On the relation between morphology and syntax.

In Gillian Ramchand & Charles Reiss (eds.) 2007. The Oxford Handbook of Linguistic Interfaces, pp. 209–238. Oxford University Press, Oxford. On the relation between morphology and syntax* Marit Julien 0.1 Introduction According to the traditional view, the relation between morphology and syntax is the following: while morphology builds up word forms—typically by combining roots with other roots and with affixes, but also by applying other operations to them, syntax takes fully inflected words as input and combines them into phrases and sentences. The division of labour between morphology and syntax is thus perfect: morphology only operates below the word level whereas syntax only operates above the word level. Moreover, these two components of grammar are ordered in strict sequence, such that the syntax takes over after the morphology has done its work. This model has formed, implicitly or explicitly, the basis of so many descriptive grammatical works that there is no point in mentioning any one of them here. Under the name of lexicalism it has also made its way into more recent theorizing. The lexicalist view has not gone uncontested, though. Already Jespersen (1924) said that for the syntax as he conceived of it, it makes no difference whether a given category is expressed as a part of a word or as a separate word. Jespersen’s insight was picked up by Chomsky (1957), who proposed that verbs combine with verbal inflectional markers in the syntax. Chomsky also extended the syntactic analysis into the derivational domain, as he argued that nominalizations are formed from sentences * The research for this chapter was financially supported by the Norwegian Research Council, grants no. 110928/520 and 141687/540. by transformation rules essentially similar to the rules that convert base sentences to derived sentences (a view developed in detail by Lees 1960). Taken together, the proposals in Chomsky (1957) implied that the atoms of syntax are morphemes, not entire words, and, consequently, that transformations operating on these atoms can have both morphological and syntactic effects. The general idea of this analysis was to become a standard assumption in generative syntax for the next decades, although the details were understood in various ways over the years. The next milestone in the present context was Chomsky (1970), where he argued that at least some of the relations that had for some time been viewed as transformational belonged instead in the lexicon.1 While still assuming that inflectional markers were manipulated in the syntax, Chomsky now took the types of complex word formation traditionally referred to as ‘derivation’ to be performed in the lexicon—that is, in a pre-syntactic component of grammar. Thus, what he argued for was a weak version of lexicalism. Stronger versions of lexicalism have later been formulated, some of them claiming that the internal structure of words is never visible to or manipulated by syntax. Among the prominent lexicalist works are Lapointe (1980), Selkirk (1982), Di Sciullo & Williams (1987), Anderson (1992), and Chomsky (1993, 1995). On the other side of the controversy, the syntactic view on word formation has been defended in Baker (1988), where derived words of various types are dealt with, and in Halle and Marantz (1993, 1994), which are attempts to formulate a general and complete theory of complex words based on the idea that every morphological element 1 But see Marantz (1997) for an updated reading of Chomsky (1970) in the light of Bare Phrase Structure (Chomsky 1995). is also a syntactic element. The theory, called Distributed Morphology, is further developed by Halle (1997), Marantz (1997), and others. In this chapter, I willl however claim that the discussion of whether complex words are formed in the syntax or prior to syntax is futile, because words as such are not formed in the grammar at all. They are not grammatical entities. Below the phrase level, syntax operates on morphemes and gives certain arrangements of these morphemes as output. Some of the resulting morpheme sequences are called words, but crucially, these sequences do not as a class correspond to one particular syntactic representation. Rather, as I argue in 0.2, words are characterized by their distributional properties. In 0.3 I show that these properties are compatible with a number of syntactic configurations. In 0.4 I deal with agreement markers, which I see as exceptional as they do not represent syntactic heads. In 0.5 I turn to an apparent counterexample to the idea that morpheme order is always determined by the syntax, namely, the Scandinavian –s(t) suffix that is used to form passives, among other things. I demonstrate that given the right syntactic analysis –s(t) is not necessarily a counterexample after all. The chapter is rounded off in 0.6 with a comment on the reality of words. 0.2 The characteristic properties of words The plausibility of the claim that words are not grammatical entities is best seen if we try to define what a word is. It then appears that although it may be easy to pick out the words in a given language, it is much more difficult to characterize these words in grammatical terms. Even if we put aside the notions of ‘phonological word’ and ‘lexical word’, and concentrate only on what we may refer to as ‘grammatical words’ or ‘morphosyntactic words’, the task does not get much easier. In the current linguistic literature one often comes across such statements as “words are morphological objects” and “words are the basic building blocks of syntax”. But strikingly, these definitions cannot be used to determine the status of elements that may or may not be separate words—they only apply after the words have been identified. Let us look at an example. Sylvain (1936) states that the markers of tense, aspect, and modality in Haitian Creole are prefixes on the verb, as shown in (1). In Spears (1990), on the other hand, all preverbal markers in Haitian Creole are analyzed as auxiliaries, and they are written as separate words, as in (2). (In addition, the two authors gloss the markers differently, as we see, but that is of less importance here.)2 (1) Haitian Creole (Sylvain 1936:87) n´ té-kwè 1PL PAST-think u t´-a-vini 2SG PAST-FUT-come ‘We thought that you would come.’ 2 The following abbreviations are used in the glosses in this chapter: ACC=accusative, ANIM=animate, ANT=anterior, AOR =aorist, CAUS =causative, COMPL=completive, CONT=continuative, DAT =dative, DEF= definite, DIST=distal, DU=dual, ERG=ergative, EX=exclusive, EXPL=expletive, FEM=feminine, FOC=focus, IRR=irrealis, FUT=future, NEG = HAB =habitual, negation, ILL =illative, NOM=nominative, IMPERS=impersonal, PASS=passive, PL=plural, IMPF=imperfective, PERF=perfective, POT=potential, PRES=present, PROG= progressive, PTC=participle, PURP=purposive, REAL=realis, REFL=reflexive, REM=remote, SG =singular, TERM=terminative. (2) Haitian Creole (Spears 1990:124) m te 1SG ANT di m t a pati say 1SG ANT IRR leave ‘I said that I would leave.’ Now how do we know whether (1) or (2) is more correct, from a grammatical point of view? According to the criteria just mentioned, the preverbal markers in Haitian Creole form a word with the verb if their position is determined by the morphology, but they are words themselves if their position is determined by the syntax. Further, they are words if they have their own syntactic representations, but not if they constitute a syntactic terminal node together with the verb. This line of reasoning could also be reversed: if the preverbal markers belong to the verbal word, they are not visible to the syntax and their positions are determined by the morphology, but if they are separate words, they are minimal syntactic units. Thus, the argumentation is going in a circle, getting us nowhere. What we really should ask is how words are recognized in the first place. Why are some morpheme sequences taken to be words, while others are not? The answer, I would claim, is that wordhood has to do with distribution. That is, morpheme sequences that have certain distributional properties tend to be seen as words. One of the relevant properties is independent distribution. Already Boas (1911) pointed out that when a certain sequence can appear, without substantial modifications, in a variety of positions relative to other elements in the sentence, “we are inclined to consider it as a separate word”, as he put it (Boas 1911:30). Thus, their independent distribution distinguishes grammatical words from smaller elements. As for the upward delimitation of words, Bloomfield (1933) argued that internal cohesion is the property that separates compounds and phrasal words from phrases. More precisely, even if both words and phrases can be built from words, with phrases it is normally the case that they can be broken up by additional words and phrases, whereas words that consist of words can be interrupted much less freely, if at all. Bloomfield (1933) also proposed another criterion for wordhood, namely, that a word is a minimal free form. However, while this is a sufficient criterion for wordhood, it is not strictly necessary. The necessary criteria appear to be independent distribution and internal cohesion, since forms that meet these two criteria tend to be regarded as words whether or not they can appear in isolation. But crucially, if a given morpheme string has independent distribution and internal cohesion, so that it is stands out as a word, it does not follow that the string is also a single terminal node in the syntax. Neither does it follow that it is more than one syntactic terminal node. What I will argue is that there is no single specific syntactic configuration that corresponds to words. This means that the concept ‘word’ has no theoretical significance in the grammar at all. It cannot then be the case that the grammar contains a word-forming morphological module. Arguably, the syntax alone determines the hierarchical and linear relations between morphemes. It does not, however, determine the realization of each individual morpheme. Hence, it must be the morphology that deals with the spelling out of morphemes and with allomorphic variation. But notably, words are not necessarily relevant in this context. Although the realization of a given morpheme may be dependent on the feature content of neighbouring morphemes, the conditioning factor is not always contained in the same word as the morpheme that undergoes the alternation. The a/an alternation in English is an obvious example; more examples are found in Julien (2002). When the trigger for a conditioned allomorphic alternation nevertheless often belongs to the same word as the target, the reason is that morphemes belonging to the same word regularly appear together. Hence, if the conditioning factor is contained in the same word as the element that undergoes alternation, each alternant will occur relatively frequently, and the alternation has a good chance of surviving. If, on the other hand, a morpheme has an allomorph whose appearance is dependent on some other morpheme not in the same word, it is possible that the alternation is seen relatively infrequently, and it will more easily get lost. A related fact is that words are often listed in the lexicon (but note that they can also be produced online), alongside the individual morphemes that they are built from, whereas phrases are listed less often and sentences only occasionally (cf. Jackendoff 1997). If we now go back to the problem that was illustrated in (1) and (2), it should be clear that it is impossible to tell what the words are simply by inspecting the morpheme strings that we see in these examples. We need to know which permutations are possible, and where additional material can be inserted. However, the primary value of such information lies in its syntactic relevance. The question of which morpheme strings are words is not really important since from the point of view of grammar, the word is merely an epiphenomenon. 0.3 The syntax of words If it is true that words are nothing but distributional units, we would nevertheless like to know what the syntactic description of a word could possibly be. In the following, I will make an attempt at answering that question. After presenting some fundamental assumptions in 0.3.1, and showing the basegenerated order of tense, aspect, and verb root in 0.3.2, I go on to give some examples of how various orders of tense marker, aspect marker, and verb can be derived. In 0.3.3 we see how head movement can lead to suffixing of tense and aspect markers, in 0.3.4 we see how phrasal movement can yield a similar result, in 0.3.5 we look at the syntax of prefixed tense and aspect markers, and in 0.3.6 we deal with a language that has prefixed tense but suffixed aspect. My conclusions concerning the syntax of complex words are spelled out in 0.3.7. 0.3.1 Preliminaries On the assumption that the left-to-right linear order corresponds to the top-down hierarchical relations, as Kayne (1994) proposed, there are four syntactic configurations that could cause the morphemes X and Y to form a word [XY]. X and Y could be parts of one complex syntactic head, as in (3a); X could be the next head up from Y, as in (3b); X could be the final element in the specifier of Y, as in (3c); and finally, Y could be the initial element in the specifier of the complement of X, as in (3d). (3) a. YP Y X b. XP X Y c. YP Y YP [… X] d. Y' Y XP X ZP [Y … ] Z' Z In each of these configurations, the sequence XY can have word properties. Internal cohesion is guaranteed in (3a), (3c), and (3d). In (3b) it follows if no phrase surfaces in Spec-YP. As for independent distribution, a complex head, as in (3a), will move as a whole if it moves at all, since excorporation is arguably not possible (Julien 2002). Hence, a complex head comes across as either a word or a part of a word (and for this reason its syntactic properties are often taken to be characteristic of the word). In the other three configurations it is less likely that XY will move as a unit, but it would be possible if YP in (3b) and (3c) and ZP in (3d) only contains Y at the point where the relevant movement takes place. Note, however, that if other constituents move around XY, XY will have independent distribution relative to those other constituents even if the sequence XY itself does not move. In short, in all the configurations in (3) the sequence XY could have the distributional properties that characterize words. What I will claim is that every morphologically complex word corresponds to one of the configurations shown in (3), or to a combination of these configurations. In the following, I will illustrate my point by presenting some examples of morphologically complex words taken from the domain of verbal inflection. 0.3.2 The base-generated order of tense, aspect, and verb In the examples in (4) and (5), the markers of tense and aspect are morphologically free elements. Moreover, in both cases they precede the verb, and the order is tense marker > aspect marker > verb root. (4) Makaa (Heath 1991:11) m´ Ng´ á wííNg 1SG REM.PAST PROG chase.away òmpy´ dogs ‘I was chasing the dogs away.’ (5) Mauritian Creole (Adone 1994:44) Lapli ti pe toñbe. rain PAST IMPF fall ‘Rain was falling.’ After surveying 530 languages, Julien (2002) concludes that the pattern shown here is by far the most common one when tense and aspect markers are realized as free elements. I take this to mean that universally, temporal heads are higher in the clause than aspectual heads, which in turn are higher than the verb, as illustrated in (6).3 3 Most likely, there are many more heads in the IP-domain—see Cinque (1999). However, for the present purpose inflectional heads other than Tense and Aspect are ignored. (6) TP T AspP Asp vP …V… Now let us assume, for the sake of the argumentation, that tense and aspect markers are always realizations of tense and aspect heads, and not base-generated on the verb. It then appears that various orders of tense marker, aspect marker and verb can be explained on the basis of the syntactic structure in (6). This is what I will go on to demonstrate. 0.3.3 Suffixed inflectional markers resulting from head movement Consider first example (7), from Macushi, a language where the unmarked word order is OVS, and markers of tense and aspect are suffixed to the verb, such that the morpheme order is verb root > aspect marker > tense marker (Abbott 1991). (7) Macushi (Abbott 1991:121) yei ya'tî-aretî'ka-'pî-i-ya wood cut-TERM-PAST-3-ERG ‘He finished cutting the wood.’ If the object surfaces above the Tense head while the subject surfaces below it, then both the OVS order and the suffixing of aspect and tense markers can be derived by moving the verb to the Asp head and the [V Asp] complex to the Tense head, as sketched in (8).4 If adjunction is always to the left (Kayne 1994), the order of elements inside the verbal word follows. (8) TP OBJ yei T' T Asp AspP T 'pî V Asp ya'tî aretî'ka SUBJ Asp' i-ya Asp vP …V… As for the subject pronoun in (7), it is phonologically weak and cliticizes onto the verb (Abbott 1991). A phonologically strong alternative is shown in (9). (9) Macushi (Abbott 1991:24) mîrîrî koneka-'pî mîîkîrî-ya that make-PAST 3-ERG ‘He made that.’ Here it seems clear that the pronominal subject is sitting in a Spec below the verbal word. In the absence of evidence to the contrary, I assume that the phonologically weak pronoun in (7) is also situated in a Spec below Tense, as indicated in (8). 4 With intransitive verbs the unmarked constituent order is SV (Abbott 1991). That is, Macushi shows an ergative pattern in the syntax as well as in the case marking of arguments. Given this, the structure in (8) seems rather plausible. It follows that in (7) we have a word that is made up of a complex syntactic head and an element sitting in a Spec position below that head. The stable part of that word is however the complex head, which necessarily has word properties. 0.3.4 Suffixed inflectional markers resulting from phrasal movement Turning now to Evenki, an SOV language, we see that the order of elements inside the verbal word here is also verb root > aspect marker > tense marker. (10) Evenki (Bulatova & Grenoble 1999:8) bu: dolboni:-Ba 1PL.EX night-ACC haBal-ža-ča-Bun. work-IMPF-PAST-1PL ‘We worked all night.’ One might guess that the complex verbal word in Evenki is formed by head movement, just like I have proposed for Macushi. However, the overall syntax of Evenki clauses is rather different from that of Macushi. Evenki is a head-final language—that is, every head in the clausal projection line is preceded by its complement. This does not only mean that bound inflectional markers appear in an order which is the reverse of the base-generated order, but also that lower verbs precede higher verbs, as illustrated in (11). (11) Evenki (Bulatova & Grenoble 1999:39) kuNaka:n ž´b-d´:-Bi: child ´j´:t-č´-r´-n. eat-PURP-REFL want-IMPF-AOR-3SG ‘The child wants to eat.’ Julien (2002) proposes that the head-finality of many SOV languages is the result of every clausal head above vP having attracted its complement to its Spec. On this analysis, the syntactic structure of (10) is as shown in (12). Note that the agreement marker that is the last element of the verbal word in (10) is tentatively placed in the Finite head, which is where subject agreement markers appear to be located in many languages (Julien 2002).5 (12) FinP TP Fin' AspP [vP SUBJ ADV V ] bu: dolboni:Ba haBal T' Asp' Asp ža T ča Fin Bun TP AspP vP We see here that the movement of each complement to the nearest Spec has made every projection in the IP-domain head-final. Since the verb is also the final element in the vP, we get a sequence of linearly adjacent morphemes, starting with the verb root, where each morpheme is the final element of the constituent in the Spec of the next morpheme. For the example in (11), an analysis along similar lines would mean that the iterative movement of complement to Spec starts from the most deeply embedded verb and goes all the way up to the highest Finite head. (The purposive marker probably represents a Mood head, while the marker glossed as REFL appears to be an agreement 5 See Rizzi (1997) and Platzack (1998) on the Finite head. marker, since it indicates that the subject of the higher verb is also the subject of the lower verb—see Bulatova & Grenoble 1999.) Note that the subject and the temporal adverbial in (10) are taken to be sitting inside vP. Concerning the adverbial, its being located inside vP is consistent with Pereltsvaig’s (2001) proposal that adverbials with accusative case are licensed in the Spec of a vP-internal aspectual head (the Inner Aspect of Travis 1992, 2000). This position is otherwise where direct objects are licensed if they serve as delimiters of the event. However, Pereltsvaig argues that certain adverbials can appear in the same position and have an the same function as event-delimiting direct objects, and consequently show up with the same case. The adverbial in (10) seems to be a good example of this. As for the subject, it has been claimed by Yanagida (1996) and Nakajima (1999) that arguments can be licensed inside vP in Japanese. Julien (2002) takes it to hold for all head-final languages. But although the arguments of the verb in head-final languages arguably do not move out of vP for case reasons, they can move to focus and topic positions in the CPdomain, above FinP. These movements can alter the order of constituents in the clause considerably, as in the following Evenki example. (13) Evenki (Nedjalkov 1997:4) Adul-il-va si gene-che-s? fish.net-PL-DEF.ACC you bring-PAST-2SG ‘Did YOU bring the fish nets? As we see, the order here is OSV, and the subject is focused. I take this to mean that the object has moved to the Spec of a Top head in the CP-domain while the subject has moved to the Spec of a Foc head below Top (see Rizzi 1997). Since adverbial phrases, and arguably even the IP, can also move to the CP-domain, many head-final languages display a wide range of word order alternations having to do with discourse functions. The sequence of heads in IP remains unchanged, though. The consequence is that the morpheme sequence consisting of the verb root and the verbal inflectional markers has word properties. 0.3.5 Prefixed inflectional markers Now consider the examples in (14) and (15), where the markers of tense and aspect precede the verb root. Strikingly, these markers appear in the order that we would take to be the base-generated one, given (4) and (5). So how come they are included in a word with the verb root? (14) Chalcatongo Mixtec (Macaulay 1993:73) a-ni-ndatu-rí uù órá TENSE-COMPL-wait-I two hour ‘I’ve already been waiting for two hours.’ (15) Shona (cf. Myers 1990) nd-a-i-teng-es-a 1s-PAST-HAB-buy-CAUS-REAL6 ‘I used to sell.’ 6 Myers (1990) glosses the final vowel in Shona verbs simply as Final Vowel. However, his description suggests that it encodes [±realis] mood, since it is –e in the subjunctive, potential, and negative, and –a elsewhere. For Shona, Myers (1990) argues that the verbal word is a phonological unit and not a morphological or syntactic constituent. From a grammatical point of view, the verb root forms a constituent with the suffixes, but not with the prefixes. Myers’s statements about the verbal word in Shona are compatible with assigning to (15) the syntactic structure shown in (16). Here the verb has moved to the Caus head and then to the Mood head, thereby forming a complex head with the causative marker and the mood marker, both suffixed to the verb. The markers of subject agreement, tense, and aspect remain in their basegenerated positions in front of the verb. (16) FinP pro Fin' Fin nd TP T a AspP Asp í MoodP Mood Caus V teng CausP Mood Caus a Caus es VP …V… The reason why they are nevertheless taken to belong to the verbal word is, firstly, that they are included in a phonological word with the verb (see Myers 1990), and secondly, that the sequence beginning with the subject agreement marker and ending with the mood marker has the distributional properties of a word. It has independent distribution—phrasal arguments and adverbials may precede or follow it—and it cannot be interrupted by phrases, only by non-phrasal adverbials, such as chimbidzó ‘quickly’ in (17). As we see, such interrupting adverbials are included in the verbal word, necessarily for distributional reasons. (17) Shona (Myers 1990:90) ndi-chá-to-chimbidzó-dzok-a 1SG-FUT-must-quickly-return-REAL ‘I’ll have to come back quickly.’ For (14), an analysis parallel to (16) goes through, except that the subject marker in (14) is probably sitting in a Spec below the surface position of the verb. Macaulay (1993) shows that enclitic subject pronouns in Chalcatongo Mixtec are in complementary distribution with postverbal DP subjects, and she takes subjects of both types to be located in Spec-VP. This means that the verb in (14) must have moved out of VP, and that the verbal word is made up of a sequence of syntactic heads—the Tense head, the Aspect head and a head containing V—followed by a Spec element—the subject marker. 0.3.6 A mixed case While the languages we have looked at so far all have aspect markers inside tense markers in the verbal word, there are also languages where tense and aspect appears on opposite sides of the verb root. Typically, the tense marker then precedes the root while the aspect marker follows it, as in (18). (18) Rikbaktsa (Boswood 1978:22) Ka-zo pitsi 1SG-father cashew pi-boro-ko. NONPAST-eat- CONT ‘My father is eating cashew nuts.’ On my analysis, the verbal word in (18) is the result of moving the verb to the Aspect head—see (19). The [V Asp] complex is preceded by the Tense head, which is not affected by any movement operations. (19) TP T pi AspP Asp V boro Asp ko vP …V… This means that just like in Shona, the verb root in Rikbaktsa forms a syntactic constituent with the suffixes, whereas the prefixes have a more distant structural relation to the root. 0.3.7 The syntactic nature of word formation The brief discussion of a few selected examples that I have presented here can of course not do justice to the full range of variation that we see in the verbal morphology of the world’s languages. It can only give an idea of what the underlying syntactic structure of a morphologically complex word might be. Still, the fact that the verbal morphology of the languages under consideration can get a syntactic explanation is no proof that syntax is the basis of all word formation. However, in Julien (2002) a survey of 530 languages from 280 genetic groups is presented, and the conclusion is that languages conform to an overwhelming degree to the hypothesis that the order of morphemes in morphologically complex words is determined by the syntax alone. At least in the domain of verb morphology, the morpheme orders attested in the 530 languages in the survey can all be derived from one single underlying syntactic structure by a syntax where movement and adjunction are always to the left (see also Julien 2003). Closer investigations of alleged exceptions reveal that in these cases too an analysis is possible which is in accordance with the idea that every morpheme is a syntactic element and its position in the surface order is a matter of syntax. If the attested morpheme orders are compatible with the hypothesis that they are derived in the syntax, the most economical grammar would be one where morpheme ordering, inside and outside words, IS in fact determined by the syntax.7 Hence, I assume that this is the case. It should also be noted that on the approach I am sketching here, there is no principled difference between affixes and clitics. Elements of both types are positioned by the syntax. However, if the syntactic frame that a given element appears in causes that element to always end up in close relation to another element of one particular category, the first element will be seen as an affix. By contrast, an element that appears next to elements of various categories will be seen as a clitic. 7 See Drijkoningen (1994) on this point. For example, an element that is always preceded by a verb-final VP, such as the aspect marker in (12), is seen as a suffix on that verb. Now imagine a structure that is similar to (12) except that the VP is not necessarily verb-final. In that case, the aspect marker will follow whatever is the last element in the VP. The aspect marker will then be seen as a clitic. But crucially, the syntactic properties of the aspect marker will not differ from what we see in (12). This suggests that the distinction between clitics and affixes does not go very deep and is not very interesting to the theory of grammar. 0.4 Agreement markers Having claimed that morpheme order is always determined by the syntax, I must now point out that there is one inflectional category that does not fit into the relatively rigid framework that the syntactic approach provides. This category is agreement. For example, subject agreement markers can be located higher than the Tense head, as in (20), where the subject agreement marker has moved with the negation to a position in front of the question marker, leaving the tense marker and the verb root behind. (20) Northern Saami I-t go manna-n NEG-2SG Q go-PAST Oslo-i? sOslo-ILL ‘Didn’t you go to Oslo?’ In (21), by contrast, the subject agreement marker follows the tense marker and the marker of grammatical mood, and we would take it to be located low down in the IP. (21) Loniu (Hamel 1994:113) sEh ya 3PL FUT EtE k-E-mE POT-NONSG-come ANIM.GOAL u 1.DU.EX ‘They will come to (visit) us.’ Taken together, these examples indicate that subject agreement markers are not in a fixed position universally. Moreover, the example in (22), where one subject agreement marker is prefixed to the negation while another is prefixed to the verb, shows that subject agreement markers are not necessarily associated with a unique head, not even within one single clause. (22) Lango (Noonan 1992:142) án à-pé à-wótò I 1SG-go.PERF Kampala 1SG-NEG kàmpálà ‘I didn’t go to Kampala.’ If object agreement is also considered, the picture becomes even more complicated. As noted in Julien (2002), of the 24 theoretically possible orderings of verb root, tense marker, subject agreement, and object agreement, only 4 are not attested; these are OAgr-T-SAgr-V, T-OAgr-V-SAgr, OAgr-T-V-SAgr, and SAgr-VOAgr-T. The 20 other orderings can all be found in one or more languages. The variation that we find in the positioning of agreement markers is such that we have to give up the idea put forth in Chomsky (1993) that clauses contain a subject agreement head and an object agreement head which are located in fixed positions universally.8 Moreover, it appears that the absence of Agr heads does not only hold for languages with weak Agr, as proposed by Chomsky (1995), but most likely for all languages. Thus, my proposal is that agreement markers, unlike markers of other categories, do not in themselves represent syntactic heads. Instead, agreement features are added to heads that also have some other content. Related ideas have been put forward by Halle and Marantz (1993), who assume that subject agreement features are added to tense heads, and by Baker (1996), who proposes that at least in polysynthetic languages, subject agreement markers are adjoined to I and object agreement markers are adjoined to Asp. However, these proposals cannot capture the full range of variation with respect to the positioning of agreement markers that we find in the world’s languages (see Julien 2002 for details). Rather, the location of agreement features must be subject to crosslinguistic variation. 0.5 An apparent counterexample In section 0.3 I presented some examples of morpheme orders that are all in accordance with the syntactic approach to word formation. Countless other examples could have been mentioned from languages all over the world. It is more interesting, though, to look at cases that appear to pose problems for the syntactic hypothesis, and see if they can be explained in syntactic terms. In the following, I will deal with an element that has been pointed to as a counterexample to the syntactic approach to word formation, namely, the –s or –st that can be 8 Cf. Iatridou (1990), Speas (1991), Spencer (1992), Mitchell (1994), and Holmberg and Platzack (1995), among others. suffixed to verbs in the Scandinavian languages,9 and that is customarily analyzed as a passive marker, besides being a marker of reflexive, reciprocal, and inchoative verbs.10 In 0.5.1 I show that the position of passive –s(t) is unexpected on the syntactic approach to complex words. In 0.5.2 I compare the –s(t)-passive to the periphrastic passive, and I conclude that the two are syntactically different. Then in 0.5.3 I present my analysis of –s(t), according to which –s(t) is an argument. In 0.5.4 I deal with the question that this analysis raises concerning V2 word order, and in 0.5.5 I point out some welcome consequences of my analysis. Finally, in 0.5.6 I look briefly at some elements outside of Scandinavian that could be taken to have the same syntax as –s(t). 0.5.1 The problem Consider first the three examples of the Scandinavian –s(t)-passive shown below. (23) Norwegian Vi plag-de-s av mygg-en. we annoy-PAST-s by mosquito-DEF ‘We were annoyed by the mosquitoes.’ (24) Swedish (Hedlund 1992:124) Hus-et måla-de-s av Kalle. house-DEF paint-PAST-s by Kalle ‘The house was painted by Kalle.’ 9 See e.g. Holmberg & Platzack (1995:28 fn. 26). 10 The form of this marker is –s in Danish, Swedish, and in some varieties of Norwegian, but –st in Faroese, Icelandic, and other varieties of Norwegian. (25) Icelandic (Anderson 1990:235) Keisar-inn klæ-dd-i-st emperor-DEF.NOM dress-PAST-3SG-st ný-jum föt-um. new-DAT.PL cloth-DAT.PL ‘The emperor was dressed in new clothes.’ As we see, the passive –s(t) follows the tense marker, and in Icelandic, even the subject agreement marker. This is unusual—the normal position for affixed passive markers is close to the verb root, inside tense and aspect, as in (26), (27), and (28). (26) Evenki (Nedjalkov 1997:218) Uluki hurkeken-du va:-p-cha-n. squirrel boy-DAT kill-PASS-PAST-3SG ‘The squirrel was killed by the boy.’ (27) Oromo (Owens 1985:172) makiináa-n ní tolf-am-t-a car-NOM repair-PASS-3FEM-IMPF FOC ‘The car will be repaired.’ (28) Seri (Marlett 1990:516) tóm ki? ?p-yo-m-p-e: money DEF 1SG-DIST.REAL-NEG-PASS-give ‘I was not given the money.’ Passive markers inside tense and aspect are compatible with the proposal of Rivero (1990), Travis (1992, 2000), and Kratzer (1996) that the active/passive distinction is encoded in a Voice head which is located below the heads in the IP- domain. In (29), I show the structure proposed by Travis, which includes an Inner Aspect head, encoding telicity, between the Voice head and the VP. (29) VoiceP Ext.arg Voice' Voice AspP Asp Int.arg VP V' In accordance with the proposals just mentioned, I will assume that external arguments are generated in the Spec of the Voice head, which could be identified with the v head shown in some of the preceding examples. In the passive, a [passive] feature in the Voice head precludes the insertion of a visible argument DP in Spec-VoiceP. In this light, the position of the suffixed –s(t) in (23), (24), and (25) is a problem. If it is a realization of the Voice head, the syntactic approach to morpheme order predicts that it should be positioned inside the tense marker, not outside it. Alternatively, it is not a realization of the Voice head. This is what I will argue in the following. 0.5.2 The two passives in Scandinavian In Scandinavian, besides the –s(t)-passive there is also a periphrastic passive construction, which clearly involves the Voice head since it only affects verbs that have an external argument (see Åfarli 1992). In Norwegian, the periphrastic passive is thus grammatical with the transitive agentive verb skrive ‘write’ in (30a) and with the transitive perception verb se ‘see’ in (30b). (30) Norwegian a. Brev-et ble skrev-et av meg. letter-DEF became write-PTC by me ‘The letter was written by me.’ b. Katt-en ble se-tt av Lina. cat-DEF became see-PTC by Lina ‘The cat was seen by Lina.’ The periphrastic passive also goes well with unergative verbs, like banke ‘knock’ in (31). However, as (32a) shows, it is ungrammatical with få ‘get’, although få is fully acceptable with the –s(t)-passive, as (32b) demonstrates. (31) Norwegian (Åfarli 1992:107) Det EXPL bli-r bank-a på dør-a. become-PRES knock-PTC on door-DEF ‘There is knocking on the door.’ (32) Norwegian a. * Tomat-er bli-r nå tomato-PL become-PRES now b. Tomat-er få-s nå tomato-PL get-s now år-et få-tt året rundt. get-PTC year-DEF round rundt. year-DEF round ‘Tomatoes are now available all year round.’ More generally, 2-place verbs with benefactive subjects and 2-place stative verbs with experiencer subjects allow the –s(t)-passive but not the periphrastic passive (Lødrup 2000). A verb of the latter type is føle ‘feel’, shown in (33). (33) Norwegian (Lødrup 2000:47) a. Bulk-en føle-s ikke i det hele tatt. dent-DEF feel-s not at all ‘The dent is not noticeable at all.’ b. * Bulk-en blir ikke føl-t i det hele tatt. dent-DEF becomes not feel-PTC at all Furthermore, an unaccusative verb like dø ‘die’ can be used with the –s(t)-passive and still retain its non-agentive meaning. An example is given in (34). (34) Norwegian (Enger 2000:11) Det dø-s EXPL altfor mye die-s too her i sogn-et. much here in parish-DEF ‘People die too much in this parish.’ When appearing in a participial passive, the same verb can only have an agentive reading, as in the example in (35). (35) Norwegian I denne opera-en in this blir det dø-dd flere gang-er hver kveld. opera-DEF becomes EXPL die-PTC several time-PL every night ‘In this opera, there is dying several times every night.’ I take the contrast between the agentive dø and the non-agentive dø to mean that some intransitive verbs allow their single argument to be generated either as an external argument or as an internal argument. That is, I assume with Baker (1997) and contra Borer (1998) that there is a tight connection between thematic roles and syntactic structure, as indicated already in (29). It follows that in (35), dø has unergative syntax. The generalization that can be drawn from the above facts is that the –s(t)-passive but not the periphrastic passive is compatible with verbs that lack an external argument. Unaccusative verbs obviously lack external arguments. Concerning verbs with benefactive subjects (i.e., verbs of possession and the like), Kayne (1993) proposed that have is a raising verb, and I assume that its inchoative counterpart få ‘get’ and other verbs of possession have basically the same syntax. As for stative verbs with experiencer subjects, Pylkkänen (2000) argues that stative psych causatives have two internal arguments. I conclude that the passive –s(t) is not a realization of the Voice head. Hence, I claim, contra Åfarli (1992), that the –s(t)-passive has an entirely different syntax from the periphrastic passive, which involves a [passive] Voice head. If this is correct, the position of –s(t) outside the tense marker need not be a problem. 0.5.3 The syntax of –s(t) Let us then consider the syntax of the element –s(t) in more detail. We have seen that when it appears on a finite verb it is the last element of the verbal word, following tense and agreement markers. If –s(t) represents a syntactic head, its position in finite verbs suggests that the head in question is at least higher than Tense—recall that agreement markers arguably do not represent separate syntactic heads. However, if an auxiliary is present in a construction with –s(t), the auxiliary will carry tense while –s(t) attaches to the main verb, as shown in (36). (36) Norwegian Genser-en må-tte prøve-s på. sweater-DEF must-PAST try-s on ‘The sweater had to be tried on.’ The morpheme order in this example suggests that –s(t) is below Tense as well as below the auxiliary. The paradox just presented forces the conclusion that –s(t) does not represent a head in the verbal projection line. If –s(t) has a syntactic reality at all, it must receive an analysis along the lines of Hedlund (1992), who proposes that the passive –s(t) is a subject, sitting in Spec-VP. My proposal is that –s(t) is a pronoun-like element which can, in principle, be inserted in any argument position in vP (which I use as a cover term for extended VPs of all types, including VoicePs). However, its features must be defective, since it is syntactically inert—it does not enter into agreement or attraction relations. As for its reference, when it is generated as the highest argument it gets a default arbitrary interpretation, comparable to arbitrary PRO or the Mainland Scandinavian non-specific pronoun man ‘one’. Thus, as many authors have noted, ‘passives’ formed with –s(t) can be paraphrased with the corresponding active clause featuring man as subject. Many authors have also noted that –s(t) apparently absorbs a theta role. This follows if –s(t) occupies an argument position. It is less clear whether –s(t) gets structural case. If it does, the case in question must be accusative, since another argument (or argument chain) will get nominative. The following Icelandic example, which seems to indicate that both nominative and accusative are available to elements other than –s(t) in –s(t)-constructions, must then receive some other explanation—for example, that the accusative case seen here is not structural. (37) Icelandic (cf. Anderson 1990:246) Hann otta-st mann-inn. he.NOM fear-st man.ACC-DEF.ACC ‘He fears the man.’ A consequence of the poor feature content of –s(t) is that it will not be attracted to the surface subject position even if it is generated as a thematic subject. Instead, some other argument will become the surface subject, as in the examples already given, or alternatively, an expletive will be inserted, as in the examples below. (38) Norwegian a. Det hør-te-s EXPL en gjøk. hear-PAST-s a cuckoo ‘A cuckoo was heard.’ b. Det danse-s ofte der. EXPL dance-s often there ‘People often dance there.’ It is also possible for –s(t) to be inserted in a lower argument position, so that it is c-commanded by a higher argument inside the same vP. In such cases, the reference of – s(t) is determined by the higher argument, and the result is a reciprocal construction, as in (39a), or a reflexive construction, as in (40a)—compare these constructions to the alternative formulations in (39b) and (40b). (39) Norwegian a. De møt-te-s hver onsdag. they meet-PAST-s every Wednesday ‘They met every Wednesday.’ b. De møt-te hverandre hver onsdag. they meet-PAST each.other every Wednesday ‘They met every Wednesday.’ (40) Norwegian a. Vi skjemme-s. b. Vi skamme-r oss. we shame-s we shame-PRES us(REFL) ‘We are ashamed.’ ‘We are ashamed.’ In Swedish, a few verbs even allow for an –s(t) object with independent reference. An example is given in (41). (41) Swedish Hund-en bit-s. dog-DEF bite-s ‘The dog bites (people).’ In these cases, the –s(t) argument is necessarily [+human] (Hedlund 1992). Hence, it is possible that –s(t) here is endowed with a [+human] feature, and that this is what prevents it from picking up the reference of the subject. It must then be a lexical property of the verbs in question that they accept a [+human] –s(t) object. Finally, there are verbs where an added –s(t) appears to be non-thematic (Anderson 1990, Hedlund 1992). Some of these verbs are inchoative, as in (42), while others are psych verbs, as the example in (43). (42) Norwegian Han har elde-s. he has age.PTC-s ‘He has aged.’ (43) Swedish (Hedlund 1992:140) Kalle vånda-s inför tenta-n. Kalle dread-s before exam-DEF ‘Kalle dreads the exam.’ I will nevertheless suggest that –s(t) fills an argument position even here. Note that the verbs in question do not have counterparts without –s(t). For the inchoative verbs, it is possible that –s(t) represents an unspecified causer, which must be present for the verb to be inchoative. Concerning psych verbs, they are known to have a rather complicated syntax (see, e.g., Pylkkänen 2000 for a recent treatment). Sometimes nominal elements show up whose function is rather unclear. Compare, for example, the two constructions in (44). While (44a) involves a subject theme and an object experiencer argument, (44b), with a subject experiencer, contains an additional reflexive pronoun that appears to have syntactic relevance for the argument structure since the theme is now realized as a PP. (44) Norwegian Hun bekymre-r she meg. worry-PRES me ‘She worries me.’ b. Jeg bekymrer meg for henne. I worry-PRES me(REFL ) for her ‘I worry about her.’ Leaving details aside, I therefore assume that –s(t) always fills a nominal position in vP, and that it cliticizes onto the verb after the verb has moved out of that vP. More precisely, I take the verb to move to the Tense head, so that the tense marker ends up as a suffix to the verb. With invisible projections omitted, the situation is then as in (45). (45) TP T V vP T st…V… Anderson (1990) argues that the inflectional markers in –s(t)-verbs are added after –s(t), since certain inflectional markers are deleted in front of –s(t)—in particular, the –r that marks present tense in Mainland Scandinavian and that appears in various places in the verbal paradigms of Faroese and Icelandic. However, although this is a reasonable conclusion from a Lexical Phonology point of view, according to which an element added postlexically should not influence elements that are added on an earlier cycle, it does not follow if we assume that phonological features are introduced at Spell-Out (Halle and Marantz 1993, 1994). On the latter approach, the form of a morpheme can be affected by any other morpheme in its environment. 0.5.4 The derivation of verb second in Scandinavian An apparent problem with the analysis of –s(t) that I have just sketched is that –s(t) moves with the verb to the second position, as (46) demonstrates. (46) Norwegian Klokk-a sju åpne-s dør-e-ne her. clock-DEF seven open.PRES-s door-PL-DEF here ‘At seven o’clock the doors are opened here.’ However, so do reflexive participles in Swedish—see (47). (47) Swedish Då satte then sat.PAST sig Kalle i soffa-n. 3.REFL Kalle in sofa-DEF ‘Then Kalle sat down in the sofa.’ Moreover, several focus sensitive adverbs are known to yield exceptions to the verb second order (see Nilsen 2002). An example is given in (48a). It seems clear that the sequence intervening between the first constituent and the verb here is too big to be a clitic on the verb, which is how Egerland (1998) analyzes the focus element bare ‘only’. But notably, when bare precedes the finite verb, as it does here, it only takes scope over that verb. Thus, [rett og slett bare varma] appears to be a constituent in this case. Also compare (48a) to (48b), where bare takes scope over the object or over the whole IP. (48) Norwegian a. Da rett og slett bare then simply only varm-a hun supp-a. heat-PAST she soup-DEF ‘Then she simply just heated the soup (i.e. she didn’t make the soup).’ b. Da varm-a hun rett og slett bare then heat-PAST she simply only supp-a. soup-DEF ‘The only thing she heated then was simply the soup.’ or ‘All she did then was simply heat the soup.’ These facts indicate that V2 in Scandinavian is the result of fronting a phrase that contains the verb and, optionally, focus elements and reflexive particles (Nilsen 2002 draws a similar conclusion).11 I would suggest that the fronted constituent is TP. Since arguments are not moved along, Scandinavian must then underlyingly be like Dutch and German, where arguments and adverbials precede the infinitival marker, which probably is no lower than T (cf. Kayne 1994:52). The Dutch example in (49) illustrates this point. (49) Dutch Jan probeert elke Jan tries dag {en/de} kikker te kuss-en. every day a/the frog to kiss-INF ‘Jan tries to kiss a/the frog every day.’ 11 This idea is also reminiscent of the proposals put forth by Hróarsdóttir (2000a, 2000b) for Icelandic and by Taraldsen (2000) for Norwegian that the VO order results from remnant VP-fronting. In the corresponding Scandinavian construction, the infinitival marker and the verb precede the object, as in (50). (50) Norwegian Jan prøv-er å kysse {en Jan try-PRES to kiss a frosk/frosk-en} hver frog/frog-DEF dag. every day ‘Jan tries to kiss a/the frog every day.’ That is, the fronted phrase minimally crosses over the objects. In embedded clauses, it moves no further, but in root clauses, it moves up to the position following the topic. That is, unless a finite auxiliary is present, in which case the auxiliary moves, as can be seen in (36) above and in (51) below. 0.5.5 Some welcome consequences of the analysis Several properties of the so-called –s(t)-passive can be explained on the assumption that –s(t) is the thematic subject. Among other things, the –s(t)-passive is known to have modal effects. For example, the auxiliary skal ‘shall’ has a deontic modal reading in the –s(t)-passive in (51a) but a future reading in the periphrastic passive in (51b). (51) Norwegian (Vinje 1987:136) a. Brev-et skal sende-s. b. Brev-et skal bli send-t. letter-DEF shall send-s letter-DEF shall become send-PTC ‘The letter must be sent.’ ‘The letter will be sent.’ On the deontic reading of shall, there must be a bearer of the obligation. In the –s(t)construction in (51a), the thematic subject –s(t) is a suitable candidate.12 In the periphrastic passive in (51b), by contrast, there is no potential bearer of the obligation, and consequently, skal gets a future interpretation.13 The auxiliary vil ‘will’ is also ambiguous between a temporal and a modal reading in Norwegian. This is illustrated in (52a). However, future vil is necessarily a raising verb. When vil has its own external argument, only the modal reading survives, as (52b) shows. (52) Norwegian a. Hun vil overføre penge-ne til min konto. she to my account will transfer money-DEF ‘She will/wants to transfer the money to my account.’ b. Hun vil she at penge-ne wants that money-DEF skal overføre-s til min konto. shall transfer-s to my account ‘She wants the money to be transferred to my account.’ In (53a) vil can be seen as a raising verb, with its surface subject originating as an argument of the lower verb. Consequently, vil can be interpreted as a tense marker here. As indicated, a modal interpretation, with the surface subject of vil as the controller of a lower argument, is also possible, but for pragmatic reasons this is not the most natural 12 In Nynorsk Norwegian, the –s(t)-passive is only considered good after modals. The reason may be that the arbitrary reference of the –s(t)-subject requires an intentional context in this variety. 13 The presence of a ‘non-speaker intention’ in –s(t)-passives that Heltoft and Falster-Jacobsen (1995) point to can probably be explained along similar lines. interpretation in this case. But interestingly, when the complement of vil is headed by a verb with –s(t), as in (53b), the modal interpretation of vil is forced upon us. (53) Norwegian (see Hegge 2004) a. Penge-ne vil bli overfør-t til money-DEF will become transfer-PTC to din your konto. account ‘The money will (or wants to) be transferred to your account.’ b. Penge-ne vil overføre-s til money-DEF will transfer-s to din konto. your account ‘The money wants to be transferred to your account.’ Apparently, the presence of –s(t) on the lower verb blocks the raising analysis of vil, which goes with the future tense interpretation. More precisely, the surface subject of future vil must either be the thematic subject of the lower verb, as is the case when vil gets a future tense interpretation in (52a), or else the thematic subject of the lower verb must be absent, as in the future tense interpretation of (53a). Now if the thematic subject of the lower verb is actually present in (53b), in the form of the clitic –s, we have an explanation for the fact that only the modal reading of vil is possible here. A similar effect of the argument nature of –s(t) is seen in (54). It seems clear that in (54a), the surface subject of the higher verb försökte ‘tried’ controls the surface subject and thematic object of the embedded predicate bli omvald ‘be re-elected’, since the thematic subject of that predicate is missing. But in (54b), the thematic subject of the embedded predicate is –s, and consequently, the reading we get is something like ‘the representative tried people to re-elect him’, which is incoherent. (54) Swedish (Engdahl 1999:7) a. Representant-en försök-te representative-DEF try-PAST bli om-val-d. become re-elect-PTC ‘The representative tried to be re-elected.’ b. ?? Representant-en försök-te representative-DEF try-PAST om-välja-s. re-elect-s ‘The representative tried to be re-elected.’ Moreover, as Hedlund (1992) notes, there is no imperative with the –s(t)-passive, although the imperative is compatible with the periphrastic passive—witness the contrast between (55a) and (55b). (55) Swedish (Hedlund 1992:125) a. Bli inte rån-ad become not rob-PTC i Chicago! in Chicago b. * Råna-s rob-s inte i Chicago! not in Chicago ‘Don’t get robbed in Chicago!’ The imperative requires a 2nd person recipient of the command. This requirement is met in (55a), but not in (55b), where –s interferes, so that the reading we get is comparable to ‘Don’t they rob you in Chicago!’. Hence, the imperative is not felicitous here. I conclude that –s(t) has certain subject properties in –s(t)-passives. This is also suggested by the fact that subject oriented adverbs in these constructions appear to target –s(t), while in periphrastic passives, they target the surface subject, if they are possible at all (see Hedlund 1992, Engdahl 1999). As for the agentive PP that is seen in some of the –s(t)-passives above, I would suggest that the relation between –s(t) and the PP is comparable to clitic doubling. For space reasons I cannot go into that here, however. 0.5.6 A look outside of Scandinavian Allegedly passive elements appearing outside of the verbal inflection is not only found in Scandinavian. The sja-marker in Russian, for example, has properties which might be taken to suggest that it is syntatically similar to the Scandinavian –s(t). In (56), we see that –sja comes last in the verbal word and that it allows an agent oriented adverb. (56) Russian (Borik 1998:18) dver’ spesial’no ne otkry-va-la-s’ door on.purpose not open-IMPF-PAST-sja ‘The door was not opened on purpose.’ And indeed, Borik (1998) suggests, albeit somewhat hesitantly, that –sja could be analyzed as an argument. On my view, the example in (57) supports this analysis, since –sja here appears to represent an internal [+human] argument that is not coreferential with the subject, just like the Swedish –s in (41). (57) Russian (Enger & Nesset 1999:37) Sobaka kusa-et-sja. dog bite-PRES.3SG-sja ‘The dog bites (people).’ The argument analysis is also mentioned by Noonan (1994) in his discussion of the so-called impersonal passive in Irish, which is formed by adding to the verb an agreement marker that is different from any other marker in the paradigm and that appears to represent an unspecified subject—see (58). (58) Irish (Noonan 1994:286) Buail-eadh le buidéal hit-PAST.PERF.IMPERS with bottle é. him ‘He was hit with a bottle’ or ‘Someone hit him with a bottle.’ Noonan rejects the argument analysis, though, on the grounds that the impersonal marker cannot be an antecedent for pronouns and anaphors, as (59) demonstrates. (59) Irish (Noonan 1994:288) * Glan-tar an duine féin. wash-PRES.IMPERS the person self (Intended meaning: ‘One washes oneself.’ Now in Scandinavian, too, we see a contrast in this respect between the passive – s(t)–, which cannot be the antecedent of an anaphor—see (60a)—and arbitrary PRO, which can—see (60b). This is unexpected if the passive –s(t) is an argument with arbitrary reference. (60) Norwegian a. * Det EXPL vask-es seg wash-s 3.REFL sjøl. self b. Det EXPL er is best PRO å vaske seg best to wash 3.REFL sjøl. self ‘Washing oneself is best.’ It can be explained, however, if we assume with Frampton and Gutmann (2000) that a Tense head can attract a default p –feature if necessary. Thus, the non-finite embedded T in (60b) attracts a 3rd person feature, shares it with PRO, and ultimately with the reflexive. The reference of the passive –s(t), which bears no relation to Tense heads, cannot be narrowed down in this way, and consequently, it is not specific enough to be shared with an anaphor. The same reasoning is valid, I believe, for the Irish impersonal passive marker. Hence, it is possible that this marker represents a nonspecific argument after all. 0.6 The reality of words I have argued in this chapter that words are simply morpheme sequences that happen to share certain distributional properties. In principle, these properties are accidental— there is no component of the grammar that specifically produces words. Nor does the grammar make reference to words as such. Below the phrase level, grammar makes reference only to morphemes. Starting with individual morphemes, the grammar can produce complex syntactic heads, which necessarily have word properties, but it can also produce other morpheme sequences that we may or may not see as words. Still, there seems to be no doubt that words do somehow exist. Notably, the term ‘word’ is perfectly meaningful even to those who have no linguistic training. As pointed out by Sapir (1921) and by numerous researchers after him, words are psychologically real. (There are however languages where no word corresponding to the English word is part of the everyday vocabulary—see Dixon & Aikhenvald 2002.) However, the psychological reality of words, and the lack of awareness of wordinternal morphemes that is also often noticed, need not mean that the elements that are commonly termed words are grammatical entities or that they form a homogeneous class in any theoretically interesting way. Popular classifications are not necessarily tenable in science—recall that whales and fish were once taken to form a class. In my view, the class of words is just as spurious. The psychological reality of words is probably a consequence of their distributional properties: since words are the minimal morpheme strings that can be used as utterances and that may be permuted more or less freely, words are the minimal linguistic units that speakers can manipulate consciously. It is then no surprise that speakers are generally aware of words. Word-internal morphemes, by contrast, cannot be consciously manipulated in the same way, and consequently, word-internal morphemes are less salient than words in the awareness of speakers. Appendix: Lesser-known languages in the examples Chalcatongo Mixtec: a Mixtecan language, spoken in Mexico. Evenki: a Tungusic language, spoken in Siberia. Lango: a Nilotic language, spoken in Uganda. Loniu: an Oceanic language, spoken in Papua New Guinea. Macushi: a Carib language, spoken in Brazil, Guyana, and Venezuela. Makaa: a Bantu language, spoken in Cameroon. Northern Saami: a Finno-Ugric language, spoken in Norway, Sweden, and Finland. Oromo: a Cushitic language, spoken in Ethiopia. Rikbaktsa: a Macro-Ge language, spoken in Brazil. Seri: a Hokan language, spoken in Mexico. Shona: a Bantu language, spoken in Zimbabwe and in neighbouring countries. Further reading Baker, Mark C. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. Cinque, Guglielmo. 1999. Adverbs and functional heads. New York: Oxford University Press. Josefsson, Gunlög. 1998. Minimal words in a minimal syntax. Amsterdam: John Benjamins. Julien, Marit. 2002. Syntactic heads and word formation. New York: Oxford University Press. Marantz, Alec. 1997. ‘No escape from syntax.’ University of Pennsylvania Working Papers in Linguistics 4(2), 201–225. References Abbott, Miriam. 1991. ‘Macushi.’ In Desmond C. Derbyshire and Geoffrey K. Pullum (eds.) Handbook of Amazonian languages, Vol. 3, pp. 23–160. Berlin: De Gruyter. Adone, Danny. 1994. The acquisition of Mauritian Creole. Amsterdam: John Benjamins. Åfarli, Tor A. 1992. The syntax of Norwegian passive constructions. Amsterdam: John Benjamins. Anderson, Stephen R. 1992. A-morphous morphology. Cambridge: Cambridge University Press. Anderson, Stephen R. 1990. ‘The grammar of Icelandic verbs in –st.’ In Joan Maling and Annie Zaenen (eds.) Syntax and semantics 24: Modern Icelandic syntax, pp. 235–273. San Diego, California: Academic Press. Baker, Mark. 1997. ‘Thematic roles and syntactic structure.’ In Liliane Haegeman (ed.) Elements of grammar, pp. 73–137. Dordrecht: Kluwer. Bloomfield, Leonard. 1933. Language. New York: Holt. Boas, Franz. 1911 (1969). ‘Introduction.’ In Franz Boas (ed.) Handbook of American Indian languages 1, pp. 1–83. Smithsonian Institution. Bureau of American ethnology, Bulletin 40. Reprinted by Anthropological Publications, Oosterhout. Borer, Hagit. 1998. ‘Deriving passive without theta roles.’ In Steven G. Lapointe, Diane Brentari, and Patrick Farrell (eds.) Morphology and its relation to phonology and syntax, pp. 60–99. Stanford, California: CSLI Publications. Borik, Olga. 1998. ‘Aspectual composition and the syntax of the Russian reflexive clitic –sja.’ MPhil thesis, University of Tromsø. Boswood, Joan. 1978. Quer falar a língua dos Canoeiros? Rikbaktsa em 26 lições. Brasília: Summer Institute of Linguistics. Bulatova, Nadezhda, and Lenore Grenoble. 1999. Evenki. Munich: Lincom Europa. Chomsky, Noam. 1957. Syntactic structures. The Hague: Mouton. Chomsky, Noam. 1970. ‘Remarks on nominalization.’ In Roderick A. Jacobs and Peter S. Rosenbaum (eds.) Readings in English transformational grammar, pp. 184– 221. Waltham, Massachusetts: Ginn and Co. Chomsky, Noam. 1993. ‘A minimalist program for linguistic theory.’ In Kenneth Hale and Samuel Jay Keyser (eds.) The view from Building 20, pp. 1–52. Cambridge, Massachusetts: MIT Press. [Also in Chomsky 1995.] Chomsky, Noam. 1995. The minimalist program. Cambridge, Massachusetts: MIT Press. Cinque, Guglielmo. 1999. Adverbs and functional heads. New York: Oxford University Press. Di Sciullo, Anna Maria, and Edwin Williams. 1987. On the definition of word. Cambridge, Massachusetts: MIT Press. Dixon, R. M. W. & Aikhenvald, Alexandra Y. 2002. Word: A cross-linguistic typology. Cambridge: Cambridge University Press. Drijkoningen, Frank. 1994. ‘Affixation and Logical Form.’ In Reineke Bok-Bennema and Crit Cremers (eds.) Linguistics in the Netherlands 1994, pp. 25–36. Amsterdam: John Benjamins. Egerland, Verner. 1998. “On verb-second violations in Swedish and the hierarchical ordering of adverbs.” Working Papers in Scandinavian Syntax 61, 1–22. Engdahl, Elisabeth. 1999. ‘The choice between bli-passive and s-passive in Danish, Norwegian and Swedish.’ Ms., Gothenburg University. Enger, Hans-Olav. 2000. ‘Verbendelsen –s i norsk bokmål: bøying eller avledning?’ [The verb suffix –s in Norwegian Bokmål: inflection or derivation?]’ Norsk Lingvistisk Tidsskrift 18, 9–35. Enger, Hans-Olav & Tore Nesset. 1999. “The value of Cognitive Grammar in typological studies: the case of Norwegian and Russian passive, middle and reflexive.” Nordic Journal of Linguistics 22, 27–60. Frampton, John, and Sam Gutmann. 2000. ‘Agreement is feature sharing.’ Ms., Northeastern University. Halle, Morris. 1997. ‘Distributed Morphology: impoverishment and fission.’ PF: Papers at the interface, MIT Working Papers in Linguistics 30, 425–449. Halle, Morris, and Alec Marantz. 1993. ‘Distributed Morphology and the pieces of inflection.’ In Kenneth Hale and Samuel Jay Keyser (eds.) The view from Building 20, pp. 111–176. Cambridge, Massachusetts: MIT Press. Halle, Morris, and Alec Marantz. 1994. ‘Some key features of Distributed Morphology.’ In Andrew Carnie, Heidi Harley, and Tony Bures (eds.) Papers on phonology and morphology. MIT Working Papers in Linguistics 21, 275–288. Heath, Daniel. 1991. ‘Tense and aspect in Makaa.’ In Stephen Anderson and Bernard Comrie (eds.) Tense and aspect in eight languages of Cameroon, pp. 3–15. Arlington, Texas: Summer Institute of Linguistics/University of Texas. Hedlund, Cecilia. 1992. On participles. Stockholm: Department of Linguistics, Stockholm University. Hegge, Per Egil. 2004. ‘Språket vårt.’ Aftenposten, 13.06. 2004. Heltoft, Lars, and Lisbeth Falster Jacobsen. 1995. ‘Danish passives and subject positions as a mood system: a content analysis.’ ROLIG-papir 54, 3–49. Holmberg, Anders, and Christer Platzack. 1995. The role of inflection in Scandinavian syntax. New York: Oxford University Press. Hróarsdóttir, Thorbjörg. 2000a. ‘Parameter change in Icelandic.’ In Peter Svenonius (ed.) The derivation of VO and OV, pp. 153–179. Amsterdam: John Benjamins. Hróarsdóttir, Thorbjörg. 2000b. Word order change in Icelandic: from OV to VO. Amsterdam: John Benjamins. Iatridou, Sabine. 1990. ‘About AgrP.’ Linguistic Inquiry 21, 551–577. Jackendoff, Ray. 1997. The architecture of the language faculty. MIT Press, Cambridge, Massachusetts. Jespersen, Otto. 1924. The philosophy of grammar. London: Allen & Unwin. Julien, Marit. 2002. Syntactic heads and word formation. New York: Oxford University Press. Julien, Marit. 2003. ‘Word order type and syntactic structure.’ In Johan Rooryck & Pierre Pica (eds.) Linguistic Variation Yearbook 1, pp. 17–59. Amsterdam: John Benjamins. Kayne, Richard S. 1993. ‘Toward a modular theory of auxiliary selection.’ Studia Linguistica 47, 3-31. Kayne, Richard S. 1994. The antisymmetry of syntax. Cambridge, Massachusetts: MIT Press. Kratzer, Angelika. 1996. ‘Severing the external argument from its verb.’ In Johan Rooryck and Laurie Zaring (eds.) Phrase structure and the lexicon, pp. 109–137. Dordrecht: Kluwer. Lapointe, Steven. 1980. A theory of grammatical agreement. Doctoral dissertation, University of Massachusetts. [Published 1985 by Garland, New York.] Lees, Robert B. 1960. The grammar of English nominalizations. The Hague: Mouton. Lødrup, Helge. 2000. ‘Exceptions to the Norwegian passive: unaccusativity, aspect and thematic roles.’ Norsk Lingvistisk Tidsskrift 18, 37–54. Macaulay, Monica. 1993. ‘Argument status and constituent structure in Chalcatongo Mixtec.’ Special session on syntactic issues in Native American languages, Berkeley Linguistics Society 19, 73–85. Marantz, Alec. 1997. ‘No escape from syntax.’ University of Pennsylvania Working Papers in Linguistics 4(2), 201–225. Marlett, Stephen A. 1990. ‘Person and number inflection in Seri.’ International Journal of American Linguistics 56, 503–541. Mitchell, Erika. 1994. ‘When AgrO is fused to AgrS: What morphology can tell us about the functional categories.’ The morphology-syntax connection, MIT Working Papers in Linguistics 22, 111–130. Myers, Scott. 1990. Tone and the structure of words in Shona. New York: Garland. Nakajima, Takashi. 1999. Word order in the Minimalist Program: a derivational approach. Doctoral dissertation, Cornell University. Nedjalkov, Igor. 1997. Evenki. London and New York: Routledge. Nilsen, Øystein. 2002. ‘Verb second and Holmberg’s generalization.’ In Jan-Wouter Zwart and Werner Abraham (eds.) Studies in comparative Germanic syntax. Amsterdam: John Benjamins. Noonan, Michael. 1994. ‘A tale of two passives in Irish.’ In Barbara Fox and Paul J. Hopper (eds.) Voice: form and function, pp. 247–277. Amsterdam: John Benjamins. Owens, Jonathan. 1985. A grammar of Harar Oromo. Hamburg: Helmut Buske. Pereltsvaig, Asya. 2001. ‘On accusative adverbials in Russian and Finnish.’ Linguistics in Potsdam 6, 155–176. Platzack, Christer. 1998. ‘A visibility condition for the C-domain.’ Working Papers in Scandinavian Syntax 61, 53–99. Pylkkänen, Liina. 2000. ‘On stativity and causation.’ In Carol Tenny and James Pustejovsky (eds.) Events as grammatical objects, pp. 417–444. Stanford, California: CSLI Publications. Rizzi, Luigi. 1997. ‘The fine structure of the left periphery.’ In Liliane Haegeman (ed.) Elements of grammar, pp. 281–337. Dordrecht: Kluwer. Rivero, María-Luisa. 1990. ‘The location of nonactive voice in Albanian and Modern Greek.’ Linguistic Inquiry 21, 135–146. Sapir, Edward. 1921. Language. London: Oxford University Press. Selkirk, Elisabeth. 1982. The syntax of words. Cambridge, Massachusetts: MIT Press. Spears, Arthur K. 1990. ‘Tense, mood, and aspect in the Haitian Creole preverbal marker system.’ In Jon Victor Singler (ed.) Pidgin and creole tense-mood-aspect systems, pp. 119–142. Amsterdam: John Benjamins. Speas, Margaret. 1991. ‘Functional heads and inflectional morphemes.’ Linguistic Review 8, 389–417. Spencer, Andrew. 1992. ‘Nominal inflection and the nature of functional categories.’ Journal of Linguistics 28, 313–341. Sylvain, Suzanne. 1936. Le créole haïtien. Wetteren, Belgium: Meester. Taraldsen, Knut Tarald. 2000. ‘V-movement and VP-movement in derivations leading to VO order.’ In Peter Svenonius (ed.) The derivation of VO and OV, pp. 97–122. Amsterdam: John Benjamins. Travis, Lisa deMena. 1992. ‘Inner Aspect and the structure of VP.’ Cahiers de Linguistique de l’UQAM 1, 130–146. Travis, Lisa. 2000. ‘Event structure in syntax.’ In Carol Tenny and James Pustejovsky (eds.) Events as grammatical objects. Stanford, California: CSLI Publications. Vinje, Finn-Erik. 1997. Moderne norsk. Oslo: Universitetsforlaget. Yanagida, Yuko. 1996. ‘Deriving surface order in discourse configurational languages.’ Formal approaches to Japanese linguistics II, MIT Working Papers in Linguistics 29, 283–302.