HARRIS THE REVOLUTIONARY: PHONEMIC THEORY BRUCE E. NEVIN Bolt Beranek & Newman/University of Pennsylvania Zellig Harris's book Methods in Structural Linguistics (Harris 1951 [1947]) is commonly thought of as a compendious handbook collecting and systematizing established methods of structural linguistics. It is not generally recognized to how great an extent he was saying something new.1 We will begin with Leonard Bloomfield. With other linguists, Bloomfield confronted the question: When we approach an unknown language, how do we know which utterances are the same and which are different? Bloomfield said that we can only tell by knowing their meanings. Therefore, because there is no antecedent science of meaning, contrast and repetition could be brought into linguistics only by a ªspecial assumption|. This presented a problem: as far as possible, science should avoid unsupported special assumptions.2 See for example Bloomfield (1933:75, 78, emphasis in original): The study of language can be conducted without special assumptions only so long as we pay no attention to the meaning of what is spoken. ____________________ 1 This is a slightly revised version of a lecture presented on August 14th 1993 at the Sixth International Conference on the History of the Language Sciences (ICHoLS VI), held at Georgetown University August 9-14 1993. I thank Dr. Kurt Jankowsky, Konrad Koerner, and Micheal Gottfried for their comments and assistance. To fit time constraints, the lecture was a severely pruned extract from a longer paper which was distributed at the conference and which will be published elsewhere. 2 Thanks are due to Victor Yngve for clarifying this issue. See Communications of the Workshop for a Scientific Linguistics numbers 8-10. -1- -1- Phonemic Theory The meanings of speech-forms could be scientifically defined only if all branches of science, including, especially, psychology and physiology, were close to perfection. Until that time, phonology and, with it, all the semantic phase of language study, rests upon an assumption, the fundamental assumption of linguistics: we must assume that in every speech-community some utterances are alike in form and meaning. One way to resolve the problem was to try to define contrast without reference to meaning, by analyzing the distributional relations among observed phonetic features in utterances. Bernard Bloch exemplified this approach: ªthe facts of pronunciation [are] the only data relevant to phonemic analysis| (Bloch 1941:95), and ªContrast between sounds can be defined, I think, on the basis of distribution alone, without the customary appeal to meaning. (1953:59[224])| Zellig Harris proposed a second way to resolve the problem. If you want to know which utterances are the same and which are different, ask+a native speaker can reliably tell you.3 On this basis, Harris devised the pair test and various substitution tests. The pair test distinguishes contrast from repetition. The substitution tests determine a ªlinguistically relevant| segmentation of utterances+linguistically relevant because the segments represent the contrasts and locate them relative to one another within an utterance. The fundamental elements are phonemic distinctions (Harris 1951:34-35): [T]he representation of speech as a sequence or arrangement of unit elements is intimately connected with the setting up of phonemic distinctions between each pair of non-equivalent utterances. If each utterance were considered by itself, it might be represented as a continuum or as a simultaneity of features which change with time. [+] ____________________ 3 Obviously, the way in which one asks demands care and tact. Indications are given in Harris (1951) and in Harris & Voegelin (1953). -2- Harris the Revolutionary However, if we match utterances [as described in chapter 4, Phonemic Distinctions], we obtain some individual difference between the members of each particular pair of utterances; that is, we obtain discrete elements each of which represents some particular inter-utterance difference [+ and] which can be combined together. These elements are phonemic distinctions, rather than phonemes; i.e. they are the difference between /k/ and /p/ (more exactly, between tack and tap, between sack and sap, etc.) rather than being /k/ and /p/ themselves. However, for convenience, we will set up as our elements not the distinctions, but classes of segments so defined that the classes differ from each other by all the phonemic distinctions and by these only. [+] In this way we define /k/ to represent all the paired distinctions in which [k] was a member, /l/ to represent all the distinctions in which [l] was a member, and so on. The classes, or phonemes, are thus a derived (but one-one) representation for the phonemic distinctions. The segmentation of Chapter 3 was carried out in order to permit the representation of continuously varying speech to express the discrete elemental phonemic differences. A phonemically written form therefore is not a direct record of some spoken form, but rather a record of its difference from all other spoken forms of the language. This has revolutionary consequences. It means that even the initial segmentation is a representation of the phonemic contrasts of the language. The further work of phonological analysis is not for the sake of defining contrast (which is already given), but only for the sake of transforming a less efficient representation into a more efficient one. ªMore efficient| or ªconvenient| means that the grammar may be more simply stated. Crucially, the initial segments are not ªphones| determined by universal phonetic considerations, although their definitions refer to phonetic characteristics. The linguistic ELEMENTS are defined for each language by associating them with particular features of speech+or rather, differences between portions or features of speech+to which the linguist can but refer. They are marked by symbols, whether letters of the alphabet or others, and may represent simultaneous or successive features of speech, although they may in either case be written successively. The elements will be said to represent, indicate, or -3- Phonemic Theory identify, rather than describe, the features in question. For each language, an explicit list of elements is defined. The statement that a particular element OCCURS, say in some position, will be taken to mean that there has occurred an utterance, some feature of some part of which is represented linguistically by this element. (1951:14) The well-known attack on structural phonemics by Chomsky and Halle does not apply to Harrisian phonemics, and misrepresents Harris's views. Chomsky said that socalled ªtaxonomic phonemics| was defined by meeting four conditions: linearity, invariance, biuniqueness and local determinacy. Linearity says two things. It says that each phoneme corresponds to one or more phones, and that the linear order of phonemes in any given utterance is the same as the linear order of corresponding phones. If phones a, b, and c correspond to phonemes x, y, and z respectively, then linear order abc can only correspond to linear order xyz,never yxz and so on. Chomsky (1964:78[407]) defines the linearity condition as follows: The linearity condition + requires that each occurrence of a phoneme in the phonemic representation of an utterance be associated with a particular succession of (one or more) consecutive phones in its representing matrix, as its ªmember| or ªrealization|; and, furthermore, that if A precedes B in the phonemic representation, then the phone sequence associated with A precedes (is to the left of) that associated with B in the phonetic matrix. The linearity condition is more simply put in Chomsky's review of Jakobson & Halle, Fundamentals of Language (1957:346-347): + speech is taken to be literally constituted of a sequence of phonemes, each with its distinctive and redundant features; accordingly, the phonetic value of a sequence of phonemes is the sequence of phonetic values of these phonemes. Harris violates the linearity condition whenever doing so is ªconvenient| for obtaining a simpler or more perspicacious -4- Harris the Revolutionary grammar. Such violations are found, for example, in his treatments of simultaneous components (starting with suprasegmentals), and in his analysis of partial overlapping. Invariance is simply a restatement of Bloch's requirement that all members of a phoneme have some characteristic phonetic feature or features in common: ªThe class of all segments + containing a given feature is a phoneme. [+] The feature common to all members of a given phoneme is the characteristic of the phoneme| (Bloch 1948, definitions 53.2, 53.12, see also postulates 11-16). Chomsky defines the invariance condition as follows (1964:79[408]): The invariance condition + asserts that each phoneme P has associated with it a certain set f(P) of defining features (that is, P=Q if and only if f(P)=f(Q)) and that wherever P occurs in a phonemic representation, there is an associated occurrence of f(P) in the corresponding phonetic representation. The invariance condition has no clear meaning unless the linearity condition is also met; I will assume, then, that it is inapplicable when linearity is violated. [+] Where linearity and invariance are both met by a taxonomic phonemic representation, the string of phones is segmented into successive segments, each of which contains, along with redundant (determined) features, the defining features f(P) of some phoneme (P), and the phonemic representation is just the sequence of these phonemes. For Harris, invariance is often ªconvenient| for simplifying the grammar, but not necessary. It may be violated if by doing so one may simplify the grammar (1951:64): It is convenient to have the definitions of the various segments within a phoneme simply related to each other. We may try to group segments into phonemes in such a way that all the segments of each phoneme represent sounds having some feature in common which is not represented by any segment of any other phoneme. He says that if a phonetic feature is shared by all the segments that are grouped as members of a phoneme, then the grammar may refer to ªthe phoneme as repesenting this common feature, rather -5- Phonemic Theory than as being a class of segments. Relations between phonemes would then represent relations between sound features| (1951:65). He discusses the obvious correspondence to distinctive feature theory. Chomsky's definition of biuniqueness (1964:80[408]) is that it ªasserts that each sequence of phones is represented by a unique sequence of phonemes, and that each sequence of phonemes represents a unique sequence of phones.| Although Harris introduced the term in his 1944 paper on phonemic long components,4 he replaced it by the phrase ªone-one correspondence| in subsequent writings. For Harris, the correspondence is not between phones and phonemes, but between ____________________ 4 Harris had introduced the term ªbi-unique relation| in his paper on phonemic long components (1944a:187-188, footnote suppressed): Finally, if we are ready to admit partial overlapping among phonemes, we may agree to have different components in different environments represent the same phonetic value. So long as we do not have a component in one environment represent two phonetic values which are not freely interchangeable, or two components or component-combinations in the same environment represent the same phonetic value, we are preserving the bi-unique one-to-one correspondence of phonemic writing. (The term bi-unique implies that the one- to-one correspondence is valid whether we start from the sounds or from the symbols: for each sound one symbol, for each symbol one sound.) Hymes & Fought refer erroneously to Harris (1944b), and Anderson (1985) evidently assimilates the error from them, although the page range for (1944b) is 196-211. I am not aware of any other place that the word occurs in Harris's publications. One can only presume that he used it in classes that Chomsky attended. In Harris (1951 [1947]), the earlier term was abandoned in favor of the phrase ªone-one relation|. -6- Harris the Revolutionary the phonemic contrasts and their representations. Harris identifies his initial elements not by consulting phonetic theory, but by consulting the linguistic intuitions of native speakers, although phonetic properties are naturally associated with the elements because this is, after all, a segmenttion of utterances. There is intrinsically a one-one correspondence between these initial elements and the contrasts that they represent. Each distinct utterance is represented by a different sequence of segments, and each different sequence of segments represents a distinct utterance. A representation for the phonological contrasts in a language must satisfy the condition of biuniqueness, in Harris's sense, or else it is not what we just said it was, a representation of the contrasts. The ªbiuniqueness condition| is not a formal condition holding between representations at two different levels of description, a phonetic level and a phonemic level. Rather, it is a methodological requirement that one not lose or obscure the fundamental observational data for a science of language, namely, which utterances are repetitions and which are not. All the operations of linguistic analysis, aimed at transforming the initial representation into a more efficient or useful one, must preserve the biunique relation between representations and contrasts. For this reason, Harris occasionally speaks of the one-one relation of phonemes to the segments of an earlier stage of analysis. In this, it may seem -7- Phonemic Theory that he is talking of a biunique relation of phonemes to phones, but he is not. Local determinacy is a qualification of the biuniqueness condition. It is essentially the prohibition against ªmixing levels| that prevailed widely in American linguistics. Chomsky denies this obvious relation, partly to vitiate counterattack, no doubt, and partly for reasons relating to a larger argument that data about language acquisition and data about linguistic analysis are incommensurate. These considerations are beyond the scope of our present discussion. Chomsky (1964:81[409]) defines local determinacy as [a one-one] correspondence such that the unique phonemic representation corresponding to a given phonetic form can be determined by ªpurely phonetic| considerations, or perhaps, considerations involving only ªneighboring sounds.| The only indication of local determinacy in Harris's work is as a practical matter: one should extend the environments that one tests no farther than is sufficient. As to mixing levels, Harris had no qualms about the scientific status of invoking morphemes and junctures and the like, because all his results refer back to the data of contrast as a touchstone of validity and scientific rigor. He was free to ªbootstrap| his analysis: a first approximation is later refined by criteria that are not initially available, tentative guesses as to the results of later stages of analysis are used as guides at an earlier stage, subject to correction when the later work is carried out more fully, and earlier results are always subject to re-evaluation in the light of results at a later stage, all of this with an eye to the overall simplicity of the grammar. This -8- Harris the Revolutionary is what Harris means when he says that operations must be ªcarried out for all the elements simultaneously| without any ªarbitrary point of departure| (1951:7). These four conditions are obligatory if one has only ªthe data of pronunciation| to work with. Without linearity, invariance, biuniqueness (in the phone-phoneme sense) and local determinacy, one is at a loss to say what might be in contrast with what. They are optional for Harris because his fundamental data are native speakers' intuitions of contrast, identified by the pair test and located in elements of a representation by the substitution tests. The question is not what contrasts with what, but rather what means are available for representing the contrasts that are present. In Harris's (1991) view, the same question faces the child learning its first language. Chomsky attacks complementary distribution, or, rather, he attacks the proposition that one can define phonemes by analyzing the distributional relations among phones. He describes complementary distribution as: basically, the principle of biuniqueness converted into a procedure. Regarded as an analytic procedure, its goal is to provide the minimally redundant representation meeting the conditions of biuniqueness and local determinacy. Talk of a procedure as ªproviding| or ªleading to| the correct representation suggests that it is a practical discovery procedure, such as might be implemented in a computer algorithm. Harris explicitly states that this is not what he is after (on the discovery procedure canard, see Ryckman 1986, Nevin 1993). More importantly, distributional procedures do not and cannot produce biuniqueness, they preserve it, exactly as operations in -9- Phonemic Theory mathematical logic preserve truth value. This is essential as one wrestles the representations into a form that supports the simplest grammar. Claiming to summarize Harris (1951: Chap. 7), Chomsky describes complementary distribution as follows (414.3, italics in original): Given a set of representations in terms of phones, let us define the distribution D(x) of the phone x as the set of (short-range) phonetic contexts in which x occurs. The relation of complementary distribution holds between phones x and y if D(x) and D(y) have no element in common. A tentative phoneme is a class of phones related pair-wise by the relation of complementary distribution. A tentative phonemic system is a family of tentative phonemes meeting a condition of exhaustiveness. We find the phonemic system (or systems) by applying additional criteria of symmetry. The most important of the distortions here are these: 1. The two items being compared, x and y, are not of equal status. 2. The comparison of D(x) and D(y) is not successive, but recursive. Chomsky describes the comparison of distributions of two phones x and y. Harris describes the comparison of distributions of a segment x and a ªtentative phoneme| in the making, y. Chomsky describes a linear sequence. A sort of distributional analysis component outputs a set of tentative phonemic systems as alternative candidates, which are subjected first to a test for exhaustiveness, then to ªadditional criteria of symmetry|. The winner output from the test component is ªthe phonemic system (or systems)|. Chomsky objects (1964:92[414-415]) that in some cases ªthe class of `tentative phonemic systems' + will not include the -10- Harris the Revolutionary optimal biunique system as a member, so that no supplementary criteria will suffice to select it from this class.| Harris describes a recursive process. The criteria are not applied to a set of alternative phonemic systems produced by distributional analysis, they are applied at every step of merging a segment with restricted distribution into a partially defined phoneme. The comparison of x and y is one step of this recursive process, where x is a segment and y is a phoneme-in-the-making. If the environments of segment x are complementary to those of phoneme y, and if the criteria indicate that an x-y grouping is superior to other concurrent possible combinations, then x is merged into y. That is, the environments of y now include those in which segment x occurred. It is to the less restricted tentative phoneme y that subsequent comparisons are made. The result in the end is of course that all the member segments of any given phoneme are complementary to each other, since complementarity is transitive over this succession of comparisons. However, the critical issue concerns not the end result but the recursive process of attaining it. At each step of the process, the environments of a tentative phoneme y are augmented by those of a new segment x. Then, before proceeding further, all the environments that formerly contained the now- included segment x must be restated in terms of the redefined phoneme. Chomsky reframes this rather obvious housekeeping step as an ad hoc procedure brought in just to save taxonomic phonemics from its flaws. -11- Phonemic Theory Some of these distortions are evident in Chomsky's discussion (1964:92[414]) of an example of phonemic overlapping due to Bloch. Alveolar flap [D] occurs intervocalically after stress in e.g. ªBetty|, and after q in e.g. ªberry|. Chomsky says: The requirement of biuniqueness is preserved if we set up the phonemes /t/, with the allophone [D] in intervocalic, post-stress position, and /r/, with the allophone [D] after dental spirants. Given a phone in a phonetic context, we can now uniquely assign it to a phoneme; and given a phoneme in a phonemic context, we can uniquely determine its phonetic realization (up to free variation). However, this solution, which is the only reasonable one [+] is inconsistent with the principle of complementary distribution. In fact, the allophones [D] and [r] of /r/ are not in complementary distribution since they both occur in the context [be-y] (`Betty', `berry'). Hence complementary distribution is not a necessary condition for biuniqueness. The sleight-of-hand here is in the phrase ªthe allophones [D] and [r] of /r/ + both occur in the context be-y]|. The [D] that occurs medially after stress in `Betty' is of course not an allophone of /r/. Chomsky's argument here goes through only if one insists that all occurrences of the phone [D] be assigned to one and only one phoneme, either /r/ or /t/. Without notice, Chomsky is assuming the strong form of the invariance condition prohibiting even partial overlapping. Underlying the invariance condition, especially in this form, is the familiar presumption that biuniqueness is a relation between phones (e.g. [D]) and phonemes (e.g. either /r/ or /t/). But Harris showed that the relation of biuniqueness instead properly holds between distinctions (e.g. `Betty' vs. `berry' vs. `throw') and various means of representing them, as follows: Word Contrasts: "Betty" "berry" "throw" -12- Harris the Revolutionary Representation #1: [beDiy] [beriy] [qDow] Representation #2: [betiy] [beriy] [qrow] The segments in [beDiy], [beriy], [qDow] constitute a less efficient representation of the contrasts between words in English (including these three words) than do the segments in [betiy], [beriy], [qrow], but both representations preserve a biunique relation to the contrasts. Chomsky is either misrepresenting Harris, or he has not understood him. Chomsky continues (414-415): Furthermore, the class of ªtentative phonemic systems| as defined in the preceding paragraph will not include the optimal biunique system as a member, so that no supplementary criteria will suffice to select it from this class. The ªsupplemental procedures| by which problems of overlapping and neutralization such as these are resolved do so not by ªselecting| one of the tentative solutions given with Harris's criteria of chapter 7, but by generating new solutions not otherwise available. These ªsupplemental procedures| are described in Harris (1951) chapter 8, Junctures, and chapter 9, Rephonemicization. In addition, Chomsky ignores Harris's description of dividing the distributional range of a segment. In this example, we divide the [D] segment into two distributionally defined (but phonetically identical) elements, and assign the one that occurs after [q] to /r/ and the one that occurs intervocalically after stress to /t/. This procedure is a further extension of distributional analysis, yielding a new system that had previously not been considered (1951:91-92): -13- Phonemic Theory The phonemic representation of a language may be simplified by means of this operation [ªdividing the segment|] when the segment A cannot be put into any phoneme without disturbing the over-all symmetry, and when it is possible to partition A into such segments A1 and A2 as would fit well into the phonemes of the language. Assignment of both A1 and A2 to some other phonemes should yield a more symmetrical or otherwise convenient phonemic stock than assigning the original A to some phoneme. (The last sentence is a criterion, not a prediction.) The thrust of Chomsky's argument against complementary distribution is that it is insufficient to determine contrast. As we have seen, that is an inappropriate goal if contrast is a primitive observation of the science. On that ground, complementary distribution is not necessary either, and indeed no criteria are necessary other than those of the pair test. Harris had said, furthermore, that complementary distribution was not sufficient for deciding how to group segments into phonemes, for example (1951:62-63): The operation of 7.2-3 [that is, analysis of complementary distribution] determines whether segment X can be associated with segment Y in a single phoneme. But it is not sufficiently selective to determine which of two complementaries, X and Z, shall be included with Y (if X and Z are not mutually complementary, so that only one of them, but not both, can be associated with Y). [+] It is therefore necessary to agree on certain criteria which will determine which of the eligible segments go together into a phoneme. I do not have space here to show how Chomsky's other examples fail. I will take a final few minutes to discuss Halle's argument ªthat it is generally impossible to provide a level of representation meeting the biuniqueness condition without destroying the generality of rules, when the sound system has an assymetry| (Chomsky's summary). The key is to realize that it is the distinctive features that are the representations of -14- Harris the Revolutionary contrasts. Harris had observed that his simultaneous components in general (1944a:205, 1951:133.3) and distinctive features in particular (unit-length components defined for the whole stock of phonemic contrasts of a language, 1951:147.2) may supplant the segmental phonemes. Chomsky proposes to eliminate the phonemic ªlevel| of representation: (i) physical phonetics (ii) systematic phonetics (distinctive features) (iii) taxonomic phonemics (segments, eliminated) (iv) systematic phonemics (morphophonemic) In fact, the representation of phonemic contrasts is simply shifted from segmental phonemes to distinctive features. What Chomsky calls ªsystematic phonetics|, using distinctive features, is no less phonemic than the segmental representation that it supplants. The effect of the actual changes in representation of sound systems is as follows: (i) phonetics (ii) contrasts: segmental representation + feature representation (iii) base representation (morphophonemic) If you state your rules in terms of distinctive features, then your phonemic representation is distributed throughout the successive stages of a derivation + in fact, everywhere that the distinctive features are used. The question ªwhere are the phonemes| is sensible only if one sees the task of phonology as defining contrast; when contrast is a primitive element given by the pair test, this is a non-issue. At first blush, the aim of Generative phonology to describe languages in terms of linguistic universals (specifically, here, a universal ªalphabet| of phonetic features) might seem to be -15- Phonemic Theory incommensurate with Harris's aim of defining elements for each language relative to one another. The resolution of the seeming conflict lies in the difference between phonetic features and distinctive (that is, phonemic) features. The phonetic features are parameters that are available for phonemic contrast. Speakers and understanders of a given language associate particular values on these parameters with the contrasts that they perceive between utterances of their language; or (arguendo) they may represent the phonemic contrasts by means of polarized differences of values along these universally specifiable phonetic parameters. The particular differentiated values that serve as distinctive features on a given universal phonetic parameter are indeed defined relative to one another for the particular language. Conversely, Harris expressed no opposition to the notion of phonological universals formulated in this way, if that was what one was interested in. It is remains an open question today, what the universal parameters for contrast are. The particular representation one might prefer for the contrasts is secondary to the contrasts themselves, as primitive elements for a science of language. Neither the features nor the segments have any privileged ontological status. It is the contrasts that are ªreal|. This broaches the peculiar ontological status of language. In the familiar Berkeleyan example, if a tree falls in a forest and no one is there to perceive it, we nonetheless assume the reality of the tree and of the event, and, insofar as sound is defined in physical terms as pressure waves in the atmosphere, there is a -16- Harris the Revolutionary sound. However, if a tape recording of the Gettysburgh Address is played in the forest and no one is there, there is only sound. There can be no language present+no words, no contrasts or distinctive features+in the absence of a hearer who controls perceptions of the words, contrasts, etc. of English. The contrasts may be represented by phonemic segments, phonemic components, distinctive features (unit-length phonemic components), or in some other way, but they cannot be defined absolutely, in purely physical terms. They can only be defined relatively, in terms of the contrasts between utterances that they represent. And the contrasts are not physically given, they are socially given. This is why the pair test is necessary for determining the phonemic distinctions or contrasts in a language. These considerations go beyond what is possible to consider in this context. The purpose here has been to to set the record straight and to explicate Harris's insights into the nature of linguistic contrast. The task of reinstating these insights appropriately in the theory and practice of linguistics today, and consideration of the consequences of doing so, must await another occasion. -17- Phonemic Theory References __________. 1985. Phonology in the Twentieth Century: Theories of rules and theories of representations. Chicago: Univ. of Chicago Press. Bloch, Bernard. 1941. Phonemic overlapping. American Speech 16.278-284. Reprinted in Joos (1957:93-96). __________. 1948. A set of postulates for phonemic analysis. Language 22.200-248. __________. 1953. Contrast. Language 29.59-61. Bloomfield, Leonard. 1933. Language. New York: Holt, Rinehart & Winston. Chomsky, Noam A. 1957. Review of Fundamentals of Language, by R. Jakobson and M. Halle. IJAL 23.234-242. Reprinted in Makkai (1972:) __________. 1964. Current Issues in Linguistic Theory. The Hague: Mouton. Chapter 4 reprinted in Makkai (1972:401- 423). Harris, Zellig S. 1944a. Simultaneous components in phonology. Language 20.181-205. Reprinted in Joos (1957:124-138). __________. 1944b. Yokuts structure and Newman's grammar. IJAL 10.4:196-211. -18- Harris the Revolutionary __________1951 [1947]. Methods in Structural Linguistics. Chicago: Univ. of Chicago Press. Reissued as Structural Linguistics. __________. 1991. A Theory of Language and Information: A mathematical approach. Oxford: Clarendon Press. __________ & Voegelin, Carl F. 1953. Eliciting in Linguistics. Nevin, Bruce E. 1993. A minimalist program for linguistics: a perspective on the work of Zellig Harris. Historiographia Linguistica XX:2/3 (1993). Ryckman, Thomas Alan. 1986. Grammar and Information: An investigation in linguistic metatheory. PhD. dissertation, Columbia Univ. -19-