky, Morris Halle, and Fred Lukoff, "On Accent and Juncture in English," For Roman Jakobson (The Hague, 1956), pp. 65-80. 4 An earlier, more detailed version of the phonemic analysis of Chamorro was submitted by Madeleine Mathiot in partial fulfilment of the requirements for the degree of Master of Science in Linguistics at Georgetown University (Washington, 1955). 5 We agree with statement of the Czech phonemicist FrantiSek DaneS about the use of instrumental phonetics in phonemics: "Today, many linguists and phoneticians, in spite of the latest improvements, look more soberly and

14

FUSED UNITS IN PHONEMICS

1.0 We shall first deal with the consistent phonetic oppositions leading to the establishment of fused-unit boundaries, viz. degrees of pause, and with the phonetic oppositions on the basis of which the phonemic prosodic features of the fused units were established, viz. different degrees of loudness and pitch. 1.1. Two degrees of pause were observed in the connected discourse: a more conspicuous longer pause (length varying from less than a second to more than a second), and a less conspicuous shorter pause (always shorter than the former). These two degrees of pause were defined to be the basic phonetic properties of TERMINAL JUNCTURE and MEDIAL JUNCTURE, respectively. Segments bounded by terminal junctures were defined as PHONEMIC PHRASES; those bound by medial juncture were defined as PHONEMIC CONTOURS.

1.2. Several degrees of loudness were observed, but upon closer examination it was found that in addition to differentially stressed nuclei within contours, contours as wholes differed in overall loudness, with this differential distributed equally over the nuclei, recritically upon the possibilities of instrumental investigation, especially in regard to connected discourse. "There are two flaws in this method; one is of a more general character, the second concerns mainly the investigation of intonation. It is true that instruments are more accurate and sensitive than the human ear..., but instrumental records and their interpretation in terms of physical acoustics do not give us a true picture of the way in which the speakers hear and understand (evaluate) their own language... "The insufficiency of the instrumental method shows up particularly in the investigation of pitch and stress and the phonetics of connected discourse in general. The instruments which we have are relatively imperfect, the investigation is cumbersome and complex and can therefore cover only a limited amount of data.... It is not completely true, either, that the instrumental method is free of accident - on the contrary, considering the relative scarcity of records ... the danger of accident is considerably greater than with the auditory method which has almost unlimited data at its disposal.... "The most complete instrumental record becomes useful to linguistics only after analysis and evaluation from the standpoint of the function and significance of the various components in terms of the system of the language. The instrumental record, however, does not by itself allow us to ascertain the significance and function of the various waves, formants, etc. - these must first be discovered, albeit only in outline, by an auditory analysis of spoken language..." (Intortace a veta ve spisovne destine, Prague, 1957, pp. 39-41).

FUSED UNITS IN PHONEMICS

15

gardless of the degree of stress of the latter. Thus, instead of positing a larger number of degrees of stress, we attributed this contourlength loudness differential to a separate phonemic feature of CONTOUR INTENSITY (classifying contours as strong versus weak), and assigned a phonemic feature of STRESS only to differences in loudness between nuclei within same contour. 6 The relation of contour intensity to stress is such that an unstressed nucleus of a weak contour is less loud than an unstressed nucleus of a strong contour, a (primarily or secondarily) stressed nucleus of a weak contour is less loud than a correspondingly stressed nucleus of a strong contour. 1.3. Four phonetic ranges of pitch were observed (from lowest 1 to highest 4). These were found to constitute regular intonation lines within each contour in terms of the vocalic nuclei contained, and the final pitch of the contour was found to be associated with contour intensity as well as with the terminal or medial juncture ending it. In terms of the association of various pitch ranges with the longer pause, three types of terminal juncture (PARAGRAPH7 FINAL, HESITATION, and TRANSITION juncture) could be established (see § 2) as against a single type of medial juncture. 1.4. Of these phonemic features, terminal junctures were considered boundary features of the phonemic phrase, medial juncture and contour intensity were considered features of the contour, and stress a feature of the vocalic nucleus. By resolving the different phonetic degrees of loudness into the separate phonemic oppositions of stress (presence vs. absence of stress) and contour intensity (strong vs. weak), we were able to establish contour intensity, together with a differentiation of the junctures, as the conditioning factors for the predictable occurrence of pitch. 6

This interpretation was influenced by Uriel Weinreich's contrastive stress, which he defines as "facultative relative loudness, actualized only if the construction as a whole is emphasized in the sentence" ("Stress and Word Structure in Yiddish," The Field of Yiddish, New York, 1954, p. 3). ' These terms were selected on the basis of an impression of the semantic function of the junctures in the utterance.

16

FUSED UNITS IN PHONEMICS

2.0. We shall now discuss in some detail the conditions for pitch predictability. As was stated in 1.3 above, the final pitch of each contour is predictable in terms of the intensity of the contour and in terms of the following juncture, while the remaining pitches are predictable in terms of the number of syllables (which is the same as the number of nuclei, the syllable boundaries being predictable in terms of the configuration of vowels and consonants). We shall therefore deal with final and non-final pitches separately and in that order. 2.1. The primary determinant of final pitch is the type of juncture, since it determines the relative pitch range, whereas contour intensity determines the variation within it. 2.1.1. Paragraph-final juncture (symbol: # ) entails pitch 1 on the final syllable of weak contours, pitch 2 on the final syllable of strong contours (symbols: 11 marks following strong contour, lp indicates long pause): [ [M

1

lp] = / */p]=/"

#1

#/

Examples: [pinikaktu1 lp] = /pinikaktu # / "my steps", versus [" mani?lu-hu2 lp] = /" mani?lu-hu # / "my siblings". 2.1.2. Hesitation juncture (symbol: .|.) entails pitch 2 on the final syllable of weak contours; it has not been found after strong contours (note that pitch 2 on the final syllable of strong contours is a phonetic characteristic of paragraph-final juncture): [

2

["

lp] = /

-I-/

•//>] = / "

#/

Examples: [para?i a lp] = /para?i .|./ "for the...", versus [" golay2 lp] = /" golai#/ "vegetable". 2.1.3. Transition juncture (symbol: |) entails pitch 3 on the final syllable of weak contours, pitch 4 on the final syllable of strong contours: [

["

3

lp] = /

«//>] = / "

1/

|/

FUSED UNITS IN PHONEMICS

17

3

Examples: [para?ilaencu Ip] = /paraPilsencu | / "to the ranch", versus [" giniPilsencu4 Ip] = /" giniPilaencu | / "towards the ranch". 2.1.4. Medial juncture (symbol: + ) entails pitch 2 on the only syllable of monosyllabic weak contours, pitch 3 on the only or final syllable of all other contours (and for all contours a slight lengthening of the final phoneme; symbol sp indicates short pause): P sp] = / - + / [

[" Examples:

[sec? 1

3

sp] = / 3

5?] = /"

+/

+/

sp] = /sae?+/ "because...", versus [" sae?3 sp]

= /" sae?+/; [zanmatuhu? 3 sp] = /zanmatuhu?+/ "when I get to...", versus [" za?ii?ati 3 sp] = /" za?ii?ati+/ "and he/she pours". Note again that while [- 2 sp] equals medial juncture, [- 2 Ip] equals hesitation juncture (see 2.1.2.). 2.2 The pitch pattern of the non-final syllables of a contour is as follows: for dissyllabic contours - pitch 2 on the penultimate; for trisyllabic contours - pitch 1 on the penult, pitch 2 on the antepenult; for contours of four or more syllables - pitch 1 on the first syllable, pitch 2 on the second syllable, and pitch 1 on every subsequent syllable up to and including the penult. Examples: (all showing transition juncture): /pa n|/ = [pan 3 ] "bread", /ta ta|/ = [ta-2ta3] "father", /b®-bali|/ = [bce-Wli 3 ] "eyelashes", /camo-ruzu?|/ = [ca1mo-2ru1zu?3] "I am a Chamorro", /palau?anma mi|/ = [pa2law1?an1ma-3mi1] "our (pi.) women". 3. The detailed relations between contour intensity and stress on the basis of different degrees of loudness were established by comparing syllables of different length and loudness, but each carrying primary stress within the contour, with a selected syllable type serving as a norm. The loudness of a primarily stressed long syllable of a weak contour - a loudness not affected by the following juncture - was

18

FUSED UNITS IN PHONEMICS

chosen for this norm. In terms of it, the various phonetic loudnesses were interpreted phonemically as follows: (a) before terminal junctures, norm loudness - weak contour (short or long stressed vowel) louder than norm - strong contour (short or long stressed vowel) (b) before medial juncture, less loud than norm - weak contour (short stressed vowel) norm loudness - weak contour (long stressed vowel) - strong contour (short stressed vowel) louder than norm - strong contour (long stressed vowel). Note that before transition juncture, (a) weak contours with short stressed vowels are consistently somewhat louder than they are before medial juncture; (b) all strong contours are consistently somewhat louder than they are before medial juncture. Examples: (phonetic symbols: - norm loudness, s below-norm loudness, " above-norm loudness): [Patti Ip] = /Patti # / "tricks", [Pitimona Ip] = /" Pitimona # / "his/her knees", [PafacicuPusp] = /?afacicu?u+/ "my work", [zama-naw sp] = /zama-nau+/ "and I went", [Puntugu sp] = /"?untugu+/ "you (sg.) know", [Puhucusp] = /' 'Puhucu- + / "I close(d)". 4.0. Since of the two orders of phonemic fused units the contour is that which, though bounded by junctures, does not itself contain a juncture, it was selected as the frame for describing vowel and consonant distribution. Within the contour, three phonemic degrees of stress could be discerned (primary, secondary, unstressed), as well as phonemic vowel length under primary stress in open syllables only. The distributional analysis also revealed two additional predictabilities of phonetic features: vowel syllabicity, and points of syllable division. 4.1. Vowel inventories under different conditions of stress are as follows:

FUSED UNITS IN PHONEMICS

Under primary

stress

i, e,

ie-

3B, SB' Under secondary

In unstressed

19

stress

i e

u, o,

uo-

3., 2L' u o

position

The primarily stressed position thus emerges as that of maximum differentiation, the unstressed position as that of minimum di differentiation. 4.2. The consonant inventory is as follows: stops p, t, k, ? b, d, g affricates c, z spirants f, s, h nasals m, n, n, r) liquids 1, r Clusters of two and three consonants were found, the latter only with liquid as the final member. 4.3. Unstressed high vowels are non-syllabic under the following conditions: (a) both before and after non-identical (primarily or secondarily) stressed vowel, and only before identical secondary stressed vowel (after identical secondarily stressed vowel, the unstressed vowel remains syllabic); (b) adjacent to unstressed /a/; (c) of two unstressed high vowels adjacent to each other, the first is non-syllabic; (d) between any two other vowels, an unstressed high vowel is non-syllabic; (e) of two unstressed high vowels enclosing stressed or unstressed

20

FUSED UNITS IN PHONEMICS

/a/, the second is consistently non-syllabic and the first is optionally non-syllabic. Conditions (a)—(c) give rise to the following phonetic diphthongs : Table for conditions (a)

I-y]

[W-] [-WÎ-] = l-ui-l [-wi] = l-uï-l

/-iè-/

l-éy-] = l-éi-l [-èy-] = /-M-/

[-wé-] = /-Ué-/ [-wè-] = l-uè-l

a [-yâ-] = /-id-/ [-yà-] = /-ià-/

t-âey-1 = i-êi-i [-ây-1 = l-éi-l [-ày-] = /-ài-]

[-W®-] = l-ué-l [-wâ-] = /-uâ-/

i

[y-1

u t-yû-] = e t-yé-] = [-yè-] = o [-yô-] =

¡-iû-l l-ié-l

l-iô-l se l-yx-] = i-ié-i

[-W]

[-îéw-] = /-«u-/ [-âw-] = /-âu-/ [-àw-] = l-ku-l

Table for conditions (b)

[y-] [-ya-] = /-ia-/

t-y]

[w-]

l-ay-] = /-ai-/

[-wa-] = /-ua-/

[-w] [-aw-] = /-au-/

Table for conditions (c)

[y-1 [-yi-] = l-ii-l [-yu-] = l-iu-l

[w-] [-wi-] = l-ui-l [-WU-] = /-uu-/

4.4. Points of syllable division were investigated by observing the informant's consistent syllabicizing. These points regularly fall: (a) before an intervocalic single consonant or (phonetically) nonsyllabic vowel; (b) between the two consonants of a two-consonant cluster, or between the first two consonants of a three-consonant cluster. 5. We now want to relate our analytic results to our initial assumption that fused units are structural units and not merely analytic constructs, by showing that they exhibit certain properties assignable only to them as wholes and not derivable from the sum of their components. These we call FUSIONAL PROPERTIES. Two classes of fusional properties can be assigned to Chamorro fused units: (a) Prosodic characteristics shared by all the components of a given fused unit - SHARED PROPERTIES;

FUSED UNITS IN PHONEMICS

21

(b) Prosodic characteristics not shared by all the components of a fused unit, but differentially statable only in terms of the fused unit as a frame - DIFFERENTIAL PROPERTIES. These two types of fusional properties are related to each other in that a shared property of a lower-order fused unit constitutes a differential property of the unit of the next higher order. In the above terms, Chamorro contours can be assigned four fusional properties: one phonemic shared property (contour intensity), two phonemic differential properties (stress and length), and one non-phonemic differential property (pitch). Chamorro phonemic phrases can be assigned only one fusional property, the phonemic differential property of contour intensity. Contours thus have many more fusional properties than phonemic phrases. Assuming that fusional properties account for the integratedness of fused units, different degrees of integratedness can be posited for units of different orders: that of the contour is much higher than that of the phrase. The fused unit with the highest degree of integratedness can be called the TYPICAL FUSED UNIT for the given language; for Chamorro, it is the contour. Languages then become comparable in terms of their typical fused units and in terms of the number and kind of the fusional properties of the latter. Within a given language, different speech styles become comparable in terms of the statistics of the typical fused unit - in Chamorro, contour length and number of contours per phrase. The style studied here (narrative monologue) is characterized by short contours (averaging 3-4 syllables) and most commonly two contours per phrase. The typical fused unit may finally serve as the analytic correlate of such impressionistic observations as the overall acoustic impression of "what a language sounds like" or the melodic and rhythmical impressions of a particular "foreign accent". Thus, the "choppy" impression of the imperfect English of Chamorro speakers can be accounted for by the segmentation into short contours and the prominence of contour-final pitches carried over from Chamorro.

ON T H E RELATIVE TRACTABI LIT Y OF MORPHOLOGICAL DATA* 0. It has been customary among linguists for some time now to tell students that "there is no such thing as a difficult language," and that the difficulty of learning of language is a subjective matter and a question of the learner's native language. At the same time, in discussing with each other the problems of their field work, linguists would readily evaluate the relative tractability of their material; on this level of the argument, linguists would agree that, for instance, some North American Indian languages present considerable difficulty to morphological analysis, while on the other hand the languages of, say, the Malayo-Polynesian family are relatively easy to analyze morphologically. These impressionistic judgments show a consistency which strongly suggests that the matter lends itself to systematic treatment, leading perhaps to a set of criteria validating such judgments with regard to morphological tractability. The procedures of morphological analysis can, roughly speaking, be summed up in the following four major steps: (1) segmenting into minimum forms (morphs); (2) assigning morphs to morphemes as their allomorphs (morpheme alternants or morphophonemic variants); (3) ascertaining morpheme distributions and establishing morpheme classes; (4) exhaustive listing of morphemes and morpheme classes. The ease or difficulty with which each of these steps can be completed depends on the readiness with which the required operations can be applied to the material of a given language; languages vary in this respect. * Originally published in Word 13.12-23 (1957). A first version of this paper was read at the meeting of the Linguistic Society of America in Philadelphia, December 28, 1956.

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

23

1. SEGMENTING INTO MORPHS. For operational purposes, a minimum meaningful form can be redefined as the smallest separable stretch of form with a consistent identifiable piece of meaning. The first step in morphological analysis thus involves two sets of operations: separating forms, and identifying the presence of a consistent piece of meaning for each. 1.1. SEPARATING FORMS. This is essentially accomplished by two closely related procedures: substitution and dropping. By dropping I mean the elicitation or comparison of examples such that a given partial present in one example will, all other things being alike, be absent in another example. (a) The substitution procedure is applied most readily when paradigmatic sets can be easily elicited, as in the following examples from Kutenai :x

Set I

hucPinaxe- "I'm going there" hincPinaxe- "you (sg.) are going there" cPinaxe- "he, etc., is going there" hucPinaxalaPne- "we are going there" hincPinakilne- "you (pi.) are going there" cPinaxe- "they are going there"

Set I unequivocally yields as separable forms hu-, hin-, and 0-(zero), for first, second, and third person respectively. It also suggests forms -ala?- and -kil- as plural markers for first and second person respectively. Set II

hucxalcPinaxe- "I shall go there" huqa-kilcPinaxe- "I used to go there" husilcPinaxe- "I am on my way there" hucxalsilcPinaxe- "I'll be on my way there"

Set II yields the separable forms -cxal- and -qa-kil- for future and past respectively, and form -sil- with something like continuative meaning. (b) The dropping procedure is applied most readily when the 1 Spoken by the Kutenai Indians of northern Idaho, northern Montana, and southeastern British Columbia.

24

THE RELATIVE TRACT ABILITY OF MORPHOLOGICAL DATA

elicitation of paradigmatic sets reveals the presence of clearly bound and clearly free forms, as in the following examples from Ponapean :2 Set I ikílarj "I see" kakilai) "you (sg., non-honorific) see" ekilai) "he, etc., sees" kilaq "see" Set I yields the bound forms i-, ka, and e-, for first, second and third person, respectively, and the free form kilaq. Set II áramaso "that man" áramasaka "people" dramas "man" Set II yields the bound forms -o, -aka for demonstrative and plural respectively, and the free form áramas. (c) The application of both procedures to the sets presented above is obviously "easy". The reason for this "ease" is equally obvious: form boundaries are clearcut, and the separation points are unmistakable. The difficulties of separating forms then must arise when boundaries are not clearcut. Let me exemplify this with Ponapean examples: ó lenpó npey "man of Ponape" meó l "it's a man" mé npó npey "Ponapeans" olsnwáy "foreigner" mé-nwáy "foreigners" In the above substitution set, the forms o-l- or ol- "man", -en- "of", -pó-npey "Ponape", -wáy "abroad" (assuming that olenwáy by analogy with ó lEnpó npey stands for something like "man of abroad"), can be clearly separated out, and the form me n- initially appears indivisible, tempting the analyst to include it in his list as a single morph. It is only after considerable additional analysis that it becomes clear that the form me-n- is the result of the contraction me+en; the form me-n- as it stands in the elicited set does not immediately suggest a division in two. Even after the division 2

Spoken by the natives of Ponape, Eastern Caroline Islands, Micronesia.

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

25

has been accomplished, the boundary between the two components cannot be clearly drawn, since only the initial and final consonants of the form mè-n- can unequivocally be assigned to the underlying components me and en, the medial vowel e- is the indivisible result of the contraction. Such phonemic units shared by two morphs, and without allowing the clearcut placement of a boundary within, I have called ambimorphemic, and the partial phonemic fusion of two morphs I call morpheme overlap. (d) An extreme instance of morpheme overlap exists in the case of portemanteau forms, where a single separable (and indivisible) form fills a place in a paradigm otherwise occupied by two separable forms which may or may not be overlapping, as in the French sequences of preposition, article, noun: ovwazç dyvwazç parlavwazç aveekbvwazç (e) The separation difficulty can thus be related to the existence of morpheme overlap, and allows the formulation of the following statement: the difficulty of separating morphs increases with the amount of morpheme overlap present in the language. 1.2. IDENTIFYING MEANING. Operationally, linguistic meaning can be handled by translation or, if the analyst is a native or nearnative speaker of the language, by paraphrase. The presence of a consistent piece of meaning in a separated stretch of form can thus be ascertained by checking the translations (or paraphrases) relevant to it. (a) The simplest case of identifying the presence of meaning exists when it is possible to elicit clearcut, consistent individual translations of each separated form, as is the case for Set I under (b) in 1.1 above. Informants will readily translate each separated form, and do so consistently: i- as "I", ka- as "you (sg., non-honorific)", e- as "he, etc.", -kilar) as "see". Note that such forms may include bound forms as well as free forms. (b) Sometimes even good bilingual informants will readily iden-

26

THE RELATIVE TRACT ABILITY OF MORPHOLOGICAL DATA

tify a form as separate and having meaning, but will be hesitant and inept in giving a translation. This happens particularly frequently when the forms in question do not correspond to readily available forms in the other language. Thus, Kutenai informants will follow the linguist in separating out forms such as Pat "indeed", pal "really", miqsan "however", but will have some difficulty in giving a translation. They will, however, agree that these forms do have separate meanings. (c) At times, informants will identify forms as separate and readily volunteer definite individual translations, but the translations will, upon closer investigation, lack consistency and appear contradictory to the linguist. Thus, Visayan3 informants will usually translate as follows: gibuhatku Paqbalay "I'm building the house" nakiPtaPku aqbalay "I'm seeing the house" They will furthermore volunteer to translate the morph -ku as "I", and the morph Pag- as "the". Later extensive analysis of the verb pattern reveals, however, that these translations are deceptive: the examples involve passive verb forms, and should "properly" be translated literally as "the house was built by me", "the house could be seen by me", with the morph -ku translated as "by me" and the morph Pag- constituting a subject marker. (d) Very frequent are the cases where translations can be obtained only for the several complete forms of a paradigm, and the translations of the separable morphs have to be identified by matching minimal differences in form against minimal differences in translation. Thus, in the Kutenai paradigm cited as Set I under (a) in 1.1, individual translations cannot be obtained for the easily separable forms hu- for first person, hin- for second person, etc., but the identification of their meanings follows readily from the translation of the examples as wholes. (e) Finally, I would like to cite cases in which the separation of forms appears quite obvious, but where the assignment of meanings presents considerable difficulties. Here belong the famous English sets such as: 3

A Philippine language.

THE RELATIVE TRACT ABILITY OF MORPHOLOGICAL DATA

receive conceive deceive perceive

revert convert

27

retain contain detain pertain

pervert subvert A casual inspection of the above collection will indeed reveal that each of these forms can by the substitution technique easily be divided into two components, yielding the suspected morphs re-, con-, de-, per-, sub-, -ceive, -vert, -tain. The difficulty arises when the presence of consistent meaning is to be identified: neither translation into another language, nor paraphrasing within English (nor, for that matter, recourse to the native speaker's intuition) yields a consistent piece of meaning for any of these separated forms. It is thus not by accident that these sets have been the subject of much controversy among linguists.4 (f) It is clear from the above that the identification of meaning involves a greater variety of problems (and hence of techniques) than the separation of forms, since meaning - as is well known - is less immediately apparent than form, and hence its presence must be ascertained by techniques that are increasingly inferential. This allows the formulation of the following statement: the difficulty of identifying the presence of meaning in separated forms increases with the number of inferential steps required. ASSIGNING MORPHS TO MORPHEMES. Two or more morphs can be assigned as allomorphs to the same morpheme if (1) their meanings fall within the same acceptably narrow range, (2) they are in free variation or complementary distribution. The two criteria are clearly interrelated: on the one hand, two morphs in the same distributional position will be considered in free variation only if their meanings can in no way be said to be in opposition; on the other hand, two morphs will be examined for complementary distribution only if they can be shown to have (or at least can be suspected of having) a significant feature of meaning in common.

2.

4

For a discussion in some detail, see Dwight L. Bolinger, "On Defining the Morpheme," Word 4.18-25 (1948).

28

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

The second step in morphological analysis thus involves two sets of operations which must be performed in close conjunction: collating morph meanings, and examining morph distributions. In testing for free variation, the sameness of distributional position is given and the non-opposition of meaning must be demonstrated. In testing for complementary distribution, the non-opposition of meaning is given (or assumed) and the exclusiveness of distributional conditions must be demonstrated. 2 . 1 . MORPHS IN FREE VARIATION. Phonemically unlike morphs can safely be considered in free variation if forms differing only by these morphs are consistently considered identical by native speakers. (a) In Kutenai, such forms as hu-cxal-?ik-ne- or hu-c-?ik-ne"1 shall eat", hin-cxal-?in-ne- or hin-c-?in-ne- "you (sg.) will be", are consistently said to "mean exactly the same"; they are often used interchangeably or given as repetitions of each other. The morphs -cxal- and -c- thus have to be considered in free variation; they are morphemic free variants in all distributions. (b) In Visayan, a number of verb stems occur in either of two phonemic shapes when followed by a suffix; the verb forms in question are completely interchangeable: gi-lisud-an or gi-lisd-an "have difficulty with" na-wala?-an or na-wal?-an "lose" gi-salipud-an or gi-salipd-an "be hidden" The morphs -lisud- and -lisd-, -wala?- and -wal?-, -salipud- and -salipd-, are morphemic free variants, but unlike the previous case, the free variation is restricted to one set of distributions only. (c) In Czech, the two locative singular morphs -e and -u appear to be in free variation after certain inanimate noun stems ending in a "hard" consonant, such as: bod-6 ~ bod-u "point (loc.)" hlas-e ~ hlas-u "voice (loc.)" jazyc-e ~ jazyk-u "tongue; language (loc.)" A closer examination of the examples reveals, however, that for instance, of the third pair jazyc-e is used more frequently in contexts where the stem is translated as "language", and jazyk-u more

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

29

frequently in some contexts, and exclusively in other contexts, where the stem is translated as "tongue" in the literal sense. The free variation is thus not only distributionally restricted, but the coincidence of meaning of the morphs involved is not clearcut. (d) The above cases can be summarized as follows: The difficulty of assigning morphs to morphemes increases with the distributional restriction of the suspected free variation, and decreases with the clarity of the meaning equivalence. 2.2. MORPHS IN COMPLEMENTARY DISTRIBUTION. Phonemically unlike morphs can safely be considered in complementary distribution if forms containing these morphs consistently share a feature of meaning assignable to the morphs in question, and are otherwise morphemically different. (a) In Ponapean, such forms as toto-k "work" and toto-q-ki "work with" share the feature of meaning "work", such forms as ssylok "travel" and ssyloq-ki "travel by" the feature of meaning "travel', assignable to the morphs toto-k and toto q-, seylok and seylo-qrespectively; they are morphemically different, since the second of each pair contains the morph -ki "by, with", the first of each pair does not. A great many such forms can be collected in Ponapean, showing that k-final morphs are replaced by q-final whenever the following morph is k-initial. These sets of morphs are thus in complementary distribution under phonemic conditions. I shall call such allomorphs morphophonemic variants. (b) In Kutenai, such forms as hu-q?umne-nala?-ne- "we are sleeping" and hu-c?inax-ala?-ne- "we are going there" share the feature of meaning "plural of first person" assignable to the morphs -nala?- and -ata?- (cf. (a) under 1.1.); they are morphemically different, since one contains the morph -q?umne- "sleep" and the other the morph -c?inax- "set out for". A great many such forms can be collected in Kutenai, showing that when preceded by a certain set of morphs, the first person pluralizer morph will be -nala?-, when preceded by another set of morphs, it will be -ala?-. The two morphs are thus in complementary distribution under morphemic conditions; I call such allomorphs morpheme alternants. (c) In Kutenai examples such as hu-n-?upx-ne-mal-ne- "we were

30

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

talking to each other", 0-cxal-cukat-e -mal-ne- "they'll get married", hin-anaq-niy-kit "if you sat down", hu-n-?aqaciy-iy-ala?-ne"we cut our hands", 0-qa-c?la?nxu-na-k-s-e- "a fallen tree was lying there", or ci?kat-a-m "look at yourself", the morphs -ne-, -e-, -niy-, -iy-, -na-, -a- will, upon further analysis, be found to be in complementary distribution, and a common grammatical meaning of "reflexivizer" can be assigned to them, even for cases where it is not obvious from the translations given above. It will then also be found that the choice between the morphs -ne-, -niy-, -na- on the one hand, and between the morphs -e-, -iy-, -a-, on the other, will depend on the same preceding sets of morphs as the choice between -nala?- and -ala?- in the examples cited above. Within the set of morphs -ne-, -niy-, -na-, and within the set -e-, -iy-, -a-, the choice is made in terms of one or several of the following morphs. In addition to this morpheme alternation, the morphs -ne-, -e- containing a long vowel are also subject to a morphophonemic variation affecting vowel length.8 (d) In English, the morphs -s, -z, -iz, -in, are generally considered to be in complementary distribution and to constitute some of the allomorphs of the plural suffix, as in cats, dogs, horses, and oxen. The choice between -s, -z, -iz on the one hand and -in on the other depends on the preceding morph, the choice within the set -s, -z, -iz depends on the preceding phoneme. We are thus dealing with two morpheme alternants, one of which in turn is made up of three morphophonemic variants. The situation appears to be clearcut so far. A difficulty arises, however, from the well-known fact that in such examples as brothers vs. brethren the two morphs -z and -in appear to be in oppositive distribution, since the morphemic environment is not obviously different (braSer- and breSr- may be considered allomorphs), and the difference in meaning between the two total forms is as easily assignable to the suffix morphs as to the stem morphs. As in the previously cited English examples, here again the set has given rise to a good deal of controversy. 5 For details of this allomorphic variation, see my "Kutenai III: Morpheme Distributions (Prefix, Theme, Suffix)," IJAL 14.185-6 (1948).

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

31

(e) The above cases can be summarized as follows: the difficulty of assigning morphs to morphemes increases with the complexity of distributional conditions and, as in the previous case, decreases with the clarity of the meaning equivalence. 3.

ASCERTAINING MORPHEME DISTRIBUTIONS AND ESTABLISHING

Once morphs have been assigned to morphemes, the distribution of morphemes as wholes, irrespective of allomorphic variation within each, is to be ascertained and they are to be grouped into classes on the basis of common distributional characteristics. In the analysis of many languages (as well as in traditional grammar), such broad classifications as stems, prefixes, suffixes, have been intuitively accepted and have allowed a fairly good pragmatic assignment to classes. Even so, in instance after instance difficulties have arisen in terms of requiring a decision between assigning a morpheme to a class of, say, proclitics or one of prefixes, one of enclitics or one of suffixes. The well-known suggestions about the possible prefix status of French preverbal pronouns, or about the possible enclitic status of the Saxon genitive in English, serve to illustrate this. Since affixes are commonly designated as "parts of the same word", and clitics as "separate words", these difficulties have served to focus attention upon the problem of word boundaries, and upon the usefulness of the word as a distributional framework for morphemes. Whereas in the case of the assignment of morphs to morphemes merely the sameness or not-sameness of the environment was a requisite criterion, in the case of the statement of morpheme distributions the environment has to be specified in detail in order to allow the precise statement of distributional differences between morphemes needed for morpheme classification. Such a precise statement of morpheme distribution requires as its proper distributional framework a larger morphemic unit, within the boundaries of which the occurrence of individual morphemes can be rigorously described. The difficulty with which morpheme distribution can be stated MORPHEME CLASSES.

32

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

will then increase in terms of the difficulty with which such a frame unit can be defined. If the definition of such a frame unit can be based on criteria of a different sort than the definition of any other larger morphemic unit, such a morphemic unit can properly be called a word. Conversely, if no such definition of a larger morphemic unit is possible, it can be said that the language in question has no words in our sense. Examples of languages in which the word can be defined in a simple manner, and by criteria different from those for the definition of the phrase or clause, are Kutenai, Ponapean, and Turkish. In these languages, the word can be defined as a sequence of morphemes and morpheme clusters in fixed order. Morpheme clusters in turn are for these languages defined as sequences of morphemes which, within a behaviorally defined unit of elicitation, the "informant word", are consistently replaceable by single morphemes. In such languages, morphemes can be grouped into classes with relative ease in terms of the position they occupy with regard to word boundaries, or with regard to some specially defined morpheme or morpheme cluster within the word (such as a stem or theme), leading to a classification of morphemes into position classes. I have tested this simple fixed-order definition of the word for French and English, with negative results, due to the absence of consistently ordered long chains of morphemes in these languages. I would assume that there are many additional languages in which the fixed-order criterion is inapplicable. In languages such as French, Czech, and Polish, word boundaries have traditionally been identified with a phonological feature fixed within the contour, namely stress. This definition has been tested by me and found unviable, since it yields a large residue of multiword contours containing proclitics and enclitics not bounded off by the required stress, and yet traditionally considered separate words. If, on the other hand, the phonological definition is carried out consistently and the traditional clitics are included into adjacent words as affixes, the statement of morpheme distributions whitin the word becomes unproductive, because it does not allow for the

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

33

formulation of morpheme classes specific enough to lead to largerunit classifications useful in syntax. My attempts to define the word rigorously for French, English and Vietnamese by criteria other than phonology or fixed order, as well as by a combination of the two, have failed; I have come to the conclusion that a proper definition of the word, that is, a consistent one without unaccountable residue, will emerge only from a complete distributional reanalysis of the morphemie structure, using a properly defined temporary framework such as the utterance. The beginnings of such an analysis were made for English, yielding an interim classification of morphemes in terms of those occurrent in utterance-initial, medial and final, defining an utterance as a unit bounded by pauses, but barring hesitation forms and citation forms. In summary, it can be said that the difficulty of ascertaining morpheme distributions and of establishing morpheme classes increases with the difficulty of defining a proper distributional framework such as the word. 4.

EXHAUSTING THE NECESSARY LISTING OF MORPHEMES AND

The criterion of exhaustiveness in the empirical principle6 requires a complete listing of morphemes before a morphological analysis can be considered done. For certain classes of morphemes, mainly those called inflectional or paradigmatic, exhaustive listing has never presented serious difficulties. For other classes of morphemes, primarily those called stems or bases, on the other hand, an exhaustive listing is ordinarily not even attempted, although partial lists may appear in lexical appendices. Let us for our purposes call the former classes of morphemes paradigmatic and the latter thematic, and examine this discrepancy in regard to listing. It will be found that one viable definition of paragidmatic morphemes is that they belong to classes with restricted membership, and of thematic morphemes that they belong to classes with unMORPHEME CLASSES.

4 See Louis Hjelmslev, Prolegomena to a Theory of Language, Francis Whitfield, transl. (Baltimore, 1953), p. 6.

34

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

restricted membership. With this definition, the criterion of consistency - which in the empirical principle overrides the criterion of exhaustiveness - allows only a listing of the restricted membership of paradigmatic classes, since an exhaustive listing of the unrestricted membership of thematic classes is a contradiction in terms. Thus, exhaustiveness will mean, for the paradigmatic classes complete listing of individual members of classes as well as of classes and subclasses, for the thematic classes complete listing of classes and subclasses, but mere exemplification - without complete listing - of individual members. It will then be found that the number of paradigmatic classes and subclasses, as well as the size of the membership of each, varies from language to language, as does the number of thematic classes and subclasses. Obviously, the more there is to list, the greater the difficulty of achieving exhaustiveness. Hence, the difficulty of the exhaustice listing of morphemes and morpheme classes increases with the number and size-of-membership of paradigmatic classes and subclasses, and with the number of thematic classes and subclasses. 5. The above criteria will allow at least a tentative evaluation of morphological raw material in terms of relative analytic tractability.7 My experience in this respect strongly points to the fact that some of the well-known Indo-European languages are clearly less tractable morphologically than some of the "exotic" languages I have dealt with. This to me also explains some of the divergences and lack of agreement in regard to the morphological analysis of languages such as English and French. The less tractable a language, the less will it be possible to apply 1 Joseph Greenberg, "A Quantitative Approach to the Morphological Typology, of Languages," in Method and Perspective in Anthropology, Robert F. Spencer, ed. (Minneapolis, 1954), pp. 192-220, has suggested some numerical criteria of use in typological classification. Some of these bear a direct relation to points covered in the present paper (thus, his index of synthesis to the separation difficulty treated in 1); note also his very pertinent discussion of the definition of the word (216-8).

THE RELATIVE TRACTABILITY OF MORPHOLOGICAL DATA

35

to it the relatively cut-and-dried sequences of steps applicable to a more tractable language, and the more will it become necessary to devise new sequences of operations framed in terms of these intractable data.

EVALUATION P R O C E D U R E I N L I N G U I S T I C ANALYSIS 1 Noam Chomsky has recently focused the attention of linguists on the question of evaluation procedure. In the chapter "On the goals of linguistic theory" in his Syntactic structures he says "An even weaker requirement would be that given a corpus and given two proposed grammars G x and G 2 the theory must tell us which is the better grammar of the language from which the corpus is drawn. In this case we might say that the theory provides an evaluation procedure for grammars." 2 And one page later he goes on: "The point of view adopted here is that it is unreasonable to demand of linguistic theory that it provide anything more than a practical evaluation procedure for grammars." I do not wish to become embroiled in the ultimate issue raised, namely, whether an evaluation procedure is indeed the only reasonable goal for linguistics. In my opinion it is not. I should instead like to discuss a more immediate question: how to provide a practical evaluation procedure, where by practical I mean something that can actually be done in practice. It appears to me first of all that an evaluation procedure for grammars - that is, entire grammars - is rather a tall order, if the procedure is to be interpreted operationally. It seems to imply that two grammars (that is, two complete descriptions from phoneme or morphoneme - to sentence) are to be compared to each other and to a corpus, in order to ascertain which is to be preferred. As to the criterion by which this is to be judged, let me quote Chomsky again: "Suppose that we use the word 'simplicity' to refer to the set of formal properties of grammars that we shall consider in 1

This paper was presented at the 1960 Annual Meeting of the Linguistic Society of America, Hartford, Conn., December 29,1960. Originally published in SIL, 15.62-9 (1960-1). 1 Mouton and Co. (The Hague, 1957), p. 59.

EVALUATION PROCEDURE IN LINGUISTIC ANALYSIS

37

choosing among them." 3 For the choice to be feasible in practice, there should first of all exist two such grammars, each meeting what Chomsky calls "the external criteria of adequacy for grammars". 4 Assuming such a condition, the two adequate grammars would then have to be compared in their entirety which, if taken seriously, might mean a comparison page by page, or statement by statement, or chapter by chapter. Each partial comparison may then result in a judgement of simplicity. If it is possible to weight each part judgement appropriately, one may assume that an overall judgement can be computed by some reasonable statistical operation. It is also thinkable that instead of this series of partial comparisons (which presupposes a matching of parts that are not necessarily susceptible to clear-cut matches), one might take each grammar separately and by some procedure to be defined when available take an independent measure of simplicity. The two measures could then be compared and a final evaluation made. I wonder whether, at the present state of the art in linguistics, any of this is very practical. If an evaluation procedure for entire grammars is deemed impractical, under what conditions does an evaluation procedure in linguistics become practical, and is it then relevant to the objectives of linguistics? Let me give a categorical and somewhat unfashionable answer to both questions. Assuming that the purpose of descriptive linguistics is the oldfashioned one of describing languages, an evaluation must be made at each step in the analytic process of the respective merits of alternative possible solutions. Given some criterion by which these solutions are arrived at and are judged equally admissible, the judgement as to the investigator's ultimate preference can be formalized on the basis of additional criteria. Such a formalized judgement can then be called an evaluation procedure; it will be applied, not to the very complex task of comparing entire grammars, but to the much more manageable objective of comparing alternative solutions to limited problems of analysis. It is not unreasonable 3 4

Op. cit., 53. Ibid.

38

EVALUATION PROCEDURE IN LINGUISTIC ANALYSIS

to expect that, as the number of variables to be judged is reduced, some determinate evaluation criterion becomes applicable. The procedure thus has some hope of being practical. The relevance of such a clearly circumscribed evaluation procedure will be insured by the anticipation that the goal of the analysis is attainable by a succession of individual analytic steps. Any operation, therefore, which permits the successful completion of a particular step by a preference judgement, will to that extent contribute to the continuity of the process of analysis: the completion of one step will permit the initiation of the next step, and the analysis can proceed uninterrupted. I want to illustrate this point by presenting a limited and rather elementary problem in morphemic analysis which might be dealt with after enough phonemic work has been done to produce a usable transcription. My argument is based on data from Palauan, a Micronesian language, collected by Robert W. Hsu.5 As a first preliminary step in morphemic analysis, systematically related sets of forms - paradigms - can be elicited by putting judiciously formulated sets of questions to an informant. If paradigms resulting from parallel sets of questions are not uniform, then as a second step these paradigms can be placed together for examination in some convenient form, such as a table. Hsu has carried out these preliminary steps and the results appear in Table 1. Now the data are ready for the actual analysis. The first analytic objective is the determination of morph boundaries, where a morph is defined as the smallest separable stretch of form with consistent identifiable presence of meaning. In applying this definition to my sample as a segmentation criterion,6 a comparison of the forms and their translations allows the reasonable and unique conclusion that each example in the table consists of two and only two morphs. It is reasonable to suspect that in each example, the left-hand morph may turn out to be a stem, the righthand morph may turn out to be a suffix. A further inspection of 5

University of California (Berkeley). For procedural details, let me refer to the immediately preceding paper in this volume, particularly pp. 23-25. 6

EVALUATION PROCEDURE IN LINGUISTIC ANALYSIS

39

TABLE 1

Sample paradigm

my your (sg.) his our (incl.) our (excl.) your P: 113, 121, X, 1132, 1111.2 I S: Y, 1132) "when-it-was-eleven there-was machine (i.e. at [district 11] there was a bulldozer)" C i n 153 t o x a #-n-?isilsahan-s-e- k-#-?itkin suyape- (P: v 2 ; 113, 121, X, 1132, 1111.2 < P: 101, 113, X / S: Y) "just-about was-very-bad what-did [the] white-man" C V 75 napit #-qaqa c ka?ika-?is #-lin-?in-s-e- (P: v : ; 113, X > S: v 3 ; Y, 1157 / P: 113, 123, X, 1132, 1111.2) "if it-were-so and his-spirit seems-to-be (i.e. if that's the case, it must have been his spirit)"

Most of these cases show one of the predicates in the absolute, the other in the obviative. As in §4.1.6, the function of obviation is to differentiate primary and secondary subject, and as in the immediately preceding sections the dependence relation between clauses as such has no bearing on obviation. Example H III 158 differs from the other examples in that both predicates stand in the obviative. The relation between the referents of the subjects is clear: we are here dealing with a primary and a secondary subject. The obviative form of the primary subject can be interpreted as standing in isolation, and its treatment deferred to §5. COMPLEX SENTENCES. The analysis so far has shown clearly that the relations of coordination and dependence between clauses as such have no bearing on obviation, and that the latter is correlated with the relation between the referents of the subject forms or implicit subjects of the clauses: in the case of a nonthird-person and a third-person subject, with the animateness of the referent of the third-person subject; in the case of two third-person subjects, with the relation of primary versus secondary subject. Complex sentences (that is, sentences consisting of more than two clauses in various coordinate and dependent relations) can therefore likewise be analyzed in terms of the relations between the referents of the subjects as stated above. I attempt here the analysis of a number of complex sentences in these terms, to show the identifying function of obviation in a normal Kutenai text. 4.3.

130

A TECHNIQUE FOR THE TREATMENT OF MEANING

(1) H V 38 hu-cxal-qakiy-ala?-is-ne- k-#-?o-la-s na-s ?al-a qalt-?is k - # qa?itxaminqa-mi-k ne- titqa-t? k-#-?upil-il (P: 111, 131, X, 1151.2, 1132, 1111.1 < P: 101, 113, X, 1132 / S: W, 1132; 41, Y p , 1157; 101, 113, X, 1191.3, 1182 < S: W; Y / P: 101, X, 1141) "we-shall-say they-having-done-it these his-children Never-Sit-Down the man being killed (i.e. we shall say that NeverSit-Down's people did it, killing this man)"

The obviative of the first person (hu-cxal-qakiy-ala?-is-ne) can here be interpreted in terms of the animate referent of the thirdperson subject in the adjacent clause (cf. §4.2.3). The obviative of the third person (k-#-?o-la-s) marks the secondary of the two third person subjects (cf. §4.2.6). (2) A II 167 #-n-?uk?nilanilpatnixuhi?k-ne- ne- k-#-t?alo-k c ki-#-?in ne?at k-#-sl-cxa-na?t-ka ni-?s k-#-?upska-s ?ìninik?-nam-is (P: 113, 121, X, 1111.1 < S: W I P: 101,113, X < P: v 3 ; 101,113, X < S: W / P: v x ; 101, 113, 141, X, 1171.1, 1162 < O: W, 1132/P: 101, 113, X, 1132/ S: Y, 1152.1, 1132) "suddenly-there-could-be-heard the making-noise and being the indeed tellingthem the coming enemies (i.e. suddenly one could hear what makes the noise telling them that the enemy is approaching [ = air raid alert])"

The obviative here marks the secondary subject (cf. §4.2.6). (3) A I I 273 na-?s #-n-?in-ne- k-#-c-la-c?inam #-n-?àqsal?it?wunwu k - # qawxakin-l-is ?à-kikl-iyi-?is (O: W, 1132 / P: 113, 121, X, 1111.1 < P: 101, 113, 151, X > P: 113, 121, X > P: 101, 113, X, 1141, 1132 / S: Y r , 1191.4, 1157) "here are about-to-leave when-they-are-several-hundred being-put-there their-names (i.e. there are several hundred here about to leave whose names are entered [on the draft rolls])"

Again the obviative marks the secondary subject. (4) C IV 371 c tax nà hin-aqalpalnPc #-cxal-qakl-is-ne- k-#-saq?an-s k-#-?ituq?lilqa-is ki-#-?in-s ?à-knumuctitt-is k-#-qa?in-s xma-k-in-qakikmil" (O: v 3 ; v 2 ; W / P : 112, X; v8 < P : 113,131, X, 1155.4,1111.1 < P : 101,113, X, 1132 < P: 101, 113, X, 1141, 1132 < | P: 101, 113, X, 1132/S: Y, 1132 < | P: 101, 113, X, 1132 < P: 11, 101, 112, X, 1131) "and then here if-you-saysomething (and) he-will-tell-you that-there-hangs-here what-is-written, that-itis [the] law, that-there-is-not [a thing] you-could-tell-him"

The obviatives of the third person here indicate secondary subjects as before; the obviative of the second person (xma-k-in-qakikmil) represents a case not previously encountered : it appears to be in agreement with the preceding obviative of the third person (k#-qa?in-s) on which it is dependent.

A TECHNIQUE FOR THE TREATMENT OF MEANING

131

(5) C IV 170 taxa-s #-sil-qal?upxa ne- Paqlsmaknik? k-#-cxal-?isqat?li?it-s c k-#-cxal-wilko--s k-#-sa-nlukqa-mi-k ne- suyape- c #-?anaxa?m-ne- c #-Ja-c?inax-e- (O: W, 1132 / P: 113,141, X / S: W; Y < P: 101, 113, 131, X, 1132 < | P: v3; 101, 113, X, 1132 | > P: 101, 113, X, 1191.3, 1182 / S: W; Y > P: v3; 113, X, 1111.1 || P: v3; 113,151, X, 1111.2) "then when-thus-knew the Indian that-it-would-be-cold and that-there-would-be-much-snow cursing the white-man and went-out and went-back (i.e. and when it turned out that this was how the Indian knew whether it would be cold and whether there would be much snow, the white man went out cursing and went back)"

The obviatives here indicate secondary subjects as before. Note that the difference between the two subject forms present (nePaqlsmaknik?, ne suyape ) is not indicated by obviation; we shall return to this case further below. (6) A159 #-qakl-ap-ne- ne- tax nasuPkin ka-nasu?kin taxa-s k-u-c-sqapqale-kn-e-mil Pat k-u-cxal-qsa-mu-nal-ka k-#-?ititla?t-il tax ne- k-#-la-wawismakniH-k ne- k- #-wanaqn4-niyi-kc-al-ka pal k- #-suka-kate- taxa-s k- #-itkin-il ?a kitla?-nam ?a qantla?-nam (P: 113, X, 1154.4, 1111.1 / S: W; v2; Y; 51, Y < O: W, 1132 I P: 101, 111, 131, X, 1191.4, 1131 < P: v^ 101, 111, 131, X, 1181, 1174.1, 1162 < P: 101, 113, X, 1141 < S: v2; W / P: 101, 113, 151, X r , 1182 / S: W; 101, 113, X, 1191.1, 1193, 1174.2, 1162 < P: vx; 101, 113, X, < | O: W, 1132 / P: 101, 113, X, 1141 / S: Y p , 1152.1; Y p , 1152.1) "told-me the then chief my-chief then I-should-change-work indeed I-should-help-them being-built then the returning the fighting-with-them really there-being-many then being-built house houses (i.e. my boss then told me I should change jobs to help build houses for returning servicemen, since they are now building many houses)"

The obviative of the first person (k-u-c-sqapqale kn-e -mil) can be interpreted in terms of the animate referent of the third-person subject (nasuPkin kanasuPkin, cf. §4.2.3); the subsequent absolute of the first person can then be interpreted as being adjacent to another first person, which does not require the obviative (cf. §4.2.1), as well as to a third-person implicit subject with inanimate referent (k-#-?ititla?t-it), which does not require the obviative, either (cf. §4.2.3). As in the preceding example (5), the three subject forms present (nasuPkin ka-nasu?kin, ne- k-#-wanaqna-niyikc-al-ka, Pa-kitlaP-nam ?a qantla?-nam) are not differentiated by obviation. On the basis of the two examples cited (and a number of additional examples in our corpus) we may state that the several subject forms of a sentence and their corresponding predicates may

132

A TECHNIQUE FOR THE TREATMENT OF MEANING

all stand in the absolute and hence be considered equivalent primary subjects. Finally, an example showing a new combination - two adjacent second persons (cf. §4.1.4, §4.2.4): (7) C V 103 c'u-qakil-ne- ni #sJ-sawsaqa- c pal k-#-qaky-am Pin tax k-in-c?isnil?upx-ne-mal napit'in-qahviy-mil k-in-c-Pupx-ne-mai" (P: v s ; 111, X, 1111.1 < S: W / P : 113,141, X < | P: v 3 ; v^ 101, 113, X, 1152.2 < O: W; v„ / P: 101, 112,131, X, 1191.3, 1174 || P: v^, 112, X, 1131 < P: 101, 112, 131, X, 1191.3, 1174) "and I-told-him the if-he-were-there and really saying this then that-you-would-see-each-other-with if you-wanted-that you-would-see-eachother-with (i.e. and I told him while he was there that people were saying: you will meet him, if you want to you will meet him)"

The obviative of the second person (hin-sl-qahviy-mii) can perhaps be interpreted as due to an adjacent subject with animate referent, or to the juxtaposition of two second person forms. 5. THE OBVIATIVE IN ISOLATION. In the preceding sections, the occurrence of the obviative has been interpreted by various relations between two or more syntactic units of the same order (phrases or clauses). I now consider the occurrences of the obviative in isolation - more exactly, the obviative as part of syntactic units in isolation. Rephrasing the definition given in §2.2, we may say that a syntactic unit stands in isolation when it is the only constituent of the unit of the next higher order (that is, the only phrase of a clause, or the only clause of a sentence). PHRASES IN ISOLATION. Examples: H IV 89, 141, 162 c?ina-mil-in" (P: X, 1131, 1113) "go-away" H III 166 quna-mit-in (P: X, 1131, 1113) "go-there" H IV 24 hin-lin-sl-qalwiy-mil-ne- (P: 112, 123, 141, X, 1131, 1111.1) "youseem-to-want-it" H IV 16 hu-l-c?ina-mil fP: 111, 122, X, 1131) "I-should-go-away" C IV 109 hu-qa?une l?at-mil-ne- (P: 111, X, 1131, 1111.1) "I-don't-knowwhat-to-call-him" C V 325 he y sa?n ?i-s c taxa-s (t; O: v 2 ; W, 1132; v3; W, 1132) "yes but this and then (i.e. yes, that's it)" C IV 326, 369 ?i-s c taxa-s (O: W, 1132; v 3 ; W, 1132) "this and then (i.e. that's it)" C V 549 taxta n4-s (O: v a ; W, 1132) "until now"

A TECHNIQUE FOR THE TREATMENT OF MEANING

133

The examples below include some cases in which, although the clauses themselves are in construction, the obviative forms have to be considered to stand in isolation, since they are unaccompanied by corresponding absolutes. CLAUSES IN ISOLATION.

C V 60 h6 y ma-k-u-huJpatnitit-mil (t; P: 12, 101, 111, X, 1131) "yes, I-didhear-it" C V 155 hu-c-la-qakik-mil-ne- (P: 111, 131, 151, X, 1131, 1111.1) "I'll-say-itagain" C n 140 Pat'u-qaPupxa-mit-ne- (P: V l ; 111, X, 1131,1111.1) "indeed I-don'tknow" C VI 346 taxa-s c m-6-qahviy-mit-ne (O: W, 1132 / P: v 3 ; 12, 111, X, 1131, 1111.1) "then (and) I-thought" C n 153 ?i-s c'u-skik-mil-ne- (O: W, 1132 / P: v„; 111, X, 1131, 1111.1) "this and I-mean (i.e. here's what I mean)" H n 31 #-su?k-s-e- ?aqlitit-?is-is" (P: 113, X, 1132, 1111.2 / S: Y, 1157, 1132) "good-were his-possessions" C V 246 na ?a-ki #-hi-n-cxa-s la?k?lak-s nasu?kin-s #-cxal-qakik-s-e- (P: v a ; v,; 113, morphophonemic prefix [cf. UAL 14.90 ri948)], 121, X, 1132 / S: W, 1132; Y, 1132 > P: 113, 131, X, 1132, 1111.2) "here also when-speaks another chief he-will-say" C IV 167 k-#-suka-kate-s k-#-sakqap-s lo-k-s sa?n k-#-snatta?e-tkin-s" (P: 101, 113, X, 1132 > P: 101,113, X, 1132 / S: Y, 1132 || P: v a ; 101,113, X, 1132) "there-being-a-lot lying-there firewood but making more (i.e. although there was a lot of firewood lying there, they should make more)" H VI 23 #-lu?-s-e- qapsin-s cin wu?u-s #-n-?in-s-e- (P: 113, X, 1132,1111.2 / S: Y, 1132 11 S: v,; Y, 1132 / P: 113, X, 1132,1111.2) "there-was-not anything, only water there-was" H VI7 #-walunis-s-e- ?aqlsmaknik?-s qu-s ?a-n lunqunik?-s ?aqlsmaknik?-s #-wa-s-e- (P: 113, X, 1132, 1111.2 S: / Y, 1132 || S: W, 1132; v s ; W, 1132; Y, 1132 / P: 113, X, 1132,1111.2) "came-packing people there more beyond people came (i.e. people came by pack, people from beyond there came)" C VI 253 Pat k-u-qal?at-ala?-is k-#-wilwum-s swin-?is-is (P: v,; 101, 111, X, 1151.2, 1132 / O: 101, 113, X, 1132; Y, 1157, 1132) "indeed our-calling-her being-big-bellied his-daughter (i.e. the one we call Bigbelly's daughter)"

In the cases in construction discussed in the preceding sections, the occurrence of the obviative was interpreted in relation to an absolute in the adjacent unit. The cases presented here, on the other hand, show obviatives without accompanying absolutes to which they could be related. An inspection of the examples reveals, however, that in a number of them such accompanying absolutes can be interpolated rather easily; the obviatives in isolation can thus be assimilated to obviatives in construction.

134

A TECHNIQUE FOR THE TREATMENT OF MEANING

In some of the cases, the interpolation can be made on the basis of the immediate context. The obviative of H IV 141 c?ina-mit-in" (P: X, 1131, 1113) "go-away" is parallel to that in H IV 139 c?ina-mil-in #-cistiyawe- (P: X, 1131, 1113 / O: 113, X) "go-awayto Passes-Too-Much-Water", where the obviative of the second person can be interpreted as due to the animate referent of the object (see under (1) in §3.2.4.1); the context makes it clear that in H IV 141 the same object can be "understood". In other cases, the obviative in isolation can be considered an "incomplete" version of a set expression which is often found also in the "complete" version. Example C IV 326 ?i-s-c taxa-s (O: W, 1132; v/; W, 1132) "this and then (i.e. that's it)' can be considered an "incomplete" version of the common ?i-s c taxa-s #-n-?in-ne- (O: W, 1132; v 3 ; W, 1132 / P: 113, 121, X, 1111.1) "this and then is (i.e. that's how it is)". A predicate can be said to be "understood". We can generalize from these more obvious cases to the less obvious, and interpolate the absolutes necessary to account for the occurrence of the obviatives. In doing this, we shall observe the safeguards mentioned by Hjelmslev in describing the procedure of interpolation (which he calls catalysis) :15 "in catalysis we must take care not to supply more in the text than what there is clear evidence for. In the case of the [Latin preposition] sine we know with certainty that an ablative is required ... [but] we are not justified in introducing by catalysis any particular noun in the ablative with the given sine. What is introduced by catalysis, then, is in most instances not some particular entity but an irresoluble syncretism between all the entities that might be considered possible in the given "place" in the chain." Applying these words to the Kutenai problem, we will not, in the less obvious cases, interpolate a specific syntactic unit in the absolute (object or predicate as above), but rather interpolate as "understood" any representative of the kind of syntactic unit in the absolute required to account for the occurrence of the obviative. For instance, in examples H III 166, H VI 16, and C IV 109 we interpolate an object with 15

Hjelmslev, op. cit. in fn. 2, pp. 60-1.

A TECHNIQUE FOR THE TREATMENT OF MEANING

135

animate referent; in H IV 24, C V 60, C V 155, CII140, C VI 346, and C II 153 we interpolate a third-person implicit subject with animate referent; in C IV 325 and C V 549 we interpolate a thirdperson implicit subject such as #-n-?in-ne- (113, 121, X, 1111.1) "it is"; in H II 31, C V 246, C IV 167, H IV 23, H VI 7, and C VI 253 we interpolate a primary subject. This allows us to assimilate all the cases of the obviative in isolation to the obviative in construction treated previously, and thus to account for the occurrence of the obviative in these examples. 6. AGREEMENT AS TO OBVIATION. I repeat the definition of agreement given in §2.2: the constituents of a sentence, clause, or phrase are said to be in agreement when all constituents sharing the same syntactic relation also share the same relation of obviation, that is, are either all in the obviative, or all in the absolute; in the contrary case, they are said to be in disagreement. To this definition should be added the following: syntactic units are in REGULAR AGREEMENT if their syntactic relation is such that it requires them to be in agreement, unless a special relation requires them to be in disagreement. I specify first the types of units which are in regular agreement, then examine the instances of disagreement and attempt to find the special relations which account for it. 6.1. REGULAR AGREEMENT. Units in regular agreement are of three types. (1) Constituents of the same phrase are in regular agreement (PHRASAL AGREEMENT). Examples: H I 78 k-#qakl-aps ia?k?lak-s nictahal-s (P: 101, 113, X, 1156.4 / O: W; 1132; Y, 1132) "being-told-by another young-man" H I 82 ... na tax na?uk?e- nictal sanla c #-mitx-ne- c #-n-?upii-ne- (S: W;v a ; W; Y; Y / P : v„; 113, X, 1111.1 | | P : v 3 ; 113, X, 1111.1) "... (this then) the-one young Piegan (and) fired and killed-him" H VI 168 #-qalsa-s-e- ni-?s k-#-?ilkake ka-s ?aqlsmaknik?-s (P: 113, X, 1132,1111.1 / S : W, 1132; 101,113, X, 1132; Y, 1132) "three-were the comingfrom-afar people"

These examples show that phrases in regular agreement are

136

A TECHNIQUE FOR THE TREATMENT OF MEANING

subjects or objects; the question of agreement does not arise within predicates, since they contain only one constituent capable of standing in the obviative, the verb, whereas subjects and objects may contain several such constituents. (2) Within the same clause, the subject and the predicate are in regular agreement (CLAUSAL AGREEMENT). Examples: H IV 165 ... #-wa-s-e- palke-s ... (P: 113, X, 1 1 3 2 / S : Y, 1132) " . . . came [the] woman ..." H V14 ?a ki # to-xa # -y unaqa?-ne- Paqismaknik?... (P: v,; v a ; 113, X, 1111.1 / S: Y) "also just-about many-were [the] people" H V 32 ... #-n-?uqoxaqa-l-is-ne- ni-?s t?inamu-?s (P: 113, 121, X, 1141, 1132, 1111.1 / S: W, 1132; Y, 1132) " . . . in-was-put the grease"

(3) Within the same sentence, all clauses and clause constituents relating to the same referent are in regular agreement (SENTENCE AGREEMENT). In the following examples, translations of constituents in agreement are italicized: H II 43 k-#-qa-kl-kqáyxu-kc-iy-am-is c k-#-wa)kwayit-s q o - s m a - # - q a latnaq-na-k-s-Pc #-qa?upx-ne- ka?a-s #-n-?a-qana-s (P: 101, 113, 132, X, 1193, 1191.4, 1152.2, 1132 | > P: v,; 101, 113, X, 1132 > O: W, 1132 / P : 12, 113, X, 1191.3, 1182, 1132, 1111.2; v3 || P: 113, X, 1111.1 < O: W, 1132 / P : 113,121,181, X, 1132)"when-people-were-playing-the-ballgameand-whenit-was-evening there he-had-sat-down-there and he-didn't-know what he-wasdoing" H II 46 #-suk}atnula-katPc k-#-la-?uwuk-s ni-?s #-ta-tuqalaqPana-s-e• (P: 113, X ; v3 | > P: 101, 113, 151, X, 1132 > O: W, 1132 / P: 113, 151, X, 1132, 1111.2) "when-he-took-a-good-look-at-him (and) getting-up-again (here) he-went-back-into-the-thicket" H III 204-5 #-litwu?t-e- cin #-q?akpakitninxu?-ne- ní-?s k-#-qásc?umqaninqap-s #-n-osanuxunqap-s-e- #-wanaqna-nam-nam-is^c (P: 113, X, 1111.2 || P: v s ; 113, X, 1111.1 < S: W, 1132 / P: 101, 113, X, 1132 || P: 113, 121, X, 1132, 1111.2 < P: 113, X, 1161.3, 1152.1, 1132; v3) "he-had-no-arrows still he-killed those being-expert, they-fled as-they-were-fighting (and)" H VI103-4 taxa-s #-laxal?upx-ne- ?iyamu-s #-suka kate-s-e- (O a : W, 1132 / P: 113, X, 1111.1 / Ot: Y, 1132 || P: 113, X, 1132, 1111.2) "then he-saw-again [the] game, there-was-a-lot-of-it" H VII 39-40 #-qawxax-i"c #-ia-hi?-s-e- pal #-q?apq?att?aqcxu-na-k-s-e-" (P: 113, X, 1111.2; v„ || P: 113, 151, X, 1132, 1111.2 || P: vr, 113, X, 1191.3, 1182, 1132, 1111.2) "he-went-there and he-went-down-there and they-were-nomore really they-had-been-smashed-to-pieces" A X 207 c tax né- qala ?unik-s #-la-qawxa?texa na-s c k-#-c-sl-qa-qaskin1-is k-#-akq?ye-t-s"c #-c-?isnilqal?ip-ne- ni-?s c ka?a-s xma- # -n-?a-qal?ip ne- swu-tmu (S: v,; v s ; W; W / O: W, 1132 / P: 113, 151, X < O: W, 1132;

A TECHNIQUE FOR THE TREATMENT OF MEANING

137

v 3 1 P: 101,113,141, X, 1141, 1132/S: 101,113, X, 1132; v3 > P: 113, 131, X, 1111.1 < O,: W, 1132; v3 / O t : W, 1132 / P: 11, 113, 121, 181, X/ S: W; Y, 1176) "and then the someone later if-he-speaks-of-it-again here and when isfinished conference and he-himself-will-die-thus here how and would-have-died the friend-with (i.e. and then if anyone speaks of it again, now that the conference is over, he, too, will die just as the two friends died)"

In the last example, note the presence of two primary subject forms,... ne- qala ... "the someone" and ... ne- swu-tmu ... "the friends"; for more details see under (7) in §4.3. 6.2. SYNTACTIC FUNCTION OF AGREEMENT. Sentence agreement allows us to discern two significant syntactic functions of agreement. (1) Agreement serves to identify actor and action reference in a compound sentence with several implicit subjects, without the need for supplying or repeating subject forms as would be done in English. In most examples cited, absence of obviation, with its function of differentiating primary and secondary subject, would result in considerable confusion. (2) Agreement serves as a negative morphological boundary signal for the sentence: the end of agreement indicates a sentence boundary. 16 For example: H III 69 #-n-osanuxun-ka?-ne- # -qáqaklinqap-s-e- ?aq?anka-?is 70 ?át #-cukat-e- ni-?s a q?anka-?is-is°c #-n-?itqayq-ne- c #-n-?it?xu?-mi-k c # lamxup-mi-k (P: 113, 121, X, 1162, 1111.1 || P: 113, X, 1132, 1111.2 / S: Y, 1157. P: v,; 113, X, 1111.1 / O: W, 1132; Y, 1157, 1132; v3 || P: 113, 121, X, 1111.1 || P: v3; 113, 121, X, 1191.3, 1182 || P: v5; 113, X, 1191.3, 1182) "Theyfled-from-him there-were-lying their-weapons. Indeed he-took (the) theirweapons and he-tied-them-up and he-lay-down and he-laid-down-his-head"

Note that the absolutes in the two sentences refer to two different entities; if the two references were included in the same sentence, they would be expressed by a primary and a secondary subject respectively, and the latter would stand in the obviative (cf. §4.1.6). Note also that the same referent "their weapons" is expressed in one sentence by a noun in the absolute, in the other " Comparable to N. S. Trubetzkoy's "Grenzsignale," cf. Grundziige der Phonologie, pp. 241 ff. (Travaux du Cercle Linguistique de Prague, Vol. 7; Prague, 1939).

138

A TECHNIQUE FOR THE TREATMENT OF MEANING

by a noun in the obviative with an accompanying adverb in the obviative. The two phrases are then in disagreement, showing that agreement stops at sentence boundary. 6.3. DISAGREEMENT. I turn now to the cases of disagreement where by the criteria of §6.1 regular agreement is expected. By analogy with regular agreement, I speak of PHRASAL DISAGREEMENT, CLAUSAL DISAGREEMENT, and SENTENCE DISAGREEMENT. The last, however - that is, disagreement between clauses and clause constituents relating to the same referent - is excluded by definition, since, in line with the function of agreement as a negative morphological boundary signal (§6.2), such a disagreement marks a sentence boundary. 6.3.1. PHRASAL DISAGREEMENT. The cases shown here are phrases with constituents that are in disagreement. Examples: H I 121 ... ki-#-?in k-#-u!aqte- titqat?-s(P: 101, 113, X / S: 101, 113, X; Y, 1132) "... being [an] old man" H IV 15 ... #-qaqalcaqakminuxunqap-s-e- na-s sit-?is" (P: 113, X, 1132, 1111.2 I O: W, 1132; Y, 1157) "... he-jumped-into (this) his-blanket" H IV 143 ... #-q&qawsaqap-s-e- ni-?s tiinamu-?is ... (P: 113, X, 1132, 1111.2 I S: W, 1132; Y, 1157) "... there-stayed (the) his-wife ..." H VI167 taxa-s #-n-?as-s-e- ni-?s c akiklu-Pis ... (O: W, 1132 / P: 113,121, X, 1132,1111.21S: W, 1132; v3; Y, 1157) "then two-were (the and) his-camps... (i.e. then he belonged to two camps)" A X 270 Pat #-qsawsaqap-mal-ne- k-#-amkuq?uqut katwumlat-s (P: v,; 113, X, 1174, 1111.1 I O: 101, 113, X; Y, 1157) "indeed he-lived-with black coats (i.e. the priests)" C IV 43 ... sa?n ni-?s Paqlsmaknik? #-l-?isnil?e tkin-s"c" ... (S: v,; W, 1132; Y I P: 113, 122, X, 1132; v3) "... but the Indian should-do-it-himself and ..." C V 209 ... ki-#-?in-s k-#-witqa qaspihik-s" (P: 101, 113, X, 1132 / S: 101, 113, X; Y, 1132) "... it-being Big Crane" C V 355 ... k-#-qakil-qakl-aps Paqlsmaknik? nasuPkin-s ... (P: 101, 113, 132, X, 1156.4 / O: Y; Y, 1132) "... he-was-told-by chief Indians (i.e. the Indian agent)" C V 419 #-?in-ne- k-#-wilqa qapsin-s (P: 113, X, 1111.2 / O: 101, 113, X; Y, 1132) "it-is great something (i.e. something important)" C V 6 0 5 . . . #-sukumxuniyaqap-se- qu-s-c sl-khikqani-Pis" (P: 113,X,1132, 1111.2 / S: W, 1132; v3; Y, 1157) "... good-seems-to-be (that and) his-talk" C VI 213 c taxa-s na-s swu-Pis #-lin-?isnilqa?n-s-e- ktitkPam-s ... (0 2 : v3; W, 1132/S: W, 1132; Y p , 1157 / P: 113, 123, X, 1132, 1111.2 / O: W, 1132) "and then this his-friend seems-not-to-be nothing ... (i.e. his friend seems to be all right)"

A TECHNIQUE FOR THE TREATMENT OF MEANING

139

Most of these cases fall into two groups: (1) examples H I 121, A X 270, C V 209, C V 355, and C V 419 show a subject or object containing a noun in the obviative, accompanied by another noun (C V 355) or a verb form in the absolute; (2) examples H IV 15, H IV 143, H VI 167, C V 605, and C VI 213 show a subject or object containing a third-person possessed noun in the absolute, accompanied by an adverb in the obviative. The nouns in group (2) follow the rule stated in §3.2.4, namely, that the obviation of the third-person possessed noun is governed by the obviation of the possessor, while the accompanying adverbs are not subject to that rule and hence take the obviation required by the syntactic function of the phrase. This accounts for the disagreement between the two phrase constituents. The cases in group (1) have to be interpreted somewhat differently. These phrases show a high degree of integratedness; many of them are single lexical units and can be considered phrasal compounds subject to obviation as wholes, rather than constituent by constituent. The obviative suffix appears as part of the last constituent only, and the disagreement between constituents is the formal marker of phrasal compounding. Example H IV 143 remains. Here the noun is in the absolute and the accompanying adverb in the obviative. Since the noun is not a third-person possessed noun, the example cannot be included in group (2). It may, however, be interpreted in the same terms as the residual cases in §6.3.2 below; see the final paragraph of that section. 6.3.2. CLAUSAL DISAGREEMENT. The cases presented here show subject and predicate in disagreement. Examples: H I 61 c pal #-n-?&qsalcukat-l-is xalcin- Pis (P: v 3 ; v^ 113, 121, X, 1141, 1132 / S: Y, 1157) "and really when-several-were-taken [of] their-horses" H H 38 ... #-la-qalaxa-s-e- xale-Pis PakuqPlalxo- ... (P: 113, 151, X, 1132, 1111.2 I S: Y, 1157; Y) "... no-more-came-back his-son Swamp-Indian ... (i.e. the Swamp Indian's son did not come back any more)" H III 10 ... taxa-s Pat #-n-?ana-s-e ?al-tat-?is ... (O: W, 1132 / P: v^ 113, 121.X, 1132, 1111.2/S: 41, Y p , 1157) "... then indeed were-hunting his-olderbrothers ..." H III 69 ... #-qi-qaklinqap-s-e- ?a q?anka-?is (P: 113, X, 1132, 1111.2 / S: Y, 1157) "... there-remained-there his-weapons"

140

A TECHNIQUE FOR THE TREATMENT OF MEANING

H I V 104 taxa-s #-sl-?as-s-e- tilnamu-Pis (O: W, 1132/P: 113,141, X, 1132, 1111.2 I S: Y, 1157) "then two-were his-wives (i.e. he had two wives)" H IV 151 #-hik?mux-ne-mu-l-is-ne- t?awu-?is Pa-k-Pis (P: 113, X, 1191.3, 1181, 1141, 1132, 1111.1 / S: Y, 1157; Y, 1157) "was-used-for-a-split his-bow [and] his-arrow" H V 11 taxa-s Pa-ki #-saklaqap-s-e- wa?nmu-?is na-s taxa-s" (O,; W, 1132 / P: v,; 113, X, 1132, 1111.1 / S: Y, 1157 / Ot : W, 1132; W, 1132) "then also there-was-there his-blood (i.e. his kin) now then" H VI 27 Pat #-qaqal?in-s-e- Pik-kpict-iyi-Pis ne- tax lawatPinak ne- pikPak PaqlsmaknikP (P: v,; 113, X, 1132, 1111.2 / S: Yr, 1191.4,1157; W; v,; W; W; W; Y) "indeed it-was-so their-favorite-meal the then from-beyond-theRockies the formerly people (i.e. it was the favorite meal of the people of old from beyond the Rockies)" A II 56 ... #-n-Pin-s-e- Pa-qate-kn-iyi-Pis ... (P: 113, 121, X, 1132, 1111.2 I S: Y r , 1191.4, 1157) " . . . it-is his-job ..." A m 31 qa-lin ni-?s xma-ki-#-?in-s ?akihviy-?is ... (O: v s ; W, 1132 / P: 11, 101, 113, X, 1132 / S: Y, 1157) "exactly here that-would-be his-heart ..." A X 95 pal #-n-Pin-s-e- nasukin-Pis k- # -wanaqna-niyi-kc-al-ka ne- tax k-#-?4kaqlilkin-mi-k (P: Vj; 113, 121, X, 1132, 1111.2 / O : Y, 1157; 101,113, X, 1191.3,1193,1174.2,1162 / S: W; v s ; 101,113, X, 1191.3, 1182) "really was their-chief those-fighting-with-them the then He-Who-Takes-His-Eye-Out (ie. the leader of the soldiers [ = Northwest Mounted Police] was He-Who-TakesHis-Eye-Out)" C IV 303 Pa-ki #-la-lo ?-is natanik^c ... (P: v 2 ; 113, 151, X, 1132 / S : Y ; v3) "also when-there-is-no-more moon and ..." C V 15 ... #-saq?an-s-e- misd l Pakikliyi-Pis^c (P: 113, X, 1132, 1111.2 / S: W; Y r , 1157; v3) " . . . there-was-hanging Michael his-name and (i.e. Michael's signature was affixed)" C V 628 .... k-#-qo qawsaq?aHqawisa-s ne- suyapi^c (P: 101, 113, X, 1132 I S: W; Y; v3) " . . . having-passed-by-there-some-time-ago the white-man (and)"

In most of these examples, the disagreement is between a thirdperson possessed noun in the absolute as subject and a predicate in the obviative. The parellelism between these cases and the ones treated under (2) above is evident; here again, the agreement is between the third-person possessed noun and the possessor (which is the primary subject and hence in the absolute), leading to clausal disagreement between subject and predicate. In H II 38, H VI 27, and C V 15, the subject includes both the possessor and the third-person possessed noun, both in the absolute, and in disagreement with the predicate. The absolute of the thirdperson possessed noun is due to its agreement with the possessor; the absolute of the latter can be interpreted in the same terms as the cases immediately below.

A TECHNIQUE FOR THE TREATMENT OF MEANING

141

The three residual cases A X 95, CIV 303, and C V 628, to which can be added the three above, show a subject in the absolute containing a noun without possessive suffix (or a subject partially in the absolute: noun in the absolute, adverb in the obviative, as in C IV 143), accompanied by a predicate in the obviative. An inspection of the contexts of these examples reveals that the nouns in the absolute can be interpreted as instances of ANTICIPATED PRIMARY SUBJECTS. That is, the reference is here to entities which in terms of the preceding context function as secondary subjects and still do so in the sentence in question, but in a later context function as primary subjects. The absolutes in the examples signal an impending change in the "role" of the entity referred to. 6.3.3. FUNCTIONS OF DISAGREEMENT. The functions of disagreement are three. (1) It marks phrasal compounds. (2) When due to the special rules of agreement affecting third-person possessed nouns, it indirectly identifies the possessor of a third-person possessed noun. (3) It indicates an impending switch in the emphasis of the narrative by marking the anticipated primary subject. 7. SUMMARY. Examining the different syntactic combinations in which obviation functions, we see that although the obviative suffix occurs in words of all three full-word classes - in "action" words (verbs) as well as "entity" words (nouns) (cf. §§0.4, 1) obviation has reference only to a relation between two ENTITIES, formally signaled or implicit. 7.1. GENERAL MEANING OF OBVIATION. Obviation serves to differentiate subject from object (§3.1, §3.2); primary object from secondary object (§3.3); and primary subject from secondary subject (§4). Summarizing these three relations, we can say that obviation refers to the relation between a more immediate and a more remote unit, that is, a relation of MARGINALITY (particularly clear in §3.2.4.V). The particular unit in which the obviative suffix is included does not depend on the referential relation here summarized, but on the grammatical and referential structure of the units that are in this relation: on the person of the subject or primary subject, on

142

A TECHNIQUE FOR THE TREATMENT OF MEANING

the animate or inaminate reference of the object or secondary subject. The placement of the obviative suffix as part of a particular unit, predictable in terms of the discussion above, is thus a POSITIONAL VARIATION of the occurrence of the obviative. On the other hand, on the basis of the referential features determining the placement of the obviative suffix, we can infer a secondary function of obviation: the placement of the obviative is the formal mark of a referential category of ANIMATENESS, which is not otherwise formally marked. 17 7 . 2 . NEUTRALIZATION OF OBVIATION. In treating agreement (§6) we have seen that certain special relations supersede the obviation regularly required by a relation of marginality such as listed in §7.1: the first component of a phrasal compound remains in the absolute (group ( 1 ) in §6.3.1), a third-person possessed noun is in the absolute if the possessor is in the absolute (group (2) in § 6 . 3 . 1 ) , the anticipated primary subject is in the absolute (§6.3.2) - all this in contexts where these units might be expected to be in regular agreement with units in the obviative. We may here legitimately speak of GRAMMATICAL NEUTRALIZA18 TION, comparable to phonemic neutralization. Carrying the parallel further, we may say that obviation is a neutralizable category, and that the cases of disagreement listed above constitute positions of neutralization for this category. In these positions of neutralization, the representative of the category is its unmarked member,19 namely the absolute. 7.3. MARK OF OBVIATION. Thus, we may say that the category of obviation has the MARK OF MARGINALITY, and that the obviative is its MARKED MEMBER, the absolute its UNMARKED MEMBER. The 17 This is an instance of what C. F. Voegelin calls INFERENTIALLY MARKED MEANING, defined as "meaning marked by the association of two or more morphemes across word boundaries; or, more indirectly, by associations of morphemes within word boundaries", "Linguistically Marked Distinctions of Meaning," in: Sol Tax, ed., Indian Tribes of Aboriginal America, Selected Papers of the XXIX Internal. Congr. of Americanists 3.227 (Chicago, 1952). 18 Cf. op. cit. in fn. 16, pp. 206 fF. For a somewhat different conception of grammatical neutralization, see Martin S. Ruipérez, The Neutralization of Morphological Oppositions, Word 9.241-52 (1953). 18 Cf. op. cit. in fn. 16, p. 73.

A TECHNIQUE FOR THE TREATMENT OF MEANING

143

presence of the obviative indicates the presence of a marginal referent (though the obviative suffix need not be included in the unit signaling this referent, cf. §7.1), whereas the absolute is neutral as to marginality (and serves to represent the category in the position of neutralization, cf. §7.2).20 7 . 4 . CONCLUSION. By applying a modification of the standard contrastive techniques of linguistics to a large body of data (nearly all the syntactic units containing obviative forms in a corpus of considerable size), supplemented by a common-sense use of the information contained in the context of the examples, it has been possible to formulate inductively a statement of the meaning of the category of obviation. As by-products of the analysis, we have been able to ascertain some interesting structural properties of the grammatical category under investigation, which at least partially confirm and extend the conclusions of previnos investigators (Jakobson, Trubetzkoy, Yoegelin, cf. fn. 17, 18, 19, 20): inferentially marked meaning, grammatical neutralization, marked and unmarked members of a category, and the relation between the last two.

20

Cf. Roman Jakobson, "Zur Struktur des russischen Verbums," Charlsteria Guilelmo Mathesio quinquagenario ... oblato (Prague, 1932), pp. 74-6.

ON STRUCTURALIST METHOD*

The basic assumption underlying structuralist method in linguistics is that of the covariance of form and meaning in paradigmatic sets and syntagmatic sequences. This covariance is exploited by controlling one of the two covariant terms as the independent variable, and studying the other as the dependent variable, under a set of conditions held as near constant as possible. This constant set of conditions is what is commonly called the substitution frame; it can be defined on the basis of the defining properties of the units under investigation. By using a constant frame a substitution class is obtained which has class properties shared by all members; each individual member will in addition have differential properties setting it off from the remaining members of the class. In linguistic field work, meaning is used as the independent variable to elicit forms. This is done by asking the informant to translate controlled sets of sequences of known meaning. These sequences are held semantically constant, with the exception of one minimal controlled variation, on the assumption that the covariant in the informant's responses of this minimal semantic variation may be a minimal formal variation. If such a covariance results, a formal paradigm has been elicited. If none results, the particular line of questioning is considered unproductive and the assumption of a paradigmatic set for which these questions were designed is abandoned. For the study of meaning, linguistic form can be used as the independent variable. The elicitation procedure is then reversed: instead of supplying meanings, the linguist supplies forms in the * Originally published in Georgetown U. Monograph Series on Languages and Linguistics 11.145-8 (1960).

ON STRUCTURALIST METHOD

145

shape of recorded spontaneous utterances, and obtains covariant meanings in the shape of informant's translations and explanations. The crux of the procedure is then to narrow down the problem to a minimal covariance, and to hold the environment sufficiently constant so that the dependent variable can be shown to vary with the independent variable and not something else. A minimal covariance can be delimited most suitably if a minimal form class is selected. Such a form class will most readily be a paradigmatic morpheme class, that is, a distributionally defined morpheme class of limited membership. The meanings of the morphemes of such a class can be called grammatical meanings. A corollary of the limited membership of paradigmatic morpheme classes is their great frequency of occurrence, which allows the selection of properly defined environments on the basis of a not too extensive corpus. Thus, grammatical meaning offers two important advantages as a starting point: formally limited covariance, and high recurrence of units. In the immediately preceding paper of this volume,1 I have attempted such a controlled study of meaning. I selected a minimal paradigmatic set, consisting of a morphologically defined morpheme class of one, and studied translation similarities and differences in controlled syntactic frames in terms of the presence of the given morpheme versus its absence. I was able to arrive at a detailed and motivated differential statement of meanings. Given this technique as a starting point, we can now consider the problem of lexical meaning. Lexical meaning can be defined as the unique meaning of a member of a form class of unlimited membership. A corollary of this membership size is a greatly reduced frequency of occurrence as compared to paradigmatic morphemes. In order to allow a detailed study of covariance, the unlimited membership of the form class must be broken up into smaller sets of a membership size approaching that of grammatical paradigms. Such sets may be found in the culture as finite terminologies and 1

Pp. 98-143 above.

146

ON STRUCTURALIST METHOD

need not be separated out by special procedures. The extralinguistic referents of the various terms may be clearly enough defined to allow a covariant analysis of form and meaning in terms of lexemes and referents. This constitutes an important economy of effort, since the reduced frequency of occurrence of individual lexical meanings in texts requires an extensive corpus to find enough proper frames for lexical comparison. In a recent paper 2 Lounsbury has analyzed such a "paradigm of terms" from a kinship system. His technique is based on the covariance of terms and referents, rather than terms and translations: he examines the particular kin types corresponding to each term and assigns a referential kin class to each of them. This class is defined by a set of distinctive features - the differential properties of the individual members of the paradigm. Metaphoric extensions of a given term are interpreted in terms of its nondistinctive features, and are not included in Lounsbury's semantic analysis.3 For those lexemes which do not fall into clearly given sets (and even for the metaphoric extensions of those that do), a preliminary technique is required for separating out paradigms of terms. Nida has proposed a method for semantic analysis4 which consists in substituting, under controlled conditions, impressionistically more general lexemes for more specific ones (for instance, "animal" for "burro", or "tree" for "oak"). His method can be used as a device for obtaining controlled semantic sets: all specific lexemes replaceable by the same general lexeme will belong to the same set. Nida's technique allows a simple operational definition of homonymy: forms apparently assignable to the same lexeme are homonyms if they belong to more than one set by virtue of multiple substitutability. A difficulty will arise when no general lexeme suggests itself for a given impressionistically related set of lexemes. In such cases, 2

Floyd G. Lounsbury, "A semantic Analysis of the Pawnee Kinship Usage," Language 32.158-94 (1956). 3 Loc. cit., 193. 4 Eugene A. Nida, "Some Problems of Semantic Structure and Translational Equivalence," in: A William Cameron Townsend en el XXV Aniversario del I.L.V., (Mexico, n.d.), pp.313-325.

ON STRUCTURALIST METHOD

147

an external common denominator will have to be found to extract a paradigm of terms not directly given by the culture. One example of this might be a category label such as "invectives", another would be a cultural frame such as a narrowly delimited trait complex or activity. The operational study of lexical meaning will obviously have to start with culturally given paradigms of terms or with substitution sets that present no special difficulties. These sets can be analyzed by comparing clearly defined referents, or by the same techniques of comparing translation similarities and differences as were used in the analysis of a simple grammatical paradigm. The result will be differentially structured, rather than atomistic, statements of the meanings of the members of the set and of the class meaning of the set as a whole. With the experience gained in the study of obvious sets, less obvious sets can be approached, and finally the different sets and sets of sets can be related to each other. Although much of this is already current in modern lexicographic practice, a better statement of overall semantic structure can be expected than the usual thesaurus-type listings of synonyms and antonyms, especially for languages spoken in cultures unlike our own.

S T R U C T U R A L I S M BEYOND LINGUISTICS*

The basic assumption of structuralism is that its particular object of cognition can be viewed as a structure - a whole, the parts of which are significantly interrelated and which, as a whole, has a significant function in the larger social setting. The cognitive elements of structuralism are thus two orders of entities - the whole and the parts, and two orders of relations - the function of the whole and the relations between the parts. Structuralist analysis seeks to delimit the entities and to describe the functions, raising the boundary problem and the problem of function as crucial analytic objectives. The analysis is possible because some of the cognitive elements are given by direct observation or may be posited on the basis of a common-sense interpretation of consistent informal observations. These can serve as the starting points for a controlled inferential method to ascertain the remaining elements, the unknown. Structuralist analysis can thus be applied to any object of cognition which may legitimately be viewed as a structure and for which appropriate analytic starting points can be found. Let me use as an example esthetic and literary structuralism as practiced by the Prague School of the 30's: in their approach the whole, the parts, and the function of the whole serve as starting points; the detailed relations between the parts are the unknown. The whole and the parts are given by direct observation: the work and its components (such as the chapters of a novel or the lines of verse of a poem) have a perceivable beginning and end. * Originally published in Georgetown University Monograph Serves on Languages and Linguistics 13.93-7 (1962). This paper is in part a restatement of views previously presented in my Introduction to the Prague School Reader (see fn. 1).

STRUCTURALISM BEYOND LINGUISTICS

149

The function of the whole - that is, of the work of art or literature - is posited to be the esthetic function, which is by the Prague School defined in opposition to the practical functions. Every object or action, language included, can be assigned a practical function - utilitarian for tools, communicative for language, and so on. If, however, an object or action becomes the focus of attention for its own sake and not for the sake of the practical function it serves, it is said to have an esthetic function; that is, it is responded to for what it is, and not for what it is for. Thus, the esthetic functions as such is not limited to works of art and literature but can appear in connection with any object or action. It comes about by virtue of what I have translated as foregrounding, as opposed to automatization. Automatization is the term used to refer to the stimulus normally expected in a social situation; foregrounding - in Czech aktualisace - on the other hand refers to a stimulus not culturally expected in a social situation and hence capable of provoking special attention. Let me paraphrase a linguistic example given by Bohuslav Havranek1: If we translate the wellknown Russian greeting "zdravstvuyte" into English by its functional equivalent of "good morning", "good afternoon", or "good evening", it will pass unnoticed as the normal greeting under the circumstances. If, on the other hand, we translate it literally as "be well", it might still be understood as some kind of a greeting - that is, it may retain its communicative function - but it will in addition provoke special notice of some sort, perhaps cause some wonderment as to the intent of the translator, or be interpreted as trying to convey the impression of a foreign environment. The free translation thus constitutes an automatization, the literal translation is an instance of foregrounding in which the wording itself, rather than the communicative content of the message, is responded to, and this property of stimulating a response in terms of itself is what constitutes the esthetic function. 1

A Prague School Reader on Esthetics, Literary Structure and Style, Paul L. Garvin, ed. (Second Printing, American University Language Center, Washington, D.C., 1959), p. 9.

150

STRUCTURALISM BEYOND LINGUISTICS

In regard to language, Jan Mukafovsky2 refers to foregrounding as the "esthetically intentional distortion of the linguistic components." To put this statement into its proper framework, we must understand that the Prague structuralists are Saussurians. In Ferdinand de Saussure's conception, the linguistic pattern - la langue is both a system of signs, and a set of social norms.3 As a system of signs, the pattern has a certain flexibility, allowing for variations within the units and in the choice and arrangement of units, to the extent that it does not conflict with the requirement of intelligibility. As a set of norms, the pattern is more rigidly circumscribed in terms of the cultural preference for, and statistical frequency of, these allowable variations. Mukarovsky's distortion is thus a distortion of the pattern qua social norm, but still within the bounds of the pattern qua system of signs, since the distorted, foregrounded units stem from the same system as their automatized counterparts, or are borrowed into it and in terms of it. The esthetic function is not limited to works of art and literature. What characterizes the latter as opposed to the random foregrounding which may occur in any social situation is, in Mukarovsky's words, "the consitency and systematic character of foregrounding. The consistency manifests itself in the fact that the reshaping of the foregrounded components within a given work occurs in a stable direction; thus, the deautomatization of meanings in a certain work is consistently carried out by lexical selection (the mutual interlarding of contrasting areas of the lexicon), in another equally consistently by the uncommon semantic relationship of words close together in the context. Both procedures result in a foregrounding of meaning, but differently for each. The systematic foregrounding of components in a work of poetry consists in the gradation of the interrelationships of these components, that is, in their mutual subordination and superordination. The component highest in the hierarchy becomes the dominant. All other components, foreground2

Op. cit. in fn. 1, p. 20. For a detailed discussion of this aspect of Saussurian thinking, see Henri Frei, "Langue, Parole et Differenciation," Journal de Psychologie, 1952, pp. 137-57.

5

STRUCTURALISM BEYOND LINGUISTICS

151

ed or not, as well as their interrelationships, are evaluated from the standpoint of the dominant. The dominant is that component of the work which sets in motion, and gives direction to, the relationships of all other components."4 While the detailed relations between the parts of the whole thus still remain to be ascertained, two basic assumptions are made of their nature, which will aid in formulating a method for their investigation: that these relations are systematic, and that they are special detailed manifestations of the function of the whole. From this set of assumptions stems the use of test frames and verification by controlled techniques as illustrated in Mukarovsky's paper on the prosodic line.5 He deals with the whole (the poem) and the parts as given by direct observation. He posits as a basic functional property of the line of verse a particular prosodic line (that is, pitch and stress distribution) characteristic not only of certain poems, but also of a given author, functioning as the esthetic dominant to which other esthetic features such as word order (which, unlike English, in Czech has a strongly reduced grammatical function and is thus available for esthetic exploitation) are subordinate. To verify this assumption, he uses the line of verse as a test frame within which he subjects word order to systematic and controlled variation. He has two controls (which he abandons only in a few instances where he explicitly states his reasons): (1) word order is varied only over the span of one half of the line, (2) rhyme is not altered. He inspects the prosodic line of the original line of verse, and compares it to the line of verse resulting from his controlled variation. The basis for the assignment of a given prosodic line to a particular line of verse appears to be (and this is the one procedural step Mukarovsky fails to state) the unemphatic fluent reading by an educated speaker (I tested this by reading the lines myself). The result of his induced variation is in each case a change of the prosodic line, often from one characteristic for the author of the parti4

Op. cit. in fn. 1, p. 23. Travaux du Cercle Linguistique de Prague 1.121-39 (1929). Garvin, op. cit., pp. 131-54.

6

152

STRUCTURALISM BEYOND LINGUISTICS

cular line of verse to that characteristic for another poet. He considers this, rightly I believe, a verification of his assumption of the dominant function of the prosodic line. Mukarovsky's methodological contribution consists in the addition of a sequential permutation test to the substitutional commutation test,6 thus doubling the inventory of basic structural techniques. The two fundamental devices of Prague School structuralism can be summarized as follows: (1) the formulation of a functional criterion serving as the basis for a theoretical conceptualization of the object; (2) the development of analytic methods exploiting the functional criterion. This dual approach has been applied by them not only to linguistics, but to literary, esthetic and folklore7 analysis as well. It can be extended further, both theoretically by the development of functional relevance criteria and analytically by the application of the basic commutational and permutational techniques, to the description of any object of cognition for which it is reasonable to assume that it is culturally structured. On the non-verbal level this includes, for instance, temporally structured cultural events (such as ceremonies) in which the whole and the substantive component parts are given by direct observation. Structuralist method can here help to study the relations of the parts to each other and to the whole, and to establish the underlying network of functional connections. This also includes, on the verbal level, any not completely idiosyncratic and therefore culturally significant texts. Here it is reasonable to assume that the documents which are the focus of interest of such language data processing problems as information retrieval and automatic abstracting, constitute such texts. It follows that structuralism has an important part to play in these efforts.

• Cf. Louis Hjelmslev, Prologomena to a Theory of Language (Translated by Francis J. Whitfield, Baltimore, 1953), p. 46. ' Cf. Peter Bogatyrev, Funkcia kroja na Moravskom Slovensku (The Function of Folk Costume in Moravian Slovakia, Turciansky Svaty Martin, 1937).

THE STANDARD LANGUAGE PROBLEM: CONCEPTS AND METHODS*

When I was on Ponape (Eastern Caroline Islands) some time ago as part of the Coordinated Investigation of Micronesian Anthropology, I was, in addition to straight linguistic research, concerned with the task of devising a unified orthography for Ponapean.1 I cooperated with a committee offivenative leaders in this work, and I went one step beyond the immediate goal in that I also attempted to introduce a suitable Ponapean terminology for use in a spelling book containing the new spelling rules. This involved in part the consolidation of existing terms, in part the creation of new terms from native and borrowed sources. My native collaborators were extremely helpful not only in agreeing on terminology, but also in helping to compose a set of what I considered concise and modernsounding statements. I was primarily responsible for the formulation, in the sense that I would make initial suggestions and after some discussion and corrections would lead the native committee to some agreement on the final wording. In working with my Ponapean friends during and after the compilation of the new orthography and spelling book, I found that they had a keen understanding of the reasons for my suggestions and were able to make intelligent decisions under my guidance whenever a problem required a choice from among several possibilities. I had a distinct sense of failure when, once the system was agreed upon and the spelling book was complete, I found that the members of the native committee had a great deal of trouble applying the * Originally published in Anthropological Linguistics, March 1959, pp. 28-31. For a detailed discussion, see my "Literacy as a Problem in Language and Culture," Georgetown University Monograph Series on Language and Linguistics 7. 117-29 (1954). 1

154

THE STANDARD LANGUAGE PROBLEM

rules and remembering the terms which they themselves had helped formulate. The gist of my Ponapean experience can be stated quire briefly: literacy is not the same as standard language. In an essentially folk cuture, where literacy was a realistic objective, I had wanted to introduce certain elements extending beyond it and into the initial phases of language standardization - which is an essentially urban phenomenon. Specifically, I had attempted to lead the Ponapeans to achieve for their language, at least to an initial degree, the standard-language properties of flexible stability and intellectualization. Flexible stability here refers to the requirement that a standard language be stabilized by appropriate codification, and that the codification be flexible enough "to allow for modification in line with culture change." Intellectualization here refers to the requirement of increasing accuracy along an ascending sclae of functional dialects from conversational to scientific. To bring about stability, the spelling book and certain official word lists were to serve as the beginnings of a codification. To insure flexibility, the native committee was to be developed into the embryo of a codifying agency, charged with working out the word lists and revisions of the spelling book and lists as required. Unfortunately, the folk culture of Ponape, in spite of the existence of native schools, a money economy, and other urban elements, did not seem to have a vital need for the various functions of a standard language. Nor did my native friends, with all their good will and interest in the matter, exhibit to any great degree the attitudes which are characteristic of both nascent and established standard language communities. Perhaps this was because they did not yet constitute the nucleus of a native urban intelligentsia. Thus, in spite of my great empathy with Ponapean culture, the presence of a single interested outsider was not sufficient to push this folk speech community further towards the urban end of the scale, although the successful completion of the spelling book was proof enough that the language qua linguistic pattern was as susceptible to standardization as any other language.

THE STANDARD LANGUAGE PROBLEM

155

Let me now turn from the particular to the general and summarize briefly the conceptual framework which I have been using. In a recent paper,2 Madeleine Mathiot and I tentatively defined a standard language as "a codified form of a language, accepted by, and serving as a model to, a larger speech community." We proposed three sets of criteria in terms of which the degree of language standardization can be discussed: "1) the intrinsic properties of a standard language. 2) the functions of a standard language within the culture of a speech community, and 3) the attitudes of the speech community towards the stadrad language." As properties of a standard language, we posited flexible stability as described by Vilem Mathesius,3 and intellectualization as discussed by Bohuslav Havranek.4 We posited four functions of a standard language: the unifying, separatist, and prestige functions which we consider symbolic functions, and the frame-of-reference function which we consider an objective function. By "unifying" we designate the function of a standard language to unite several dialect areas into a single standard language community, by "separatist" its function to set off a speech community as separate from its neighbors. The remaining two functions refer to the prestige resulting from the possession of a standard language, and to the function of the standard language to serve as a frame of reference for correctness and for the perception and evaluation of poetic speech. As typical standard-language attitudes we listed language loyalty (as discussed by Uriel Weinreich),5 pride, and awareness of the norm. 2

"The Urbanization of the Guarani Language - A Problem in Language and Culture," Proceedings of the 5th International Congress of Anthropological and Ethnological Sciences, Philadelphia, 1956, pp. 783-90 (Philadelphia, 1960). 3 "O pozadavku stability ve spisovnem jazyce (The requirement of stability for a standard language)," in: Spisovnd cestina a jazy/cova kultura, Boh. Havrdnek and Milos Weingart, eds., pp. 14-31 (Prague, 1932). 4 "Okoly spisovneho jazyka a jeho kultura (The purposes of a standard language and its cultivation)," ibid., pp. 32-84. 6 Languages in Contact (Publications of the Lingusitic Circle of New York, No. 1), pp. 99-102 (New York, 1953).

156

THE STANDARD LANGUAGE PROBLEM

We differentiate between the former two by thinking of language loyalty as a protective and defensive attitude, and of pride as a positive attitude. These attitudes are assumed to be linked to the four functions: language loyalty to the unifying and separatist functions - and with them to the broader attitude of nationalism, pride to the prestige function, and awareness of the norm to the frame-of-reference function. In interpreting my Ponapean example, I have used these concepts to arrive at an intuitive explanation of an informally observed event, an explanation which has the status of an articulate opinion and can lay claim to a certain, perhaps high, degree of plausibility. Let us accept that the requirements of a good methodology are, (1) to discover which elements are relevant to the problem, and (2) to subject these elements to rigorous treatment. By yielding a plausible interpretation, my conceptual framework has met the requirement of relevance. To achieve rigor as well as relevance, I must show how these concepts can be used as a starting point for a controlled procedure which will yield products having the status of analytic results and meeting the more rigorous requirement of verifiability. To develop such a procedure, I shall assume that my three sets of criteria reflect the nature of my subject matter, and that by studying them in detail I will arrive at an orderly description of a language situation in terms of the degree of standardization. I can delimit my universe of data by differentiating between, on the one hand, the two sets of criteria which can be investigated directly - namely the properties of the standard language and the attitudes towards the standard language, and on the other hand the third set of criteria, the functions of the standard language, which can only be inferred from the observed attitudes and other cultural observations. I can select the specific techniques from which to make up my operational steps by nothing two characteristics of standard-language properties and attitudes: (1) They are continuous, sliding-scale features rather than discrete,

THE STANDARD LANGUAGE PROBLEM

157

yes-no features. That is, a given language situation can be described as meeting the criteria for a standard language to a given degree, rather than absolutely. (2) They are internally structured, that is, they are composed of multiple interrelated factors. The major scales corresponding to standard-language properties and attitudes can then be studied by formulating a series of detailed partial scales for each in terms of the various component factors. These partial scales will be comparable to the sclaes of Redfield's folk-urban continuum; they will have to be sufficiently narrowly defined to be operationally viable. The factors entering into the scales of flexible stability and intellectualization are focused on the linguistic variables of vocablulary and style; the factors entering into the attitudinal scales are focused on behavioral and situational variables. Of these variables, vocabulary is clearly definable in terms of lexical units and easily amenable to statistical techniques. The remaining variables are not a priori definable for language as such or for culture as such, but must be stated specifically in terms of the particular linguistic and cultural pattern. Thus, the only concrete techniques that can be listed at this point are the relatively trivial ones of specially slanted vocabulary counts. Synonym counts can serve as a measure of the stability factor in flexible stability, on the assumption that the number of synonyms for a given technical term is an inverse index of stability. Counts of specialized vocabulary, both that available in the formal code and that in actual use, as well as the ratio of both, can be worked into a measure of intellectualization. Techniques dealing with the style will have to be based on the results of morphemic analysis (primarily syntactic for those languages to which the morphology-syntax division applies); those dealing with behavioral and situational variables will have to be based on cultural and social psychological analysis. To state such techniques in broad outline would be to belabor the obvious; to state them in detail would border on the fictitious. Once the partial scales have been formulated and applied to or-

158

THE STANDARD LANGUAGE PROBLEM

ganize the data, they can be considered units of a dynamic structure and structuralist method can be applied to relate them to each other, weight them, and relate them to the whole - that is, the particular major scale of which they form part. In doing this, the two basic techniques of structuralist analysis - the technique of "frame and substitution" and the analysis of sequential co-occurrence - will have to be formulated in terms of the relations assumed to constitute the internal structure of each major scale, and will have to be appropriately modified to accomodate the type of units dealt with. Additional structualist techniques may evolve from the results of this analysis, and of the preceding investigation of the partial scales. Thus, contrary to the common assumption that structuralism is limited to the analysis of static systems only, it becomes applicable to the study of a dynamic problem.

JANUA LINGUARUM STUDIA MEMORIAE NICOLAI VAN WIJK DEDICATA

Edited by Cornells H. van Schooneveld SERIES MINOR I . ROMAN JAKOBSON a n d MORRIS HALLE: F u n d a m e n t a l s o f L a n g u a g e .

1956. 97 pp. Gld. 6.— 3. EMIL PETROVICI: Kann das Phonemsystem einer Sprache durch fremden Einfluss umgestaltet werden? Zum slavischen Einfluss auf das rumänische Lautsystem. 1957. 44 pp. Gld. 4.— 4. NOAM CHOMSKY: Syntactic Structures. Third printing 1963. 118 pp. Gld. 8.— 5. N. VAN WIJK: Die baltischen und slavischen Akzent- und Intonationssysteme. Ein Beitrag zur Erforschung der baltisch-slavischen Verwandtschaftsverhältnisse. 2nd ed. 1958. 160 pp. Gld. 15.— 6. BERNARD GEIGER, TIBOR HALASI-KUN, AERT H. KUIPERS AND KARL

H. MENGES: Peoples and Languages of the Caucasus. A Synopsis. 1959. 77 pp., map. Gld. 8 — 7. ERNST PULGRAM: Introduction to the Spectrography of Speech. 1959. 174 pp., 31 figs., 2 tables. Gld. 12.— 8. AERT H. KUIPERS: Phoneme and Morpheme in Kabardian (Eastern Adyghe). 1960. 124 pp. Gld. 16 — 10. URIEL a n d BEATRICE WEINREICH: Y i d d i s h L a n g u a g e a n d

Folklore.

A Selective Bibliography for Research. 1959. 66 pp. Gld. 6.— I I . E . and K. DELAVENAY: Bibliography of Mechanical Translation. — Bibliographie de la traduction automatique. 1960. 69 pp. Gld. 10.— 12. CARL L. EBELING: Linguistic Units. Second printing 1962. 143 pp. Gld. 12.— 13. SHELOMO MORAG: The Vocalization Systems of Arabic, Hebrew, and Aramaic. Their Phonetic and Phonemic Principles. 1962. 85 pp., two folding tables. Gld. 15.— 14. DWIGHT L. BOUNGER: Generality, Gradience, and the AU-or-None. 1961.46 pp. Gld. 5.50 15. ALPHONSE JUILLAND: Outline of a General Theory of Structural Relations. 1958. 58 pp. Gld. 7.50 16. Sens et usages du terme Structure, dans les sciences humaines et sociales. Édité par ROGER BASTIDE. 1962. 165 pp. Gld. 16.—

17. w. SIDNEY ALLEN: Sandhi. The Theoretical, Phonetic, and Historical Bases of Word-Junction in Sanskrit. 1962. 114 pp. Gld. 16.— 19. WALERIAN i w i E C Z K O w s K i : Word Order Patterning in Middle English. A quantitative study based on Piers Plowman and Middle English Sermons. 1962. 114 pp. Gld. 16.— 20. FINNGEIR HIORTH: Zur formalen Charakterisierung des Satzes. 1962. 152 pp. Gld. 15.— 22. E. F. HADEN, M. s. HAN and Y. w. HAN: A Resonance Theory for Linguistics. 1962. 51 pp. Gld. 7.— 23. SAMUEL R. LEVIN: Linguistic Structures in Poetry. 1962. 64 pp. Gld. 8.— 24. ALPHONSE JUILLAND and JAMES MACRIS: The English Verb System. 1962. 81 pp. Gld. 8.— 25. IVAN FONÄGY: Die Metaphern in der Phonetik. Ein Beitrag zur Entwicklungsgeschichte des wissenschaftlichen Denkens. 1963. 132 pp., 5 figs. Gld. 14.— 26. H. MOL: Fundamentals of Phonetics. I: The Organ of Hearing. 1963. 70 pp., 28 figs. Gld. 9.— 27. LASZLÖ ANTAL: Questions of Meaning. 1963. 95 pp. Gld. 10.— 28. ROBERT M. w. DIXON: Linguistic Science and Logic. 1963. 108 pp. Gld. 10.— 29. PUNYA SLOKA RAY: Language Standardization. Studies in Prescriptive Linguistics. 1963. 159 pp. Gld. 16.—

M O U T O N & CO • P U B L I S H E R S • THE H A G U E