Bruce E. Nevin
In the early hours of Friday morning, May 22 1992, after a pleasant working day as usual, Zellig Harris died in his sleep. He was 82 years old. He was born in 1909 in Balta, Ukraine. His family came to the United states when he was four. I am told that he chose the name Zellig Sabbettai at that time. I like to think that the semantics of happiness and steadfastness were on his mind. Certainly they were keynotes of his life.
When he died, he was just finishing a book on politics that he had been planning for most of his life. With the 1992 publication of his book A Theory of Language and Information (Oxford), he had wrapped up the central part of his life's work on language. He seems to have felt at liberty to take up this other unfinished business. I understand from Paul Mattick, Jr., who was Harris's friend and neighbor for many years in New York, that this last book is about how to get from capital ism to socialism. This is surely not a conventional take on either capitalism or socialism, not so much because Harris was an anarchist, but because he was always an originative thinker and because he worked deeply what ever ground he chose to till. With this book in mind, he reviewed the literature because (characteristically) he would write it only if what he wanted to do in it had not already been done by others.
Harris described himself as a methodologist rather than a linguist: While absolutely accurate, this could be misleading, and this in part bespeaks prevalent misapprehensions about the place and importance of method in science. He said that linguists would not be interested in his work, though people interested in language would be, and that his work was not part of linguistics as it is institutionally defined.
Yet surely he was a linguist by most of the operational definitions one might come up with. He had done extensive fieldwork on a variety of languages, Cherokee and Swahili for example. When he was doing the final revision of the 1992 Oxford book, he undertook to test his theory against every language of which he had some control, reportedly 44 languages, and consulted with colleagues who were native or fluent speakers of various languages. He checked a wide range of language families and language types. He spent months reading grammars from morning to night, and evaluating whether his theory had a reasonable account for what he found there. He was clear that no scientific conclusions were warranted from this exercise, and so he gives but slight notice of it in the book, but he wanted to feel reasonably secure that his conclusions were not idiosyncratic to English, French, German, Korean, and the few other languages that had been the primary bases for their development. I am told that he was pleased with the results.
His contributions to linguistics were numerous and weighty. He founded the first linguistics department in the United States, that at the University of Pennsylvania. He introduced the algebraic representations and abstract mathematical treatment which have become so much norms of the field that it is difficult now to appreciate how much he did so over the kicking and screaming protests of his peers. He invented X-bar notation, though of course not by that name, to cope with the now well-known ineptness of immediate constituent analysis with the head-of relation. He showed the need for discontinuous morphemes in grammatical analysis and developed ways to accommodate them, including morphemic long components. He charted a way out of difficulties experienced by Bloch and others in phonology, by making it clear that contrast is the fundamental "difference that makes a difference" (in G. Bateson's phrase) for setting up phonemes, rather than phonetic identity, a ghost that has risen to haunt generative phonology more than once. He developed the notion of phonemic long components, partly as a way of assimilating the advantages of distinctive feature analysis without the pitfalls attendant on their being defined in terms of phonetic content. He invented transformational analysis in context of developing discourse analysis to get at the information content of texts. He invented string analysis to complement (not rival) immediate constituent analysis. Indeed, he thought of all of these as complementary tools of analysis, not as rival characterizations of grammar in themselves.
The distinction between tools of analysis and grammar formalisms deserves emphasis for another reason that relates closely to Harris's concern for appropriate scientific method. It was only after many years of work with these tools that Harris presented a grammar of a language as a whole (1982), though the outlines were evident earlier (e.g. 1968, 1969). He did not make programmatic predictions to which grammars were expected to conform, and then test rival predictions with example and counterexample. He did not set up language-like mathematical systems and then try to see where language fit within them. Instead, he treated the objects and relations of language as mathematical objects and relations, seeking a mathematical structure that sufficed exactly for language.1
Many of these and other contributions still await recognition and exploitation in the field of linguistics as institutionalized today. Obvious examples include sublanguage analysis and sublanguage grammar, operator grammar based on word dependency, discourse analysis for information content, and his theory of information as an account of a central aspect of semantics. For example, string grammar and its natural extension into transformational grammar is the basis of the very successful work of Naomi Sager and others at NYU in information formatting of sublanguage texts, applied there mainly to medical informatics (Sager 1981, Grishman and Kittredge 1986). Stephen Johnson (1987) has implemented a system for representing the information content of texts, based on operator grammar. The complementarity of string relations (adjunction) and immediate constituency (phrase structure) with respect to the head-of problem is the basis and origin of Joshi's Tree-Adjoining Grammars (TAGS), and a similar complementarity of dependency and adjunction helps make possible the exceptional simplicity, elegance, and power of operator grammar (1982, 1990, 1992). Harris brought the long line of research in transformational structure back to its roots in discourse analysis, creating the field of sublanguage grammar (1990, Kittredge and Lehrberger 1982). Successes of this sort are still little noticed within linguistics.
It is characteristic of Harris that there was no vanity or self importance in him. He knew that his work was of lasting importance, and treated it as such, but he was no guru or empire builder seeking followers, and would not accept any such role being projected onto him. Harris's words about Bloomfield and Sapir in his review of the Bloomfield anthology (1973) apply well to himself:
A clue as to the basis of this choice against fame and influence may perhaps be found in his advice to a student starting out in his first teaching position, many years ago. "Don't invite anybody over for dinner," he said, "and don't accept any invitations. If you get involved in the social life of an academic, you won't be able to get any work done." The work came first.
I studied with Harris from 1966 through 1970. I was an undergraduate much of that time, but that did not matter to him. He had a sink-or-swim approach like that of Sapir (Darnell 1990). Often he would start out his seminar by asking if there were any questions. If there were none, he would just start talking about what he was working on. Sometimes he would bring in some of the innumerable envelopes in which he organized examples. The presentation, however extemporized, was always comprehensive and well organized. A colleague of a later generation than mine tells of a session devoted to a masterful demonstration of how the theory applied to a variety of Semitic languages. During my tenure, his seminars were much concerned with theoretical issues attendant on preparing his 1968 book and the emergence of operator grammar from the network of "elementary sentence-differences" in English, with relevant data as needed. The process was not a lecture or monologue, but a continuing conversation with his students, trying out alternatives, posing and working out problems for a mathematical characterization of language. Perforce, one became aware of his enormous intellectual capacity and brilliance, but as an appreciative beneficiary rather than, as is too often the case in academic institutions, as its foil or even victim. With intensive reading outside, one began to catch on and to participate.
I recall telling him at the end of one seminar meeting in my first year that I would try to disprove his theory. Of course, this troubled him not a bit. I worked up a problem in Modern Greek that I thought might be problematic for his approach. (I had lived in Greece for a couple of years, and spoke the language, but I worked with an informant for this project.) When my results turned out actually to corroborate the point I had intended to challenge, he merely thanked me for the data on Greek. A year or two later, I had come up with a proposal to analyze definitions in a dictionary to extract semantic primitives by something like componential analysis, much as Martha Evens and now others have done. Although the notion of semantic features, as elements of a metalanguage or "mentalese" prior to language, is inimical in concept and method to his work, he said (and this is an exact quote) "Others have tried this and have failed, but you are welcome to try." I offer this in refutation of the sometimes heard view that Harris was dictatorial. I ran into conflicts in such matters with others, never with Harris.
I have also heard it asked why he never retorted to attacks on his work. I think it did not matter to him. He did not expect his methods and results to be understood and taken up by everyone in the field of linguistics. Maybe his attitude differed in the 1940s, when he wrote the structural restatements and the manuscript eventually published as Methods in Structural Linguistics (1951a).2 Maybe his expectations of the field changed after some of Chomsky's followers began making him out to be one of the "bad guys" (the informal title of a course at MIT. I don't think so, based on his writings and on the testimony of some who were his students then. I never heard him comment on the commonplace attribution to Chomsky of the discovery of transformational grammar and the "transformational revolution. Hockett (1968) attributes to Harris "nothing, or a long silence, after 1957," showing ignorance not only of things like string analysis, for which he might perhaps be excused, but even ignoring the paper "Transformational Theory" prominently published in Language in 1965. I showed this passage to Harris, and he shrugged. It did not matter.
In particular, I never saw any evidence that Harris opposed or blocked Noam Chomsky's ambitions. In my experience it would have been entirely out of character for him. And after all it was Harris who proposed Chomsky to speak in his stead to the 1962 International Congress. A similar canard regarding Bernard Bloch has recently been laid to rest in an editorial in Language. One must, I think, be alert to the social psychology that leads people sometimes to rewrite history so that their avatar is depicted as an embattled hero. Now, an old Indian friend once told me that one cannot point a finger without having three other fingers of the same hand pointing back, so I hasten to add that this is not the picture I intend to paint here of Harris. He accomplished what he intended to quite well, thank you very much, and seems to have been quite happy in the process. The point is precisely that he seemed in no way embattled by attacks and uncomprehending misconstruals of his work.
And uncomprehending misconstruals abound. Frawley's review in Language of the grammar (Harris 1982) identifies operator grammar with predicate calculus, though Harris clearly delineates critical differences between language (a fortiori operator grammar) and language-like mathematical systems, including predicate calculus. Frawley can see in this comprehensive grammar only an attempt in 1980 to do 1960s generative syntax, because he is unable to step out of the Generativist paradigm sufficiently to understand Harris's work on its own terms. Another review (Wheeler 1984) asserted that Harris's grammar was unable to account for certain familiar semantic problems—middle voice, the semantics of find vs. seek, and quantifier scope in examples like "someone was opposed by everyone." In my review (1984) I showed how Harris did in fact account of each of these problems in the book. Kac, in his Language review of the first volume of selected writings (Harris 1970), asked "why bother?" and indeed, from within the Generativist paradigm that must be the only plausible question.3 It is only in setting aside paradigmatic blinkers that one can see, having these writings in one place, how consistent and self coherent Harris's program has been over the years. Transformational grammar was not a revolutionary break but part of a continuous evolution.
All of this requires more extensive discussion than is possible (or suitable) here. I will mention only one other misconception about Harris's work, not because it is in any way fundamental but because it is so commonplace. I probably will be greeted with disbelief when I say that discovery procedures were not his aim.4 (Jim McCawley's witticism about Harris and discovery procedures in the collection traditionally circulated in May really reverses the roles of the teller and the butt of the joke.) It is not hard to see how linguists have come to this mistaken belief. Discovery procedures are an abiding fixture for linguistics as institutionally defined. When Methods was published (195la) linguists sought an aid to fieldwork and the writing of linguistic descriptions. More recently, discovery procedures have been institutionalized as a whipping boy. This has colored perceptions of Harris's intentions and results.
For Harris, it was certainly of interest and value when redundancy among objects on one level of linguistic representation could be used in a practical way to determine boundaries of objects on the next, but this was a corroboration, not an aim. The "constructional procedure" described in the 1955 paper "From Phoneme to Morpheme" was implemented in FORTRAN in the early 1960s and proven to work, and Ralph Grishman has had some preliminary success in implementing programs to discover word classes and rules of sublanguage grammar from sublanguage texts. But in general Harris did not think that discovery procedures were feasible. In particular, he told me he thought that grammatical analysis could not be done solely with a corpus or by asking informants, one had to control the language oneself. And then one had to work over the data to tease out pattern and wrestle that pattern into coherent form, a lengthy and demanding process, as probably most of us know from experience. So much for the popularized image of feeding in a corpus, turning a crank, and having a grammar reel out the other end.
In the introduction to Methods in Structural Linguistics (1951a), Harris states clearly that these methods are not discovery procedures. He affirms that one uses many means to come up with proposals for describing what is going on in a language.-hunches, guesses, heuristic rules of thumb, typological generalizations, proposed universals, comparison with related languages or with earlier stages of the language, and so on, more art than science (or rather, more art than engineering). Harris was acutely aware of the danger of swamping one's control of the language by growing familiarity with marginal examples. Language is after all a social institution, continuously in change as it is constantly recreated in the crucible of use. The aim of the methods was not to substitute for these informal ways of coming up with possible analyses, but to verify, for any given result, whether the result had a valid relation to the data of the language. Of those who have actually read the book, how many have said (and some have in fact said to me) "he didn't really mean that." But if nothing else, Harris was always careful to say exactly what he meant.
This concern for verification arises out of a deeper concern which becomes more explicit in Harris's later work. This is a critical point for linguistics as a science. For any other science, there is a standpoint external to the science domain for its metascience. In particular, practitioners in physics, chemistry, even in mathematics, rely on the "background vernacular" of language to ensure communication about shared meanings and ultimately to validate the relation of conclusions, however reached, to the observations on which they are based. Not so for a science of language. Harris recognized that there is no vantage point outside of language from which to describe language (and, conversely, each language observably contains one of its own metalanguages). I'll repeat that, because it is I think a key to understanding what Harris was about, and because it is easy to overlook its importance.5There is no vantage point outside of language from which to describe language. By contrast, Generativist theory postulates a universal metalanguage, external to language, that is part of one's biological endowment. (I personally do not find this biologicist, neophrenologist doctrine of mental organs credible, but the issue rests not on opinion but on facts yet to be determined.) Harris's stance seems to me perfectly consonant with the argument made by Anderson (1981) (with a different end in view, to be sure) that one cannot derive linguistic structure from the findings of some study bearing a metascience relation to linguistics.
Harris was interested in how language can carry or transmit information, and this is the thread that underlies the really remarkable consistency in his work over more than 50 years. Intuitively, we know that differences in form must correlate somehow with differences in meaning, but the correlation seems messy and inconsistent in the observed data of language (say, in a body of writings or of phonemic transcriptions, including whatever utterances the investigator may come up with in the ad hoc search for examples). The untidiness of the correlation is commonly misattributed to one of the correlates, the stream of overt linguistic forms. What Harris found was that there is indeed a close correlation of linguistic form with a component or aspect of meaning, which he termed linguistic information. He found how the apparently messy, inconsistent stream of words can be the product of two concurrent systems: a system of word dependencies and a system of reductions. The word dependencies correlate with perceptions in a subject-matter domain (a science subfield is an especially coherent example). The reductions change word shapes (often to zero), motivated in part by issues of redundancy and efficiency and in part by historically contingent social convention. The reductions are analogous in a general way with the Generativist notion of transformations as rules of grammar. They introduce degeneracies such as ambiguity and paraphrase, and otherwise obscure the correlation of form with meaning, but without destroying that correlation. They apply (usually optionally) as an operator word enters on its arguments.
The important point here is that one of the products of Harrisian linguistic analysis is a semantic representation, in the form of the operator-argument dependencies found in an instance of language use. This semantic representation does not capture all aspects of meaning, only the linguistic information. Other aspects of meaning may be communicated by a variety of means, including but scarcely limited to the (quasi-gestural) ways in which we use language, and it is the correlation of language with these other aspects of meaning that is fluid and loose. By contrast, the transmission of linguistic information, and in important ways its very creation, is a capacity exclusive to language.
Given that structure (differences of form) correlates with meaning (linguistic information), it is of critical importance that the machinery of description not import any structure extraneous to that found in language. Harris's endeavor was always, then, to determine a "least grammar," a description that required an absolute minimum of primitive objects and relations. Any additional objects or relations in the description introduce extrinsic structure that obscures the informational structure in language. This could be the basis for a telling critique of various other theories of language. Harris chose not to make such a critique. When I asked him once about certain aspects of Generativist theory, he would only comment, with evidence of mild amusement (though whether at it or at me for asking was unclear), that it did seem to be overstructured. Like Sapir and Bloomfield, Harris had an interest in problems of international communication and in proposals for an international auxiliary language.6 And like Sapir and Bloomfield he had in particular a long-standing interest in international cooperation and communication in science. This is probably one reason much of his material for linguistic analysis was drawn from technical journals of various sciences. Another and more immediate reason is that technical sublanguages of science have more well-defined semantic and syntactic restrictions.
All of this and more is reflected in the extended and detailed analysis of the form of information in a science sublanguage presented in Harris et al. (1989). This book describes the grammar of a sublanguage of immunology during a specific period in the development of that field, based on discourse analysis of sublanguage texts from that period and adequate for making explicit the information structures in arbitrary other texts in that sublanguage. The analysis shows how the structure of the sublanguage changed concurrently with a change in immunologists' perceptions in the domain of their science. A difference in informational structure correlates with a difference in meaning. The informational structures that are clearly represented in the binary array resulting from discourse analysis are still present in the actual form of the source texts as written, albeit obscured under reductions in word shape that are partly motivated by considerations of informational efficiency and avoidance of redundancy, and partly dictated by conventions of language use as a human social institution.
Harris was first and foremost a scientist. He so arranged his life as to enhance the autonomy of his work economically, politically, and socially. He was Principal Investigator for a long series of grants from the NSF, NIH, and other agencies, many of them concerned with what might now be called informatics of science. Throughout his life he was involved with scientists and science. His wife was a physicist at the University of Jerusalem, and had been Albert Einstein's assistant at Princeton. A brother was an immunologist (he is an author of some of the work analyzed in the 1989 book). Harris's perspective on what constitutes a science and on what is appropriate for conduct of a science is thus not merely philosophical or political, but deeply grounded in the real world of scientists doing science. In defense of science he occasionally did inveigh against the rough and tumble of polemic attack and retort as something inappropriate for science, but beyond this he would not participate in it. That too would be a distraction from the work.
Harris's characterization of Sapir and Bloomfield in the 1973 review quoted previously applies equally to himself, as indeed in many ways his work carries forward a synthesis of theirs. To paraphrase a passage from the review, he was, to the good fortune of those who knew him, and I hope of himself, an extremely decent person of high integrity; he had utter and explicit contempt for the posturings and status in this society as well as for its vast injustice and inequality. He was a person not with ambition, least of all with ambition in the terms of this society, but with satisfaction in what he was producing. Those who remember Zellig Harris know this about him.
The publications for which he is best known mostly give the reader only a narrow vantage on their author. There is unpublished work that one hopes might be made available, for example his work in historical linguistics. Harris was a cultured man whose broad interests and competencies glimmer through in his writings. An example, stunning for its breadth and clarity, is his review of Sapir's Selected Writings (Harris 1951b). After one of the Bampton Lectures at Columbia in 1986, a young member of the audience approached him and asked what he would take up if he had another lifetime before him. He mentioned poetry, especially the longer works of 19th century poets like Browning. He mentioned music. And he mentioned sign language.
He had a long and very productive life. He had brought his life's work to a successful culmination. With the completion of his book on politics, I imagine Death coming to him, as to the chess playing knight in Ingmar Bergman's film The Seventh Seal, and him saying "OK, I'm ready now."
It was a privilege to know him and to learn from him. He is an abiding inspiration.
Anderson, Stephen R.1981. Why phonology isn't ‘natural’. Linguistic Inquiry 12:493-539.
Corcoran, John 1972. Harris on the structures of language. In Ploetz (1972).
Darnell, Regna. 1990. Edward Sapir. Berkeley: University of California Press.
Grishman, Ralph, and Richard Kittredge. 1986. Analyzing Language in restricted Domains: sublanguage description and processing. Hillsdale, NJ: Erlbaum.
Harris, Zellig S. 1951a. Methods in Structural Linguistics. Chicago: University of Chicago Press. (Reissued as Structural Linguistics.)
———.1951b. Review of Selected Writings of Edward Sapir. Language 27:288-333. (Reprinted in Harris 1970).
———.1955. From phoneme to morpheme. Language 31.2:190-222. (Reprinted in Harris 1970. A report of a computer verification was printed in the Transformations and Discourse Analysis Papers series in the mid 1960s and is reprinted, with incorrect attribution, in Harris 1970.)
———.1965. Transformational theory. Language 413: 363-401. (Reprinted in Harris 1970, 1981.)
——— 1967. "Morpheme Boundaries within Words: Report on a computer test". (= Transformations and Discourse Analysis Papers, No.73.) Philadelphia: Univ. of Pennsylvania. (Repr. in 1970.68-77.)
———.1968.Mathematical Structures of Language. New York: Wiley.
———.1969. The two systems of grammar: report and paraphrase. Transformations and Discourse Analysis Papers 79. (Reprinted in Harris 1970,1981)
———. 1970. Papers in Structural and Transformational Linguistics. Formal Linguistics Series, ed. by Henry Hiz. New York: Humanities Press.
———.1973. Review of A Leonard Bloomfield Anthology, ed. by Charles F. Hockett. IJAL 39.4:252-255.
———.1981. Papers on Syntax. Synthese language library, vol.14, ed. by Henry Hiz. Dordrecht: Reidel. (Contains some of the papers in Harris 1970).
———. 1982 . A Grammar of English on Mathematical Principles. New York: Wiley.
———.1990. Language and Information. New York: Columbia University Press.
———.1992. A Theory of Language and Information. Oxford: Clarendon Press.
———, Michael Gottfried, Thomas Ryckman, Paul Mattick, Jr., Anne Daladier, T. N. Harris & S. Harris. 1989. The Form of Information in Science: analysis of an immunology sublanguage. Boston Studies in the Philosophy of Science. Dordrecht: D. Reidel.
Hockett, Charles F.1968. The State of the art. TheHague: Mouton.
Johnson, Stephen S. 1987. An Analyzer for the Information Content of Sentences. PhD. dissertation, Computer Science, NYU.
Kittredge, Richard, and Jack Lehrberger (eds). 1982. Sublanguage: studies of language in restricted semantic domains. Berlin: de Gruyter.
Nevin, Bruce E. 1984. Review of A Grammar of English on Mathematical Principles. by Zellig S. Harris. Computational Linguistics 10.34:203-11.
———.1988. Review of Language and Information, by Zellig S. Harris. Computational Linguistics 14.4:87-90.
Partee, Barbara. 1975. Comments on C.J. Fillmore's and N. Chomsky's papers. In The Scope of American Linguistics, ed. by Robert Austerlitz. Lisse: Peter de Ridder.
Ploetz, Senta (ed.). 1972. Transformationelle Analyse. Frankfurt: Athenaum.
Ryckman, Thomas 8.1986. Grammar and Information: an investigation in linguistic metatheory. Ph. D. dissertation, Columbia University.
Sager, Naomi S. 1981. Natural Language Information Processing. Reading, MA: AddisonWesley.
Wheeler, Eric. 1984. Review of Zellig S. Harris, A Grammar of English on Mathematical Principles. In Computers in the Humanities 17.3: 88-92.
* Published in California Linguistic Notes 23:2.60-64 (Spring-Summer 1992), a slight revision of a post in Linguist List 3.445 (Saturday, 30 May, 1992).
1. Thus, one does not see the argument, familiar elsewhere, that the unacceptability of a given starred example that is generated by the unconstrained theory must be due to its being blocked by Constraint X or by Principle Y. There is no problem of overgeneration.
2. The title was to have been Methods in Descriptive Linguistics, but the publisher substituted the buzzword "Structural." I recall him saying, amusedly, "I don't remember whether they asked me or not." In the reissue, the abbreviation to just Structural Linguistics—especially egregious because of a widespread identification with the tarbrush caricature "taxonomic linguistics"—was apparently also done without consultation, though if the Press had sent him a query they may well not have received a reply. Answering professional correspondence was not high on Harris's list of priorities.
3. In correspondence, Kac has objected that he had more to say than just this in this review.
4. For an extended treatment of this, and indeed many of the other issues discussed here, see Ryckman (1986).
5. This is the basis for the fact, as noted by Partee (1975), that there are no independent or extratheoretic criteria for distinguishing syntactic primitives from semantic primitives, or indeed for determining primitives of either type.
6. A paper on this that appeared in a 1962 volume on avoiding World War III, which is reprinted in Harris 1970, is also of interest for its insight into Harris's political thought.