Morphological theory and English

Links & Letters 1, 1994 71-84 Morphological theory and English Andrew S~encer L Department of Language and Linguistics University of Essex ABSTRACT This paper presents a review of a number of recent issues in the field of generative morphology, with their implications for the description of English. After an introduction to the field two types of question are considered. First, 1 examine the nature of word structure and illustrate two competing approaches, one of which assurnes that words have a constituent structure (much like the phrase structure of syntax) and the other of which rejects this assumption. Then we look at the way morphologicai structure interacts with syntax. We examine the extent to which syntactic principles can account for the behaviour of certain types of compounds and aiso the expression of syntactic arguments in nominaiizations. l. Introduction: morphology in generative grarnmar This review will deal with a number of trends in recent morphological theory as they affect our understanding of English morphology. Descriptions of English morphology can be found in Adams (1973), Bauer (1983), Spencer (forth~omin~) and Szymanek (1989); Selkirk (1982) contains a good deal of factual information, too. The standard descriptive source (with historical information) is Marchand (1 969). Morphology entered the domain of generative linguistics with Chomsky (1 970), Halle (1 973), Siegel (1 979) and Aronoff (1 976). Morphology stands at the interface between the lexicon, phonology and syntax, and many of the most significant questions concern the way that morphological representations interact with representations at other linguistic levels. At the same time, important quest&ns have been raised aboutthe nature of morphological units and morphological processes.Generative grammar seeks to provide an explicit, formal theory of language structure. Originally this meant constructing sets of rules, which are ultimately formalizable as mathematical expressions, but which in practice are usually stated in a relatively informal notation. Such a set of rules is a grammar, and this is held to underlie the native speaker's tacit (unconscious) knowledge of hislher language. More recently, grammars have been viewed as sets of principles and constraints on the well-formedness of linguistic expressions, but the overall conception remains essentially the same. It is very important that such a grammar provide a representation of al1 the grammatical expressions of the language (i.e., sentences in the case of syntax, - 72 Linls & Letters 1, 1994 Andrew Spencer words in the case of morphology) and fail to prove a representation of those expressions not permitted in the language. We say that a grammar generates the well-formed expression (i.e., correctly formed sentences or words). There are three main aspects of word structure: (i) the phonological structure of words and morphemes; (ii) the internal structuring of words, i.e., the way that component morphemes combine, and also the relationships that hold between sets of morphologically related words; (iii) the relationship between word structure and syntactic processes and representations.In this review 1 shall be concerned with questions (ii) and (iii), basing myself on phenomena in English which have been the subject of recent theoretical discussion. 2. The internal structure of words There have been two basic approaches to account for the structure of a word such as untied. First, we can store individual morphemes un-, tie , -d, along with their meanings or grammatical function, and then combine them to give first [un + tie], then [[un + tie] + d]. On this conception a morpheme is essentially like a morphologically simple (monomorphemic) word, stored in the lexicon. This is basically the model proposed by Halle (1973), one of the first explicit statements of generative morphology. The other conception is to regard the verb stem un-tie as the result of a morphological operation of prefixation performed on the root (or word) tie. The past tense is then formed by virtue of another operation of -d suffixation. In this view it is the lexical roots like tie which are stored in the lexicon, and the inflectional and derivational morphemes are added by specifically morphological operations. This model was first argued for by Aronoff (1976) (though he only discussed derivation; Beard, 1987, proposes a model including inflection). Finally we can adopt Halle's model for derivational morphology, but handle inflection in something closer to Aronoff's way. An influential approach along these lines is the model of inflection proposed by Anderson (1 982). These competing views mean that one of the main controversies in generative morphology is whether word structure should be thought of as like syntax, and involving the successive concatenation of morphemes, the way that words are concatenated to form sentences, or whether this analogy with syntax is misleading. An important notion here is that of 'morpheme'. In structural theories of morphology a word form such as cats consists of a root morpheme cat and a plural morpheme -s. This sort of description is often referred to as an Item-and-Arrangement model. However, we can also say that the abstract lexeme CAT, which has the basic phonological shape /kat/, is modified by addition of the -S plural marker. This sort of approach is often called the Item-and-Process approach. For a simpleminded example such as this there Morphological Theory and English Links & Letters 1, 1994 73 seems to be little difference between the two formulations. However, consider an irregular plural form such as men. What is the plural morpheme in this word? These types of cases are notoriously difficult for Item-and-Arrangement models, but submit to description under an Item-and-Process view, where we can simply say that there is an idiosyncratic pluralization process which involves changing the vowel. This sort ofsituation lessens the vaiue of the whole concept of 'morpheme'. Aronoff (1 976) argued further against the morpheme concept. He pointed out that many morphological processes seem to apply to words rather than to morphemes. A survey of these issues can be found in Scaiise (1 984) and Spencer (199 1). One of the points Aronoff makes is that it is not always possible to take the morpheme as a sign, that is, as a unit of meaning. Sometimes, we must recognize meaningless morphemes which nonetheless combine to form meaningful words. The prime example of this is a fairly large class of verbs in English of latinate origin consisting of a prefm and stem of the type commit, receive, import, detain. It seems clear that these are prefixed since the same prefixes, con-, re-, in-, de-, per-, pro-, ex-, mns- and so on recur. At the same time, a small number of recurrent stems can be identified, including -mit, -ceive, -port, -tain, -fer, -late, -duce, -mote. However, it is impossible to provide a unitary meaning, or even collection of meanings, either for the stems or the prefixes'. Yet a word such as transmit or collate has a definite meaning. Aronoff argues that these latinate stems are morphemes rather than just meaningless morphs on the grounds that they undergo distinctive ailomorphy. Thus, from -ceive we form adjectives such as receptive, deceptive, perceptive, and nominaiizations such as reception, deception, perception. Here it is clear that there is a distinct allomorph, -cept, for these formations.Now,the only thing which can exhibit allomorphy is a morpheme,ergo stems such as-ceive are morphemes,albeit meaningless ones. These and related arguments lead to the conclusion that we should regard at least certain types of morphological structure as the result of processes which stems undergo, rather than as 'things' (i.e., morphemes, which have a separate meaning and which are concatenated to form expressions with a complex derived meaning). This approach to derivationai morphology is reminiscent of the 'Word-and-Paradigm' approach to inflection. In this approach inflectionai formatives are not lexical entries (i.e., not morphemes) but are realizations or exponents of sets of morphosyntactic or morpholexicai features. This is motivated by two types of phenomenon, cumulation and extended (and overlapping) exponence. Cumulation occurs when a single formative signals more than one grammatical property or 'meaning'. For instance, in the Catalan verb form porta 'he 1. These are different from so-cded Cranberry morphs' such as the meaningless element cranin cranberry. They recur in a whole series ofwords and have a definable morphological properties, just like bona fide morphemes. 74 Links & Letters 1, 1994 Andrew Spencer l carries, is carrying', the desinence -a realizes ('is an exponent of') not just 3sg subject, but also tells us that the verb is indicative mood (cf. pres. subj. port-i and imperf. subj. port-és-0, with a zero agreement marker), and present tense (ct. port-ava, port-á, port-ar-á). Moreover, this desinence can be regarded as coding the fact that the verb is 1st conjugation (cf. perd tem, sent, serveix). Extended exponence refers to the fact that a single meaning or grammatical property can be signalled simultaneously by several different formatives in the word form. For instance, in the conditional form portaria we have an ending which under other circumstances would be theimperfect indicative of a 2ndI3rd conjugation verb. However, in conjunction with the future stem port-ar (itself related to the infinitive) the -ia ending signals conditional mood. Thus, we can't say that there is a morpheme -ia with the meaning 'conditional', nor can we say that there is a suffix -ar with this meaning. However, together the two formatives do indeed express the property 'conditional'. Considerations such as these have led morphologists such as Anderson (1982) to abandon the morpheme concept altogether (especially for inflectional morphology). Anderson argues instead that we must assume that the grammar takes a word stem together with an abstract description of the inflectional form. We must then construct a set of rules to generate the word form corresponding to that inflectional description. To give a very simple example we might have a rule for English saying that a verb marked [3sg present indicative] has a suffix /z/ attached to it. Hence, the stem untie will become untie-s. In the case of irregular forms a special rule has to be written for each word (e.g. sing [past tense] - sang (ablaut)). Diametrically opposed to this type of approach are what we might cal1 the Word Syntax approaches, which retain Halle's (1973) assumption that morphemes are lexical entries. In these approaches (e.g. Lieber, 198 1, Selkirk, 1982, Di Sciullo and Williams, 1987) a word has a constituent structure, similar in kind to the constituent structure of a sentence. Such a structure can be represented by familiar tree diagrams. (They can also be generated by a phrase structure grammar.) An example would be (l), a representation of the word indeciphera bility: Noun in de cipher abil i'7 Each morpheme has its own grammatical category. We can imagine such a tree being constructed from the root cipher outwards to give [de+cipher], [[de+cipher] +able], etc. Each added affm defines the category of the resulting word, and for this reason is regarded as the head of the word. Because of the resemblance to syntactic struckres such an approach is often referred to as Word Syntax. This approach is also supposed to apply to inflected forms. For instance, the representation for cats will be (2): (2) Noun, pl. cat S Here the plural suffix is the head of the word and thus characterizes the whole word as a plural form. Technically, the label pl.' in (2) is a morphosyntactic feature, and the noun cats is said to inherit this feature by a process of percolation. Feature percolation is supposed to take place just from heads. If we regard grammatical category labels like 'Noun', 'Adj.' as features of morphemes, then we can say that the way that -able in decipherable characterizes the whole word as an adjecive is parallel to the way that -S defines cats as a plural noun. This can be seen in (3), where percolation is shown by an arrow: (3) (a) (b) Adj N I I de cipher able cat S However, with inflection there is already a problem. Inflections do not change grammatical category, so the fact that cats is a noun is a property of the stem cat, not of the plural suffix. But this suggests that a feature ('N') has come from a non-head, cat, rather than the head. A slightly more complete example from Catalan makes this more obvious: cat S 76 Links & Letters 1, 1994 Andrew Spencer (4) v indic im p f indic impf V impf I port av a The sufix -avis the head ofthe stemportav-, and this stem therefore inherits the tense Imood Iaspect features of the suffix by percolation. However, after suffixation of -a, the stem is the non-head. The only way for the word as a whole to be marked as imperfect indicative, therefore, is for these features to percolate from the non-head. A different problem for the notion of constituent structure in words concerns the noiion of adjacency (cf. Siegel, 1977) in morphology, a specific example of a locality effect in grammar. To understand this notion, consider the case of nominalizations of verbs formed by suffixation of -ize, to give -izdtion. Now, -ize attaches to certain classes of adjectives and aiso to a smailer number of nouns, as in industrial-ize, regular-ize as opposed to motor-ize, hospital-ize. Other things being equai, it would theoretically be possible for the nominaiization rule to be sensitive to this distinction, so that, say, -ation only suffixed to deadjectival -ize verbs. In other words, the -ation rule would be able to see the boxed part of the structures of reguhrize and hospitalize as in (5), and fail to apply where the boxed section is not an adjective: 1- regular 1 ize 1 hospitai 1 ize It is an intriguing fact about morphological processes that in general they are unable to make this kind of distinction (see Raffelsiefen, 1992, for a reanaiysis of putative counterexamples). A number of authors propose principies to prevent such reference (e.g. Siegel, 1977, Williams, 198 1). However, Morphological Theory and English Links & Letters 1, 1994 77 this is unsatisfactory. Why build an elaborate tree structure only to impose a restriction preventing any of it from being accessible? Another set of problems besetting theories of morphology relying on constituent structure is that known as 'bracketing paradoxes'. A bracketing paradox arises when a morphologicai construction seems to require one constituent structure anaiysis at one leve1 and a different anaiysis at a different level. A familiar example (though still controversiai; see, e.g., Sproat, 1992) is that of the word unhappier (Pesetsky, 1985). The comparative -er suffix only attaches to a stem if the resulting word forms a single stress foot (i.e., if it has no secondary stresses). In practice, this means single syllables or disyllabic stems v. ending ih syllabic conionants or -i Phonologically, then, we Would expect unhappier to have the structure of (ha): Here, -er has attached to happy, which is permitted (hence we have happier). Then, we prefix un-. However, when we look at the semantics, we see that unhappier means 'more unhappy'. This means that the comparative morpheme takes the negative morpheme in its scope. This implies the structure (6b): These are incompatible, hence the 'paradox'. There are, in fact, a good many types of construction where this kind of anomaly arises, and there is no reason to believe they are al1 of the same kind, and should be anaiysed in the same way. Indeed, one class of bracketing paradoxes seems to be the result of a special type of word formation process in English. This is the type illustrated by the expression transformationalpmmarian. Though this can (facetiously) be interpreted to mean 'a grammarian who is transformationai', the normal interpretation is 'speciaiist in transformational grammar'. On this reading, we obtain a morphological constituent structure in (7a) which is at variance with the phonological bracketing of the phrase, (7b): (7) a. [[transformational grammar] ian] b. [transformational [grammar ian]] The justification for (7a) is that -jan means (roughly) 'speciaiist in ...', and hence modifies the entire phrase mnsformational grammax In Spencer (1 988; cf. also Spencer, 199 1, ch. 10, Carstairs-McCarthy, 1992: 92-97; and see Beard, 199 1, for a different analysis) 1 point out that this example represents a widespread feature of English in that we can form expressions meaning 'person who doeslis associated with X' for a great variety of types of X, more or less independently of the morphological structure of X. Moreover, 78 Links & Letters 1, 1994 Andrew Spencer this process (of 'personal noun formation') engenders a more drastic type of problem. Consider the following cases: (8) a. moral philosopher b. baroque flautist c. electrical engineer monumental mason historical linguist East German The problem with the exarnples in (8) is that their sources (in (9)) seem to be morphologically inappropriate in various ways. (9) a. moral philosophy b. baroque flute c. electrical engineering monumental masonery historical linguistics East Germany Yet we cannot argue that the examples of (9) are derived from those of (8) morphologically, because the specialist meaning found with (8) is obviously derived from the specialist meanings associated with (9), not vice versa. In (8a) we see a case where an a%x -y is replaced by -er, and in (8b) the stem undergoes drastic allomorphy. The most telling examples are those of (8c), where we seem to have lost an affix in the course of the derivation. Clearly, these cases show that we don't even have a paradoxical bracketing: no sensible constituent structure of any kind can be given to these examples. The solution 1 advocate is to say that there is a process of personal noun formation which is independent of the morphological processes which realize it. In this respect, we can liken personal noun formation to plural formation: (more or less) every noun in English can have a plural, even though the plural morphology varies considerably. This is effectively to say that there is a separation of the abstract morphological process from the various morphological operations which might give rise to it. In fact, the way such expressions are constructed turns out to depend on the contents of the lexicon of English. We can form a personal noun in just those cases where the head noun in its unmodified form has some sort of personal form. For instance, corresponding to linguistics we have linguist. Therefore, when we form an expression theoretical linguistics and store this as a set phrase in the lexicon, this licenses the construction of the personal noun based on its head, viz. theoretical linguist. This can be thought of as a type of productive analogical formation, as diagrammed in (10): Morphological Theory and English Links & Letters 1, 1994 79 linguistics linguist This is only possible, however, when the three licensing expressions are - - lexicalized. If ké try to construct a personal noun from a straightfonvard syntactic phrase, such as 'difficult linguistics' we are unable to form a personal noun. Hence, dzficult linguist cannot possibly mean 'one who specializes in difficult linguistics', nor can wooden jkzutist mean 'one who plays a wooden flute'. This is because the positions in diagram (10) corresponding to that of theoretical linguistics is missing for these cases. This type of phenomenon constitutes a case ofword formation (derivation) which is based on the lexicon, and can be regarded as a type of paradigmatic word formation. This means that we no longer associate a specific meaning with a specific affixal morpheme and compute the meaning of the whole word as a combination of the meaning of the stem plus that of the affix. Instead, we assume an abstract derivationprocess (personal noun formation) and a relatively complex set of morphological operations which realize this process. These operations are defined in part by the idiosyncratic properties of lexical items. In this regard, we analyze the personal noun formation process in much the same way that inflectional processes are analyzed in a Word-and-Paradigm model. The approach is essentially the same as van Marle's (1985) analysis of feminine noun formation in Dutch, and it also owes a good deai to BeardS Separation Hypothesis.' Carstairs-McCarthy (1992) and Spencer (1991) discuss further possible cases of this kind. 3. The morphology-syntax interface Having investigated the extent to which the internai structure of a word reflects the internai structure of a sentence we can now turn to the way that a word's structure can affect its syntactic properties. This is a particularly lively and contentious area of current research. A question from the morphology-syntax interface which has aroused considerable interest in the theoretical literature concerns the way that the argument structure of a verbal stem can be satisfied in and outside of compounds. Argument structure (valency) refers to the types of subjects and complements a predicate can take. Thus, intransitive (sleep), monotransitive (hit), and ditran- 2. Sturnp (1991) presents sorne interesting criticisrns of this treatrnent of personal nouns.