WORD-STRUCTURE by MARK H, ARONOFF B.A., McG41l University 1969 SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY at the MASSACH’S2TTS INSTITUTE OF ‘T3CANOLOGY August, 1974 Signature of Author Certified by. Drewice Hy Thesis Supervisor hecested ty tt Chairman, Departre ital Committe Archives CMLL SEP 6 1974for M. and G. And he said, Draw not nigh hither: put off thy shoes from off thy feet, for the place whereon thou standest is holy ground. Bxodus 3:5ABSTRACT Word-Structure Mark H. Aronoff "Submitted to the Department of Foreign Literatures and Linguistics on August 12, 1974 in partial fulfillment of the requirements for the degree of Doctor of Philosophy." This work deals, in the main, with that aspect of word-structure referred to commonly as derivational morphology, though other related areas are touched upon in the course of discussion. Much, though by no means all of the material discussed is drawn from English, The framework presented pretends, however, to universal scope. We divide the question of word-structure into two subparts, that of word-formation, the coining of new words, and that of word~ the provision of structure to already existing words. We say that there is a lexicon or dictionary, the main provision for entry into which is that an entry mist be a word, and be arbitrary (unpredictable) in at least one aspect of its meaning or form. New words are coined by the application of general rules called Word Formation Rules. Such a rule forms a new word from an already existing one, one in the lexicon, by performing an operation of a specific sort on that existing word. Existing words are analysed by applying to them the same Word Formation Rules, but as redundancy rules, ise. as rules for determining how 2 word might have been formed. Not all new words ere added to the lexicon. Whether a given word is edded depends on wether it is arbitrary, and this is correlated with the productivity of a rule by which it is formed. Because of the way in which Word Formation Rules are formulated (one affix, one rule), it is necessary to posit a class of readjustment rules, which operate on the output of the Word Formation Rules, and whose output itself is the input to the phonology, and to the Lexical Insertion Transformation. The rules of derivational morphology are completely separated from the other sats of rules of a grammar. They operate on words to produce words or to provide structure to existing words. The figure below gives a simplified picture of the interaction of the derivational morphology with other parts of the grammar.i —-- Adjustment Rule: (iss of Exceptional Words (Lexicon) \ a v SS Lexical Insertion Transformation Phonology Derivational Morphology - ) Word Formation Rules 3 Thesis Supervisor: Morris Halle Title: Professor of LinguisticsForeword I have been able to trace the roots of ay interest in morphology as far back as high school. There, for four years I studied Hebrew grammar, which I consistently failed, Since my first introduction to formal linguistics I have been trying to atone for this failure and the present work marks yet another point along that frightful path. It is quite clear to me, and will become so to the reader, that I have not progressed far in that endeavor. Yet we know that one should not even attempt to study the true mysteries unless he has reached middle age, is happily married, and has a full stomach, and we know further that of the three who saw the light only Akiba was capable of return— ing to darimess intact. Therefore, I mst not despair. Comfort can perhaps also be gained from the words of that great sage, the nameless sage who was the teacher of ben-Moshe, who said that of all the morphological mysteries those of the Semites were the most dread- ful. "Once we have understood the others, we may begin to look at these." Though I cannot claim to have achieved understanding, even of the others, I do have some conviction that I am proceeding in a proper direction. We find comfort in precedent. It is convenient, when introducing a notion which is not uncontroversial, to defend the introduction with an allusion to its conmonness in older thought. This may reflect a deep ecclesiastian conviction. It is more usually viewed as a sign of modesty, Modesty, though, is a convenient cover in many instances for a less virtuous attitude. When something is not ours, we can 5casily disclaim ultimate responsibility for it. With this in mind, Jet me note that the basic view of the workings of morphology presented in this work, that words are formed from words, is not new. However, to my knowledge, there have been no previous attempts to integrate it into the general framework which I am presupposing, that of generative transformational grammar, I believe that this framework is essentially correct. The truth or falsity of my views mst be proved within it, and not within some more general theory of epistemology, and all responsibility for the assertion of these views therefore rests with me, Nevertheless, I mst acknowledge ny predecessors and my debts to them. I have benefitted greatly from the work of Hans Marchand, especially his The Categories and Types of Prosent-Dgy English Word— Formation (1969). His view of the workings of word-structure are a principal source of mine, though the framework in which he is working 4s radically different. The book has also been an invaluable source of data. There is no more complete work on the subject. I have also been influenced by two works on morphology which are closer in general Morris Halle's Prolegomena to a Theory of Yord- Formation (1973). and Dorothy Siegel's regrettably still unpublished perspective to min Sone Lexical Transderivational Constraints in English,(1971). With regard to less central matters the sketch of English phonology presented in The Sound Pattern of English (1968) has been as indispensable as it must be to any work remotely connected with that domain. On the most general plane I must cite two works, Noam Chomsky!s Aspects of the Theory of Syntax (1965), and Ludwig Wittgenstein's Philosophical Investigations (1953), which I can only hope not to have 6abused. Most important and most ephemeral is the knowledge I have gained from personal contact. In my years at MIT I have talked about linguistics with many people. Among these I am especially indebted to Richard Oehrie, Alan Prince, Dorothy Siegel, and Edwin Williams, and to the members of my thesis committee, Professors Ken Hale, Morris Halle, and Paul Kiparsky. Ms. Siegel has been working on very similar questions to those that heve interested me. Such a situation often leads to conflict, yet we have been able to cooperate most fruitfully. I will note he: only that many of my conclusions with regard to the Anteraction of morphological operations, boundaries, and the rules of the phonology are psrallel and similar to hers, and will direct the reader to her discussion of these matters which is contained in her disserta‘ion (Siegel, 1974). Professor Halle has taught me much of what I know about language and linguistics. I hope I will prove worthy of his attentions. Throughout most of ay stay at MIT, I received the financial support of the Canada Council and the Department of Education of the Province of Quebec. I must thank them for their generosity, and only hope that the future will allow me to pay the debt that I owe to my native land. Finally, I must thank Ms. Dorothy H. Brandl for the time and care she took in typing this work.ABBREVIATIONS The following abbreviations will be used for works frequently cited in the text. Marchand, Hans, English Word-Formation. (Munchen, 1969). Chomsky, Noam, and Morris Halle, The Sound Pattern of English. (New York; Harper and Row, 1968). Walker, J., ts Khyming Dictionary, revised and enlarged by Lawrence J. Dawson. (ew York: B B! Dutton, 1936).Chapter 1 TABLE OF CONTENTS Ground and Teleology. . + e+ eee eee eee eee 1,1, Derivation and Inflection 1.2. Other Types of Morphology 1.3. A Brief Survey of the Recent History of the Study of Morphology... ee ee eee eee Footnotes to Chapter 1... ~~~ Basics 21. 2.1.1. 2.1.2. 2.1.3. 2.1.3.1. 2.1.3.2. 2.1.3.3. 2.1.3.4. 241,356 2.1.4. 2.1.4.1. 2.1.4.2. 2.1.5. 2.2. 2.2.1, 2.2.2. 2.2.3. Trouble with Morphemes . Minimal Signs... . Words 2... eee ee Morphems 2... es Cranberry Morphs . . . « Other Berries... + PrefixStem (+latinate) A Similar Case... ss Defining the Morpheme . Trouble with Words. « Cranberry Words... . The Numerous Verbs An Historical Note on Infleetion Word Formation Possible and Actual Words Types of New Words... - ++ + What are New Words Coined From . 9 Page a Sty 7 9 2h 25 25 26 27 28 29 29 31 x ” 38 39 40 4a wa 452.2.3.1. Oddities 2. eee ee ee ee eee 46 2.2.3.2. Words from Morphemes. 2... -- e+e -2-- 2.2.3.3. Word Based Morphology .-- +--+ +++ + 48 2.2.3.4. Word Formation Rules ...+-+eee2+- 48 2.2.3.5. Assumptions about the Lexicon ..... +--+ 50 2.2.4. Evidence for the Proposal ..-- +++ +++ 52 2.2.4.1. The Phonological Cycle... +++ see eee 51 2.2.4.2, Irregular Back-Formations ..- +++ e+e 58 2.2.5. Counter-Bvidence. 6. ee es ee ee eee 61 2.2.6, Word-Structure 2.2 eee eee ee eee 64 Footnotes to Chapter 2 wee ee eee eee 68 3. Productivity 2 see eee eee eee eee eee 3.1. Preliminaries see eee eee eee ees 3.2. fress and Hty. 6 ee eee eee ee B21, Gaps ve eee ee eee ete eee eee 6 3.2.2, Semantion ss see eee eee creeees 7 3.2.3. Phonology see ee eee ee eee ees 3.2. Gaps Again. ee eee ee ee ee ee ee 82 3.2.5. Semantics Again... sees eee e eres 86 3.3. Blocking see eee eee e reece eres 87 Blt. Conclusion. - eee e eee rece eres Footnotes to Chapter 3. - ee eee eee rere ees | Preface to Part 2 .. + a Word Formation Rules... -.- ++ sees eee reee 9 41. Syntax and Semantics.» ++ +e ee eee ee 10L Weld. The Bases eee ee ee eee DOL 4.lsl.l. The Unitary Base Hypothesis... +--+ +++ 102 Wel.2. The Output... ee eee ee eee 10H 104.1.2.1, 4.1.2.2, 4.2, 42.1, 42.1.1, 4.2.1.2. 4.2.1.3. Morphological Restrictions on the Base... Abstract Morphological Features... ... Restrictions Statable on Individual Morphems 22.2 +e ee eee a Encoding Morphological Restrictions... « 4,2,1,3,1, Ordering of WFRs 6. ee eee ee ee ee 4,2.1,3.2. Unordered WFRs see ee 4.2.2, 4.2.2.1, 4.2.2.2, 4.2.2.3, 42036 4.2.3.1. 43. 4.3.1. 4.3.1.1. 43.1.2, 4.3.1.3. HDL 431.5. 4,361.6. 463.26 43.2.1. 4.3.2.2. ‘The Morphological Operation... +2... Copying Rules «2 pe ee eee ee eens Infixing eae ae ae ee Consequences». 2 ee ee ee ee ee ee The Place of the Morphological Operation in the Grammar eee re Reduplication Paradoxes . . . ee eee Phonology . eee ee tee tee eee Phonological Conditions ....-. 20s An Aside Concerning Negative and Positive Conditions . 2... eee ee ee Conditions on the Underlying Form of the Base Stress Sensitive Affixes ........ 6% + Boundaries and Cycles .......--- # Boundaries and Cycles «2-1. eee ee A Theory of Boundaries and Brackets... . - Problems... ee eee ee ee ee eee A Condition on the Surface Form of the Output 2 ee ee ee tee ee eee ee A Global Phonological Condition. ...... n 105 107 109 109 109 113 u7 ng 29 133 134 a 146 150 152 167 167 168 169 170 172 175 175 176 176 1774.3.2.3. W362 Ad, A Transderivational Constraint ........ Boundary Problems... - ee eee eee Sumary ......-20. Footnotes to Chapter 4 ..... Adjustments Rules. .....-- 5.1. 5.1.1. 5.1.2. 5.1.3. Sale 5.1.5. 5.1.6. 5.1.7. 5.2. 5.2.1. 5.2.1.1. 5.2.1.2. 5.2.1.3. 5.2.1.4, 5.2.2, 5.2.2.1. 5.2.3. 52H, Footnotes to Chapter 5 Bibliography Truncation Rules . . . PO eee ee eee ih} DOOOGoU oOo GDOUoobDoOG a Sere eee aaa a) TrunceWFRs 2. 1. ee Truncation and Phonology Russian Truncation . . German ge- Deletion . . Allomorphy Rules. ADlomorphs of Ation . . Stems of the Form Xtate ‘The Marked Roots . . « Root Allomorphy ... Other Allomorphy ... Allomorphy and Other Parts of the Grammar. . . Biographical Sketch. ......-- 178 180 183 185 188 189 89 gz 195 198 199 200 205 207 208 209 210 212 212 218 228 229 232 236 238 243CHAPTER 1 GROUND AND TELEOLOGY We will be concerned in this work with the internal structure of words, a subject which, in the linguistic literature, is called mor- ology. The notion word has long concerned students of language. It's defi- nition is a long-standing problem in linguistics, and entire volumes have been devoted to the subject (e.g., Worth, 1972). A reasonably detailed procedure for isolating phonological words (units which may be considered as words for phonological purposes) is provided in SPE (pp. 366-70). Further refinements of this approach are discussed in Selkirk (1972). Syntactically, Postal (1969) puts forth a persuasive argument that the which is a word, as a syntactic unit, corresponds to the Anaphoric Island, syntactic string, the internal elements of which cannot participate in anaphora. Though semantic definition of the notion is a traditional goal, it has not, to my knowledge, been achieved. To say that morphology is word structure is not to say that all of the structure of the word is encompassed in the domain of morphology. There 4s a branch of phonology, termed phonotactics, or morphene structure, which concerns itself with the determination of possible sequences of sounds in a given language, "possible phonetic words." This is not morphology. Morphology treats words as signs, not just as forms, but as meaningful forms. It is therefore concerned with words which are not simple signs, 3 two distinct but made up of more elementary ones. This concern encompa: 3but related matters: firstly, the analysis of existing composite words, and secondly, the formation of new composite words. A unified theory of morphology should be capable of dealing with both of these ereas in a uni- fied and coherent manner, though it may not be possible or even desirable, as we will argue below, to treat them in exactly the same manner. On the subject of unified theories, it should be stressed that mor- phology, as defined, is a small sub-system of the entire system of (a) language. A theory of morphology must be integrated or at least integrable Anto a fairly specific general theory of language. Morphology, as a sub- system, and a sub-theory, may have its ow peculiarities; a system can be unified without being completely uniform. However, it does not exist in_ vacuo. The present work is conceived in the general framework of ‘Transformational Grammar as outlined in such works as Chonsky (1965) and SPE, More particularly, it presupposes the Lexicalist Hypothesis of Chomsky (1970) and at least the spirit, if not the letter of Kiparsky's views with regard to phonological abstractness, discussed in Kiparsky (1973). 1.1. Derivation and Inflection There are traditionally two types of morphological phenomena, derivational and inflectional. The distinction is delicate, and sometimes elusive, but nonetheless important. Inflection is generally viewed as en- compassing the "purely grammatical" markers, those for tense, aspect, person, number, gender, case, etc. Within a Lexicalist theory of syntax, (cf. Chomsky, 1970), inflectional morphemes would be dominated by the node X, and, perhaps, higher nodes (ef. Siege1,1974) while derivational morphemes would be dominated by the node X. Derivational morphology is thus restricted to the domain of Lexical Category. ubIt 4s generally true, and in accord with the Lexicalist formalism, that derivational markers will be encompassed within inflectional markers. In the English word comparttmenttal#izetd, for example, the last morpheme, #d, is inflectional, and all those internal to it are derivational. The two sets may not be interspersed. Thus the word compart+nenttalfiztattionts is possible, though the word *compart+menttalfizeddtation#s is not. One peculiarity of inflection is that it is paradigmatic. Thus, every English non-modal verb exhibits a paradigm consisting of the following forms: Vo VAs Vid) Vd, Ving For example: sigh sighs sighed (has) sighed sighing go goes went (has) gone going ‘The verb go exhibits suppletion, the filling of one of the slots of the paradigm by a phonologically unrelated form. Since derivational morphology is not paradigmatic, it does not show any suppletion, i. ©. it does not concern itself with phonologically dissimilar, out semantically related forms. Sometimes a paradigm is defective, lacking a form, The missing form iz almost always the uninflected one. So, in English, we have scissors, pants, trousers, but not *scissor, *pant, *trouser, except, of course, in derived forms, where the constraint on the mixing of morphologies still holds, as the following examples demonstrate: scissorlike “scissorslike trovserleg *trousersleg! A fuller description of some of the properties of inflectional morphology can be found in Bloomfield (1933). An independent character- ization of the properties of derivational morphology is more difficult. Nida (1949) suggests the following: if, in a syntactic class (defined by 5substitution in his system, and in corresponding ways in other theories), we find items which are monomorphemic, then the polymorphemic items in that class are derived by the system of derivational morphology. The most immediate problem for such a definition is the existence of suppletive forms, such as went above, which, by Nida's criterion, would force us to include the past tense suffix in derivational rather than inflectional morphology. This is where the paradigm enters. We find that the past tense is a paradigmatic category, and therefore must be inflectional. We might also invoke more abstract syntactic evidence to show that though ‘ent is monomorphemic, on the surface, there is evidence for an abstract. past tense morpheme. This is more difficult, though perhaps possible. In any case, Nida's simple criterion, as he notes, must be amended to exclude clearly suppletive forms which are members of paradigms. More difficult are cases of syntactically or semantically arbitrary forms. Consider the noun police in the following sentence: (1) The police have arrested six people already. The verb shows us that the noun is syntactically plural. Unlike words like sheep, which are sometimes plural and sometimes singular, police, in this sense, never appears in a singular context. We cannot, therefore, argue in any straightforward way for the existence of a zero plural marker, as in sheep. Nor can a paradigm help here, since there isn't any. Recourse might be had to syntactic argument, but here there is the danger that a system which permits abstract morphemes in this instance, where we believe we are not dealing with derivation, might also permit us to posit abstract morphemes in other cases, where we feel we are dealing with derivation. Consider the set of agentive occupational nouns, exemplified below: 16(a) (b) (e) (2) baker cook chef packer pilot chauffeur painter coach smith hunter mechanic tanner surgeon ‘The items in column (a) exhibit a clear suffix: -er; those in (b) might be derived by zero suffixation, from the corresponding verb. The items in (c), however, correspond to police, in having no corresponding verb from which they may be derived, This is clearly the sort of case Nida had in mind. We wish to use the fact of the existence of (c) as evidence that the other items are proruced by the system of derivational morphology, and not by inflection. However, if we can argue for an abstract morpheme in the case of police, why not here, Clearly, we mst constrain our system of argumentation (syntactic theory) in such a way as to exclude the possibility of positing an abstract morpheme here, but not in police, In any case, it shows that Nida's criterion, even when extended to take suppletion into account, is not sufficient to capture our intuitive notion of what derivational category is, unless we have a better idea of what a possible syntactic derivation is. I do think that once we have clarified that issue, Nida's criterion, in its simple form, will prove to be correct. 1.2. Other Types of Morphology. Derivation and inflection do not exhaust the domain of morphology. There are "grammatical" morphological phenomena which cannot be subsumed under inflection. The best lmow, of these is that of incorporation or cliticization. In Classical Hebrew, for example, definite pronominal objects, under specific conditions (basically, when they are anaphoric, rather than deictic) are incorporated into the verb, forming, phonologically, a single unit with it, | There is no question here of inflection, since 7this specific form of the verb only occurs when we would otherwise expect a definite pronoun object. A similar situation holds in English (ef, Selkirk, 1972). A slightly more complicated example along the same lines comes from Syriac. Here, in addition to pronoun object cliticization, we have the copying of a pronoun for any definite object, other than anaphoric pro- nouns. The copied pronoun is cliticized to the verb, giving the same verb form as that containing the pronoun object. Clearly, the copying and the cliticization are both syntactic facts, and they are not paradigmatic. Sometimes other material than pronouns can be incorporated into the verb. In Navaho, a specific adverb may scmetimes occur inside the verb, and sometimes elsewhere in the sentence, but never in both places in the same sentence, This fact can be most easily captured by a syntactic movement rule, There is no traditional term for this third type of morphology. It is clearly "syntactic", and can be grouped together with inflection, as opposed to derivation, on that ground. There is often a clear historical connection between pronoun copying and cliticization and verbal agreement, and it may very well be that all agreement arises by a falling away from and generalization of pronoun cliticization, This would of course strengthen the contention that this third type of morphological phenomencn and inflection are really of the same nature, and opposed to derivation. We will accept this opposition in the greater part of the body of this work, and restrict the scope of further discussion to the domain of derivational morphology. This restriction will be relaxed only in regard to the interaction of phonology and morphology, where morphology en- compasses both inflectional and derivational markers. 181.3. A Brief Survey of the Recent History of the Study of Morphology. Morphology is not something new, or like syntax, something mich talked about for many years, but little studied or understood. The early Indo- Europeanists, Bopp for instance, were interested almost solely in mor- phology, and morphology has remained one of the mainstays of the philological tradition (cf. the extensive bibliography in M), American descriptivists, though their tools were better adapted to phonological and morphophonemic purposes, did do much substantive work in the area of morphology as we have defined it. In the specific area of English morphology, I have already cited M, end my debt to that work. Jespersen also devoted a volume of his Modern English Grammer to the subject. Of the more recent work, I will note Zimmer's work on affixal negation (1964) which is notable for its concern with semantics and the very general and difficult problem of productivity. Within the Generative framework, morphology was, for a long time, quite successfully ignored, There was a good ideological reason for this: post Syntactic Structures linguistics, in its zeal, saw phonology and syntax everywhere, with the result that morphology was lost somewhere in between. For proponents of early Generative Grammar, grammar consisted of syntax and phonology. Phonology, at last freed from its phonemic linkers, encompassed all of morphophonemies and phonemics in a grand system of ordered rules. Syntax took care of everything else: "all of the granmatical se- quences of morphemes of a language" (Chomsky, 1957, p. 32). Within such a framework, morphology is not a separate study. In fact, though some of the earliest studies in transformational syntax were specifically restrict~ ed to the domain of the word (e.g. Lees, 1960), this domain was not considered to differ in any real way from that of the sentence. Even very recently, the school of Generative Semantics has insisted that the 19word is fundamentally no different from any other syntactic unit, thus espousing a position which, in its essence, as that of early Generative Grammer, denies the independence of morphology. Recently, a substantial interest has arisen in the peculiarities of inflection as a separable syntactic phonomenon. The first study in this area was that of Bierwisch (1967). It has been followed by others, of which I will note Wurzel (1970) and Kiefer (1970, 1973). I will not discuss these works here, as their domain stands outside that established for the present work. The Return of Morpholo; Morphology found its way back into generative linguisties through several rear doors, almost simultaneously. The first hints that there might be something between syntax and phonology are found in SPE, There the question is first raised of whether the output of the smtactic component is in fact the input to the phonological component. It is noted that there are "certain discrepancies", and that ", . .the grammar mst contain cer‘ain rules converting the surface structures generated by the syntactic component into a form appropriate for use by the phonological component." The rules in question divide surface structure into phono- logical phrases. They are called readjustsent rules, and are supposed generally to “involve elimination of structure." An illuminating dis- cussion of such rules is contained in Selkirk (1972), But these are not the only rules called readjustment rules, There are in addition rules which "eliminate grammatical formatives in favor of phonological matrices", converting, for example,[(sing) past], into sung, and {tmend] y past]y into mended, The term readjustment rule is obviously being used broadly, for these last rules are clearly rules of inflectional mor- 20morphology. Yet a third type of readjustment rule is in no way connected with elimination of structure. This sort applies "...to specific derivable formatives; for example the rule (110): (110) = t-p4/ _ mi__+ive "(SPB p. 223). ver_+ion Rule (110) is a very different sort of morphological rule. It is a rule of allomorphy, which spells out the form of particular morphemes in specific morphological environments. We see, then in SPE, the beginnings of a recognition of the independence of certain classes of phenomena from syntax and phonology. The term readjustment rule is not a particularly well-defined one, but among the rules so termed, we do find a significant number which are plainly morphological. SPS inadvertently created in its wake a second entrance for mor- phology. The purely formal spirit of Chomsky's and Halle's approach to Phonology in general, and of the sketch of English phonology presented in ‘SPS, in particular, prompted a reaction. It was felt by many scholars, most prominently Kiparsky, that by disregarding concrete evaluation measures, Chomsky and Halle were often led to propose phonclogical systems which were too abstract, and abused the classificatory funetion of the phonetic features, These criticisms can be seen, historically, as a reaction to the excesses of revolutionary fervor. Remember that Chomsky and Halle were fighting against a teory which term ed phonological only the most apparent of alterations, and put all others into one morphophonemic bag of lists, without regard for the differences in regularity among them, The revolutionary step of these pioneers was to pull down the phonemic barrier, and declare all alterations to be the pro- vince of phonology. But, said their critics, surely not all connections are aphonologically regular? Most of those which were earlier included under the morphophonemic label can indeed be treated as phonologically governed rules, but there is some limit. There are alterations which are just not determined by purely phonological features. A further step, one which the critics have by and large not taken, is to ask whether some of these alternations, which are not phonologically determined, are in fact not part of the phonology at all. I will argue below that a class of rules which a more tightly constrained theory rejects as not optimal phonological rules, can be fruitfully included in a theory of morphology. Thus, because of a desire to place restrictions on the power of phonological theory, we find that certain phenomena now lie outside the domain of the theory. Many of these phenomena can be seen as morphological. ‘The same sort of pattern is found as we found in SPE. A similar retreat took place at about the same time in syntax. In an attempt to restrict the power of grammatical theory, certain phenomena were removed from the domain of the syntax. In contrast with phonology, however, where the realization that the system as it stood could not be sufficiently constrained came gradually and inexorably, with very little objection on anyone's part to at least the spirit of the trend, and curiously few suggestions as to what should be done with the discarded material, or even what it was, morphology sprung out of syntax's thigh full-blown, and caused a great to-do when it did so, The birth of mor— phology, or at least the declaration of its domain, is simultaneous with, and contained in, Chomsky's "Remarks on nominalization" (1970). This paper presents a new theory of syntax, in which all of derivational morphology is isolated and removed from the syntax; it is instead dealt “i expanded lexicon, by a separate component of the grammar. 22This distinction legitimizes the field of morphology as an independent entity. “Remarks on nominalization" was long and bitterly opposed, mainly, I believe, on esthetic grounds. Where previous and rival theories view language as one vast domain, encompassed by pervasive constraints (cf. Postal, 1972), Chomsky prefers to see language as divided into smaller well-distinguised units, each governed by its own, perhaps idiosyncratic, rules. As the reader will discover, I am more inclined tovard the latter perspective, even within the narrow field of morphology. Chomsky did not propose a theory of morphology, he merely suggested that there should be one, and that its properties, if he is correct in dividing morphology from syntax so sharply, should be very different from those of an adequate theory of syntax. I will attempt to elaborate such a theory, The theory which I will present bears, indeed, little resemblance to any vrevalent theory of syntax. It will also encompass many phonological phenomena which cannot be easily incorporated into a reasonably narrow theory of phonology, and provide what I think is e unified account of morphological phenomena within a generative grammar. This unity is important. Critics of the new esthetic accuse its Proponents of excessive rug-sweeping, clearing away so much data in the name of restricting the power of a grammar, that the describable residue becomes miniscule. However, if we cen show that what has been swept aside can be gathered up again, then we are vindicated in our vision. 23FOOTNOTE the behavior of pants is exceptional: pantspocket *pantpocket pantsleg *pantlegCHAPTER 2 BASICS 2.1. Trouble with Morphemes. The units into which words are analyzed, out of which they are com posed, are termed morphemes, We will be concerned in this section with some problems concerning the defining characteristics of the morpheme, concentrating on a central premise of the approach which has been most pervasive in American linguistics. This premise is the definition of ‘the morpheme as "the smallest individually meaningful element in the utterances of a language," (Hockett, 1958, p. 123). Accepting this premise entails that every polymorphemic word be a compositional entity. It is compositional in two senses, both semantically and structurally, the semantics being a function of the morphemes and the structure, just as a sentence is semantically compositional. Recent work has revived the truism that every word has its own individual idiosyncratic traits, some of which can be very erratic and elusive. (We are speaking here of derivational word, inflection does not suffer from this erraticness), If it is true of words that they are minimally meaningful, then what about the morpheme? Does it have no status at all, or can we define it other than semantically? The point of this chapter is to show that the latter question can be answered in the affirmative. Specifically, we will isolate a class of morphemes, show that there is no way in which the menbers of this class can be said to have any meaning at all, and then demonstrate that there are phonological criteria which allow us to 25isolate occurrences of these meaningless morphemes. The importance of this demonstration is two-fold. First, it shows us that any theory of "minimally meaningful element" is misguided. Secondly, it shows that despite this fact, it is still possible to construct a theory in which the morpheme plays a central role. 2.1.1, Minimal Signs, In order to understand what is at stake here, we mist first have a better understanding of what is meant by "minimal meaningful element." This entails a short review of Saussure's concept of the minimal sign. ‘The sign is one of the most basic concepts of linguistics, and the literature on the subject is vast, nor is the definition of the sign a closed matter. I will adopt in this discussion what I think is an orthodox view where orthodox does not, of course, mean totally uncontrov- ersial. The minimal meaningful unit of a language is the basic, minimal, is an arbitrary union of Saussurean sign (cf, Saussure, 1949). the semantic and the phonetic. So the sign deg has a meaning and a sound; one cannot exist without the other; they are arbitrarily united, Arbit- rarily, because there is nothing in the sound which dictates its meaning, and vice versa, except social convention. The sound may change, and the meaning remain, or the meaning change, and the sound remain. Out of the minimal signs we can construct composite signs. These signs are not arbitrary, Their meanings may be predicted from their structure and the minimal signs out of which they are constructed. Sentences are composite signs. It 1s sometimes argued that there are different degrees of arbit- rariness. A sign like dog is completely arbitrary. However there are 26other signs for which there is felt to be some intrinsic connections between their sounds and their meanings. Onomatopoetic words, and those which involve phonetic symbolism (cf. M 398 ff.), like slurp and quack, are said to be partially motivated (non-arbitrary) because o1 this in- trinsic connection. The class of partially motivated signs also in- cludes composite items whose meanings can be partially, but not completely, derived from the meanings of their parts. Thus a sign which formally consists of the signs atb, but whose meaning must be represented as ABH, that 1s, the meanings of a and b plus something else specific 4n addition, 4s sometimes said to be partially motivated. I will hold with Srussure, and contra Bally (1940) and M, that only fully motivated signs are to count as non-minimal, that partial motivation is not significant. Thus, any sign which is the least arbitrary is considered to be part of the basic inventory of signs.) Most of what follows is devoted to deciding what sorts of elements form this basic inventory. 2.1.2. Words. That there are minimal signs which are polymorphemic was first stressed as an important fact, at least within the framework of Gener- ative Grammar, by Chomsky (1970). Chomsky notes that much of derivational morphology is semantically irregular, and should not be handled in the syntax. Out of this remark there developed two hypotheses. The Strong Lexicalist Hypothesis of Jackendoff (1973) excludes all morphological phenomena from the syntax. This means that the syntax carnot relate some and any, or ever and never, and that inflection, if it is referred to in the syntax, must be handled by some sort of filter, The Lexicalist Hypothesis which is more widely accepted than this one,but which to my knowledge has never been explicitly formulated in print, is that deri- 2vational morphology iy not ever dealt with in the syntax, but inflection is, along with other such "morphological" matters as Do Support, Affix Hopping, Clitie Rules, i.e. all of "gramratica2 morpholoyy." This seems to be the position of, for example, Chomsky (1973).” This latter Lexi- calist Hypothesis, which I will assume, does not say that derivational processes are alvays irregular and that their semantics is always non- compositional. Nor does it exclude from the domain of the syntax only irregular derivational pheononena (as Chonsky (1970),says one might do). It says rather that derivational pheonomena are always separate from the syntax, regardless of their regularity, Postal (1969) presents very convincing evidence for this hypothesis.) Translated into a Saussurean framework, the hypothesis saya that for the purposes of syntax, the word? (sans inflection) is the minimal sign.? This hypothesis says nothing about intra-word phenomena and relations; they may or may not be regular. Of course the main reason for the adoption of the hypothesis in the first place was semantic irregularity, and we must develop a theory of deri- vational morphology which allows for, and hopefully even predicts and accounts for, this observed irregularity. I will now present evidence that the word is a minimal sign, not merely for the purposes of the syntax, To do this, I will show that below the level of the word we encounter morphemes whick, while they must be assumed to be real Linguistic elements, have no meaning which can be assigned independently of each of the individual words in which they occur. This demonstration is not a novelty. The most extended and formalized argument of the point I am making that I know of is in Hervey and Mulder (1973). 2.1.3. Morphenes. 282.1.3.1. Cranberry Morphs. There is a class in English of hapax legomena, morpheme: which only occur in one English word. They are often called cranberry morphs. Consider the following list: qa) cranberry boysenberry huckelberry mulberry bilberry Since the words in our list are all names of berries, we may isolate this last unit as a meaningful morpheme. We are left with the items in (2): (2) terant Pooysent thuckle# mult doilt None of these items occurs either independently, or in any other words than those in (1). There is thus no non-circular way of assigning mean- ings to the morphemes in (2). Their meanings are intimately connected with those of the individual words in which they occur, As Hervey and Mulder note "both + each can be identifiable as a sign only and b if the other is also identifiable as such." One can, of course, ignore problems of circularity, and assign a meaning to the item in question. It is then merely an accident that this fully meaningful item occurs only in one word. However there are cases in which such a simple out is not possible. 2.1.3.2. Other Berries, As noted above, it is possible to assign a meaning to items such as feran#, simply because they do occur only in one word. With other names of berries, however, this simple device will not work. Consider the following list: (3) strawberry blueberry blackberry gooseberryBy removing berry again, we can isolate the morphemes in (/+): (4) fstrawh fbluet #black# #eooset As opposed to the items in (2), these occur elsewhere than as parts of the names of berries, in fact they occur as independent words. However, when they do appear as independent words, they have meanings which can in no way relate to the meanings they might be assigned in (4). For example, one might think that a blackberry is black, however, not all black berries are blackberries, and furthermore, many blackberries are green or red (a fact also noted by Hervey and Mulder), There is therefore no way to assign a meaning to the item black which will be valid both when it occurs as an independent word and when it occurs in the word blackberry, The same holds for blueberry, and the connection between geese and gooseberries, or between straw and strawberries is not very apparent. The problem here is that we can not resort to the simple ruse of assigning the items in (4) constant meanings, for they do occur elsewhere than in the words in (3), but with meanings which a e totally incompatible with those we would like to assign to them on the basis of the meaning of the corresponding word in (3). This problem of a morpheme having different meanings in different words can be gotten around without giving up the claim entirely that morphemes are meaningful. The basic tack is to give morphemes under~ determined meanings, with contextually determined allomeanings. This is essentialy the tack which Chomsky (1970) takes. In order to handle Adiosyneratic semantic differences in verb-noun pairs like refuse-refusal, he says that "the lexical entry may specify that semantic features are in part dependent on the choice of one or another of these categorial features" (noun or verb), To the extent that these dependencies are 30regular, and syntactically motivated, there is virtue in such a device, or a similar redundancy convention, but to the extent that they are idiosyneratie, which many of them are, the device merely serves to obscure the truth, that it is the words that are idiosyncratic. ‘Though this system may allow us to preserve the idea that morphemes are meaningful, it is only at the level of the individual word that these meanings can be fully specified. In the particular case with which we are dealing, the device of underspecification and contextual filling leads to a particularly un- satisfying result, Since, as noted, sone blackberries are red, and since something cannot be both black and red at the same time, the two alloneanings of #black# will be contradictory, and share almost no semantic features (tcolor?), Allowing a devi.e which permits such a situation 4s very dangerous; it essentially gives homophony as the only criterion for deciding whether two things are instances of the same meaningful entity.6 One might also go entirely the opposite route. Thus one could clain that the various instances of #black# are completely unrelated, each a different morpheme, This rids us of the problem of morphenes with under— specified meanings, though we are still left with the circularity problem; is it the word or the morpheme which specifies the meaning? The next sot of data bears on this theory, 2.1.3.3. Prefix-Stem (+latinate). The last two sets of data consisted of what are traditionally called "partially motivated" forms, There was one element, berry, whose meaning was relatively constant, and another, which in a sense told us what sort of berry we were dealing with, but which never occurred, or never occurred anwith the same sense, outside of the particular word with which we were dealing. This next set of data differs from these in having no such fixed element. The set of data consists of the latinate verbs with bound stems and prefixes which are always stressed on the stem. This class is marked phonologically, in the system of SPE, by the presence of a special boundary, =, between the prefix and the stem, Examples of such verbs are refuse, convene, inject. I will not discuss verbs such as suffer, proffer, differ, which diverge in their stress patterns from other pre= fixed verbs with bound stems, and for which no = boundary is posited. Nor will I discuss verbs such as re#fuse (fuse again), in which there is a # boundary in the system of SPE. For this class (x = y) it 1s possible to demonstrate that neither the prefix nor the stem has any fixed meaning. First the stem. Consider the verbs in (5): (5) Xefer — Xemit Xzceive refer remit resume receive reduce defer demit deceive deduce prefer presume infer induce confer commit: consume conceive conduce transfer transmit transduce submit: subsume admit assume adcuce permit perceive let us presume for the moment that the prefixes in (5) have constant meanings, much as the berry of (1) and (3). Is it possible to extract ‘any common meanings, however minimal, from the different occurrences of each stem? At first glance, if we merely compare pairs of verbs, onemight be tempted to think so. Confer and transfer might appear to share something, similarly remit and submit, conceive and perceive, assume and presume, induce and deduce. However, if we attempt to extend matters beyond these select pairs, and extracting the common sense fron each, assign it to the other verbs in the particular stem, the result is nonsense. What, even vague, sense does prefer share with confer and transfer? or commit with remit and submit? or receive with conceive and Berceive? or consume with presume and assume? or reduce with induce and deduce? None. There is no meaning which can be assigned to any of these stems, and combined with the presumable constunt meanings of the prefixes in a consistent way, to produce the meanings of all the verbs in that stem. Each stem occurs in different verbs, but never with the same sense. Rather the sense is determined by the individual verb. As suggested above, one might attempt to reduce the whole problem to cranberries, (with, of course, the accompanying problems of that class) by calling each occurrence of a given stem a different morpheme. This system completely denies any linguistic reality to the stems, and re- places each of them by a list of homophones, each with its own meaning, and each occurring with only one, perhaps even two, prefixes, In such a system, one would not have, for instance, a stem mit which occurred in all the relevant words in (9), rather one would have many homophonous stems, This system would be fine if these stems had nothing at all in common, The problem is that all occurrences of the stem mit do share a common feature which is not predictable from any general phonologicel properties of the sequence [mit]. As carefully documented in Chapters 4 and 5, all instances of the latinate stem mit exhibit the same, phonologically arbitrary, variant (allomorph), before the suffixes tion, tory, tor, tive, table. The details of the rele- vant argument are given in Chapter 4, For the moment we can look at‘the following paradign: (6) permit, permission permissive remit remission remissory excrete exeretion excretive assert assertion assertive digest digestion digestive prohibit prohibition prohibitive The last colum reveals the difference between verbs of the form and other verbs with final t, before the suffixes in question. Mit always takes the form mis here, and the change of t to s in this environ- ment is confined to this one stem, There are no exceptions to this rule either way. This regularity, or the factors which condition it, cannot be phono- logical, but must be stated on another linguistic level, the level of the stem, or morpheme. Proof of this assertion is the fact that other in- stances of the phonological sequence(mit), not instances of the latinate stem mit, do not show up as [mis] in the relevant environment. So we find vomit/vomitory. In the word vomit, there is no reason to pre- sume that we are dealing with a prefix vo and a stem mit, in fact there is good reason to believe that we are not; vo never shows up as a prefix elsewhere, and the stress pattern gives us no evidence of a boundary, or at least of the sort of boundary for which there could be evidence, The alteration in question 4s therefore restricted to the latinate stem mit. This means that all the items which, in the theory in question, were more homophones, mity,...,mit,, must be at some level instances of the sano thing. Otherwise there is no way to express the fact that all occurrences of mit exhibit the same allomorphy. There is good evidence that the evel at which the rule embodying the facts in question mst be stated 3h4s that of the morpheme. First of all a feature such as latinate, which governs, among other things, what sorts of affixes can be attached to a word, is a property of morphemes. Secondly, the sort of rule that changes t to s in the relevant environnent here is a rule which applies ‘to a morpheme, and not to any other linguistic level, lower or higher. Mit 4s therefore a morpheme, though it has no meaning. Nor is mit the only case, As we shall see in Chapter 5, there are many stems which undergo rules of allonorphy. It appears, then, that there is something fundamentally wrong with the theory of many homophonous mits, for there is good evidence that we are indeed dealing with one morphems. This turns us back to the allo-~ meaning theory, with its problems of under-specified meanings and circularity, or to the theory that morphemes are not mininal signs. ‘The allomeaning theory had some plausibility with reference to pro- ceding sets of data, (cranberry, blackberry, ete.) mainly because, as noted, we always had one constant element, with a relatively per- spicuous meaning, and we could as a result attribute the residue of the meaning of each word to the problematic morpheme. Here, however, when we look at the prefixes, we find that just like the stems there is no constant meaning which can be attributed to any of the prefixes. How, then are we to segment the meaning of the individual vords in a principled manner? Consider the following list: (7) resk con=X repel compel Ampe1 remit commit denit refer confer infer defer resune consume 35receive conceive deceive reduce conduce induce deduce Though it is more likely that one could attribute more commonality of meaning to occurrences of some of these prefixes than one could to any of the stems, there is no general meaning which can be assigned to any of them, Thus one might try to assign to re a meaning "back", and def- initely many of the verbs of the form re=X have something to do with "back" (cf, Williams, 1973). What about receive, though? or reduce in the following sentence? (8) The government reduced the size of the quart fron 32 to 31 ounces, in an effort to stop inflation. Since the quart never was less than 32 ounces, there is no way in which back" can be involved in the meaning of reduce here. Now, since we know from (7) that re= has no fixed meaning, and we Imow from (5) that duce has no fixed meaning, how are we to segment the meaning of renee into two parts, one associated with re=, and the other with duce, in a principled manner? We can't. The word principled 4s important here. One can a priori split any word in two and give each part a meaning. I can divide apple into a and pl, and give each of them part of the meaning of the whole word. We don't want to do this, for, ‘as noted above, allowing such an analysis reduces the predictive power of a theory to nil. It is unfalsifiable. Thus the fact that the allo- meaning theory mist be made so strong in these cases that its empirical validity is reduced to nil, forces us to fall back on the only position Jeft to us, There are morphemes which have no meaning. The hypothesis that morphemes are the "minimal meaningful elements of language" cannot be maintained in any of its even most contorted variants. In many cases this role of the minimal sign mst be moved one level up, to the level 36