Inflectional morphology

3 Inflectional morphology Balthasar Bickel and Johanna Nichols Introduction The prototypical inflectional categories include number, tense, person, case, gender, and others, all of which usually produce different forms of the same word rather than different words. Thus leaf and leaves, or write and writes, or run and ran are not given separate headwords in dictionaries. Derivational categories, in contrast, do form separate words, so that leaflet, writer, and rerun will figure as separate words in dictionaries. In addition, inflectional categories do not in general alter the basic meaning expressed by a word; they merely add specifications to a word or emphasize certain aspects of its meaning. Leaves, for instance, has the same basic meaning as leaf, but adds to this the specification of multiple exemplars of leaves. Derived words, by contrast, generally denote different concepts from their base: leaflet refers to different things from leaf; and the noun writer calls up a somewhat different concept from the verb to write. That said, finding a watertight cross-linguistic definition of ‘inflectional’ which will let us classify every morphological category as either inflectional or derivational is not easy. Nor can ‘inflectional’ be defined simply by generalizing over attested inflectional systems or paradigms; the cross-linguistic variation in both forms and categories is too great. Rather, we define inflection as those categories of morphology that are regularly responsive to the grammatical environment in which they are expressed.1 Inflection differs from derivation in that derivation is a lexical matter in which choices are independent of the grammatical environment. 1 Bickel’s research was supported by grant 8210-053455 from the Swiss National Science Foundation. Nichols’s work on Ingush and Chechen was supported by NSF grant 96-16448. Some of her work on verbal categories was supported by NSF grant 92-22294. We are indebted to Fernando Zúñiga, David Peterson, Enrique Palancar, and Louis Boumans for comments on an earlier draft. This chapter was circulated in Spring 2001 on the AUTOTYP project website (http://socrates.berkeley.edu/∼autotyp). In this we follow S. R. Anderson (1992:74–85), but we extend the definition to cover not only syntactic but also more generally grammatical sensitivity, as explained below. For a different approach to the definition of inflection, based on prototype theory, see chapter 1 of this volume. 169 170 Balthasar Bickel and Johanna Nichols The relevant grammatical environment can be either syntactic or morphological. The syntactic environment is relevant, for example, when morphological choices are determined by agreement. Many languages require determiners and adjectives to agree in form with the head noun in an np, as in the following German examples:2 (1) German a. ein-e gut-e Lehrerin a.nom.sg.fem good-nom.sg.fem teacher(fem).nom.sg ‘a good (female) teacher’ b. ein-es gut-en Lehrer-s a.gen.sg.masc good-gen.sg.masc teacher(masc).gen.sg ‘of a good teacher’ Morphological choice – case, number, and gender in ein- ‘a’ and gut- ‘good’ – here depends directly on the syntactic environment, specifically on the status of these words as modifiers of a head noun. In (1a), the head noun Lehrerin has feminine gender and is inflected as nominative singular. This determines feminine nominative singular forms of the article and the adjective. In (1b), the head noun is masculine and in the genitive singular case, and this triggers masculine genitive singular forms of the article and the adjective. The choice of these article and adjective forms is thus an automatic response to the form and nature of the head noun. In contrast, the choice of derivational categories – in this example, between Lehrer and Lehrer-in – is a purely lexical matter which specifies the reference of the head noun. The effect that derivational morphology has on syntax is at best indirect, by reassigning words to different parts of the lexicon: the suffix -in, for example, reassigns Lehrer ‘teacher’ to the class of feminine nouns, and this property shows up in agreement. Note that it is not the derivational suffix -in that triggers agreement, but the more general notion of feminine gender, which mostly includes nouns without such a suffix (e.g. Schule ‘school’ would trigger exactly the same determiner and adjective forms in (1a) as Lehrerin). Other examples of inflectional categories sensitive to syntax are case assignment (government), tense choice in complex sentences (sequence of tenses), switch reference, and many more which we will review in this chapter. Often, however, inflectional categories are sensitive not so much to the syntactic environment as to the morphological environment in which they appear. As an example of this, consider aspect in Russian, which consists of a highly irregular morphological distinction between what are called perfective and imperfective verbs, e.g.: 2 See the list of abbreviations at the beginning of the volume. Inflectional morphology (2) Imperfective pisat govorit kupit delat saditsja otcvetat staret pit Perfective napisat skazat pokupat sdelat sest otcvesti postaret vypit 171 ‘write’ ‘say’ ‘buy’ ‘do’ ‘sit down’ ‘bloom’ ‘get old’ ‘drink’ That Russian aspect is inflectional is shown by the fact that it figures in a morphological rule: the future tense is formed analytically (periphrastically) if the verb is imperfective, but synthetically if it is perfective. For example, in the future tense the third person singular form of the imperfective verb pit ‘drink’ is budet pit ‘(he or she) will be drinking, will drink’, i.e. the future is expressed analytically by combining an auxiliary verb budet ‘(he or she) will’ and an infinitive pit ‘drink’. The same future tense of the perfective verb vypit ‘drink, drink up’, by contrast, is expressed by the synthetic word form vypet ‘(he or she) will drink, will drink up’. Thus, the realization of future tense forms is determined by the aspect of the verb. In other words, aspect is part of the structural context of the future tense formation rule in the same way as gender of the head noun is part of the structural context of the agreement rules illustrated by example (1) above. Again, derivational categories are different. German, for example, has verb morphology that is in many ways similar to that of Russian, and it even has pairs of verbs that look similar to the perfective versus imperfective contrast of Russian; compare Russian pit ‘drink (ipfv)’ versus vypit (lit.: ‘out-drink’), ‘to drink up, drink to the end, empty (pfv)’, and German trinken ‘drink’ versus austrinken (lit.: ‘out-drink’), ‘to drink up, drink to the end, empty’. The difference is that, in German, there is no syntactic or morphological rule that refers to this opposition: all tense forms, for example, are formed in exactly the same way. The choice between trinken and austrinken is simply a lexical one, so the difference is one of derivation. The difference between inflection and derivation often coincides with differences in morphological typology: inflection is often more transparently and more regularly marked than derivation. Also, inflectional categories are typically more general over the lexicon than derivational categories. While these are typologically significant tendencies, they are by no means necessary or universal. Russian aspect, for example, is very opaque and irregular. Sometimes, as in the example of pit and vypit above, it is marked by a prefix, but sometimes it is signalled by a stem difference or by suppletion (e.g. ipfv otcvetat vs pfv otcvesti ‘to bloom’; ipfv govorit vs pfv skazat ‘to say’). Transparency 172 Balthasar Bickel and Johanna Nichols of marking has to do not with inflection versus derivation but with the choice between what we will describe below as concatenative and nonlinear, and also with that between flexive and nonflexive morphology, structural distinctions that will be reviewed in section 1. The other frequent concomitant of inflection, generality over the lexicon, is not a necessary correlate either. It is possible for inflectional categories to be restricted to a subset of lexemes. The Nakh-Daghestanian languages Chechen and Ingush, for example, limit verb agreement to about 30 per cent of the verbs, yet the category is as sensitive to syntax as verb agreement is in English or Russian. Case morphology is sometimes different for different parts of the lexicon, e.g. following, as in some Australian languages (Silverstein (1976)), a nominative–accusative schema for pronouns and an ergative–absolutive schema for nouns; and in many languages, case paradigms are often defective (lacking some cases) for some nouns but not others. These and other examples will be discussed below. In the following, we will concentrate mainly on the formal aspects of inflection – i.e. how and where inflectional categories such as case or agreement are expressed – and on how such categories interact with syntax. The content of inflectional categories is dealt with in detail in other chapters, (see vol. i, chapter 5, on mood and illocutionary force, and chapters 4 and 5 of this volume on gender, and tense, aspect and mood, respectively), and we limit ourselves to a brief survey of those categories that are not covered or only partially covered in this work. The chapter is organized as follows. In section 1 we discuss the difference between inflectional and lexical categories, review the notion of clitic, and dissect the traditional typological parameters of morphology, i.e., phonological fusion, flexivity, and semantic density (exponence, synthesis). Sections 2 to 6 are devoted to further parameters of typological variation: the place and position of inflectional markers, paradigm and template structure, and obligatoriness of marking. In section 7 we briefly review the content of a few inflectional categories, and in section 8 we summarize some of the ways in which inflection interacts with syntax, concentrating on agreement and case marking. 1 Formatives and morphological types 1.1 Words versus formatives At the heart of inflectional morphology are what we will call formatives. Formatives are the markers of inflectional information. (In (1) above, the endings -e, -es, -en, and –s are all formatives.) They are different from words in that Inflectional morphology 173 they cannot govern or be governed by other words,3 cannot require or undergo agreement, and cannot head phrases: formatives are morphological entities, words syntactic. In the better-known Western European languages, formatives are typically realized through bound morphology and words through phonologically independent elements. Case markers (formatives), for example, are often tightly fused endings (e.g. English he versus hi+m), while adpositions, words which govern case and head PPs, are often free-standing units (e.g. with him, where with governs objective case on the pronoun). However, this need not be the case, and indeed often is not. In East and Southeast Asian languages, case formatives are generally realized in the form of phonologically free units, sometimes called ‘particles’. In Lai Chin, a TibetoBurman language of Burma, for example, phonologically bound affixes all have a CV shape (i.e. they are monomoraic or ‘light’), whereas independent words all follow a CVC or CV syllable canon (i.e. they are bimoraic or ‘heavy’). Case markers, unlike agreement prefixes, follow the pattern of words: (3) Lai Chin (Tibeto-Burman; W. Burma) Tsew Máŋ niʔ ʔ a-ka-thoʔŋ t. erg 3sg.a-1sg.p-hit ‘Tsew Mang hit me’ It is a general characteristic of these languages that the phonological notion of the word is largely at odds with grammatical considerations: not only is the case formative niʔ an independent phonological word, but so are both parts of the proper name it marks in the example (Tsew and Máŋ). It is as if the rhythmical articulation of speech goes its own ways – ways that are quite distinct from the conceptual and syntactic segmentation, in which for instance Tsew Máŋniʔ is a single, indivisible unit (a single grammatical word, as we will see). Turning to words in the sense of syntactic units, we find variation in their phonological independence no less than for formatives. While words are often realized as free morphemes, many languages allow them to be (morpho-) phonologically incorporated into other words, and a number of languages have large sets of what are called lexical affixes which have their own syntactic properties (e.g. assigning specific cases and semantic roles to nps in the clause). These are all issues of derivational morphology and compounding and are discussed in chapters 1 and 6 of this volume. Another common instance of 3 We use the term govern in the traditional sense of determination by one word of the grammatical form (i.e., the inflectional categories) of another. For instance, English prepositions govern the objective case of pronouns: with me and not *with I. Russian prepositions lexically govern different cases on their objects: s ‘with’ takes the instrumental (s drugom (with friend.instr), ‘with a friend’), bez ‘without’ takes genitive (bez deneg (without money.gen), ‘without money’), and so on. In contrast to agreement, the governed category is not contained in the governing word: contrast (1) above, where the gender is contained in the head noun that triggers gender agreement (Lehrer is masculine, Lehrerin feminine). 174 Balthasar Bickel and Johanna Nichols phonologically bound words is cliticizing adpositions. This is a widespread phenomenon, for instance, in Slavic and Indo-Aryan languages. Many Russian prepositions, for example, are proclitic and behave much like prefixes: they are subject to word-internal voicing and pretonic vowel reduction rules, e.g. ot=druga4 ‘from friend:gen.sg’ is realized as [addrugə ], just as the singleword expression otdaj ‘give back’ is realized as [addaj]. That prepositions are grammatical words on their own, however, is still evident from the fact that they govern case, cf. ot=druga ‘from (the/a) friend’, with ‘friend’ in the genitive, versus s=drugom (phonetically, [zdrugə m]) ‘with a friend’ where ‘friend’ is in the instrumental case. Yet another instance of a phonologically bound word arises from incorporation, to which we will briefly return below. Words often develop into formatives through grammaticalization. It is no surprise, therefore, that there are many transitional cases where the distinction between, e.g., pronouns and agreement formatives, or between adpositions and case markers, is blurred. See Hopper and Traugott (1993) and Lehmann (1995) for surveys of grammaticalization phenomena. 1.2 Clitics As we saw in the preceding section, the word versus formative distinction is a purely syntactic one and crosscuts the phonological difference between free and bound units. The traditional notion of a word conflates the syntactic and phonological criteria: it implies that words are both syntactically and phonologically independent units and that affixes are in both respects dependent units. With regard to the word, a distinction is often made between grammatical word (in our terms, word as opposed to formative) and phonological (or prosodic) word (free as opposed to bound unit). The same distinction could be made for affixes as well: a grammatical affix would be a formative, a phonological affix any bound unit (a bound formative, a lexical affix, an incorporated noun, etc.). However, for most practical purposes it is safe to talk about formatives and affixes without qualification. ‘Formative’ then refers to any inflectional exponent whether bound or free, and ‘affix’ refers to any bound unit whether grammatical or lexical. A third notion besides word and affix that is often invoked is that of clitic. The term is used in two quite different senses. In one sense, clitics are simply phonologically bound words, i.e., syntactic units like the Russian prepositions that, as we saw above, are phonologically dependent on their objects. In the other, typologically more important but often less straightforward, sense, clitics are categorially unrestricted bound formatives, 4 Here and in the following, we mark clitic boundaries by ‘=’; affix boundaries are marked by hyphens. Inflectional morphology 175 i.e., formatives that are unrestricted as to the syntactic category of the word they attach to. In this they contrast with affixes, which are usually more selective in what host they take. Case affixes, for example, are usually restricted to nominals, tense affixes to verbs. A clitic like the Turkish interrogative =mi (and its vowel-harmonic variants), by contrast, attaches to whatever word it marks as a question, regardless of that word’s syntactic category, e.g. sen=mi ‘me?’ (pronoun), yarn=m ‘tomorrow?’ (adverb), or gördün=mü ‘did you see?’ (finite verb: gör-dü-n ‘see-past-2sg’). An important way in which formatives can come to be categorially unrestricted is that they can be affixed to phrases (constituents) rather than to words, and then it does not matter what kind of word happens to be in the place at the edge of the phrase where the formative is attached. A classic example is the English genitive -s, which is suffixed to the right edge of an np regardless of what element is found there. The rightmost word can even be a verb form, as in examples like [np [np a guy you [V know]]’s idea]. In many languages, this pattern is more general, comprising all case markers. In the Papuan language Kâte, for example, case formatives cliticize to any word that ends an np (np-final words are boldfaced): (4) Kâte (Finisterre-Huon; Papua New Guinea; Pilhofer (1933)) a. [np e=le fiʔ ]=ko mi fe-naŋ ! 3sg=dest house=adl neg climb-1pl.hort ‘Let’s not climb into his house!’ (p. 113) b. [np ŋ iʔ moʔ -moʔ =sawa]=tsi e-mbiŋ man indef-indef=restr= erg do-3pl.rem.pt ‘Only some of the men did it’ (p. 110) c. [np ŋ iʔ wiaʔ e-weʔ ]=tsi dzika ki-tseyeʔ man thing do-3sg.rem.pt=erg sword bite-3sg.rem.vol ‘The man who did these things should be killed’ (literally ‘should bite the sword’) (p. 142) In (4a), the adlative =ko is cliticized to a noun; in (4b), the ergative =tsi is attached to an indefinite pronoun which already hosts another clitic (=sawa ‘only’); and in (4c), we find the same ergative marker on a finite verb form, indicating the function of the internally headed relative clause.5 Another common type of phrasal clitic is bound articles (determiners, specifiers) that attach not only to nominals but also to verb forms, where they function as nominalizers or relativizers. This phenomenon is particularly common in many North and Central American languages. Phrasal clitics typically have scope over the whole np they are attached to, i.e. they modify the whole np expression although formally they are not copied 5 See vol. ii chapter 4 for more on relative clauses. 176 Balthasar Bickel and Johanna Nichols onto each element. The ergative in (4c), for example, specifies that the whole expression ‘man who did these things’ is an agent, but, formally, the ergative appears only on the last element (eweʔ ‘did’). Phrasal scope is an important issue in np morphosyntax and we will return to it in section 8.2. However, it is important to note that, while phrasal scope is a common concomitant of clitics, this property is not a sufficient criterion for clitichood. To decide whether something is a clitic, it is imperative to carefully analyse the category structure of the language. An element is a clitic only if it can attach to hosts of diverse categories. In all of the preceding examples of clitics, they attach directly to the phrase or word they modify. However, since clitics are category-neutral, this is not a necessary condition. Clitics can also be detached from the element they modify. In North Wakashan languages, for example, case formatives (=i ‘subject’, =x.a ‘object’, =sa ‘instrumental’) and determiners (=da) regularly attach to the preceding phrase: (5) Kwakw’ala (Wakashan; NW America; S. R. Anderson (1985b)) nep’id=i=da gə nanə m=x.a gukw =sa t’isə m throw=subj=det child=obj house=instr rock ‘The child threw a rock at the house’ Here the instrumental formative on ‘rock’ is cliticized to ‘house’, whose object marker is in turn cliticized to the preceding word ‘child’. While uncommon, such patterns are also occasionally attested in Australian languages (Evans (1995b)). Some languages have detached clitics whose position appears to be syntactically unconstrained: they can attach to any constituent in the clause, depending on the information structure. Such is the case in Tsakhur, discussed by Kibrik (1997), where the auxiliary complex =wod can adjoin to any of the three words in (6). If the clitic attaches to an np, that np is focussed (indicated by small caps in the translation). If the clitic follows the verb, the entire proposition is focussed. (6) Tsakhur (Nakh-Daghestanian; NE Caucasus; Kibrik (1997:306)) a. MaIhaImaId-e Xaw alyaʔ a =wo=d m.-erg house(iv):nom build =aux=iv ‘Muhammed is building a house’ b. MaIhaImaId-e Xaw =wo=d alyaʔ a m.-erg house(iv):nom =aux=iv build ‘Muhammed is building a house’ Inflectional morphology 177 c. MaIhaImaId-e =wo=d Xaw alyaʔ a m.-erg =aux=iv house(iv):nom build ‘Muhammed is building a house’ A similar situation is found in the Tibeto-Burman language Belhare, where the reported speech marker =phu/=bu can occur after any part of speech in the clause, sometimes even on two at once (Bickel (2003)). While Tsakhur and Belhare illustrate unconstrained clitic placement in the clause, some languages spoken in the Kimberley region of Australia exemplify the same pattern on the np level. Case markers in these languages can appear on any element of the np, whether it is the head or not: (7) Gooniyandi (Bunuban; NW Australia; McGregor (1990:227))6 a. ngooddoo=ngga garndiwiddi yoowooloo that=erg two man ‘by those two men’ b. marla doomoo=ngga fist clenched=erg ‘by a fist’ The most frequent position for detached clitics, however, is what is traditionally called the Wackernagel position (named after the famous Indo-Europeanist who first described the phenomenon in 1892). This position is especially common for clause- and verb-level inflectional properties such as tense, mood, and agreement. In the best-known examples, the Wackernagel position is right after the first accented phrase or subconstituent of it. This is characteristic, for instance, of South Slavic, Wakashan, and many Uto-Aztecan languages: (8) 6 Luiseño (Uto-Aztecan; S. California; Steele (1976)) a. ʔ iviʔ ʔ awaal =up waʔ i-q dem dog =3sg.pres bark-pres ‘This dog is barking’ b. ʔ iviʔ =up ʔ awaal waʔ i-q dem =3sg.pres dog bark-pres ‘This dog is barking’ c. hamuʔ =up wiiwiš kwaʔ -q already =3sg.pres w. eat-pres ‘She is already eating her wiwish’ McGregor (1990) calls the case clitics ‘postpositions’ because they have phrasal scope. As discussed above, we restrict the term adposition to syntactic words, which govern case and head adpositional phrases. See section 8.2 for further discussion. 178 Balthasar Bickel and Johanna Nichols In (8a), the tense- and agreement-indicating clitic =up attaches to the first np, in (8b) to the first subconstituent of this np. Example (8c) shows that the host phrase need not be an np, but can just as well be an adverbial phrase. In Luiseño, and also in South Slavic languages not illustrated here (but see Spencer (1991:355ff.)), the definition of the Wackernagel position rests on the prosodic criterion of accent: the first accented string, whether constituent or word. In other languages, the Wackernagel position is defined syntactically and limited to complete phrases. As a result, in such languages clitics cannot attach to subconstituents of phrases. In Warlpiri, a Central Australian language, clitics occur after the first complete syntactic phrase: (9) Warlpiri (Pama-Nyungan; C. Australia; Hale, Laughren, and Simpson (1995); T. Shopen (p.c.)) a. kurdu yalumpu-rlu =ka=jana jiti-rni jarntu wita child dem-erg =pres[3sg.a]=3pl.p tease-npt dog little b. jarntu wita =ka=jana jiti-rni kurdu yalumpu-rlu dog little =pres[3sg.a]=3pl.p tease-npt child dem-erg c. jiti-rni =ka=jana jarntu wita kurdu yalumpu-rlu tease-npt =pres[3sg.a]=3pl.p dog little child dem-erg ‘The child is teasing the little dogs’ In all of these examples, the clitic complex =ka=jana follows the first constituent (nps in (9a,b), a verb in (9c)), but it would not be possible for the clitics to follow part of a constituent, e.g. kurdu ‘child’ or jarntu ‘dog’ alone in (9a) and (9b), respectively. On the level of phrases, second-position clitics are found in Wakashan languages of North America. In Nuuchahnulth (previously known as Nootka), for example, determiner phrase (DP) formatives like the definite article =ʔi often follow the first word of the phrase they modify: (10) Nuuchahnulth (Wakashan; NW America; Nakayama (1997)) a. hin=a či
[DP minwaʔ ath=ʔ i] (p. 190) there:mom=go.out.to.meet British.soldier=def ‘They went out there to meet the British soldiers’ b. ʔ u-ch.i=n
[DP
u=aq=ak=ʔ i h.a kw a
] (p. 107) her-married.to=mom nice=very=dur=def girl ‘He got married to the very beautiful girl’ Since in (10a) the head noun minwaʔath ‘British soldier’ is the only word in its DP, the article cliticizes to this word. In (10b), however, the article is found on the preceding modifier
u=aq=ak ‘very nice’ because this is now the first Inflectional morphology 179 word in the DP. (Note, incidentally, that the pattern is the same on the clause level: aspectual formatives like =n
‘momentaneous’ and entire words like =ači
‘go out to meet’ are clitics in the clausal Wackernagel position.) Wackernagel formatives are typically clitics, but not always. In many Kru languages of Western Africa, for example, negation is marked by a phonologically free, tone-bearing second-position particle ni: (11) Bete (Kru; Ivory Coast; Marchese (1986:197)) ná dı̄bà ni fl lı̄ kɔ̀kɔ my father neg eat chicken ‘My father doesn’t eat chicken’ Similarly, what are traditionally called clitics in Tagalog are mostly free formatives in the Wackernagel position: as phonologically independent units, they do not lose stress or show any other reduction that is associated with phonological affixes or clitics (S. R. Anderson (1992:204)). As illustrated by the following example, pronominal ‘clitics’ like siya ‘he’ are fixed in their Wackernagel position: (12) Tagalog (Schachter and Otanes (1972:183)) a. nakita siya ni Pedro saw:p.voice 3sg.nom gen p. ‘Pedro saw him’ b. *nakita ni Pedro siya saw:p.voice gen p. 3sg.nom ‘Pedro saw him’ Despite this special positioning, pronouns like siya are phonologically independent words, not clitics. Free Wackernagel formatives often develop into bound clitics. Indeed, after pronouns, the Bete negation particle (see (11) above) reduces to a high tone clitic, which triggers vowel lengthening so as to have a place for realization (i.e., `ɔ = } is realized as `ɔɔ). (13) Bete (Marchese (1986:197)) ním 3sg=neg drink ‘He doesn’t drink’ ɔ̀= In some languages, there is considerable variation in the phonological dependence of Wackernagel formatives. Consider the following examples from Toura, a Mande language spoken in the same area as Bete: 180 Balthasar Bickel and Johanna Nichols (14) Toura (Mande; Ivory Coast; Bearth (1971)) a. nέ ké ló-ı̀ı̄ boı́ child ind go-progr field ‘The child is going to the field’ b. nέ=` lò boı́ child=act go.decl field ‘The child goes to the field’ c. kó ló boı́ 1pl.opt go field ‘Let’s go to the field’ Interacting with verbal morphology, the Toura detached formatives express a variety of tense–aspect and modal notions and are placed in the Wackernagel position. Some of the formatives, such as the indicative mood particle ké in (14a), are phonologically free. Others, e.g. the ‘actual’ (‘act’) mood marker in (14b), are tonal clitics. After pronominal subjects, mood-indicating formatives are completely fused with their host (14c): compare kó ‘we (optative)’ in (14c) with such forms as kwéé ‘we (actual, resultative)’ or kwéè ‘we (actual, ingressive)’. 1.3 Degree of fusion In the preceding section we noted that formatives are often phonologically fused to their host, and that there is a gradient in how tightly they are fused. This is a general characteristic of morphology, and it is suitable here to set up a scale of phonological fusion:7 (15) Fusion isolating > concatenative > nonlinear 1.3.1 Isolating At one end of the spectrum is complete isolation, where formatives are fullfledged free phonological words on their own. This is common in many Southeast Asian languages, and we saw an example in the Lai Chin ergative case marker in (3) above. Most languages, however, have at least some isolating formatives or ‘particles’. They are particularly frequent as markers of negation, mood, and various evidential and illocutionary categories (conveying such notions as the source of evidence or the firmness of assertion). 7 The scale is also useful in derivational morphology, cf. chapter 1 in this volume. Inflectional morphology 181 1.3.2 Concatenative (bound) Concatenative8 formatives are phonologically bound and need some other word for their realization. They include inflectional desinences as well as cliticized formatives. The hallmark of concatenation is that formatives are readily segmentable. The paradigm example is Turkish number and case formatives, e.g. ad-lar ‘name-pl’, ad-n ‘name-gen’, ad-lar-n ‘name-pl-gen’, where each formative is a clear cut sequence of phonological segments. In this regard, concatenative formatives are similar to isolated (independent) formatives. However, unlike these, concatenative formatives trigger some phonological and morphophonological adjustments in the word they build up together with their host – and the more such adjustments there are, the tighter the degree of fusion. In Turkish, a well-known phonological adjustment is vowel harmony: when the stem vowels have front instead of back articulation, the affixes follow suit: cf. el-ler ‘hand-pl’, el-in ‘hand-gen’, el-ler-in ‘hand-pl-gen’ versus ad-lar, ad-n, ad-lar-n just above. Another, cross-linguistically very frequent, concomitant of concatenative morphology is assimilation. This involves the spreading of phonological features across formative boundaries and can be illustrated by another example from Turkish: the past tense marker -ti assimilates in voice to the preceding consonant, cf. git-ti ‘go-past’ versus gel-di ‘come-past’. Dissimilation, i.e. prohibition against the same features in adjacent segments, is less common. An example is found in Belhare, where the coronal glide in the non-past marker -yu forces a preceding /t/ to lose its coronal point of articulation. As a result, this stop is realized by the default consonant of the language, the glottal stop; cf., e.g., khaʔ-yu ‘s/he’ll go’ from khat- ‘go’ and -yu ‘nonpast’. Another process sometimes affecting concatenative formatives is elision. In Turkish, for example, stem-final /k/ is deleted in polysyllabic words when followed by a vowel-initial suffix: e.g. çocuk-un ‘child-gen’ is realized as /çocu n/. Vowels are particularly prone to elision. In Belhare, for example, /i/ regularly deletes before /u/, cf. -chi-u → ch-u in tar-he-ch-u-ŋ a ‘bring-past-du-3p-[1]excl’, i.e. ‘we (two, without you) brought it’, with plain -chi in ta-he-chi-ŋa ‘come-pt-du-[1]excl’, i.e. ‘we (two, without you) came’. A final type of effect to be noted results from general prosodic constraints. Often, epenthetic elements are inserted when the concatenation of an affix would result in a structure that violates the language’s syllabic 8 An alternative term is agglutinative, but, as we will see in section 1.4 below, this term traditionally has connotations that go far beyond phonological boundness. We avoid the simpler term bound because it is already functionally overloaded in other parts of grammatical description. 182 Balthasar Bickel and Johanna Nichols templates. In the Austronesian language Lenakel (spoken in Vanuatu), for example, a prefix–stem sequence like r-va ‘3sg-come’ is broken up by an epenthetic vowel /-i/ so as to fit into the CV(C) syllable canon of the language, resulting in r-iva ‘s/he came’. Where the syllable canon is satisfied, there is no epenthesis, cf. r-imarhap-ik ‘s/he asked’ from r--im-arhap-ik ‘3sg-past-ask’ (Lynch (1978)). Prosodic constraints can also lead to the truncation of extrasyllabic material. The Belhare temporary aspect marker -hett, for example, is reduced to -het unless there is some additional suffix whose syllable onset the second /t/ could form: cf. ta-het ‘come-temp’, i.e., ‘s/he is coming’ versus ta-hett-i ‘come-temp-1pl’, i.e., ‘we are coming.’ 1.3.3 Nonlinear Despite (morpho)phonological adjustment rules that blur formative boundaries, concatenation results in linear strings of segmentable affixes. Nonlinear formatives, in contrast, are not segmentable into linear strings but are instead realized by direct modification of the stem, i.e. by a simultaneous realization of formative and stem. The best-known instance of this is morphology in Semitic languages. In Modern Hebrew, for example, inflected word forms are the result of superimposing on a consonantal skeleton (e.g. g-d-r ‘enclose’) various vocalisms indicating tense, mood, or voice: e.g. a-a ‘active’ (gadar ‘he enclosed’) versus u-a ‘passive’ (gudar ‘he was fenced in’), or -o- ‘future, imperative’ (gdor ‘enclose it!’) (Glinert (1989)). Similar in nature but more common is the superimposition of prosodic formatives (tone, stress, length) onto word stems. Many Bantu languages, for example, distinguish temporal and modal values by purely tonal patterns. In Kinyarwanda (Overdulve (1987)), one set of subordinate verb forms (called ‘conjunctive’, used mainly for complement and adverbial clauses) is distinguished from indicative forms by high tone on the agreement-marking prefix, another set (‘relative’, used mainly for relative clauses) by high tone on the last stem syllable: cf. conjunctive múkora ‘that we work’, relative mukorá ‘which we work (at)’, and indicative mukora ‘we work’ (all with agreement prefix mu- ‘1pl’). A different type of non-concatenative formative involves substitution or replacement of a stem segment. Replacive formatives are common, for instance, in Nilotic languages, where the plural of nouns is often formed by replacing the stem-final vowel by one of a set of plural-marking endings, e.g. in Lango (Lwo; Uganda; Noonan (1992)): bùrâ ‘cat’ versus bùrê ‘cats’, or láŋô ‘Lango’ versus lə́ŋ´ ‘Langos’. This is sometimes accompanied, as the latter example shows, by tonal substitutions and ablaut. In Ute (Uto-Aztecan; Givón (1980)), substitution of an individual phonological feature is recruited for case marking, cf. nominative ta’wáci
‘man’ with devoicing of the final vowel versus accusative ta’wáci ‘man’ without devoicing. Inflectional morphology 183 Still another type of nonlinear formatives is subtractive formatives. This is a rare phenomenon, but it is attested in the morphology of aspect in Tohono ‘O’odham (previously known as Papago; Uto-Aztecan; S. California; Zepeda (1983:59–61)), e.g. him (ipfv) vs hi (pfv) ‘walk’, hink (ipfv) vs hin (pfv) ‘bark’, ʔeipig (ipfv) vs ʔeip (pfv) ‘peel’, med. (ipfv) vs me (pfv) ‘run’, etc. Each perfective form is derived from the imperfective by subtracting whatever happens to be the final consonant. (In some cases, a side effect of this is compensatory lengthening of the root vowel.) A final type of nonlinear formatives to be mentioned is reduplication. An example of this widespread phenomenon is given by Ancient Greek perfect tense forms. Under reduplication, the first consonant of the stem is repeated together with a supportive vowel /e/, e.g. dé-deikha ‘I have shown’ from deı́knūmi ‘I show’, me-mákhēmai ‘I have fought’ from mákhomai ‘I fight’, dé-drāka ‘I have done’ from dráō ‘I do’, etc. Reduplication can also be analysed as the prefixation of a syllabic skeleton Ce-, where the value of C is determined by the stem. On such a view (especially prominent in the theory of Prosodic Morphology; McCarthy and Prince (1995)), reduplication would be a (very tightly fused) concatenative affix rather than a nonlinear formative: the Ceskeleton would be a well-segmentable prefix and the value of C would result from a simple phonological spreading rule, similar in fact to consonant harmony. Either way, it is evident that reduplication involves a tighter interlacing of formative and stem material than what is common in canonical exemplars of concatenative morphology. The degree of fusion is not as high, however, as with the other subtypes of nonlinear fusion, and on the scale of fusion in (15), reduplication holds a position between concatenative and nonlinear morphology. This completes the scale of fusion. It is important to note that the scale applies to individual formatives, or sets of formatives, and not, as is sometimes suggested, to languages as wholes. Isolating formatives, for example, are found almost everywhere: virtually all languages have at least a few phonologically unbound particles, regardless of the kind of formatives they employ in the rest of their morphology. But mixtures of formative types can also be more intricate. For instance, while in Arabic and Kinyarwanda most verbal categories (aspect, mood, etc.) are expressed by nonlinear formatives, person and number inflection is realized through concatenative affixes in both languages. Given such distinctions, it clearly makes little sense to talk about concatenative or nonlinear languages per se. However, languages differ in the degree to which they employ one or the other type of formative, and from this point of view, Kinyarwanda is more nonlinear, as a whole, than, say, Turkish, which has only rudimentary and non-productive traces of nonlinear morphology borrowed from Arabic (Lewis (1967: esp. 27f.)). 184 Balthasar Bickel and Johanna Nichols 1.4 Flexivity (variance, lexical allomorphy, inflectional classes) Another important parameter along which formatives vary typologically is flexivity. Flexive9 formatives come in sets of variants called allomorphs. Allomorphs are selected on lexical, i.e. item-based, principles. One example is Lango plural marking discussed above: some nouns take endings in -ê, some in -ı́, and so on. Conservative Indo-European languages have sets of case allomorphs which are selected depending on the declension class to which a noun belongs. Thus, the Latin nominative singular formative is -s after most nouns, but some nouns select an ending in -m (most of what are called the neuter o-stems) and yet other nouns have a zero ending (the a-stems, among others); cf., e.g., diē-s ‘day’ versus v¯nu-m ‘wine’ versus poēta-ø ‘poet’. Instead of the formatives themselves, it can also be the stems that show item-based alternations in flexive morphology. In German, for example, some verbs show characteristic ablaut or umlaut patterns, where person- and tenseindicating formatives trigger different vocalisms. From tragen ‘carry’, we get first person singular present trage ‘(I) carry’, second person singular present trägst ‘(you) carry’, and third person singular past trug ‘(s/he/it) carried’, each with different stem vowels. The set of verbs exhibiting such alternations is lexically restricted (to what are traditionally called ‘strong’ verbs). Thus, other verbs (called ‘weak verbs’), such as nagen ‘gnaw’, show forms like nage (1st sing. pres.), nagst (2nd sing. pres.) and nagte (1st sing. past) without stem alternation. A similar but more complex example of this is provided by Dumi, a Tibeto-Burman language of the Himalayas (van Driem (1993)). In this language, verbs divide into eleven conjugation classes, each characterized by a distinct ablaut pattern. A selection is illustrated in table 3.1. Verbs of conjugation class ii (e.g. dzeni ‘to speak’ in table 3.1) have one stem form in the first person singular and another one in the first person dual and plural non-past. Verbs of class iii (e.g. botni ‘to shout’) have also two stems, but in this case it is the first person singular and dual that share the same stem, distinct from the first person plural. Verbs of class iv (e.g. l n ‘to commence’) have three different stem forms. Conjugation and declension classes are an important and frequent characteristic of inflectional paradigms, and we will return to them in section 4.1. The hallmark of flexive formatives is that their variation is item-based, i.e. allomorphs are selected by some lexical contexts but not others. Some 9 The original, nineteenth-century term is ‘(in)flectional’ (German flektierend), but this term is also (and nowadays more commonly) used in opposition to ‘derivational’ rather than as a concept in morphological typology. To avoid confusion of ‘flexive’ and ‘inflectional’, we use flexivity (rather than ‘flection’) as the abstract noun. Comrie (1981a) suggests ‘fusional’ but this conflates flexivity with phonological fusion, a distinction for which we argue below. Inflectional morphology 185 Table 3.1 Dumi nonpast verb inflection (selection) 1sg 1du.incl 1du.excl 1pl.incl 1pl.excl ii: dzeni ‘speak’ iii: botni ‘shout’ iv: l n ‘commence’ dze-tə dzi-ti dzi-t dz -k t dzi-kta bus-tə bus-ti bus-t boʔ-kti boʔ-kta lo-tə lu-ti lu-t-i l -kti l -kta stem forms are selected by one formative but not another, or some forms of formatives are selected by some words but not others. In contrast, nonflexive formatives are invariant across the lexicon and do not trigger formative-specific or lexeme-specific stem alternation.10 The kind of variation they show is due to general morphophonology or phonology: examples are Turkish vowel harmony and Belhare dissimilation, discussed in section 1.3.2 above. Note that the distinction between flexive (item-based, allomorphic) and nonflexive (general, morphophonological) variation is independent of whether the variationtriggering context is defined morphologically or phonologically (see Kiparsky (1996)). Examples of morphologically triggered allomorphy were discussed in the preceding paragraphs. An example of phonologically triggered allomorphy comes from Warlpiri. The Warlpiri ergative desinence is -ngku after disyllabic stems (cf. kurdu-ngku ‘child-erg’) and -rlu after longer stems (cf. nyumpala-rlu ‘you(dual)-erg’: Nash (1986)). Although the triggering context is phonologically defined, the allomorphy does not result from a general phonological rule that systematically associates the number of syllables with the choice between /ngk/ and /rl/; the variation depends on a binary division of the lexicon into two inflectional classes, and the formative is thus flexive.11 Since the nineteen century, morphological typology has tended to integrate these various differences into a single scalar hierarchy: (16) isolating > agglutinative > flexive > nonlinear (or introflexive) These have generally been presented as whole-language typologies, with prototypical examples probably being (respectively): (17) 10 11 Chinese > Turkish > Latin > Arabic Apart from irregular verbs; nearly every language has a few irregular or exceptional stems whose forms do not follow the morphological rules, but these are not at issue here. This kind of phonologically defined inflectional class distinction is common in many Australian languages. Examples from Papuan languages are discussed in detail by Aronoff (1994). 186 Balthasar Bickel and Johanna Nichols This scale conflates the concatenative/nonlinear and flexive/nonflexive parameters. However, from a broader typological perspective, flexivity is orthogonal to fusion, and all possible combinations of values on the two parameters are attested, although not all are equally common. The commonest combination is flexive–concatenative (and the traditional notion of flexive or ‘(in)flecting’ is often restricted to just this combination). Latin and Dumi illustrate this type: while they display lexical allomorphy of stems and/or formatives, the formatives are all more-or-less well-segmentable affixes, undergoing various (morpho-)phonological rules. Latin case declension, for example, shows various patterns of assimilation and elision. Thus, the Latin nominative singular allomorph -s triggers regular (pan-lexical) voicing assimilation (e.g. leks ‘law’ from leg-s), vowel raising (lupus ‘wolf’ from lupo-s), and simplification of consonant clusters (dens ‘tooth’ from dent-s). Likewise, Dumi stem–suffix boundaries are subject to various morphophonological adjustments (van Driem (1993:91–5)): an example in the paradigm selection in table 3.1 is the stem-final glottal stop in boʔkti ‘we (incl.) shout’ and boʔkta ‘we (excl.) shout’ which is a regular morphophonological variant of /t/ before /k/ (cf. infinitive bot-ni ‘to shout’). Flexive–nonlinear formatives are abundant in Afroasiatic languages, especially in Semitic languages, and the prominent role that these languages played in early typology has motivated the label introflexive for just this combination of parameter values. In Semitic languages, the verb lexicon is compartmentalized into several inflectional classes traditionally called binyanim (singular binyan), and these classes determine much of the allomorphy of agreement and tense–aspect morphology. In Modern Israeli Hebrew (Glinert (1989); Aronoff (1994); Orin Gensler (p.c.)), for example, the past versus future opposition is expressed by different vowel and consonant alternations dependent on the binyan (as well as on subclasses of these): cf. gadar ‘he enclosed’ and yi-gdor ‘he will enclose’ in the first binyan versus kipel ‘he folded’ versus ye-kapel ‘he will fold’ in the second binyan. In the first (subclass of the first) binyan, past is characterized by a-a and future by -o- vocalism, while in the second binyan, past has i-e and future a-e vocalism. In addition to this, there is allomorphy of the agreement prefixes in the future tense: yi- in the first, ye- in the second binyan. (In the past tense, third person masculine agreement is zero-marked.) Flexive–isolating formatives are by far the rarest combination, which is to say that lexical allomorphy is much more common within phonological (prosodic) words than across phonological word boundaries. But examples are found in some Pama-Nyungan languages in Australia. Yidi has a set of suffixed formatives which Dixon (1977) calls non-cohering because they constitute their own phonological word, i.e. are isolating. Some of these are at the same time flexive since they show lexical allomorphy based on verbal conjugation class: Inflectional morphology 187 the verbal comitative,12 for example, has two allomorphs, -ŋa ∼ lmaŋa. The disyllabic allomorph is selected by what are called l- and r-stems, and it commences its own phonological word, cf., e.g., [word magil][word maŋ a l] from magi-lmaŋa-lnyu ‘climb.up-appl:com-pt’. The phonological autonomy of the formative is shown by the fact that it counts as its own domain for (i) stress assignment rules, according to which primary stress falls on the first or the first long-vowelled syllable of the word, and (ii) two rules that operate only in phonological words with an odd number of syllables: a penultimate lengthening and a final syllable reduction rule, both operating here on the trisyllabic sequence [ma.ŋ al.nyu], which is reduced to [ma.ŋ a l]. Another example comes from the Mesoamerican language Sierra Otomı́, in which tense–aspect, person, and sometimes deixis are marked in a phonologically free formative that precedes the lexical verb word. These formatives show flexivity conditioned by four lexical classes of verbs: (18) Sierra Otom́ (Otomanguean; Mexico; Enrique Palancar (p.c.), from Voigtlander and Echegoyen (1985)) dı́ pε̌ʔ tsʔ i ‘I keep (it)’ (conjugation class i) dı́n nú ‘I see (it)’ (conjugation class ii) dı́dı́ hóki ‘I fix (it)’ (conjugation class iii) dı́dı́m pε‘pfi ‘I work’ (conjugation class iv) 1sg.prs [verb] n o n f l e x i v e i s o l at i n g : Nonflexive formatives are often isolating; and the most common type of isolating formative is nonflexive. An example is case in Lai Chin as in (3). In Lai Chin there is no allomorphic variation for the ergative marker niʔ; it is the same for any noun in A function. n o n f l e x i v e c o n c at e n at i v e : When nonflexive formatives are concatenative, they are traditionally called agglutinative. This combination of parameter choices is also very common, one of the best-known examples being Turkish morphology, discussed above in section 1.3.2. n o n f l e x i v e n o n l i n e a r : Finally, nonflexive nonlinear formatives are common with suprasegmental (tonal or accental) morphology. An example is Kinyarwanda tense and mood inflection, as discussed in section 1.3.3. In the discussion of fusion, we noted that languages sometimes use concatenative techniques for some categories and nonlinear techniques for others. Similar splits are found in flexivity. Thus, while Russian case desinences are mostly dependent on lexical declension classes and are therefore flexive (e.g. 12 The suffix has an applicative function, turning a comitative np into a direct object. Dixon classifies this form as derivational, but on our criterion it is inflectional because its occurrence is an obligatory response to at least some syntactic environments. An example where -ŋ a is used in response to such an environment appears in (63) below. 188 Balthasar Bickel and Johanna Nichols dative sing. in -u with o-stems like stol-u ‘table’, but in -e with a-stems like kryš-e ‘roof’), the dative, instrumental, and locative plural formatives are invariant, nonflexive formatives (e.g. dat. pl. stol-am ‘table’, kryš-am ‘roof’).13 1.5 Semantic density The difference between flexive and nonflexive is often conflated with the question of whether grammatical and semantic categories are realized through separate formatives or whether they accumulate in a single formative, i.e with the question of the semantic density of formatives. However, there is no logical necessity for flexivity or, for that matter, phonological fusion (concatenative versus nonlinear) to covary with semantic density (cf. Plank (1999)). There are two dimensions of semantic density that need to be distinguished. One is density on the level of the formative. This is traditionally called exponence. The other dimension is density on the level of the word. This is traditionally called synthesis. (For more on semantic density of words see Talmy, chapter 2 in this volume.) 1.5.1 Exponence Exponence refers to the degree to which different categories, e.g. number and case, or person and tense, are grouped together in single, indivisible formatives. Two prototypes are typically distinguished: cumulative and separative formatives. Cumulative formatives are common in Indo-European, where number and case, for example, are most often cumulated into a single set of formatives. Thus, in Russian one gets gen. sg.-a ∼ -i, but gen. pl. -ov ∼ -ø ∼-ej (allomorphs dependent on lexical declension class), where there is no correspondence whatsoever between categories and parts of formatives (segments), i.e. no part of, say, genitive plural -ov that can be identified with genitive case or plural number. A concept related to cumulative formatives is portmanteau formatives. Like cumulative formatives, portmanteau formatives express more than one category, but each of the categories expressed corresponds to a separate formative that also exists in the language. For example, the French portmanteau form du ‘of the’ has corresponding formatives de ‘of’ and le ‘the’. By contrast, there are no case-only or number-only formatives corresponding to the cumulative genitive formatives of Russian. The opposite of cumulative formatives is separative formatives. Separative formatives encode one category at a time. In Turkish, for instance, case and number are, as we saw, each expressed by their own suffix, e.g. gen. sg. 13 Such splits are not random. See Plank (1999) for a preliminary survey. Inflectional morphology 189 -in, gen. pl. -ler-in (all with vowel-harmonic alternations). There is some tendency for nonflexive concatenative (‘agglutinative’) morphology to go with separative exponence as in these Turkish examples and for flexive formatives to be cumulative as in Latin or Russian, but this need not be so. The Turkish first person plural ending -k (as in gör-dü-k ‘see-pt-1pl’, i.e. ‘we saw’) cumulates person and number, but is invariant across the lexicon and thus clearly nonflexive. And flexive formatives can be separative. In the preceding section we saw that Dumi person, number, and tense formatives are flexive in that they select lexically defined ablaut classes. But this does not entail that the three categories are always expressed cumulatively: in a desinence like -tə, for instance, -t marks nonpast tense separatively from -ə for first person singular (cf. -ø-ə ‘1st person singular past’). Thus exponence type is independent of flexivity. And it is independent of fusion: although cumulative exponence is best known from bound morphology (e.g., Russian case– number exponence as mentioned above), some West African languages have isolating (free) formatives cumulating person agreement and tense/aspect/mood values. This is illustrated for Hausa with two examples in the completive aspect: (19) Hausa (Afroasiatic; West Africa; Newman (2000:569)) a. Mūsā yā tàfi Bicı̀ m. 3sg.masc:compl go b. ‘Musa went / has gone to Bichi’ b. yârā sun ga mac¯`jı̂-n? children 3pl.compl see snake-art.pl ‘Did the children see the snake?’ 1.5.2 Synthesis and wordhood The second dimension of semantic density, synthesis, applies to the level of the word. It is customary to distinguish three prototypes on a scale from analytic to synthetic to polysynthetic, measured by the number of formatives and lexical roots that are bound together in one word: one or very few formatives and at most one root in the case of analytic words, a moderate number of formatives together with one root in synthetic words, and an abundant mixture of formatives and lexical roots in polysynthetic words. The relevant notion of word here is the grammatical word, not the phonological word. The grammatical word is defined as the smallest unit of syntax, technically the terminal node or minimal projection (X0 ) in phrase structure. In He worked, for instance, he and worked are grammatical words, one simple (he), one complex (worked, containing the root work and the past tense suffix -ed). The formatives that are combined into a single grammatical word (work+ed) cannot be interrupted by phrasal constructions. They exhibit only morphological and phonological 190 Balthasar Bickel and Johanna Nichols dependencies (such as allomorphy selection and phonological fusion), but never enter into syntactic dependencies such as agreement or government. They usually have fixed morpheme order, while the ordering of grammatical words with respect to each other is commonly (though not always) freer. Typically, grammatical words are also phonologically coherent, but, as we saw in the Yidi and Sierra Otomı́ examples in section 1.4, the phonological word can be a smaller unit than the grammatical word. Phonological words can also be larger units than grammatical words; common examples of this arise from cliticization. Russian prepositions, for instance, form a single phonological word with the noun they govern. As we saw in section 1.1, however, the relationship between preposition and noun is still one between independent grammatical words. Analytic words comprise just one or a very limited number of formatives or they comprise just one lexical root. Examples are the words he (one pronominal root and one nominative case formative) and worked (one lexical root and one past tense formative) we looked at just before. Sometimes analytic words combine syntactically in the expression of inflectional categories. This is called periphrastic expression. An example is the expression of tense and aspect values by means of auxiliary constructions in European languages. The English future (will go), for instance, involves two distinct grammatical words, each comprising only one formative (the auxiliary will) or one root (go). The two words occupy variable phrase-structural positions (Your friend will go vs Will your friend go?) and the expression is interruptible by phrase-heading expressions (He will definitely go). (Note that analytic words can be phonologically bound: English auxiliaries typically cliticize to preceding words (he’ll go). They are no less grammatical words for being phonologically bound, however.) Words such as the auxiliary have in English, which comprises two formatives, a tense-indicating root and an agreement marker (cf. has vs have), are traditionally classified as analytic just like single-formative auxiliaries. The notion of synthetic words is usually restricted to words with more elaborate formative sequences, but the difference between synthetic and analytic is one of degree, and any categorial distinction ultimately misses the point. When flexive formatives are involved, synthetic words typically comprise two or three formatives along with a lexical root, e.g. a verb root and formatives expressing aspect, tense, and agreement, or a nominal root and formatives expressing case and number. An example of this is found in Russian verb forms like vyp et ‘will drink’, which express tense (future), aspect (perfective), person (third), and number (singular). Nonflexive concatenative (i.e. ‘agglutinative’) morphology usually allows longer and more complex synthetic words. An extreme example of this is Turkish word forms like the one in (20), which includes no less than ten formatives suffixed to the stem tan- ‘know’. Inflectional morphology (20) 191 Turkish tan-ş-tr-l-a-ma-dk-lar-n-dan-dr know-recip-caus-pass-pot-neg-nzr-pl-3poss-abl-3cop ‘It is because they cannot be introduced to each other’ (literally, ‘[it] is from their not being able to be made known to each other’) Synthetic words mostly involve bound (concatenative or nonlinear) formatives, but, as pointed out before, phonologically isolating formatives can also combine into single grammatical words and can thereby constitute complex synthetic words. Indeed, many isolating formatives in Southeast Asian and East Asian languages form a single grammatical word together with the lexical root they modify. In Lai Chin, for instance, formatives indicating agreement, tense and mood are phonologically free, i.e. isolating, but any sequence of a verb and one or more of these formatives constitutes a single, uninterruptible word from the point of view of syntax: (21) Lai Chin (Ken Van Bik (p.c.)) a. na-tuk nhaa 2sg.a-hit.with.stick:
2 3pl.p ‘You will hit them’ b. na-kan-tuk 2sg.a-1pl.p-hit.with.stick:
2 ‘You will hit us’ làay fut làay fut In strings of formatives like these, the ordering of formatives is rigidly fixed (*natuk laay nhaa), and this contrasts with the relatively free ordering of grammatical words in Lai Chin sentences. Moreover, the third person plural object agreement marker nhaa is obligatory and is in direct opposition with the first person plural object agreement marker which is a phonologically bound prefix (kan-). Further, as shown by the contrast in (22), no phrasal constituent can intervene: (22) Lai Chin (Ken Van Bik (p.c.)) a. *na-tuk nhaa, ʔ ùy tsaw làay 2sg.a-hit.with.stick:
2 3pl.p dog fut Intended: ‘You will hit the dogs’ b. na-tuk nhaa làay, ʔ ùy tsaw 2sg.a-hit.with.stick:
2 3pl.p fut dog ‘You will hit the dogs’ These facts suggest that the sequence natuk nhaa làay ‘you will hit them’ forms one single, synthetic grammatical word, just like the expression ʔùy tsaw, which 192 Balthasar Bickel and Johanna Nichols is a single lexical item meaning ‘dog’. Thus, even though at first sight one is tempted to compare the syntactic status of làay to that of the English auxiliary will and the status of nhaa to that of the English pronoun them, làay and nhaa are formatives within a word, and not grammatical words in syntactic combination. This is all completely independent of the fact that Lai Chin grammatical words often comprise several phonological words as shown in section 1.1 above. While synthetic forms comprise only formatives and one lexical word (the stem), matters are different with polysynthesis, which brings together not only formatives but also incorporated stems and lexical affixes into a single grammatical word (an X0 in phrase structure). This phenomenon is widespread in North American languages (for which it was first described by Du Ponceau in 1819), but it is also found elsewhere. The following examples of polysynthetic words are from Siberia and Papua New Guinea, respectively: (23) Telqep Chukchi (Chukotko-Kamchatkan; Siberia; Dunn (1999)) utt-ə n-ejmew-jə w-ə -ninet=ʔ m wood-caus-approach-collective-epen-3sg.A:3pl.p=emph ‘He brought them wood’ (24) Yimas (Lower Sepik-Ramu; Papua New Guinea; Foley (1991)) paŋ kra-kaykaykay-kwalca-mpi-kulanaŋ -tal-kia-ntu-ŋ kt 1pauc.s-quickly-rise-seq-walk-start-at.night-rem.pt-pauc ‘We few got up at night and quickly started to walk’ In these verb forms, not only grammatical information like person, number, and tense, but also various lexical concepts like ‘wood’ or ‘at night’ are expressed by bound morphology. Polysynthesis often involves grammatical words that are phonologically coherent, but, as with synthesis, not necessarily. Indeed, unlike the Chukchi example in (23), a Yimas string like the one in (24) consists of several phonological words,14 defined by stress and allophone distribution (Foley (1991:80–7)), but the string nevertheless forms a single grammatical word in syntax (i.e. a V0 or minimal projection constituent). Its grammatical wordhood is evidenced, among other things, by the fact that the string involves purely morphological, non-syntactic dependencies: the appearance of the paucal suffix -ŋkt, for example, is contingent on the presence of a person-indicating prefix, here paŋkra- ‘we few’. The suffix cannot appear if the person reference is established by means of syntactically independent pronouns rather than prefixed formatives. The first person paucal pronoun, for example, is incompatible with the paucal suffix because the pronoun projects its own analytic grammatical word. (First person 14 This has also been shown for polysynthetic words in the two North American languages Cree (Algonquian) and Dakota (Siouan); see Russell (1999). The analysis of Algonquian and similar languages (e.g. Kutenai) as polysynthetic has become a matter of debate, however. See, e.g., Goddard (1988) and Dryer (2000) for controversial discussion. Inflectional morphology 193 reference is expressed periphrastically for first person paucal, compensating for the lack of a corresponding synthetic form.) (25) Yimas (Foley (1991:223)) paŋ kt ŋ kul-cpul(*-ŋ kt) 1pauc 2du.p-hit(*-pauc) ‘We few hit you two’ If suffixing -ŋkt were possible here, this would mean that the second word was agreeing with the first and that the relationship between the two was therefore one of syntactic agreement. By analogy, one could then argue that -ŋkt appears in (24) above because of agreement with paŋkra-; the relationship between these two elements would then be a syntactic relationship holding between two distinct grammatical words. A case could then be made for analysing the expression as analytic. But the fact is that the distribution of -ŋkt is not governed by agreement between grammatical words but is instead subject to morphological rules that are operative within, rather than across, grammatical words. One of the typologically most important characteristics of polysynthesis is that pronominal and even lexical arguments are incorporated into their governing verb. The Yimas words in (24) and (25) exemplify incorporated pronouns: in (24) the first person paucal prefix paŋkra- functions as an affixed subject pronoun. In (25), the second person dual prefix -ŋkul functions as an incorporated object pronoun, while the subject pronoun paŋkt ‘we few’ is not incorporated. The Chukchi example in (23) illustrates incorporation of a lexical argument. The direct object utt-‘wood’ is incorporated into the verb (as a regular response to low discourse saliency of the object; see Dunn (1999)). Incorporated elements are no longer grammatical words heading their own constituents in the clause. They typically lose many of their syntactic abilities and could thus be called semi-words. We will briefly come back to pronoun incorporation in our discussion of agreement systems in section 8. 2 Locus Locus is the term we propose for what has been known as head/dependent marking (Nichols (1992)). The essential distinction can be illustrated by examples from Hungarian and English (the relevant formatives are in boldface): (26) Hungarian (Uralic) az ember ház-a the man house-3sg ‘the man’s house’ (27) English the man-’s house 194 Balthasar Bickel and Johanna Nichols In both of these, the possessed noun ‘house’ is the syntactic head of the construction and the possessor is non-head. Hungarian puts an inflectional suffix on ‘house’ (the head) while English puts it on the possessor. The Hungarian inflectional suffix is a possessive suffix which agrees in person and number with the possessor; the English one is a case clitic and not an agreement marker. As these examples show, the syntactic relation of adnominal possession can be reflected by placing a formative on either the head or the non-head of the phrase. The inflectional categories differ, but not because the syntactic relation they reflect differs; rather, certain inflectional categories have affinities for one or another locus. Person and number, for instance, are almost always on heads and almost always due to agreement, while case is on non-heads and is not always (and in fact not often) due to agreement. The locus of marking can be not just on the head or the non-head, but also on both or on neither. The following examples give some idea of the variety of locus types and the variety of inflectional categories that mark them, using possessive nps (Nichols (1992:49ff.)). On head (h e a d m a r k i n g ): (28) Tadzhik (Indo-European; J. R. Payne (1980:167–8)) xona-i padar house-ez father ‘father’s house’ (29) Abkhaz (West Caucasian; Hewitt (1979:116)) à-č’k◦ ’ə n yə -y◦ nə̀ art-boy 3sg-house ‘the boy’s house’ In (28), the formative -i on the head noun ‘house’ indicates that there is a dependent present in the np but that it does not agree with it. This construction is known as izafet or ezafe in the grammatical traditions of many Turkic and Iranian languages (and glossed here as ‘ez’). In (29), the dependency relation is indicated by possessor agreement, again marked on the head. This is the inflectional category generally known as possession or possessive affixes, common in languages of Siberia, the Himalayas, and the Americas. For more on possessor agreement, see sections 4.1 and 8.1 below. On dependent (d e p e n d e n t m a r k i n g ): (30) Chechen (Nakh-Daghestanian; Caucasus) dee-n aaxcha father-gen money ‘father’s money’ Inflectional morphology 195 On both (d o u b l e m a r k i n g ): (31) Nogai (Turkic; Baskakov (1963:539)): men=im kullyg-ym 1sg=gen work-1sg ‘my work’ On neither (j u x ta p o s i t i o n ): (32) !Kung (Khoisan; S. Africa; Snyman (1970:92)): dzheu = | xanu woman book ‘woman’s book’ On neither (d e ta c h e d m a r k i n g ): (33) Tagalog (Austronesian; Philippines; Schachter and Otanes (1972:116, 123)): a. nasa mesa=ng libro on table=link book ‘the book on the table’ b. libro=ng nasa mesa book=link on table ‘the book on the table’ This Tagalog example is another instance of a Wackernagel position clitic on the np level (cf. example (10) in section 1.2). We call this marking detached because the clitic is not attached to either the head (libro ‘the book’) or the dependent (nasa mesa ‘on the table’). It is placed between the two. Marking can also be split. Many languages use two different loci of marking to implement what is often termed ‘alienable’ versus ‘inalienable’ possession. The ‘inalienables’ are often nouns such as kin terms and body parts (called ‘inalienable’ because they typically cannot be sold or given away) and the ‘alienables’ are the rest. It is common for ‘inalienable’ possession to be headmarked and ‘alienable’ not, as in (34): (34) Amele (Madang; New Guinea; Roberts (1987:139)) (‘mouth’ and ‘son’ are inalienable) a. ija na jo 1sg of house ‘my house’ b. Naus na jo n. of house ‘Naus’s house’ 196 Balthasar Bickel and Johanna Nichols c. ija co-ni 1sg mouth-1sg ‘my mouth’ d. Naus mela-h-ul n. son-3sg-pl ‘Naus’s sons’ or for ‘inalienable’ possession to have no marking and ‘alienables’ to have case marking or the like: (35) Dyirbal (Dixon (1972:61, 105)) a. balan ugumbil mambu [inalienable] det woman back ‘the woman’s back’ b. bayi waŋ al baŋ ul yaa-ŋ u [alienable] det boomerang det.gen man-gen ‘the man’s boomerang’ The different locus types can also be distinguished in the marking of clause relations. Here are examples of languages that mark the relations of subject and object only on the verb (head marking, as the verb is the head of the clause): (36) Abkhaz (West Caucasian; Georgia; Hewitt (1989:67)) a-p◦w ə s a- aca a- arp Ø-yə -zə -lə - w a-yt det-woman det-man det-shirt 3sg.p-3sg.m.io-for-3sg.f.a-wash-aor ‘the woman washed the shirt for the man’ only on the arguments (dependent marking, as the arguments are the dependents): (37) Martuthunira (Dench (1994:75)) ngayu tharnta-a nhuwa-lalha parla-ngka 1sg.nom euro-acc spear-past hill-loc ‘I speared a euro in the hills’ on both: (38) Belhare unchik-ŋ a yeti n-thuu-t-u? 3nsg-erg what.nom 3nsg.a-cook-npt-3p ‘What do they cook?’ and on neither: Inflectional morphology (39) 197 Thai (Jenny (2001), from a popular Thai song) phruŋ 2 nii3 chan4 cə rak3 khun tə lɔɔ t1 pai tomorrow 1fam prospective love 2hon whole continuative ‘Tomorrow I will love you forever’ Certain grammatical categories favour particular loci, and the traditional terminology for various grammatical categories contains implicit reference to locus of marking. Case, for instance, is always marked on dependents, and in fact case can be defined as dependent-marked affixal indication of clause and phrasal relations. The same information can perfectly well be marked on heads, but then it is not called case. In the following Georgian examples, the form of the first person agreement prefix indicates the role of the first person referent: subject in the first example, object in the second. (40) Georgian (Kartvelian; Caucasus) a. v-xedav 1sg.a-see ‘I see (him/her/it)’ b. m-xedav 1sg.p-see ‘You see me’ In the following examples from a Mayan language, the agreement markers are glossed with case names: abs = absolutive and erg = ergative. (41) Jacaltec (Mayan; Mesoamerica; Craig (1977:122, 111)) a. x-Ø-haw-il naj asp-abs.3-erg.2-see 3sg ‘You saw him’ b. xc-ach w-abe asp-abs.2 erg.1-hear ‘I heard you’ 3 Position By position we mean the location of an inflectional formative relative to the word or root that hosts it. The formative may precede the host, follow it, occur inside of it, be detached from it, or various combinations of these. There is a standard terminology which accounts for most of these positions together with the formative type and degree of fusion. Table 3.2 expands this terminology somewhat. Latin prepositions or truncated adverbs label the position categories. Types that may not be self-evident or have not been illustrated earlier are explained and exemplified in what follows. 198 Balthasar Bickel and Johanna Nichols Table 3.2 Typology of positions and formatives. * = example below in this section Position Formative type and/or degree of fusion Prae Preposed free formative * Proclitic Prefix Initial reduplication (cf. Ancient Greek example in section 1.3.3 for illustration) Substitution (cf. section 1.3.3) Ablaut (i.e. bare ablaut; if ablaut is triggered by an affix, the combination of affix and ablaut constitutes simulfixation, described below) Infix (including Interposition *) Endoclisis * Subtraction (cf. Tohono ‘O’odham example in section 1.3.3) Prosodic formatives (cf. Kinyarwanda example in section 1.3.3) Final reduplication Suffix Enclitic Postposed free formative Simulfix, simulclitic, etc. (including circumfix) * Detached (word or formative, cliticized or free; see sections 1.2 and 2 for discussion) In Post Simul None of the above Examples: F r e e f o r m at i v e s Like affixes, free (or isolating) formatives are typically fixed in their position. Plural words and other grammatical number words (Dryer (1989)) are often free formatives. The singular and plural words of Yapese, shown in the following examples, are in a fixed position in the nominal modifiers. (42) Yapese (Austronesian; Dryer (1989:868) from J.T. Jensen (1977:155)) a. ea rea kaarroo neey art sg car this ‘this car’ b. ea pi kaarroo neey art pl car this ‘these cars’ E n d o c l i s i s A clitic inserted into a word constitutes endoclisis. The phenomenon is rare, but well documented for Udi by Harris (2000). In (43), the person–number agreement marker is a clitic (
= first element of split simplex stem; see Harris for the full argument that =z= is a clitic): Inflectional morphology (43) 199 Udi (Nakh-Daghestanian; Caucasus; Harris (2000)) kaghuz-ax a=z=q-e letter-dat
=1sg=receive-aor ‘I received the letter’ I n t e r p o s i t i o n Interposition is a typologically and historically distinct subtype of infixation. In general, infixation places formatives into a phonologically or prosodically defined environment (e.g. after the stem’s onset consonant(s), or after the first syllable), but in the case of interposition, the environment is more nearly morphological, reflecting morphologized infixation or petrified derivational morphology or compounding. Interposition typically involves formatives placed between the two parts of a bipartite stem. A bipartite stem is a stem where only part, but not the whole, is the target of morphological rules (affixation, reduplication, mutation, particle hosting, etc.), and the location of the boundary between the two parts is morphologically defined, i.e. neither semantically (e.g., by scope as in the juxtaposition of independently inflected stems) nor phonologically (e.g. by syllable structure as with infixation) (Jacobsen (1980); DeLancey (1996, 1999)). Interposition in verb stems is particularly well known in languages spoken in the American Pacific Northwest, but it is also attested in various Caucasian and Himalayan languages: (44) Washo (Jacobsen (1980)) suʔ m-te -ı́tiʔ throw-pl-down ‘to throw down repeatedly’ (45) Andi (Nakh-Daghestanian; Caucasus; Gudava (1959:197)) a-b-ch-o wash-gender.agreement-wash-past ‘(I/you/he/she/we/they) laundered, washed (it)’ (46) Belhare15 a. la-ŋŋ -u-yakt-he dance-3nsg.s-dance-ipfv-past ‘They were dancing’ b. tha-tok-ka-tok n-ca-he (< tha-tok- ‘to know s.o.’) know-know-recip-know 3nsg.s-aux-past ‘They knew each other’ 15 Phonologically, these strings bracket into two or more prosodic words: [laŋ ][ŋ uyakthe], [tha][tokka][tok], but, syntactically, they are indivisible wholes, i.e. single grammatical words; cf. the discussion of synthesis in section 1.5.2. 200 Balthasar Bickel and Johanna Nichols In all these examples a formative is affixed only to the second part of the stem. In (46b), we find in addition, a process of reduplication affecting only one part of the stem (tok-). Sometimes, the two parts of bipartite stems have independent meanings and the properties of independent verb stems, as in the Washo example, where šuʔm means ‘throw’ and determines the transitivity of the complex, and ı́tiʔ means ‘down’ and shows the morphophonological behaviour of independent stems. But despite its position on only one stem part, the plural affix has scope over both parts simultaneously, and the stem as a whole behaves as a single grammatical word (a terminal node) in the syntax (see Jacobsen (1980)). The elements ach- ‘wash’ in Andi and lau-‘dance’ in Belhare are simplex expressions that cannot be further analysed into component parts, at least not synchronically. It is chiefly verbs that are bipartite, but bipartite nominal stems that undergo interposition are attested in Limbu (Tibeto-Burman, Nepal). The third person singular possessive form of teʔlphuŋ ‘garments, clothing’, for instance, is kudeʔl-ku-bhuŋ (van Driem (1987:27)), with the possessive marker ku- occurring not only at the beginning of the word but also at the beginning of its second (etymologically separate) part. (This example also illustrates simulfixation, as is discussed just below.) S i m u l f i x at i o n : This term, which was first proposed by Hagège (1986:26), involves several tokens of a single morpheme, realized at different places in the word. The most common subtype is circumfixation (as, e.g., the circumfix ge- . . . -t marking German participles such as ge-lieb-t ‘loved’), but there are other options. The formatives can be both suffixes, both prefixes, or one can be internal, the other external. The Belhare perfect exemplifies concatenative simulfixes of which both pieces (-ŋa and -ha) are postposed: (47) Belhare khai-ŋ a-ŋŋ -ha go-perf-1sg-perf ‘I’ve gone’ Combinations of internal and external marking are abundant in Germanic languages, e.g. in words such as English children, whose plural number is marked by both ablaut (internal) and a suffix (postposed). A more complex example of this kind is found in Lak, where in some verbs gender is marked both by initial mutation (b/d/Ø) and, internally, by ablaut of the medial consonant (v/r). (48) Lak (Nakh-Daghestanian; Caucasus; Zhirkov (1955:93, 1962:418)) a. b-u-v-na b. d-u-r-na Inflectional morphology 201 c. Ø-u-v-na g e n d e r . a g r e e m e n t -go-g e n d e r . a g r e e m e n t -p a s t ‘went’ (different genders) The Limbu example used above to illustrate interposition (ku-deʔl-ku-bhuŋ ‘his/her clothes’) also illustrates simulfixation: it has one token of the possessive formative preposed and one interposed into a bipartite stem. The apparent position of affixes in a word can be deceptive, so that what appears to be (say) an infix to the naked eye proves to be a prefix or suffix when the morphological analysis has been done. For example, Tagalog infixes have been successfully analysed as prefixation under prosodic constraints against closed syllables (see McCarthy and Prince (1995), and Crowhurst (1989) for critical discussion): cf. um-ibig ‘love’ versus s-um-ulat ‘write’ and gr-um-adwet ‘graduate’. Here, the actor-voice prefix um- is forced to shift to after the first onset in order to avoid the ungrammatical closed syllables *( um) (as in *umsulat, *um-gradwet) or *( gum) (as in *gumradwet).16 Another potential source of confusion in the analysis of affix positions is internal constituent structure within inflected and derived words. In the following examples from the Daghestanian language Kubachi Dargi, the gender formatives b and w appear both at the beginning and in the middle of the word: (49) Kubachi Dargi (Nakh-Daghestanian; Magometov (1963:76)) a. b-e n-ka-b-išši-j gender-in-down-gender-go-inf ‘insert, put in’ (B gender) b. w-e n-ka-w-išši-j gender-in-down-gender-go-inf ‘go in’ (W gender)17 This is not simulfixation, however, but simultaneous prefixation to both a verbal preverb and the verb root. 4 Paradigms Inflectional systems are typically organized into paradigms of variable size, ranging from, e.g., the two-member paradigm of English verb agreement, with third person singular versus everything else (e.g. goes vs go) to large case paradigms. Plank (1991:16) notes that very large case inventories are found only in languages with separative exponence and do not occur in languages with chiefly cumulative exponence (see section 1.5.1). 16 17 Following standard conventions, ‘’ stands for syllable and the parentheses are syllable brackets. The verb is ambitransitive, and is interpreted as transitive (semantically causative) when it agrees in the inanimate B gender but as intransitive when it agrees in the animate W gender. 202 Balthasar Bickel and Johanna Nichols Table 3.3 Latin noun paradigms Singular: Nom. Voc. Acc. Gen. Dat. Abl. Plural: Nom. Voc. Acc. Gen. Dat. Abl. ‘wolf’ ‘war’ ‘road’ ‘foot’ ‘attack’ lupus lupe lupum lup¯ lupō lupō bellum – bellum bell¯ bellō bellō via – viam viae viae viā pēs – pedem pedis ped¯ pede impetus – impetum – – impetū / -e lup¯ lup¯ lupōs lupōrum lup¯s lup¯s bella – bella bellōrum bell¯s bell¯s viae – viās viārum vi¯s vi¯s pedēs – pedēs pedum pedibus pedibus impetūs – impetūs – – – The organization of inflectional forms into paradigms brings with it a series of properties not typically found in other parts of morphology: inflectional classes, syncretism, defectivity, suppletion, deponence, and eidemic resonance. Case inventories and the terminology for them will be discussed briefly at the end of this section. 4.1 Inflectional classes Case paradigms are paradigms par excellence and display most of the important properties of paradigms. Tables 3.3 and 3.4 show Latin and Chechen case paradigms, respectively. (Gaps in some of the Latin paradigms illustrate defectivity, discussed below.) The Latin nouns shown in table 3.3 fall into distinct declension classes based on the stem-final (traditionally,‘thematic’) vowels (-u∼-o vs -a vs -u) or consonants (-d in ped- ‘foot’) and the considerable allomorphy of the endings (e.g. nominative singular -s vs -m vs zero). The Chechen nouns in table 3.4 have mostly the same endings but considerable variation of stems. The noun ‘daughter-in-law’ has stem ablaut, and most nouns have stem extensions in the plural paradigms: -ar- in ‘daughter-in-law’, -arch- in ‘pig’, -o- in ‘mother’, -an- in ‘grief’. The -i- found in several oblique cases in the singular of ‘grief’ and ‘pig’ is another extension, absent in the nominative, ergative, and (synchronically, though probably not diachronically) allative. Extensions are lexically conditioned and carry no meaning (though they may have their origins in frozen derivational or inflectional suffixes). The Chechen system of extensions Inflectional morphology 203 Table 3.4 Chechen noun paradigms (all-Latin no-diacritics transcription; see http://socrates.berkeley.edu/∼chechen for this transcription) Singular: Nom. Gen. Dat. Erg. All. Ins. Lat. Csn. Plural: Nom. Gen. Dat. Erg. All. Ins. Lat. Csn. ‘window’ ‘daughter-in-law’ ‘mother’ ‘grief’ ‘pig’ kor kuoran kuorana kuoruo kuorie kuoraca kuorax kuoral nus nesan nesana nesuo nesie nesaca nesax nesal naana neenan naanna naanas neenie neenaca neenax neenal baala baalin baalina baaluo baalie baalica baaliax baalial hwaqa hwaqin hwaqina hwaquo hwaqie hwaqica hwaqiax hwaqial kuorash kuoriin kuorashna kuorasha kuorashka kuorashca kuoriax kuorial nesarii nesariin nesarshna nesarsha nesarshka nesarshca nesiax nesial naanoi naanoin naanoshna naanuosha naanoshka naanoshca naanoix naanoil baalanash baalaniin baalanashna baalanasha baalanashka baalanashca baalaniax baalanial hwaqarchii hwaqarchiin hwaqarchashna hwaqarchasha hwaqarchashka hwaqarchashca hwaqarchiax hwaqarchial is a modest version of the elaborate systems found in Daghestanian languages (Kibrik (1991)), distant sisters of Chechen. The notion of declension class, or more generally inflectional class, was devised traditionally to handle paradigms like the Latin ones, where at first glance there seem to be different series of endings (-us, -um, -¯, -ō in ‘wolf’; -a, -am, -ae, -ā in ‘road’; -Ø, -em, -is, -¯, -e in ‘foot’, etc.). In fact, though, there are two sets of differences, one resulting from the vowels (traditional ‘thematic vowels’) that expand the word stem (-u ∼ -o in ‘wolf’ vs -a ∼ -ā in ‘war’ vs Ø in ‘foot’ vs -u in ‘attack’) and one resulting from differences in the endings themselves (e.g. nominative singular -s or -Ø or -m; genitive singular -¯ or -(i)s, nominative plural -i or -ēs); these two kinds of differences can also occur simultaneously (e.g. nom. sing. in Ø with a-stems, but in -s or -m with others). The thematic vowels are rather like stem extensions; this means that the Chechen and Latin case paradigms differ in degree of morphophonemic transparency (Latin being less transparent) rather than in morphological type. A full taxonomy of variation in stem and ending adequate to typologize inflectional paradigms would be a three-way distinction of variation for both stems and endings: lexically conditioned, i.e. lexeme-based, allomorphic variation; category-based allomorphic variation, i.e. allomorphy dependent on specific 204 Balthasar Bickel and Johanna Nichols Table 3.5 Typology of inflectional classes Formative: Stem Lexeme-based allomorphy Category-based allomorphy No regular allomorphy Lexeme-based allomorphy Category-based allomorphy No regular allomorphy Latin nouns Latin and Polish verbs [unattested in our sample] Polish nouns, Anêm possession Newar verbs Chechen verbs, nouns, Dumi verbs (1.4) Belhare verbs Germanic weak verbs, Ossetic sg./pl. case Finnish nouns Table 3.6 Belhare verb paradigm (selection). The k∼g alternation in yak- and -ka is morphophonologically conditioned; -ʔ and -yu mark nonpast (the allomorphy is determined by prosodic structure), -he past, -ŋ e resultative, and -kone inconsequential 1sg 2sg 3sg nonpast past resultative inconsequential yau-ʔ-ŋ a yau-ka yak-yu yag-he-ŋ a yag-he-ga yag-he yau-ŋ e-ŋ a yau-ŋ e-ga yau-ŋ e yak-kone-ŋ a yak-kone-ga yak-kone inflectional categories but general across all lexemes; and no allomorphic variation. Lexeme-based allomorphy of stems, or stem classes Stem classes are present when stems differ (because of ablaut, stem extensions, stress shift, etc.) when inflected for the same category, and the differences are lexically (and not [morpho-]phonologically) conditioned. Examples are the Chechen and Latin paradigms in tables 3.3 and 3.4 above. In Chechen, for example, the vowel ablaut in ‘daughter-in-law’, or the choice of stem extensions (-ar-, -an-, etc.) in the plural, is a purely lexical and unpredictable matter. In Latin, as argued above, the traditional declension classes are in fact lexical differences of thematic vowel (obscured by regular morphophonology such as vowel raising in the nominative singular lupus < lup-o-s or monophthongization in the genitive singular lup¯ < lup-o-i; see section 1.4 above). c at e g o r y - b a s e d s t e m a l l o m o r p h y In some languages, all stems have the same allomorphy, selected by specific morphological categories or paradigms. Belhare verbs all undergo the same stem alternations from person to person and from tense to tense. The verb yakma ‘to stay overnight, find shelter’, for example, has the two stem forms yak- and yau-, and table 3.6 Inflectional morphology 205 Table 3.7 Verb paradigms in Latin and Polish Latin ‘love’ 1sg 2sg 3sg 1pl 2pl 3pl Polish ‘write’ Present Perfect Present Past amo amās amat amāmus amātis amant amāvi amāvisti amāvit amāvimus amāvistis amāverunt  pisze piszesz pisze piszemy piszecie pisza pisaem pisaeś pisa pisaliśmy pisaliście pisali shows how they are distributed over a selection of forms. The primary stem here is yak-, and the secondary stem -yau is derived from this by imposing a CVV syllable structure: the original root coda /k/ is vocalized while retaining its tongue and velum positions (i.e. its point of articulation and nasality/orality), e.g. yak- ∼ yau- ‘stay overnight’, yaŋ- ∼ yaũ- ‘carry by hand’. Bilabials are exempted from this and remain unchanged (e.g. lap- ‘catch’). CV roots are fitted into the CVV shape by epenthesis of /i/ or, after /i/, /u/ (e.g. so- ∼ soi‘wait’, khi- ∼ khiu- ‘quarrel’, etc.). These rules hold across the lexicon; the stem allomorphy is entirely regular and exclusively depends on the person and tense choice: the secondary stem occurs before the nonpast allomorphs -t and -ʔ, and before the resultative (and perfect) markers -ŋe (and -ŋa), among others. N o s t e m a l l o m o r p h y Stems need not behave differently when inflected for the same categories. The noun stems of Finnish, for example, and most noun stems of Polish, behave essentially alike and are essentially unchanged (except for automatic phonological and morphophonemic alternations) when inflected for case. For Finnish paradigms, see Eliot (1890:26ff.); Serebrennikov and Kert (1958); Branch (1987). F o r m at i v e c l a s s e s When inflectional formatives have lexeme-based allomorphy we have formative classes. For example, the Latin nouns shown above have different sets of endings. C at e g o r y - b a s e d f o r m at i v e a l l o m o r p h y The verbs of IndoEuropean languages generally have different person–number agreement suffixes in the present and past tenses, but these differences are the same for all verbs (with few exceptions). For example, consider the Latin and Polish conjugations in table 3.7. In Latin and Polish, different agreement classes co-occur with differences in stem classes: while amāre ‘love’, a class i verb in Latin, has the stem amā- in the perfect (amā-v-i), other classes have different perfect stem forms, which are most often irregular (e.g. agere ‘to guide’: ēg-; rı̄dēre ‘to laugh’: rı̄s-, etc.). In Polish most verbs have -e- in most paradigm forms, as in table 3.7, but 206 Balthasar Bickel and Johanna Nichols Table 3.8 Latin noun paradigm (singular only) Nominative Accusative Genitive Dative Ablative ‘case’ ‘mode’ ‘gender’ cāsus cāsum cāsūs cāsu¯ casū modus modum mod¯ modō modō genus genus generis gener¯ genere a smaller (though still large) class of verbs has -i: lubie, lubisz, lubi, etc., ‘love’. These languages are different from Dolakha Newar (Tibeto-Burman; Nepal; Genetti (1994)), where tense-based agreement allomorphy combines with stem alternations that are phonologically defined (similar in spirit to what we described for Belhare) and do not require the discrimination of arbitrary lexical classes. Tense-based regular agreement allomorphy is to a limited degree also characteristic of Germanic languages (cf., e.g., German third person singular lieb-t ‘loves’ in the present vs lieb-t-e ‘loved’ in the past), but stem allomorphy is restricted to a set of irregular verbs traditionally called ‘strong’ verbs as opposed to the regular ‘weak’ verbs. N o f o r m at i v e a l l o m o r p h y Finnish nouns all have the same set of case suffixes, and likewise for nouns in Hungarian, Turkish, and Basque. All variation there is phonologically or morphophonologically conditioned, i.e. the same across the (regular) lexicon. Where there are inflectional classes, an important consideration is identifying the inflectional form or forms from which all or most of the others can best be predicted. This is the reference form(s) or principal part(s) (Wurzel (1987a), 1987b); Carstairs-McCarthy (1991)), and it should be included in dictionaries, glossaries, and practical descriptions. Latin dictionaries, for example, list the nominative and genitive forms of nouns, and from these one can infer all other case forms. Thus, while in all of the following nouns the nominative ends in -us, they have different case paradigms, and this is predictable from the genitive form that goes together with the -us nominative in each case: cāsus ‘case’ has genitive cāsūs, modus ‘mode’ has genitive mod¯, and genus ‘gender’ has genitive generis; cf. table 3.8. Note that other case combinations, e.g. nominative and accusative, would not unambiguously identify the paradigms. The nominative (citation form) plus the genitive (principal part), however, serve to completely identify the rest of the declension. Case paradigms are the prototypical declension classes, but a number of languages around the Pacific Rim have declension classes defined by allomorphy of possessive inflection. Languages in our sample with this kind of declension Inflectional morphology 207 Table 3.9 Anêm possessed noun paradigm (selection) (Thurston (1982:37)). -ng-, -g-, and -d- in the last three words are stem extenders. The final elements are person–number–gender suffixes 1sg 2sg 3sgm 3sgf ‘water’ ‘child’ ‘leg’ ‘mat’ kom-i kom-ı̂ kom-u kom-ı̂m gi-ng-e gi-ng-ê gi-ng-o gi-ng-êm ti-g-a ti-g-ı̂r ti-g-ı̂ ti-g-ı̂ mı̂k-d-at mı̂k-d-ir mı̂k-d-it mı̂k-d-it classes are Amele (Madang family or perhaps Rai Coast-Mabuso, New Guinea: Roberts (1987)), Anêm (New Britain family, New Britain: Thurston (1982)), Äiwo (Reefs-Santa Cruz, southeastern Pacific: Wurm (1981)), Chichimec (Otomanguean, Mexico: Lastra de Suárez (1981)), Cayuvava (isolate, South America: Key (1967)), and Limbu (Tibeto-Burman, Himalayas: van Driem (1987)). Languages with typical alienable/inalienable possession might be described as having two declension classes defined by possessive inflection, but the six languages listed here have three or more declension classes, usually with considerable and complex allomorphy of the possessive affixes or stem alternations triggered by these. Amele has 31 declension classes of inalienables (Roberts (1987)) and Anêm about 20 created by a combination of different person–number suffixes and different stem extensions (Thurston (1982:37–8)); cf. table 3.9 for illustration. This is lexeme-based flexivity of both formatives and stems, similar in kind to Latin case inflection: both the shape of the stem (with extensions -ø, -ng, -g, -d) and the shape of the formative depend on the particular lexical declension class of the root. 4.2 Syncretism Every one of the Latin nouns in table 3.3 has at least one instance of syncretism, or falling together of case endings: an example is dative and ablative lupō of ‘wolf’. Chechen has virtually no syncretism in its noun paradigms. Syncretism is sometimes an accident of sound change, but more often it seems to be driven by purely morphological considerations. It is not at all obvious that syncretizing cases are semantically or syntactically similar; for some discussion see Plank (1991:19) or Blake (1994:44ff.). Hjelmslev (1935, 1937) and Jakobson (1971a [1936], 1971b [1958]) assume that syncretism follows, and reveals, the basic structural components of case meanings such as markedness of categories (markedness is defined in section 5). An instance of syncretism to which 208 Balthasar Bickel and Johanna Nichols functional motivation is often attributed is the nominative–accusative syncretism of neuter nouns in Indo-European languages (as in bellum ‘war’ in table 3.3). The motivation lies in the fact that neuters are almost all inanimate, hence presumably more likely to function as objects than as subjects of transitive verbs (as shown by discourse studies in many languages; see Dubois, Kumpf, and Ashby (2003)); hence there is little need for these nouns to distinguish subject and object case forms. Plank (1991:19–20) suggests ordering the cases of a language so as to put syncretizing forms adjacent to each other to the extent possible. This procedure yields the following order for Latin: Vocative, Nominative, Accusative, Ablative, Dative, Genitive. 4.3 Defectivity and suppletion Some words simply lack certain paradigmatic forms. Latin impetus ‘attack’, in table 3.3 above, forms only a few of the cases (Rhodes (1987)). Bagvalal place names, as mentioned in section 4.6 below, lack a nominative case. A more common kind of defectivation is lack of an entire category, or neutralization of categories, in the presence of some other: e.g. Swahili verbs lack a contrast of simple and imperfective aspect in negative forms, though they have it in affirmative forms (e.g. w-a-soma ‘they read’ with wa-na-soma ‘they are reading’ but only ha-wa-soma ‘they don’t read, they are not reading’). Category-based defectivity is not random; see Aikhenvald and Dixon (1998) for a preliminary survey. Gaps in paradigms are sometimes compensated for by (etymologically) different words. The lacking plural forms of Latin impetus ‘attack’, for example, are frequently supplied by incursiōnēs ‘attack’. When this is regular and obligatory, the result is known as suppletion. Examples are the Latin past and perfect stems tul- and lat- which are in paradigmatic opposition to the infinitive stem fer- ‘carry’; or the English past tense went in opposition to the other tense forms based on go. Suppletion of formatives (e.g Latin nominative in -s vs -m vs -ø) is usually called (lexical) allomorphy (cf. above). 4.4 Deponence A deponent word lacks the usual inflectional forms for a specific paradigm and instead takes on the forms of another. Deponent verbs in Latin and Greek are stranded passives, i.e. they have only passive forms, but they are used with active syntax; an example is Latin eum sequor ‘I follow him’, with sequ-or inflecting like a passive (cf. ag-or ‘I am being driven’) but with a transitive object eum ‘him’ in the accusative. This is the traditional sense of the term ‘deponent’. Inflectional morphology 209 Table 3.10 Chechen deictic prefixes hwadwahwalwa- toward speaker away from speaker up down Corbett (2000a) and Baerman (2006) show that the phenomenon is more general and gives other examples: Russian nouns like zhivotnoe ‘animal’, which is a syntactic noun with the declension of an adjective; Mohawk (Iroquoian) syntactic nouns with verb morphology such as ra’swà:tha’ ‘fireman’ (lit. ‘he extinguishes’); in Limbu and Belhare, a small number of syntactically transitive verbs are inflected as if they were intransitive, and vice versa. The Limbu verb form mε ʔru ‘s/he is fat’, for example, is a regular transitive verb form indicating a third person singular actor (zero prefix) and a third person singular undergoer (-u suffix). But syntactically and semantically, this is an intransitive predicate (Michailovsky (1985, 1997); also cf. Bickel and Nichols (2001)). 4.5 Eidemic resonance As pointed out by Hockett (1987), all morphology rests fundamentally on a basic notion of what he called resonance: parts of words resonate with each other and can therefore be extracted as meaningful formatives or morphemes. For example, English cooks and runs resonate in that they contain the similar sounds /s/ and /z/, associated with the identical meaning component ‘third singular subject in the present indicative’, and from this we can extract a morpheme -s. This is the most straightforward example, but in addition the forms of a paradigm often resonate with each other through alliteration, rhyme, or other paronomasia without entailing any general and consistent semantics or morpheme extractability. Rather, the resonances serve to structure paradigms, compartmentalize the lexicon, and provide psycholinguistic processing cues. Following Bickel (1995) we call this eidemic resonance. Eidemic resonance is probably best attested in small closed lexical paradigms such as personal pronouns (e.g. French singular object pronouns me, te, le, se, which rhyme and have the same syllable structure), basic kin terms (e.g. mama and papa, with the same vowels and syllable structure and similar consonants: Jakobson (1941)), essential deictics (e.g. this, that, there, etc., as the only English words with initial /ð /), and the like, but also occurs in inflectional paradigms. In Ingush and the predominant pronunciation of lowlands Chechen, there is a closed set of deictic prefixes which are in part inflectional (table 3.10). 210 Balthasar Bickel and Johanna Nichols Table 3.11 Warrgamay (Pama-Nyungan, Australia; Dixon (1980:287, 329)) Role ‘woman’ 1sg 1pl a s p ŋ ulmburu-ŋ gu ŋ ulmburu ŋ ulmburu ŋ aja ŋ ayba ŋ anya ŋ ali ŋ ali ŋ ali-nya Type: Ergative 3-way Accusative All four have pharyngeal segments or pharyngealization (spelled ‘w’ in this transcription) and /a/ vocalism and are monosyllabic. The local prefixes, which follow these, are varied in form and number of syllables, lack pharyngealization, and are an open set. 4.6 Case inventories and case terminology Case inventories range from two cases to dozens, and are usually displayed in paradigms (see section 4.1 above for some case paradigms). The various caseinflecting words of a language do not necessarily all have the same inventory of cases. In many languages of the Pama-Nyungan family of Australia, nouns have ergative case paradigms while personal pronouns have three-way or accusative paradigms. The examples from Warrgamay in table 3.11 show the three possibilities in one language. The distribution of alignment across parts of speech is motivated by expectations of agency on the indexability hierarchy (Silverstein (1976); DeLancey (1981)). The higher a referent is on this hierarchy, e.g. I in contrast to stone, the more likely this referent is to be agent. Therefore there is less need of explicit agency-marking in the form of an ergative case (because agency is already expected), and at the same time more need of explicit patientmarking in the form of an accusative case (because patienthood is not expected). And, vice versa, the lower a referent is on the hierarchy, e.g. stone in contrast to I, the more there is a need for explicit ergative-marking (the unexpected role) but the less there is for explicit patient-marking (the expected role). As a result, high-indexable referents tend toward zero vs accusative marking and low-indexables toward ergative vs zero marking. Apart from these well-motivated splits in morphological alignment, there are many instances where different words or word classes have different inventories or numbers of cases. In Chechen, for instance, nouns distinguish eight basic cases while attributive adjectives distinguish only nominative vs oblique: Inflectional morphology (50) Nominative Genitive Dative Ergative etc. ‘good’ dika dikacha dikacha dikacha 211 ‘person’ stag stegan stegana steguo This could also be described as syncretism of all oblique cases in the adjective. In various Nakh-Daghestanian languages, place names and other local nouns are often adverbs or oblique case forms in origin, and they tend to have defective declension and restricted syntactic functions. Daniel (2000) describes Bagvalal (Nakh-Daghestanian) place names as a word class midway between nouns and adverbs, with a highly defective declension lacking a nominative. In Russian, a number of nouns distinguish, in addition to the basic six cases of Russian, a second prepositional (or locative) case and/or a second genitive (or partitive) case. It might be said that the vast majority of Russian nouns (including all derived nouns) syncretize these two but a number of (underived) nouns distinguish one or the other (or both) of them. A very few nouns distinguish a separate ‘counting case’ used on nouns quantified by the numerals 2, 3, or 4, while the vast majority use the genitive for this purpose. (The ‘counting case’ differs from the genitive only in stress placement.) These various minor cases are found only on nouns; pronouns and adjectives distinguish only the basic six cases. These Russian examples differ from the others discussed in this section in that they are almost always judged to be ‘extra’ cases in a few paradigms rather than defectivity of the others. Standard schemas exist for names of cases in elaborate case systems; see Melčuk (1986); Hjelmslev (1935); Blake (1994); and grammars of various Nakh-Daghestanian and Uralic languages. In such languages the local cases tend to fall into neat series based on topography and directionality vs rest: inessive (‘in’), illative (‘into’), elative (‘out of’); adessive (‘on, at’), allative (‘onto’), ablative (‘away from’); superessive (‘on top of’), superlative (‘onto the top of’), superelative (‘off the top of’); etc. There is less uniformity of opinion and practice concerning terminology for the more grammatical cases and in smaller case systems. Cases are usually named for what is taken to be their primary function. Nominative is the classical term for the basic case or citation form (cf. Latin nomināre ‘to name’), and the term is still used in this sense in most Greek-derived and Russian-derived grammatical and linguistic traditions, while many western linguists use it only for S = A subject cases and use absolutive for S=P cases. Accusative and ergative are standard for P and A cases respectively. Dative is commonly used for a case that marks indirect objects and often some subject-like experiencers. The term is also sometimes used for primary objects, which comprise the P of monotransitives and the 212 Balthasar Bickel and Johanna Nichols Table 3.12 Russian noun paradigm ‘lake’ Nominative Genitive Dative Accusative Instrumental Prepositional ‘book’ Singular Plural Singular Plural ozero ozera ozeru ozero ozerom ozere ozera ozer ozeram ozera ozerami ozerax kniga knigi knige knigu knigoj knige knigi knig knigam knigi knigami knigax Goal argument of ditransitives (see vol. i, chapter 4), while accusative is the traditional label for direct objects, which comprise the P of monotransitives and the Theme of ditransitives. Genitive is most common for the default adnominal case, though possessive is also found. 5 Markedness and obligatoriness Morphological forms are defined through oppositions: we know that the form rivers is marked by a suffix -s ‘plural’ because we know that rivers, like hundreds of other such nouns, stands in opposition to river, without an -s suffix. It is a frequent characteristic of such oppositions that, as in this example, one member is zero-marked, i.e. has no overt marker of its own. Another frequent example for zero-marking is the nominative or absolutive case of nouns. More unusual are paradigms with zeros in other places, e.g. the genitive plural of many Russian nouns (table 3.12; zero-marked forms are boldfaced). Zero-marking is sometimes context-specific: the Belhare locative case is regularly marked by the suffix -(C)e, e.g. mi-e ‘at, to, on, in the fire’, but a few location-denoting nouns such as place names or words like khim ‘house, home’ or gaũ ‘village’ have zero-marked locatives if (and only if) they function as the goal argument of a verb of directed motion. (51) Belhare a. Dhankuta-Ø khar-e-ŋ a Dh.-loc go-past-[1sg]excl ‘I went to Dhankuta’ b. Dhankuta-e yag-he-ŋ a. Dh.-loc stay-past-[1sg]excl ‘I stayed in Dhankuta’ Inflectional morphology 213 In (51a), the place name Dhankuta has a zero locative ending because it serves as the goal argument of the verb. In (51b), locative case must be overtly marked, in contrast, because the place name is in an adjunct rather than argument function. In the terminology first established by the Prague School of linguistics, a member of a paradigm is unmarked (German merkmallos) if it does not have a semantic or syntactic value of its own on a par with the other members of the paradigm and acquires a value only through opposition with other forms.18 Zero-marked nouns in English, for example, have a singular value only through opposition with nouns marked as [+plural]. Where the opposition is neutralized, as in generic statements, the zero-marked form can be used with a non-singular value. This is why The kangaroo is native to Australia has the same truth value as Kangaroos are native to Australia. Unmarkedness tends to go together with zero marking (cf. Haiman (1985:147–51)), but the correlation is not universal: even though the genitive plural forms ozer ‘of the lakes’ and knig ‘of the books’ in table 3.12 are zero-marked, there is no context of neutralization and indeed no reason to assume that they are functionally unmarked members of the paradigm. Languages differ greatly in the number of contexts in which an opposition is obligatory and in which, as a corollary, the use of unmarked forms implies the opposite value of marked forms. While English obligatorily requires number marking for all but the generic statement context and reference to amorphous masses (e.g. sugar, water, mud), many languages draw the line between animate or human referents and the rest, requiring number marking only for nouns referring to animate beings. When referring to a group of girls, for example, one must say in Belhare kaepma-chi ‘girl-pl’; use of kaepma would entail, as in English, reference to one single girl. By contrast a word like phuŋ ‘flower’ can have either singular or plural value, and, although grammatical, phuŋ-chi ‘flower-pl’ is a rare form. Some languages go further than this, and do not require number marking in any context. This is typical for languages with numeral classifiers and many others. In Yucatec (Mayan, Mexico; Lucy (1992)), for example, a word like pèek’ ‘pig(s)’ or máak ‘man, men’ can have either singular or plural value. The use of an explicit plural suffix (-ób) is reserved for emphasis, contrast, or clarification. Optional number marking of this kind is common in languages all around the Pacific Rim. When analysing a language, it is very important to take note of differences between contexts requiring obligatory marking and contexts allowing optional marking because it is these contexts that determine the actual value of an unmarked (and often also formally zero-marked) form in discourse. If 18 Such oppositions are called privative and are contrasted with equipollent oppositions where both members are equally specified. See Baltaxe (1978) and Anderson (1989) for historiographic and theoretical surveys. 214 Balthasar Bickel and Johanna Nichols the context requires an obligatory opposition, the unmarked form will have the opposite value to the marked form (e.g. a singular value in opposition to a marked plural form). If the opposition is optional, no such implication arises, and the unmarked form can have either value (e.g., a singular or plural value). 6 Layered (hierarchical) versus templatic morphology Strings of inflectional formatives often have a layered, or hierarchical, or nested structure which can be represented as a branching tree or bracketed structure. Such a string is said to be configurational, i.e., it has a regular constituent structure. In a hierarchical string, dependencies between formatives are chiefly between adjacent ones, the choice of an allomorph can depend on a more inner formative but usually not on a more outward one, there is a single root or head, and in general the position of each formative depends on its function (or the function of its agreement trigger). An example is the following set from Quechuan (Stump (1996:236) citing Muysken (1986)): (52) Quechuan (S. America; Muysken (1986:636)) a. riku-na-chi-ku-n-ku see-recip-caus-refl-3-pl ‘Theyi caused them to see each otheri ’ b. riku-chi-na-ku-n-ku see-caus-recip-refl-3-pl ‘Theyi caused each otheri to see them’ c. riku-na-ku-chi-n-ku see-recip-refl-caus-3-pl ‘They caused themi to see each otheri ’ The relative ordering of the reciprocal, reflexive, and causative formatives determines their relative scope: (52 ) a . [[riku-na]-chi]-ku-nku ‘[[see each other]-cause]-themselves’ b . [[riku-chi]-na]-ku-nku ‘[[see cause]-each other]-themselves’ c . [[riku-na]-ku]-chi-nku ‘[[see each other]-themselves]-cause’ Some of the clearest examples of layered structure come from multiple case marking (see section 8.2 below, where these examples are discussed further): Inflectional morphology (53) 215 Huallaga Quechua (Quechuan, Peru; Weber (1989)) haacha-wan-naw mutu-n machiita-wan axe-com-sim chop-3 machete-com ‘He chops with a machete as though it were an axe’ The ordering of the comitative (‘com’) and similarity (‘sim’) cases on ‘axe’ reflects their relative scope: (53 ) [[haacha-wan]-naw] ‘[[axe with] as though]’ A more complex example comes from Kayardild: (54) Kayardild (Tangkic, Australia; Dench and Evans (1988:34–5)) maku-ntha yalawu-jarra-ntha yakura-naa-ntha woman-obl catch-past-obl fish-abl(prior)-obl dangka-karra-nguni-naa-ntha mijil-nguni-naa-nth. man-gen-instr-abl(prior)-obl net-instr-abl(prior)-obl ‘The woman must have caught fish with the man’s net’ The case suffixes on ‘man’ in this example are assigned for the following reasons: the genitive reflects the noun’s own function as possessor (of the net); the instrumental is in agreement with ‘net’, which ‘man’ modifies; the ablative is in agreement with the verbal tense and indicates prior time reference; and the oblique is in agreement with the case of the entire clause. Thus the word has the following bracketed structure: (54 ) [[[dangka-karra-] nguni-] naa-] ntha [[[man-gen-] instr-] abl(prior)-] obl Dench and Evans (1988) show that, in several of the many Australian languages exhibiting multiple case marking, local processes of metathesis, haplology, syncope, etc. superficially obscure the neat nested structure of the case strings, but these processes operate on, and thus require, the original nested assignment of the case suffixes. Hierarchical morphology in verb agreement systems is illustrated by Abkhaz. The structure of Abkhaz prefix strings is shown in (55) and table 3.13. The prefix strings include three different positions for agreement with the direct object (‘P’) or intransitive subject (‘S’), indirect object (‘IO’), and transitive subject (‘A’). The agreement morphemes used in the three different positions are essentially identical (except for minor allomorphy). In using essentially the same set of agreement morphemes and assigning different functions to different positions, Abkhaz agreement morphology is reminiscent of English clause relations, where nps are assigned different grammatical functions by different positions in the clause (and minor case on pronouns). Abkhaz could 216 Balthasar Bickel and Johanna Nichols Table 3.13 Abkhaz verb agreement 1sg 2sgm 2sgf 3sghuman 3sgm 3sgf 3sgnonhuman 1pl 2pl 3pl s(ə)- ∼ z(ə)w(ə)b(ə)d(ə)- (only in S/P slot) y(ə)l(ə)- (only in IO and A slots) y(ə)- ∼ (n)a◦ (a)- ∼ a◦ - ∼ aaš◦ (ə)- ž◦ (ə)y(ə)- ∼ r- ∼ d(ə)- thus be said to have word-internal configurationality, with relative positioning in the prefix layers determining function: (55) Structure of Abkhaz prefix strings (tam = tense–aspect–mood): s/p-io-preverb-a-stem-tam-final The S, P, IO, and A slots are filled with markers from a general person and number paradigm, as given in table 3.13 (adapted from Hewitt (1979)). In the following examples, the function of b(ə)- ‘you (fem. sg.)’ is determined by its position: (56) Abkhaz (Northwest Caucasian; Hewitt (1979)) a. bzə̀ya bə -z-bò-yt’ well 2sg.f-1sg-see-fin ‘I love you’ (p. 105) b. b-ca-r, də -b-bò-n. 2sg.f-go-if 3sg.hum-2sg.f-see-fin ‘If you had gone, you would have seen him’ (p. 173) In (56a), bə- is in the S/P position of a transitive verb form, so that it is in object (P) function. In the form bcar ‘if you had gone’ in (56b) b- is again in the S/P position, but since the verb is intransitive, it is assigned the S function. In the transitive form dəbbòn ‘you would have seen him/her’, b- follows another agreement marker and this shows that it is in the A slot, therefore in transitive subject function. Layered morphology contrasts typologically with what is called templatic morphology (Simpson and Withgott (1986); see also Spencer (1991:208ff.); Inkelas (1993); Stump (1996); Hyman (2003)). In templatic morphology the structure of the string of formatives is flat and departs in a number of ways from layered structure: there can be more than one root or head, dependencies can obtain between non-adjacent formatives, allomorphy of more inward formatives pf2 N- ‘3nsg’ N- ∼ miN- ‘neg’ pf1 mi- ‘3nsg’ -yuk ‘definitive’ -yakt ∼-ya(u) ‘ipfv’
sf1 sf3 -chi ‘du’ -i ‘1/2pl’ sf2 -(h)e ∼ -att ‘past’ -t ∼ -yuk ‘npt’ -a ‘subjunctive’ -n(i) ‘neg’ sf4 -ŋ (a) ‘excl’ -k(a(k)) ‘2’ sf5 Table 3.14 Belhare intransitive verb agreement of selected tense/aspect/mood forms (pf = prefix position, sf = suffix position,
= verb stem, N = nasal morphophoneme) 218 Balthasar Bickel and Johanna Nichols can be sensitive to more outward formatives, and the position of formatives in the string can be determined by their formal categories, or by phonological principles, rather than their syntactic or semantic functions. Templatic morphology is characteristic, for example, of verb agreement in Algonquian, Bantu, and Kiranti languages, where it regulates the sequencing of inflectional formatives. Table 3.14 illustrates the templatic structure of Belhare (Kiranti) intransitive verbs (see Bickel (1995, 2003), for a complete analysis). As is typical for templatic morphology, there are many long-distance dependencies across several affix positions. For instance, the allomorphy of the past tense marker -(h)e ∼ -att in suffix position sf2 is regulated by whether or not there is a negation marker in sf4 (-n(i)), and these are often not adjacent (e.g. n-ta-at-chi-n neg-come-pt-dual-neg ‘we two didn’t come’, with an intervening sf3 filler -chi ‘dual’). The appearance of the negative prefix in pf2 (N-) is contingent on the simultaneous presence of the sf4 negation marker (-n(i)). (There are transitive negative forms with only the sf4 negation marker, but none with only the pf2 marker.) In templatic morphology there is often a tendency for different affix positions to be characterized by the same categories: e.g. in table 3.14, all fillers of the sf1 and sf2 slots are tense, aspect, mood markers, and all fillers of the sf5 position are person markers. However, positions are not always homogeneous. The pf2 position, for instance, includes both person and negation markers. The rationale for assigning morphemes to templatic position is purely formal: fillers of the same position cannot co-occur in the same string. Therefore, a third person nonsingular negative form, as in (57a), requires the use of the pf1 filler mi‘3nsg’. Although they are semantically compatible, the markers N- ‘3nsg’ (as in 57b) and N- ‘neg’ (as in 57c) cannot co-occur and are therefore assigned the same affix slot (the negative allomorph miN- only occurs in infinitives): (57) Belhare a. mi-n-ta-at-ni 3nsg-neg-come-past-neg ‘they didn’t come’ b. n-ta-he 3nsg-come-past ‘they came’ c. n-ta-at-ni neg-come-past-neg ‘s/he didn’t come’ The ordering does not reflect any syntactic functions, as it does in the hierarchical morphology of Abkhaz, but is purely morphological (and arbitrary). Inflectional morphology 219 Occasionally, templatic ordering leads in some languages to functionally indeterminate structures, as in Maithili, where the ordering of non-nominative, honorificity-indicating agreement suffixes is rigidly fixed and allows for a variety of interpretations: (58) Maithili (Y. P. Yādava (p.c.)) dekhau-l-i-au-nh show-pt-1nom-2nonhon-3hon ‘I showed him/her to you’ ‘I showed you to him/her’ ‘I showed his/her X to you’ The sequence -i-au-nh is the only one that is possible in Maithili with three simultaneous agreement markers, and this is largely due to prosodic constraints requiring verbal desinences to consist of an end-stressed light-heavy syllable sequence (Bickel, Bisang, and Yadava (1999)). It is probably not uncommon for templatic morphology to be determined or at least historically motivated by prosodic and other phonological principles, but research on this area has just begun; see, e.g., Hyman (2003) on the sonority hierarchy as a driving source for suffix ordering in Bantu. However, templatic versus layered properties are likely to hold of individual formatives rather than of the entire string. Judging from examples in the literature, templatic properties seem to be typical of formative strings that include inflectional elements, are head-marking or detached, and are in Prae or Wackernagel position, though sometimes (as in the Belhare example mentioned above) they are in Post position. Layered properties are most common in suffixed formatives (though in Abkhaz, above, a prefix string is layered) and in dependent-marking morphology, with Australian multiple case marking surely the most extreme example. We tentatively raise these generalizations as hypotheses. Regardless of whether formatives follow the principles of templatic or layered arrangement, they tend to abide by universal semantic ordering principles, which interact with whatever other syntactic, morphological, or phonological principles determine formative order in the given language: (59) Universal affix ordering in layered morphology a. verbs: voice/aspect > modality > status/tense > evidentials/illocutionary force (Foley and Van Valin (1984); Van Valin and LaPolla (1997); Bybee (1985)) b. nouns: number > case (Greenberg (1963)) 220 Balthasar Bickel and Johanna Nichols These principles are often seen as absolute universals, but there are exceptions, and their status rather seems to be one of default principles that apply only in the absence of overriding constraints (chiefly phonological or prosodic constraints). 7 Two examples of common inflectional categories: person and number Categories that are commonly inflectional and treated in other chapters of this work include gender, deixis, tense, aspect, mood, illocutionary force, and voice oppositions of various kinds. Nominalization, causative, reflexive, reciprocal, middle, and negation are categories which, if not always strictly inflectional, at least frequently have their overt marking worked into inflectional paradigms. Two common inflectional categories treated elsewhere in this chapter are agreement and case (section 8). The rest of this section briefly describes two major inflectional categories that are covered only partially or not at all elsewhere in this chapter or this work. 7.1 Person Person concerns the grammaticalization of conceptual distinctions between participants involved in speech activities. From a pragmatic point of view, many such distinctions play a role in communication, e.g., the difference between those persons who actually attend a speech act and those who are merely referred to, between those to whom an utterance is targeted and those who happen to hear it as bystanders, etc. (see Levinson (1988) for an analysis of such notions). Grammars typically conflate such distinctions and reduce the system to three terms grammaticalizing the roles of speaker (first person), addressee (second person), and other (third person), respectively. While this triad is the most common system worldwide, other ways of dividing up the conceptual space of person are also found, and we briefly discuss them in the following. Note, however, that person systems other than the standard triad often apply to verbs only, or pronouns only; it is not uncommon to find splits here across parts of speech. 7.1.1 Exclusive versus inclusive Many languages distinguish between an exclusive and inclusive conception of the first person, and in many cases these are subcategories of plural (or dual) number marking. An example is found in So, a language spoken in the Uganda– Kenya border area. Exclusive here refers to the speaker and his or her group, but excluding the addressee(s). The inclusive forms, by contrast, explicitly include the addressee(s) along with the speaker and his or her group in the notion of ‘we’. Inflectional morphology 221 Table 3.15 So pronouns (Kuliak, E. Africa; Serzisko (1993)) Singular Plural 1 aya 2 3 piya ica exclusive: inia inclusive: isia pitia itia Table 3.16 Belhare intransitive verb agreement (
= stem, N = nasal morphophoneme) excl incl 2 3 Singular Dual Plural -ŋ a -chi-ŋ a -chi -chi-ga N-
-chi -i-ŋ a -i -i-ga N- -ga
- Some languages treat the exclusive versus inclusive distinction on a par with the basic second versus third distinction rather than as a subcategory of plural first persons. In such a system, exclusive and inclusive have singular values, just as the other persons do. Table 3.16 is an example from Belhare intransitive verb agreement (cf. Table 3.14 for the templatic arrangement of affixes, and table 3.6 for a sample paradigm in the singular). For the exclusive (‘speaker(s) but not addressee’) this works without complications, since restricting the reference to one person simply means reference to the speaker. The inclusive, by contrast, does not allow a true singular value because it comprises both the speaker and the addressee and thus requires at least two referents. While Belhare sidesteps this issue by not having an overt inclusive marker at all, many languages of Siberia, North America, and Northern Australia use a different kind of number system to accommodate the inclusive as a basic person category: instead of distinguishing singular versus non-singular, these languages distinguish minimal versus augmented number (McKay quoted by Dixon (1980:351–6)). Table 3.17 illustrates this in a Siouan language of North America. Minimal means singular for exclusive (ha- ‘I’), second person (ra‘you [sg.]’), and third person, but for the inclusive person minimal entails dual number reference, i.e. h˜- ‘thou and I’. Augmented is plural for all persons (h˜-wi ‘you and I’, ha- -wi ‘we, excluding you’). In Northern Australian languages, a third term, unit augmented, is sometimes distinguished. This translates as 222 Balthasar Bickel and Johanna Nichols Table 3.17 Hocak (a.k.a. Winnebago; Siouan) subject agreement (root xé ∼ xa ‘bury’: Lipkind (1945)) excl incl 2 3 Minimal (-ø) Augmented (-wi) ha-xé h˜-xé ra-xé xé ha-xa-wı́ h˜-xa-wı́ ra-xa-wi xa-wı́ ‘I bury him’ ‘thou and I bury him’ ‘thou buriest him’ ‘he buries him’ ‘we (they and I) bury him’ ‘we (you and I) bury him’ ‘you bury him’ ‘they bury him’ Table 3.18 Rembarrnga pronouns (N. Australia: Dixon (1980:351–6) after McKay excl incl 2 3 masc 3 fem. Minimal Unit augmented (-pparraʔ) Augmented (-ə) ŋənə yəkkə kə nawə ŋ atə yarr-pparraʔ ŋ akorr-parraʔ nakorr-parraʔ parr-pparraʔ parr-pparraʔ yarr-ə ŋ akorr-ə nakorr-ə parr-ə parr-ə trial for the inclusive and dual for the other persons, as in Rembarrnga (see table 3.18). The inclusive minimal form yəkkə refers to a simple set of speaker and addressee and thus has a dual referent; the unit-augmented form ŋakorrparraʔ adds to this one more referent and therefore has a trial referent (I, you, and one other person); the augmented ŋakorrə finally adds further referents, and thus has a plural value (I, you, and several others). For all other persons, the minimal has a singular value (thus, ŋənə ‘I’, nawə ‘he’, etc.), the unit-augmented forms have a dual value (thus, yarrpparraʔ ‘the two of us, without you’, parrpparraʔ ‘the two of them’, etc.), and the augmented forms have a plural value (yarrə ‘we, without you’, parrə ‘they’). The diagnostic feature of augmented number systems is an additional dual or trial number found only with first person inclusive forms (e.g. Hocak h˜- ‘1 dual inclusive’, but no form glossed ‘1 dual exclusive’). When the description leads one to positing such an additional number, a reanalysis in terms of augmentation is usually called for (cf. Dixon (1980)). It is important to note that in all of these systems in which inclusive and exclusive are independent person categories there really is no generalized first person singular concept, no term corresponding to English I or So aya. Reference to speaker alone is always achieved indirectly by minimizing or singularizing the Inflectional morphology 223 category of the exclusive person. Only in languages where inclusive/exclusive is a subtype of first person plural (as in So), and of course in languages like English which lack any inclusive/exclusive distinction, is there a true generalized first person singular pronoun. 7.1.2 Conjunct/disjunct systems While the distinction between first and second person as indices to the speaker and addressee, respectively, is the most common type worldwide, typological research has established that this is not the only one possible. A few languages in Asia and South America have grammaticalized a completely different categorization, at least in verb agreement. One person, usually labelled ‘conjunct’,19 refers to the speaker in statements and to the addressee in questions (excluding rhetorical questions, which are really statements in function). Thus, the conjunct person form wonā in Newar, the Tibeto-Burman language of the Nepalese capital Kathmandu, can mean ‘I went’ or ‘did you go?’. This is in opposition to what is called a disjunct form, wona, which is used for all other situations, i.e. meaning ‘you went’ or ‘s/he went’ or ‘did s/he go?’ or, where this makes sense in context, ‘did I go?’. What is at the functional core of the conjunct person category is the indexing of what Bickel (2001) calls the informant, i.e. the person who the speaker supposes or claims to be the immediate supplier of the information. In statements, this is the speaker himself or herself, but in questions this role of informant is attributed to the addressee. The disjunct person indexes any participant who is not the informant in the speech situation. Conjunct/disjunct systems are sometimes geared toward agents in the sense of volitional instigators of situations. In Newar (A. Hale (1980); Hargreaves (1991)) and some other Tibeto-Burman languages, conjunct person marking generally applies only to such referents and therefore only to volitional or controlled verbs.20 In other languages, however, the distinction applies to other arguments as well, and one occasionally finds it applied to both actors and undergoers marked differently. The South American language Awa Pit, for instance, has agreement differentiation in conjunct marking: 19 20 The term is from A. Hale’s (1980) pioneering description of the phenomenon in Newar. The less than ideally transparent terminology derives from the use of conjunct forms in reported speech where the form marks coreference (referential ‘conjunction’) of the subject with the speaker referent reported in the matrix clause (i.e. it has the same effect as a logophoric marker). Alternative terms found in the literature are locutor, egophoric, subjective, and congruent; cf. Curnow (2002). In Tibetan, this has to do with the historical source of the distinction, which is an epistemological category focussed on agency. See DeLancey (1990, 1992) and Bickel (2000b) for discussion of this; and Dickinson (2000) for a study of epistemological categories and conjunct person in Tsafiki (Barbacoan, Ecuador). 224 Balthasar Bickel and Johanna Nichols (60) Awa Pit (Barbacoan; Ecuador and Columbia; Curnow (2002)) a. k-in-ka=na, na=na Santos=ta dawn-when=top 1sg[nom]=top S.=acc izh-ta-w see-past-conjunct.subject ‘At dawn I saw Santos’ b. shi ayuk=ta=ma libro ta-ta-w? what inside=loc=q book put-past=conjunct.subject ‘Under what did you put the book?’ c. Juan=na (na=wa) izh-t-i-s j.=top 1sg=acc see-past-conjunct.undergoer ‘Juan saw me’ d. nu=wa=na m-in=ma pyan-t-i-s? 2sg=acc=top who=q hit-past-conjunct.undergoer ‘Who hit you?’ e. p-ina alu ki-mat-i-zi very rain do-pfv-past-disjunct ‘It rained heavily’ In (60a) and (60b), the verb is marked for a conjunct person subject: in (60a), a statement, it indexes the speaker; in (60b), a question, it indexes the addressee. The examples in (60c) and (60d) illustrate the conjunct person in undergoer function, again indexing the speaker in a statement (60c) and the addressee in a question (60d). Example (60e) illustrates disjunct marking, which signals that the conjunct person is neither subject nor personally affected by the situation. 7.1.3 Person and the indexability hierarchy In most languages, the person triad and the conjunct/disjunct opposition are not disjointed sets of terms but form a tightly structured hierarchy which is responsible for various morphosyntactic effects. At the core of the hierarchy is the distinction between speech-act participants and third person referents, but the hierarchy is often elaborated in distinguishing, among third persons, between human and non-human referents, or between animate and inanimate referents. Sometimes other parameters, such as anaphoricity or definiteness, gender, kinship, number, possession, size, and discreteness or segmentability, affect the structure of the hierarchy as well. The hierarchy has many effects ranging from number differentiation to splits in case-marking patterns, and we will review some of them below. We refer to the hierarchy as the indexability hierarchy (Bickel (1999)) since its basic variable is the ease with who a referent can be identified – or ‘indexed’ – from within the speech-act situation. Identification is easiest for speaker and addressee, who are necessarily Inflectional morphology 225 co-present, and it is easier for human referents than for other animates because humans tend to be topics in ordinary discourse and are therefore cognitively more accessible. Singular and individualized referents are generally easier to point at unambiguously than groups or masses, so that in many languages they figure higher on the indexability hierarchy. Alternative terms like animacy, agency, generic topicality, egocentricity, or empathy hierarchy that have been proposed in the literature (cf., among many others, Comrie (1981a); DeLancey (1981); Givón (1994))21 capture some, but not other aspects of the hierarchy. Note, however, that there is considerable (but at present ill-understood) cross-linguistic variation in the details of how the hierarchy is set up among third person referents, and different parameters may prove relevant in different languages. While such details vary, one way of distinguishing among non-speech-act participants is particularly noteworthy from a typological point of view: some languages expand the indexability hierarchy beyond the traditional person triad by adding a fourth (or obviative) and sometimes even a fifth (or further obviative) person.22 Such extensions are best known from Algonquian languages but they are also attested in a few other North American languages. Depending on a number of syntax and discourse factors, nps in these languages appear in discourse as either third or fourth (or fifth) person. In Cree, fourth person (also called obviative) is marked by the suffix -a; third person (also called proximative) is zero-marked. This difference has a reflex on verb agreement. Agreement in Cree and other Algonquian languages is in person–number but it does not indicate role. To indicate the roles, verbs are marked as what is called ‘direct’ or ‘inverse’: a direct marker signals that the A argument is higher on the indexability hierarchy than the P argument, while an inverse marker establishes the reverse role assignment, with a person lower on the hierarchy acting on a person higher. This mechanism applies equally to positions in the hierarchy. Thus, if a third person acts on a fourth person (downwards, as it were), the verb will be marked as direct. If a fourth person acts on a third person (upwards, as it were), the verb will be marked as inverse. The same logic applies when, for example, a first and a third person are involved. Again, if the action goes ‘down’ the hierarchy (first acting on third), the marking is direct. If the action goes ‘up’ the hierarchy (third acting on first), the verb is marked as inverse. The following examples illustrate this. 21 22 The hierarchy was first extensively discussed by Silverstein (1976), but there are many precursors, to say nothing of the very fact that person categories are referred to by the numbers 1, 2, 3 in both the Graeco-Roman and the Indic linguistic traditions (although in different order: for the Indian grammarians, the speaker was ‘3’). Note that the label ‘fourth person’ is sometimes used in a different sense. In descriptions of Eskimoan languages, for example, it is the traditional label for reflexives. 226 Balthasar Bickel and Johanna Nichols (61) Plains Cree (Algonquian; N. America; Dahlstrom (1986)) a. e -wa pam-a -ya hk-ik b. e -wa pam-iko-ya hk-ik det-see-dir-1pl.excl-3pl (conj) det-see-inv-1pl.excl-3pl (conj) ‘They (3) see usexcl (1)’ ‘Weexcl (1) see them (3)’ d. e -wa pam-iko-t c. e -wa pam-a -t det-see-dir-3[sg][-4sg] (conj) det-see-inv-3[sg][-4sg] (conj) ‘He (3) sees him (4)’ ‘He (4) sees him (3)’ In (61a), the direct marker -a signals that a first person acts on a third person. In (61b) this is reversed, and it is the third person that acts on the the first. This is exactly parallel to (61c) and (61d), respectively, but here the relationship is between a third and a fourth (obviative) person (zero-marked here): in (61c) this relationship is direct, so that the third (proximate) person acts on the fourth; in (61d) the relationship is inverse, so that the fourth person acts on the third. Determining which referent is third and which one is fourth (obviative) depends by and large on topicality or other prominence in discourse. But there are also purely syntactic factors involved: a possessor, for instance, is always higher on the hierarchy than its possessed object (Wolfart (1978)). Algonquian languages differ in how syntactic and discourse factors compete in determining person assignment (Rhodes (1990); Mithun (1999:76f.)). Scenarios involving speech-act participants only (‘I saw you’, ‘you saw me’) often enjoy a special status on the hierarchy. Sometimes speech-act participants are ranked: in Plains Cree, for instance, the second person takes preference over the first in triggering person marking (in independent mood forms). But the inverse/direct marking does not apply in I/you and you/me scenarios, and instead there are portmanteau morphemes signalling ‘1>2’ (-iti) or ‘2>1’ (-i) (where ‘>’ indicates a transitive relationship with the first term as subject and the second as object).23 Portmanteau morphemes for these person sets are a widespread phenomenon worldwide (as noted by, among others, Hagège (1982:107); Heath (1991, 1998); Bickel (2000b); Jacquesson (2001)). Kiranti and many other Tibeto-Burman languages, for instance, have dedicated agreement markers for the ‘1>2’ relation (e.g. Belhare nise-na (see-1>2) ‘I saw you’). Some languages, such as the Indo-Aryan language Maithili, neutralize scenarios here and have only one form covering both ‘1>2’ and ‘2>1’ relations (e.g. dekhl-i ‘I saw youhon. ’ or ‘Youhon. saw me’: Bickel et al. (1999)). The reason for blurring the nature of the relationship or coding it by a portmanteau morpheme is probably, as Heath (1991:86) suggests, that such scenarios are ‘doubly dangerous’ since ‘they not only combine the most pragmatically 23 Alternatively, one could analyse -iti and -i as markers of inverse and direct relations, specialized for scenarios involving only speech-act participants (Dahlstrom (1986)). For discussion, see Bickel (1995). Inflectional morphology 227 Table 3.19 Old Church Slavic number paradigm (Huntley (1993:140)) ‘woman’ Singular Dual Plural Vocative Nominative Accusative Genitive Dative Instrumental Locative ženo žena ženo ženy ženě ženojo ženě ženě ženě ženu ženama ženama ženu ženy ženy ženŭ ženamŭ ženami ženaxŭ sensitive pronominals’ but ‘also combine them into a syntagmatic structure and thereby necessarily focus on the speaker–addressee relationship’. Another type of person that is often specially marked is generic or nonspecific person. English uses second person pronouns in this function, e.g. You win a few, you lose a few. Some languages have a dedicated generic person form which is grammatically third person in verb agreement, e.g. German man, French on, Hausa a(n) (Newman (2000:486)), or the Slave (Athabaskan) prefix ts’(Rice (2000:187)). In other languages it is the first person inclusive category that is used for generic reference. For instance, the Belhare form hiu-t-i ‘cannpt-1pl[incl]’ can either specifically mean ‘us’ including the addressee(s) (‘we can (do it)’), or it can be meant in the generic sense of ‘one can (do it)’. 7.2 Number Number is, minimally, an opposition of singular to plural.24 Less common numbers are dual (two individuals), trial (three individuals), and paucal (a few individuals). Old Church Slavic makes a singular/dual/plural opposition in nouns, pronouns, adjectives, and verbs (see table 3.19). In a number of languages, verbs make an aspectual or aspect-like distinction of single versus multiple action, often in addition to singular versus nonsingular agreement. An example from Chechen is in table 3.20 (semelfactive = single action; pluractional = multiple action). Number-like categories include distributives (which imply a plurality of separate individuals) and collectives (which imply a number of individuals viewed as a set). 24 See Corbett (2000b) for an exhaustive treatment of number. 228 Balthasar Bickel and Johanna Nichols Table 3.20 The Chechen verb ‘drive’. 1x = once, Nx = many times Singular Plural Semelfactive Pluractional loallu ‘one drives one 1x’ loaxku ‘one drives many 1x’ loellu ‘one drives one Nx’ loexku ‘one drives many Nx’ Number often shares formatives or at least paradigms and position slots with person, and number agreement is systematically marked in the great majority of languages having person agreement on the verb. On other parts of speech, number is more likely to be optional or missing entirely. It is fairly common for number not to be marked overtly on nouns. It may be marked instead on an article or plural word (illustrated for Yapese in (42) above), and many languages have number marking on verbs although the nouns with which the verbs agree in number have no overt number marking themselves; an example of such a language is Lakhota (Siouan, North America). In a number of languages, verbs make more number distinctions than do nouns (e.g. verbs in Yimas distinguish singular/dual/paucal/plural while nouns distinguish only singular/dual/plural). Where present in a language, number marking is likely to be optional on nouns, especially those in the lower reaches of the indexability hierarchy; or it may be available only to animate or human nouns or other high-indexability nouns (see section 5 above). Personal pronouns are more likely than nouns to make number distinctions, and pronominal formatives more likely to distinguish number than independent pronouns. These and other patterns of optionality and limitation in number categories are briefly reviewed in Nichols (1992:144ff.). An unusual marking of number is number toggling (or ‘inverse number marking’) in the Kiowa-Tanoan languages (Wonderly Gibson, and Kirte (1954); Watkins and McKenzie (1984:78ff.); Weigel (1993)), in which nouns have inherent number, every noun being either singular or plural, and the suffix -go (and its allomorphs) toggles singular to plural and vice versa. Number intersects with person in various ways, and this has impacts on the referential value of number categories. One instance of this is the effect of exclusive versus inclusive distinctions on number, which in some cases yields, as we saw in section 7.1.1, a distinction between minimal and augmented rather than between singular and plural. Another effect is that nonsingular in the first person usually means ‘the speaker and his/her group’ rather than a multitude of simultaneous speakers (Jespersen (1924b/1969: 192)). Some languages allow this use of nonsingular forms with other nouns as well. Belhare ama-chi, for instance, does not refer to several mothers but rather to ‘my mother and her people’ (e.g. sisters, friends, etc., depending on the situation). This type of nonsingular number, known as associative number, is a distinct category of its Inflectional morphology 229 own in a few languages (Moravcsik (1994); Corbett and Mithun (1996)): in Hungarian, it is marked by the suffix -ék (Jánosék ‘John and his associates)’, distinct from the ordinary plural -ok (Jánosok ‘several Johns’). Similar contrasts are found in Pomoan and Eskimo languages. Associative numbers are usually confined to names, kin terms, titles, and occupations and do not usually extend to common nouns. However, with inanimate nouns, a similar notion is sometimes expressed by echo words, in which a word is repeated with some mutation. In many Eurasian languages, this involves replacing the initial consonant, cf. Nepali raksi-saksi ‘raksi (a distilled alcoholic beverage) and things that go with it (snacks, etc.) or are similar in kind (beer, etc.)’ with default mutation to /s/, or Turkish çocuk-mocuk ‘children and all that goes with them (toys, games, etc.)’ with default mutation to /m/. Most South Asian languages extend echo-wordformation to other parts of speech, e.g., Hindi nahā-vahā ‘bathe and do whatever goes with this (dry, get dressed again, etc.)’ or jaldi-valdi ‘fast, etc.’. In these cases, the semantic effect is sometimes more generally one of inspecificity than of association. See Abbi (1994:27–33) for a discussion of semantic variation in South Asian echo words. 8 Morphology in syntax 8.1 Agreement Agreement is the phenomenon by which a word carries morphological features that originate somewhere else. For instance, a verb agrees in person with its subject or a modifying adjective agrees in case with the head noun. There are two fundamentally different types, based on where the features originate: head-driven and dependent-driven agreement. Head-driven agreement consists in percolating features from the phrasal head to its dependents, e.g. from the noun heading a noun phrase to some or all of its dependents. The result of this is dependent marking in the sense defined in section 2. Consider (1) in the introductory section, from German, or the example from Hindi in (62). In this language, agreement targets not only adjectives but also the adnominal postposition kā ‘of’: (62) Hindi (Indo-European; South Asia) a. lar.k-õ=k-ā chot-ā boy-pl.obl=of-masc.sg small-masc.sg ‘the small room of the boys’ b. lar.k-õ=k-e chot-e boy-pl.obl=of-masc.pl small-masc.pl ‘the small rooms of the boys’ kamr-ā room(masc)-sg.nom kamr-e room(masc)-pl.nom 230 Balthasar Bickel and Johanna Nichols If the head noun is nominative masculine singular, adjective and postposition end in -ā (62a); if the head noun is nominative masculine plural, adjective and postposition end in -e (62b). Head-driven agreement usually involves gender, number, and/or case and chiefly affects nps. On the vp and clause level, head-driven agreement is sometimes found in the form of transitivity or tense agreement. Transitivity agreement is illustrated by the Australian language Yidi, where it is required across the verbs in a complex predicate vp: (63) Yidi (Pama-Nyungan, NE Australia; Dixon (1977:252)) guwal dyara -l gali-ŋ al-nyu, bulmba. name[abs] put-past go-appl:com-past place[abs] ‘[He] gave names to all the places as he went along’ (p. 522) In this example, the intrinsically intransitive verb gali- ‘go’ receives a comitative applicative marker that increases its valence and thus allows the verb to match the valence of the head verb dyara- ‘put’. Tense agreement is illustrated by the Uto-Aztecan language Luiseño: (64) Luiseño (Uto-Aztecan, S. California; Steele (1990)) noo=n=il čaqalaqi-qus. hengeemal-i 1sg=1sg=past tickle-past boy-acc ‘I was tickling the boy’ (p. 3) Both the auxiliary (=nil) in the Wackernagel clitic position and the lexical verb (čaqalaqiqus.) are marked as past tense, and they must agree in this marking. Dependent-driven agreement is the mirror image of head-driven agreement, with features copied from a dependent usually to the head. Classic examples are the registration of possessors on the head noun in an np (as in the Hungarian and Abkhaz examples, (26) and (29) in section 2 above), or the registration of arguments on a verb. The following Belhare examples illustrate both: (65) Belhare a. ŋ ka-ha a-tak 1sg-gen 1sg.poss-friend ‘my friend’ b. un-chik-ŋ a ŋ ka ma-ŋ -ni-at-ni 3-nsg-erg 1sg[abs] 1sg.p-3nsg.a-see-past-neg ‘They didn’t see me’ In (65a), the head tak ‘friend’ of the np registers the person and number of its possessive dependent. In (65b), the verb ni- ‘see, know’ agrees with both the A-argument unchikŋa ‘they’ and the P-argument ŋka ‘me’. Dependent-driven agreement typically targets the head only. But occasional examples of multiple Inflectional morphology 231 targets are attested. Consider the following examples from Archi (agreement formatives are boldfaced): (66) Archi (Nakh-Daghestanian, NE Caucasus; Kibrik (1994:349)) a. buwa-mu b-ez dit abu   ◦ alli abu father-erg iii-1sg.dat early:iii bread(iii):abs.sg make:iii ‘Father made the bread for me early’ b. nenabu   ◦ alli abu 1incl.erg:iii bread(iii):abs.sg make:iii ‘We made the bread’ In (66a), the absolutive argument ◦ alli ‘bread’ is in gender iii and this feature is matched by nearly all constituents of the clause, including not only the head of the clause, i.e. the predicate (abu ‘made.it’) but also other dependents such as adverbs (ditabu ‘early’) and pronominal arguments (bez ‘me’). Whether or not a constituent undergoes agreement depends on the availability of morphological slots on it. Nouns do not have such a slot, which is why buwamu ‘father’ in (66a) does not show agreement, unlike the pronoun nenabu ‘we(incl.)’ in (66b). (Note that agreement markers are infixed in most instances.) Another case of multiple agreement targets is found in Coahuilteco, an extinct language isolate of southern Texas. In this language, subject agreement is manifested on the verb and on dependent object nps (including embedded clauses). Thus, both the verb form and the shape of the accusative suffix (boldface) are determined by the person of the subject referent: (67) Coahuilteco (isolate; N. America; Troike (1981)) a. Dios tupo -n naxo-xt’e wal wako God dem-acc.1 1pl.s-annoy caus ‘We annoyed God’ b. Dios tupo -m xa-ka wa xo e? God dem-acc.2 2s-love aux q ‘Do you love God?’ c. Dios tupo -t a-pa-k’tace y God dem-acc.3 3s-sub-pray:pl ‘that (all) pray to God’ Dependent-driven agreement is by and large limited to features specifying referents, and this is why cross-reference is often used as an alternative term. Typical examples involve inflection of nouns or verbs for person, number, and gender of referents. Nonreferential features like case are rarely affected by dependent-driven agreement (but see Bickel et al. (1999) for an example from Maithili). Clause-level categories like mood are equally rare in dependentdriven agreement. However, in some languages, question words sometimes 232 Balthasar Bickel and Johanna Nichols trigger interrogative mood marking on the verb. This is obligatory in Greenlandic Eskimo (Sadock (1984)) and Hausa (Newman (2000:493)), and is an optional possibility in Japanese (Hinds (1984)): (68) West Greenlandic Eskimo (Eskimo-Aleut, Greenland; Sadock (1984:200)) kina maanii-ppa? who be.here-3sg.interrogative ‘Who is here?’ In these languages, interrogative mood also appears in polar (‘yes/no’) questions, where it is not triggered by question words. The Papuan language Tauya, by contrast, has a dedicated mood (-ne) for parametric (‘wh’) questions, distinct from the mood marking polar questions (-nae ∼ -nayae). Thus, the parametric mood only appears as the result of agreement: (69) Tauya (Adelbert Range, Papua New Guinea; McDonald (1990)) we fofe-ʔ e-ne? who come-3sg.fut-parametric.interrogative ‘Who will come?’ Dependent-driven agreement, especially on the clause level, is often sensitive to the nature of the relationship between the dependent and the head. One distinction is that between grammatical and pronominal agreement.25 Grammatical person/number agreement marks a relationship between the verb and argument nps. This is illustrated by the examples in (65b) through (67) above, or, indeed, by the subject agreement found in the English translations of these examples. Pronominal agreement, in contrast, does not mark a relationship between verb and argument nps; rather, the agreement morphology absorbs argument positions and consequently the agreement-triggering nps can no longer overtly appear in these positions. Put differently, grammatical agreement points to an argument while pronominal agreement is the argument. This is the case, for example, in Irish: (70) 25 Irish (McCloskey and K. Hale (1984)) (pronominal agreement) a. chuirfinn (*mé) isteach ar an phost sin put:1sg.cond 1sg in on art job dem ‘I would apply for that job’ b. churfeadh Eoghan isteach ar an phost sin put:cond e. in on art job dem ‘Owen would apply for that job’ This distinction has a long tradition (but terminology varies). The idea was first introduced by Du Ponceau (1819) and von Humboldt (1836) and had a veritable renaissance in the mid-1980s (see, among others, Jelinek (1984); Mithun (1985); Van Valin (1985); Bresnan and Mchombo (1987)). Inflectional morphology 233 In (70a), the verb is inflected for first person singular. This inflection absorbs the subject argument position, and therefore no np (mé ‘I’) can fill this position in the clause. If the verb is not inflected for person and number, as in (70b), subject nps (here, Eoghan) can occur overtly. Similar patterns are found all over the world, e.g. in many languages of the Americas (cf. Popjes and Popjes (1986) on a Jê language; Abbot (1991) on a Carib language; and Galloway (1993) on a Salishan language) and in several Semitic languages. The ban on overt agreement-triggering nps is often not general but concerns a specific phrase-structural position reserved for true arguments. In Chichewa, object nps can co-occur with pronominal agreement markers if they are moved out of their canonical postverbal argument position into topic (or afterthought) position: (71) Chichewa (Bantu, E. Africa; Bresnan and Mchombo (1987:751)) a. ??ndi-kufúná kutı́ [VP mu-wa-páts-é a-lenje] mphâtso 1sg.s-want comp 2sg.a-3.pl(ii).p-give-sub ii-hunter gift ‘I want you to give them a gift, the hunters’ b. ndi-kufúná kutı́ [VP mu-wa-páts-é mphâtso] a-lenje 1sg.s-want comp 2sg.a-3pl(ii).p-give-sub gift ii-hunter ‘I want you to give them a gift, the hunters’ Example (71a) is unacceptable because the primary object alenje ‘the hunters’ occupies the vp-internal argument position that is already filled by the agreement marker wa-, which denotes a class II (= plural animate) noun in primary object (‘P’) function.26 Moving the np out of the vp into an afterthought (or fronted topic) position as in (71b) resolves this problem. A similar possibility is given in many Amazonian languages, e.g. in Yagua (Peba-Yagua family; Everett (1989)) or Maxakalı́ (Jê; Rodrigues (1999)). When nps are removed from argument positions, their relation to agreement markers is no longer one of feature-matching. Instead, it is one of anaphoric resumption. In this respect, pronominal agreement markers resemble cliticized or incorporated pronouns. However, unlike pronouns, pronominal agreement markers are formatives, not grammatical words. One effect of this is that they have more referential possibilities than pronouns. For instance, they can have indefinite reference (‘someone, something’) without any special marking. Ordinary pronouns (like he, she, it) usually do not have this option. See Evans (1999) for detailed discussion. The diagnostic feature of pronominal agreement is that nps in the same argument role as the agreement markers are banned from syntactic argument (actant) positions in the clause. Whether or not overt nps occur at all in the sentence is a different issue. In most languages, nps are completely optional in all 26 The notion ‘primary object’ is discussed in vol. i. chapter 4, section 2.3. 234 Balthasar Bickel and Johanna Nichols positions, regardless of whether the language has grammatical agreement (e.g. Latin, Belhare, or Maithili) or pronominal agreement (e.g. Maxakalı́, Yagua, or Chichewa). Note, however, that there are split systems. Agreement systems can be grammatical in some part (say, subject agreement, or object agreement with animate nps) and pronominal in other parts (e.g. object agreement, or object agreement with inanimate nps). Such splits generally reflect ongoing processes of grammaticalization: pronominal agreement involves the same kind of anaphoric links that are found in discourse in general, and, over time, these links can become strengthened and grammaticalized. This results in grammatical agreement systems. See Givón (1976, 1984) for exemplification and discussion. Grammatical agreement systems are all based on relating features in the agreement trigger and features expressed by the agreement morphology. In most cases, this relation consists in unifying (or merging) the features so as to create one single referential expression: even though in e.g. he walk-s there are two different referential indexes, one implied by the np and one implied by the agreement desinence -s, there is only one single referent expressed. This and similar agreement systems are what we call integrative agreement systems. In addition, there also exist a s s o c i at i v e agreement systems (Bickel (2000a)), which employ different ways of relating features. In associative systems, which are characteristic of many Tibeto-Burman and Australian languages, the features of the agreement trigger enter into a variety of relations with the features expressed by agreement morphology. A particularly rich example is found in Lai Chin: (72) 27 Lai Chin (Tibeto-Burman, W. Burma; Bickel (2000a)) a. a-maʔ a-ni 3[sg]-dem 3[sg]s-laugh ‘S/he laughs’ (identity) b. a-háw daʔ nà-n-raʔ ? 3[sg]-who q 2-pl.s-come ‘Who of you came?’ (part of) 27 c. tsó n piak tu niʔ làw ka-thloʔ vé teacher erg field 1[sg]a[-3sg.p]-work even ‘Even as a teacher I can work the field’ (apposition) d. ka-lùŋ na-r
`ŋ 1[sg]poss-heart 2[sg]s-suspicious ‘I suspect you’ (other relation) In keeping with the isolating morphology of this language, words like tsón piak tu niʔ ‘teacher erg’ are unitary from the point of view of syntax and lexicon but not from the point of view of phonology. Spaces demarcate phonological, not grammatical, word boundaries. Inflectional morphology 235 Only in example (72a) do features merge into unified reference to a single third person. In (72b), the subject argument aháw ‘who’ represents a subset of the referents expressed by the corresponding subject agreement prefix nàn- ‘you (pl.)’. In (72c), the subject tson piak tu niʔ ‘teacher’ is understood as a secondary predicate (a copredicate) of the subject (A) prefix ka- ‘I’. The most complex relation is found in (72d), where the subject np, of which ‘r
`ŋ ‘be suspicious, be green’ is predicated, is kalùŋ ‘my heart’. As a subject, this np triggers agreement in the corresponding subject agreement slot on the verb. However, it is not the third person singular feature of this np (nor the possessor’s features) that are registered there, but rather the features of the referent with regard to whom the predication holds, here na- ‘you (sg.)’. In systems like these, the feature specification in the verb agreement morphology is independent of the specifications in the agreement-triggering nps. The two feature sets are then related to each other through the agreement relation itself, and this is done in the various ways indicated in (72) above. Integrative systems, by contrast, involve one unitary set of features and the agreement relation merely assures this unity; it does not create it. 8.2 Case spreading and stacking Cases and adpositions can also appear on words secondarily, i.e. not because they are directly assigned but because they are assigned to some other word with which the host stands in some syntactic relationship. There are two types of secondary case assignment: spreading and stacking. Both contrast with inert behaviour, where no secondary cases appear. Inert behaviour is the simplest situation and the most common type cross-linguistically. Copying and agreement of cases and adpositions can generically be called spreading. Spreading of cases within the np is common in Utian and IndoEuropean languages: (73) Southern Sierra Miwok (Utian, California; Broadbent (1964)) a. cyty-ʔ naŋ a-ʔ good-nom man-nom ‘a/the good man’ b. ʔ i-s-ʔ ok cyl a-s that-instr-that awl-instr ‘with that awl’ (74) Latin a. ascia nova axe.nom new.nom ‘a/the new axe’ 236 Balthasar Bickel and Johanna Nichols b. asciā novā axe.abl new.abl ‘with a/the new axe’ In a language with inert cases, case would be marked only once for the np here. In Belhare, for example, Latin asciā novā ‘with the new axe’ would translate as uchoũat phendikŋa, where the instrumental case suffix -ŋa appears only once on the head; in fact spreading would be ungrammatical (*uchoũatna phendikŋa ‘new-instr axe-instr’). When case is inert, it has scope over the whole phrase. Although the instrumental is not marked on the adjective in a Belhare np, the adjective is still in the scope of this case marker, and it therefore refers to the quality of the instrument ‘axe’ here. The adjective does not constitute an independent nominative np. Because of their phrasal scope, inert case markers are sometimes analysed as cliticized adpositions, on the assumption that phrasal scope means that markers are attached to the whole np (a phrase) rather than to the head noun (a word). However, if carried through its logical conclusion, such an analysis would suggest, counterintuitively, that the English plural is a cliticized postposition: it too has phrasal scope and the plural does not spread onto adjectives (as it does in German, cf. gross-e Häuser with big-Ø house-s, where gross ‘big’ is marked as plural in German – cf. gross-es Haus in the singular). Phrasal scope is a result of morphological inertness; it does not require adpositions, i.e. syntactically independent words. Spreading of adpositions is rare. An example is preposition repetition in Old Russian (Klenin (1989)): (75) Old Russian a. za ego djadeju za Matfěem” after his uncle.instr after Matthew.instr ‘after his uncle Matthew’ b. pro kolokol” pro nemec’skyi about bell.acc about German.acc ‘about (the) German bell’ In (75a), Matfěem” is in apposition to djadeju ‘uncle’, and in (75b) nemec’skyi ‘German’ is an adjective modifying kolokol” ‘bell’ and agreeing with it in gender and number. In both, the preposition preceding the head noun spreads to its modifier. np-internal spreading can be subject to various restrictions. In several Finnic languages, spreading is limited to only some of the cases and found on only some adjectives. In Chechen, as shown in section 4.6 above, attributive adjectives distinguish only nominative versus oblique cases, which is to say that all oblique cases syncretize in spreading. Inflectional morphology 237 It is common for case to be inert on continuous nps but spreading on discontinuous nps. In many languages, case agreement is found only when the phrase is discontinuous, i.e., interrupted by other sentential material that does not belong to the phrase. This is true of many Australian languages: (76) Warlpiri (Pama-Nyungan, C. Australia; Hale et al. (1995:1434)) a. [np [N maliki] [A wiri-ngki]] =ji yarlku-rnu dog big-erg =[perf-]1sg.p bite-past ‘A big dog bit me’ b. [N maliki-rli] =ji yarlku-rnu [A wiri-ngki] dog-erg =[perf-]1sg.p bite-past big-erg ‘A big dog bit me’ In (76a) the np is continuous, so there is no case agreement, but in (76b) case agreement is a mandatory means for identifying the discontinuous parts of the np. Stacking of cases within nps is not uncommon; for surveys, see Plank (1995). Often one of the cases is due to copying and one to assignment, as in Old Georgian: (77) Old Georgian (Kartvelian; Fähnrich (1991:197)) a. saxl-man israeyl-isa-man house-erg Israel-gen-erg ‘the house of Israel’ b. arkw dze-ta israeyl-isa-ta speak son-obl.pl Israel-gen-obl.pl ‘speak to the sons of Israel’ The genitive case in both examples is assigned by the adnominal construction, and the ergative in (77a) and the oblique in (77b) are assigned to ‘house’ and ‘son’, respectively, and spread to ‘Israel’. Since stacking is most common in adnominal constructions, cross-linguistically it is the genitive case – the universal default adnominal case – that is most prone to have another stacked onto it. Clause-level stacking of case suffixes is illustrated by Huallaga Quechua and Kayardild. The Quechuan example involves copredicatives, as is relatively common; the Kayardild one has ordinary clause members (see the discussion of it above in section 6). (78) Huallaga Quechua (Quechuan, Peru; Weber (1989:221)) (= (53) above) Haacha-wan-naw mutu-n machiita-wan axe-com-sim chop-3 machete-com ‘He chops with a machete as though it were an axe’ 238 Balthasar Bickel and Johanna Nichols Table 3.21 Behaviour of words and formatives with regard to assignment, spreading, and stacking. Blanks mean that we have no examples of that phenomenon Assigned (inert): Spreading: Stacking (79) np clause np clause np clause Syntactic word Formative Engl. of, etc. Engl. to on IO, etc. Old Russian prep. IE prep./preverb adnominal genitive case on arguments IE case agreement IE predicate nominals Old Georgian, etc. Kayardild modal case IE prep./preverb Kayardild (Tangkic, Australia; Dench and Evans (1988:34–5))28 (= (54) above) maku-ntha yalawu-jarra-ntha yakura-naa-ntha woman-obl catch-past-obl fish-abl(prior)-obl dangka-karra-nguni-naa-ntha mijil-nguni-naa-nth man-gen-instr-abl(prior)-obl net-instr-abl(prior)-obl ‘The woman must have caught fish with the man’s net’ Stacking of syntactic words appears to be less common than stacking of cases. For example, where two prepositions would be assigned by the syntax in Russian, the first is deleted. This happens in time expressions, as in (80), where v ‘in’ would ordinarily be assigned to this kind of time adverbial, and here its object happens to be a more or less fixed expression starting with a preposition, bez chetverti . . . ‘a quarter to . . .’. (80) Russian on prishel (*v) bez chetverti sem’ he came at without quarter 7 ‘he came at a quarter to 7’ Perhaps this is preposition stacking with obligatory syncope.29 Table 3.21 summarizes the behaviour of formatives and words with regard to assignment, spreading, and stacking. 28 29 For glossing of cases and the interlinear (prior) see ex. (54) above. At one time, preposition stacking must have been possible in Russian, for there exist compound prepositions such as iz-za ‘because of’ (lit.: ‘from-behind’), iz-pod ‘of, from’ (lit.: ‘from-under’). Both govern the genitive (as iz does) and not the instrumental (as za and pod do). Inflectional morphology 9 239 Conclusions Morphological typology played a pioneering role in the development of typology in the nineteenth century, but in the second half of the last century, the traditional approaches came under heavy criticism for conflating parameters (see the discussion in section 1), and the field was often questioned for its general usefulness (e.g. by Comrie (1981a)). However, advances in the theoretical understanding of the word – specifically, the systematic breakdown of this notion into phonological and grammatical words – have now made it possible to put morphological typology on a more precise foundation. In addition, since the 1990s, there has been a renewed interest in the theory of inflection classes and this has improved the understanding of one of the most intricate problems in morphology: allomorphy and the nature of paradigms. Further, the grand renaissance of grammaticalization studies in the early 1990s has brought with it much insight into the diachrony of morphology and its functional and cognitive dimensions. Together, these three strands of development have led to a rapid and theoretically diverse expansion of the field of morphology. Along with this, morphological typology has begun to survey the languages of the world with new tools and analytical notions. We hope this chapter has shown that morphological typology can in turn improve descriptive analysis by paying close attention to all parameters along which inflectional morphology varies. 10 Suggestions for further reading General surveys of theoretical issues in inflectional morphology are Spencer (1991) and Carstairs-McCarthy (1992). Spencer (1991) in particular, contains a helpful discussion of the interaction of syntax and morphology, which has been one of the traditional controversies of grammatical theory. See also S. R. Anderson (1992) for a word-based approach. For a basic introduction to morphology, see Haspelmath (2002); for general reference, consult Spencer and Zwicky (1998) or Lehmann, Mugdan, and Booij (2000). Some of the typological distinctions we draw here are treated under various technical terms in generative frameworks, and are not always easy to recognize: much discussion of synthesis and notions of wordhood (section 1) is currently covered by literature on complex predicates, e.g. Alsina, Bresnan, and Sells (1997) or Ackerman and Webelhuth (1998), and on what is called the principle of lexical integrity (e.g. Mohanan (1995); Bresnan and Mchombo (1995)). On the phonological word, see in particular Hall and Kleinhenz (1999); on grammatical word notions, see Di Sciullo and Williams (1987). The properties of layered morphology as distinct from templatic morphology (section 6) are attributed to the Mirror Principle, which states that the sequence of 240 Balthasar Bickel and Johanna Nichols morphological operations mirrors syntactic tree and scope structure (Baker (1985)). See Alsina (1999), Rice (2000), and Stump (2001) for some recent controversial discussion. Pronominal agreement markers (section 8) are typically analysed in terms of movement from syntactic argument positions to their morphological host. Grammatical agreement is analysed, by contrast, as basegeneration of markers (clitics, affixes) at the host; since such markers co-occur with nps, the phenomenon is then also referred to as ‘clitic doubling’ in the literature. See Spencer (1991:384–90) for a useful summary.