Lexicalizing nuber and gender in Lunigiana Knut Tarald Taraldsen CASTL, University o Trosø Abstract In this article, I present an analysis o gender and nuber arking on nouns in a group o Italian dialects. These dialects share the property that the ural orphee is in both the einine and the asculine gender in both declension classes. But there is an asyetry: in contexts where urality is arked on a deteriner, the ural arking does not appear on nouns or adjectives in the einine gender, but does appear on asculine nouns and adjectives. I argue that this asyetry can be understood once it is recognized that a vocabulary ite can lexicalize ore than a single terinal, and that lexicalization is governed by the Superset Princie, i.e. i the lexicon associates a vocabulary ite with a eature set F, it can lexicalize any constituent with the eature set F provided F is a superset o F. 1. Introduction Current late insertion accounts o the relationship between syntactic structure and vocabulary ites (orphees) tend to share the two assuptions in (1) and (2): (1) Vocabulary insertion targets only terinal nodes. (2) The Subset Princie A vocabulary ite A associated with the eature set F can reace a terinal X with the eature set F i and only i F is a subset o F. However, a growing body o conceptual and epirical considerations suggests that (1) and (2) should be reaced with (3) and (4), as argued extensively by M. Starke (CASTL research seinars); c. Abels and Muriungi (2008), Caha (2007), Caha (2009), Muriungi (2008), Taraldsen (to appear) as well as McCawley (1968), eelean and Szendröi (2007) and Weeran and Evers-Vereul (2002): (3) Vocabulary insertion targets subtrees (4) The Superset Princie I a grateul to Rita Manzini or useul observations and to Lucie Medová or oratting the paper. c 2009 Knut Tarald Taraldsen. ordlyd 36.2, special issue on anosyntax, ed. Peter Svenonius, Gillian Rachand, Michal Starke, and Knut Tarald Taraldsen, pp. 113 127. CASTL, Trosø. http://www.uuit.no/baser/nordlyd/
Lexicalizing nuber and gender in Lunigiana A vocabulary ite A associated with the eature set F can reace a subtree X with the eature set F i and only i F is a superset o F. The eature set o a subtree will be the set o eatures associated with its terinal nodes. 1 The purpose o this reark is to provide an additional epirical arguent or oving ro (1) (2) to (3) (4). 2. Feinine ural arking in Lunigiana Manzini and Savoia (2005:III, 618) report that various Italian dialects in the Lunigiana area never use the standard Italian -e as a arker o.. Instead, one inds -ja on deteriners, nouns and adjectives which would have -a in the.sg, e.g. δona woan vs δonja woen in the Colonnata variety. Like Manzini & Savoia, I think the null hypothesis should be that -ja is decoposable as + -a, where -a is the usual einine gender arker also ound in the singular ors, 2 while is the ural aix also ound in the ural o asculine nouns, both in Lunigianese, e.g. kwanti oi how any en (Filattiera), and in Standard Italian. So, both on the standard approach and on ine, lexical insertion will target two syntactic nodes separately: (5) -a I take it that the coniguration in (5) is created by successive oveent o the to the speciier o the head through the speciier o, as shown in (8)): 1 I eatures are privative and associated one-to-one with syntactic heads ordered by a ixed hierarchy, a lexical entry will siy associate a vocabulary ite with a syntactic structure Σ, and (4) is equivalent with (i): (i) The Superset Princie A vocabulary ite A associated with the structure Σ can reace a tree X i and only i X is a subtree o Σ. 2 Harris (1991) argues that noinal endings like -a, which he calls class arkers, relect gender only indirectly. This is largely based on the observation that there are nouns in -a, e.g. artista artist and pirata pirate, which trigger asculine agreeent on deteriners and adjectives, there is a einine noun ano hand in -o (otherwise liited to asculine ors), and the e-class contains both asculine and einine nouns and adjectives. I could adopt Harris s conclusion here without any consequences or the analysis. The eatures and that appear throughout should then be thought o as whatever eatures one ight use to characterize the dierent class arkers. However, y analysis o the e-class in section 8 sees to reove the otivation this class ight provide or distinguishing class-arkers ro gender-heads, there is only one noun like ano, and the peculiarity o artista etc. is liited to the singular. Thus, I reain relatively unconvinced by Harris s arguents, and treat -a etc. as gender-heads. 114
Knut Tarald Taraldsen (6) That is, I assue that the heads and appear in their underlying order. One reason or assuing this is that it sees natural to assue that the gender arker, being directly selected by the, should be below uber (). I also assue that traces are ignored when the lexicalization procedure parses a syntactic structure. 3 Then, [ [ ]] is a subtree in (6), and it could be targeted by vocabulary insertion o -e with the lexical entry in (7a): (7) a. -e {,} -e This would produce the Standard Italian. ors. Lunigianese, however, doesn t have the Standard Italian -e. So, instead, and are lexicalized separately, as would, o course, be entirely possible also on standard accounts. otice that i traces are ignored by the lexicalization process, and will end up oring a constituent or the purposes o lexicalization even i has a coeent stranded by -oveent in (6). In act, Cinque s successul (2005) account o Greenberg s Universal 20 in ters o Poveent iies that the coeents o are evacuated ro a P beore the P starts raising across adjectives, nuerals or deonstratives, and the positions the coeents o are evacuated to ust obviously be above the position has raised to in (6). 3. Masculine ural arking in Lunigiana The ural ors o asculine nouns have a single aix in Lunigianese, e.g. kwanti oi how any en, instead o the two aixes seen in the einine ural. That is, there is no separate aix coparable to the -a o the einine paradig spelling out the gender arker. Yet, on the assuption that all nouns conor to the sae structural teate, one would expect that the position lexicalized by -a in the einine ors should be present in the structure o the asculine ors as well. 3 Any theory has to have a way o encoding the act that the in (6) is going to be lexicalized in its derived position rather than in the position o the trace (or copy). Saying that traces are ignored when the lexicalization procedure parses the input tree, aounts to saying that this procedure only sees syntactic eleents in the positions where they are lexicalized. 115
Lexicalizing nuber and gender in Lunigiana I (3) is valid, one can in act analyze a asculine ural like oi en as in (8b), positing the lexical entry in (8a): (8) a. {,} o That is, the surace discrepancy between einine urals and asculine urals would not be attributed to an underlying structural dierence, but rather to the irreducibly idiosyncratic properties o VIs (vocabulary ites). The act that associated with the eature set {, } also lexicalizes just in the einine ors, would ollow ro the Superset Princie, but is inconsistent with the Subset Princie. There are two other analytical options, both copatible with (1) (2): One could assue that the two heads hosting and -a in the einine urals use into one in the asculine paradig, or assue that the gender arker is Ø in the asculine paradigs. 4 Taking the irst o these two options one will have to block usion in the einine paradig, since the einine urals would otherwise also surace with just or just -a rather than a. 5 Thus, there will be two dierent coponents o the graar in which unpredictable properties are stipulated, the lexicon and a orphological coponent coprising usion. This appears to be an unwarranted weakening o the theory in view o the act that the alternative analysis in (8) successully relegates unpredictability to the lexicon alone. As or the second option consistent with (1) (2), there is no general arguent against null orphees as such, and, as we will see in the next section, the or o singular asculine nouns and adjectives in Lunigianese sees einently copatible with the hypothesis that the asculine gender arker is lexicalized by Ø in these varieties. 4. Masculine singular nouns in Lunigiana Most asculine nouns do not have an exponent o gender in the singular, i.e. no -o coparable to the Standard Italian.sg. -o: (9) a. o an a an radel brother a brother (Lunigianese) 4 I agree with Rachand (2008) and Fábregas (2007) that every node ust be lexicalized, but this ust still allow lexicalization by Ø. 5 Taking a as a single orphee would see to beg the question why this orphee looks exactly like the concatenation o an independently existing ural arker and a gender arker. 116
Knut Tarald Taraldsen As or those ew that do, I assue that the inal vocalic eleent is epenthetic, ollowing R. Manzini (p.c.). This is obviously consistent with saying that the gender arker is lexicalized by -Øon asculine nouns and adjectives, i.e. -Ø {}. But (3) (4) also allows one to assue that the asculine gender is lexicalized by the root along with, e.g. /o/ = [ [ ]], in the singular (assuing also privative, although that is not critical): 6 (10) o In the asculine ural ors (produced by -oveent without piedpiping), however, and do not or a subtree: (11) Thereore, no root can lexicalize and together in the ural ors. But can lexicalize together with i it has the entry in (8a) (repeated below): (12) a. {,} o As already pointed out in section 2, the Superset Princie akes this ully consistent with the hypothesis that lexicalizes just in. ors. Since the Superset Princie would also allow to lexicalize just in the asculine singular, as in (14), we also need to say why the.sg is in act not oi, but o: (13) o When a VI with the eature set F reaces a subtree T whose eature set is a proper subset o F, a nuber o the eatures in F ails to ind a atch in T. This is exicitly allowed by the Superset Princie, but we ay still assue that the lexicalization procedure seeks to iniize the nuber o unatched eatures when aced with a choice between two copeting lexicalization patterns. I so, the pattern in (10), in which no 6 otice that this rests on the assuption that lexical VIs are introduced exactly the sae way as unctional VIs. In particular, they are not conined to special root phrases, but reace phrases built up ro syntactic heads at the end o the syntactic coputation. 117
Lexicalizing nuber and gender in Lunigiana eature associated with o is unatched, is preerred over (14), in which both the eature associated with o and the eature associated with ail to be atched in the substructures targeted by lexicalization. 7 At this point, then, two analytical options reain alive or the.sg ors o Lunigianese. The gender arker ay either be lexicalized by the noinal root or by a null orphee Ø. But I will show that denying that the gender arker is Ø in asculine nouns in Lunigianese leads directly to an exanatory account o the acts we will exaine in the next section, whereas the copeting analysis doesn t. 5. Silent in Colonnata In the Colonnata dialect o Lunigianese, the ural on the einine noun sees to be in coeentary distribution with the ural arking on deteriners and quantiiers, i.e. when the noun cooccurs with an article or quantiier arked with (relected only in the palatalization o the l- o the article in (14a)), only the gender arker -a appears on the noun: (14) a. λa the.. the woen tantja so.any.. δona woan. so any woen δona woan. Otherwise, the noun has the ull. inlection a, e.g. tre d donja three woen. I take this to be an ellipsis phenoenon in the speciic sense that whenever the eleent lexicalizing is spelled out on a deteriner or a quantiier, it is not also pronounced on the noun or an attributive adjective. Siilarly, is not pronounced on a past particie agreeing with a. object clitic, as in (16a). But ro this perspective, it is surprising that is in act always spelled out on a. or (with the exceptions discussed in section 6): (15) kwanti oi how.any. an. how any en (16) a. a λ o caata S O.. have.1.sg called. I have called the einine. (Filattiera) 7 In act, Miniize unatched eatures can be shown to underlie the various blocking eects discussed in Caha (2009) and Taraldsen (to appear). 118
Knut Tarald Taraldsen a j o caati S O.. have.1.sg called. I have called the asculine. (Colonnata) In particular, this is surprising i is just speciied as {}, as it would have to be on an analysis o. a adhering to (2). But on an account assuing (4), the Superset Princie, rather than (2), we can assign the lexical entry in (8a), as the einine vs. asculine asyetry discussed in section 2 requires, i the asculine gender arker is not lexicalized by Ø in Lunigianese. This allows us to take advantage o the act that even with einine nouns the gender arker is always spelled out. The ollowing sections will show how this works. 6. Why ellipsis preserves the in the. Suppose now the structure in (15) is ebedded under a deteriner with its own arking. On the basis o (8), we now expect the head in (15) not to be pronounced. Yet, as (15) shows, the noun will have a inal. We know ro (14) that although the exponent o is not pronounced in the presence o an inlected deteriner, the gender arker (-a) is. The hypothesis that lexicalizes both and the asculine gender arker, i.e. {, }, enables us to use this act to exain why asculine ural nouns and adjectives retain even in the contexts where the einine ors lose it. Since ellipsis never aects the gender arker, suppose that -ellipsis relects the existence o a lexical entry Ø {} associated with a recoverability condition. Then, elliptical einine urals arise ro the lexicalization pattern in (17): (17) δon Ø -a That is, there will be no ural, because the only piece o structure that could be lexicalized by it is silent. otice that whenever the recoverability condition is et, Ø will always be preerred over or the lexicalization o in the einine or, since Ø is a perect it or, while has one eature () not atched by the target. But when the head is lexicalized in a asculine or, a syste iniizing unatched eatures will select the lexicalization pattern depicted in (18), assuing, as beore, that has the lexical entry in (8a) ( {, }): 119
Lexicalizing nuber and gender in Lunigiana (18) o Thus, we have the result that the o the. ors is retained under ellipsis or exactly the sae reason as the gender-arking -a o the. ors: Ø doesn t lexicalize the gender arker. 7. The case or (3) (4) at this point To assess the iport o the Lunigiana acts, it is necessary to begin by considering how analyses consistent with (1) (2) would account or the sae acts. Fro section 3, we know that an analysis respecting (1) should assue that the asculine gender eature is lexicalized by Ø in Lunigiana in order to aintain orphosyntactic parallelis between the einine urals in a and the asculine urals in. In section 4, we saw that the shape o asculine singular nouns is copatible with that assuption. But in section 5, we noticed that although the ural arker disappears ro the einine ors in certain contexts, it reains in the asculine ors in the sae environents. I the asculine ural ors are parsed as in (19b), aintaining total parallelis with the einine ors exeiied in (19a), an ellipsis rule targeting the eature should ake the disappear in the asculine ors as well: (19) a. δon -a o Ø Accordingly, upholding (1) coes at a price. Rather than orulate the ellipsis rule in the siest or copatible with the einine paradig, we ust add that is only deleted when preceding. But the analysis is unable to tie this up with any other property distinguishing asculine ro einine nouns. Abandoning (1) in avor o (3), however, we can deny that the gender arker is ever lexicalized by Ø in Lunigiana, taking to lexicalize the constituent [ [ ]] in asculine ural ors like oi. Correspondingly, the Superset Princie predicts that will appear in asculine ural ors (as a lexicalization o ) in the contexts where it disappears in the einine ors even i we aintain the siest possible stateent o the ellipsis rule. The contrast between the asculine ural and the einine in ellipsis environents siy derived ro the lexical entry or. So, the case or (3) (4) is not ade by showing that no theory consistent with (1) (2) can provide a descriptively accurate account o the Lunigiana acts considered in the preceding sections. What we have seen is rather that 120
Knut Tarald Taraldsen a descriptively accurate analysis consistent with (1) (2) requires a brute orce stipulation to handle the einine asculine contrast in the ural ors. A theory based on (3) (4) delivers a uch ore elegant analysis o the sae acts, relegating idiosyncracies to the one coponent where idiosyncracies cannot be avoided, i.e. the lexicon. This is a perectly valid line o arguent in y opinion, and one that has been proitably pursued in the developent o linguistic theory. And soe Lunigiana acts that reain to be discussed will strengthen y case. 8. Another declension class Manzini & Savoia s description indicates that even soe. ors ay in act lose the in Lunigianese dialects. But the only exaes given are the adjective zoven young (Standard Italian giovani) in Filattiera and the noun can dogs (Standard Italian cani) ro Bedizzano. R. Manzini (p.c.) adds brev short (Standard Italian breve). These contrast with. adjectives like bravi good, clever, belli beautiul and nouns like oi en in the sae contexts. I the exaes are representative, we have a contrast between declension classes. Adjectives like zoven and nouns like can all into the class o asculine nouns and adjectives whose sg ors end in -e rather than -o in Standard Italian. In Lunigiana, they apparently have no inal vowel in the singular. As a irst step towards an analysis, I adopt the view that the -e appearing in the singular ors o nouns and adjectives in the Standard Italian e-class is epenthetic. This leads to the urther conclusion that the gender arker ust be lexicalized by the root in the -e class o Standard Italian. Feinine nouns in this declension class lexicalize the einine gender arker, and asculine nouns lexicalize the asculine gender arker, along with the other eatures lexicalized by any noinal root. To extend this to adjectives, we ust represent one o the two genders as an aggregate o two privative eatures. Taking the einine as the arked gender, I decopose it into {,}. 8 An e-class adjective like breve short ust then have the entry breve {A,, }, and the Superset Princie will allow it to lexicalize the gender 8 When the einine is decoposed as {, }, and, correspondingly, the.sg gender arker is assigned the entry -a {, }, -a becoes a candidate or lexicalizing the gender arker in an elliptical asculine ural, as in (i): (i) o Ø -a But since one o the eatures associated with -a is unatched in its target in (i), this option is blocked in avor o (18). 121
Lexicalizing nuber and gender in Lunigiana arker both in the einine and the asculine ors. This orced ove will be seen to have another benign consequence later on. On this view, the derivation o a Standard Italian asculine singular like cane proceeds in the two steps shown in (20): (20) a. can (lexicalization) can cane (epenthesis) This accounts or the act that the einine and the asculine ors have the sae singular ending in the e-class. We ight also capture this by saying that the gender arker is lexicalized by Ø in both genders in the e-class. However, the einine and the asculine ors also syncretize in the ural in this class. Both the einine noce nut and the asculine cane or their urals with : noci, cani. Positing zero alloorphs or the gender arker turns out not to yield an account o this. The act that the einine nouns in the e-class or their urals with is unexpected i the structure underlying these ors are exactly like those otivated on the basis o the Lunigianese. a in section 2, i.e. [ [ [ / ]]]. To see this, consider irst what we need to say to ake the ural o donna coe out as donne rather than donnia in Standard Italian. Assuing the einine ural arker -e in Standard Italian donne woen to coe ro the entry -e {,, }, both lexicalization patterns in (21) are a priori possible: (21) a. donn -e donn -a To block (21b), we need to assue that the entry or is {, }. Then, a lexicalization procedure seeking to iniize unatched eatures in the VIs will preer (21a) over (21b), since the gender eature associated with ails to ind a atch in its target in (21b), while no eature o -e is unatched in (21a). In this light, consider the options in (22), with a einine e-class noun, assuing Ø {, }: (22) a. noc -e noc Ø Given {, }, (22a) will be preerred over (22b) or exactly the sae reason (21a) is preerred over (21b), incorrectly predicting that the 122
Knut Tarald Taraldsen ural or o noce should be noce rather than noci. Thereore, I suggest that the urals o e-class nouns in Standard Italian have a dierent structure ro the urals in the o/i-class and the a/e-class, which will continue to be the one arrived at in section (2), i.e. [ [ [ [ ]]]]. More precisely, I propose that the oveent o across at the second step o the derivation pied-pipes the gender arker in the e-class, leading to the roll-up derivation in (23) (where traces are ignored): (23) The derivation or nouns in other classes will not involve pied-piping: (24) Treating the e-class as special in exactly this way, we are iediately able not only to rule out urals in -e or the einine e-class nouns, but also to account or the asculine einine syncretis in the ural. otice irst that -e cannot lexicalize any o {,, } together in the rolled-up structure [[ ] ] ], since they do not or non-trivial constituents in this structure. That is, the assuption that lexicalization targets subtrees rules out the patterns in (25), aong others: (25) a. -e noc -e noc Thus, -e can lexicalize in einine e-class nouns only i it lexicalizes only the single head, as in (26): (26) noc -e 123
Lexicalizing nuber and gender in Lunigiana (28) shows two o the a priori possible lexicalization patterns copeting with (26), given the lexical entries in (27): (27) a. noc {,, } -a {, } c. -e {,, } d. {, } (28) a. noc noc -a The version (28b) is dispreerred relative to both (28a) and (26), since the root noce is associated with two eatures not atched in its target, while all eatures o noce are atched in (28a) and (26). Reacing -a with -e in (28b) or with -e wouldn t change this, and (28a) is thereore the only copetitor to (26). Coparing now (28a) to (26), we see that -e has two eatures not atching the target in (26) ( and ), while has only one (). Thereore, (26) in act blocks (28a), and we correctly predict that the ural or o noce is noci, relying on the decoposition o gender needed to accoodate the e-class adjectives. What we have seen, is that an analysis based on (3) (4) and assuing roll-up derivations or e-class nouns in Standard Italian accounts or the einine asculine syncretis in the ural without any appeal to contextsensitive alloorphy. Instead, we assue that nouns and adjectives in this class trigger pied-piping. Yet, within the space o analytical options copatible with (1) (2), there is one that achieves the sae desideratu: Delete in e-class nouns and adjectives. However, we will now see that there is one act which this alternative analysis can only handle at a cost, while the analysis assuing (3) (4) autoatically predicts it. 9. Back to Lunigiana Suppose we say that the Lunigiana varieties also have a separate class with the properties I have attributed to the Standard Italian e-class except there is no epenthetic -e. I an adjective like brev or a noun like can belongs to this class, the structure underlying its ural or will be the one created by the roll-up derivation, i.e. [[[[ ] ] ] ] or einine nouns/adjectives and [[[ ] ] ] or the asculine ones. This gives rise to the lexicalization patterns we have already seen or Standard Italian: 124
Knut Tarald Taraldsen (29) a. A can brev Since the root lexicalizes the gender arker both in the einine and the asculine ors, we have two correct predictions. The ural ending in the einine ors is rather than a, and in the contexts licensing -ellipsis, the disappears both in the einine and the asculine ors, giving /. brev and can, as desired, since atches all eatures o Ø (with the entry Ø {}), but leaves one eature o () unatched. The latter prediction cannot easily be reicated in an analysis consistent with (1) (2). As already noted, such an analysis could capture the einine asculine syncretis in the e-class urals by deleting the eature ro the gender arker. But in order to prevent the ural ro being targeted by -ellipsis in the asculine ors o the class corresponding to the Standard Italian o/i-class, e.g. oi en or belli beautiul, such an analysis needs to ake ellipsis apicable only in the context o a gender arker associated with. So the question now arises why -ellipsis is apicable at all with e-class nouns and adjectives. One could o course sidestep this proble by saying that what is deleted in the e-class is the whole gender arker, both in the einine and the asculine ors, and that -ellipsis apies except in the neighborhood o a gender arker bearing the eature. 9 But negative context speciications o this sort extend the power o distributional stateents in such a way that one can restrict the distribution o a or to a set o paradig cells not corresponding to a natural class. Consider, or exae, the ollowing distribution o the alloorphs A and B next to eleents o a category C subclassiied by two binary eatures X and Y: (30) +X -X +Y BC AC -Y AC AC The distribution o A in (30) cannot be given by a stateent deining its context as a natural class, e.g. A Z/ {C, +X}. But it can be stated as A Z except in {C, +X, +Y }. However, a distributional pattern like (30) can also be handled by blocking. One would posit A Z/ {C} and B Z/ {C, +X, +Y }, relying on soe version o the elsewhere princie to choose B over A next to a C speciied as {+X, +Y}. The act that an analysis consistent with (1) (2) needs a negative context speciication or -ellipsis in Lunigiana thereore eans that adopting such an analysis 9 Deleting just instead o would also account or the einine asculine syncretis in the e-class, and would predict -ellipsis in the asculine e-class ors with the asculine rather than the einine decoposed as {, }, but would require positing a special. alloorph or the e-class. 125
Lexicalizing nuber and gender in Lunigiana leads to a theory that have two dierent ways o handling paradigs like (30). But parsiony would dictate that there should only be one. The alternative analysis based on (3) (4), as developed over the preceding sections, is in act a blocking account o -ellipsis in the Lunigiana varieties. In the ellipsis contexts, appears instead o Ø just in case there is an extra eature () which can be lexicalized by, but not by Ø. 10. Suary I believe that the preceding sections have shown that adopting (3) (4) enables one to develop a coherent and relatively elegant account o the Lunigiana acts. Copeting accounts adhering to (1) (2), on the other, can achieve descriptive adequacy only by resorting to various ad hoc easures soe o which seriously reduce the appeal o the underlying general theory o the syntax lexicon connection. O course, I would view this conclusion lightly, i a theory based on (1) (2) were known to provide ore insightul analyses in other cases, or could be argued to be ore constrained. But I know o no case where an analysis copatible with (1) (2) has been shown to be superior to any analysis based on (3) (4). As or restrictiveness, I note that analyses oicially vindicating (1) (2) typically all back on orphological processes like usion to deal with acts that on the ace o it contradict the basic preise. This, o course, both underines any clai to restrictiveness and coicates the architecture o the syste by adding a orphological coponent which becoes totally redundant once (3) (4) are adopted. Reerences Abels, Klaus and Peter Muriungi. 2008. The ocus particle in Kîîtharaka: Syntax and seantics. Lingua 118: 687 731. Caha, Pavel. 2007. Case Moveent in PPs. In ordlyd. Special issue on Space, Motion and Result, edited by Monika Bašić, Marina Pantcheva, Minjeong Son, and Peter Svenonius, vol. 32, pp. 239 299. Caha, Pavel. 2009. The anosyntax o Case. Ph.D. thesis, CASTL Trosø. Cinque, Guglielo. 2005. Deriving Greenberg s universal 20 and its exceptions. Linguistic Inquiry 36 3: 315 332. Fábregas, Antonio. 2007. An exhaustive lexicalization account o directional coeents. In ordlyd. Special issue on Space, Motion and Result, edited by Monika Bašić, Marina Pantcheva, Minjeong Son, and Peter Svenonius, vol. 32, pp. 165 199. Harris, Jaes W. 1991. The exponence o gender in Spanish. Linguistic Inquiry 22 1: 27 62. Manzini, Maria Rita and Leonardo M. Savoia. 2005. I Dialetti Italiani e Roanci: orosintassi generativa. Edizioni dell Orso: Alessandria. 126
Knut Tarald Taraldsen McCawley, Jaes D. 1968. Lexical insertion in a graar without deep structure. In Papers ro the ourth regional eeting o the Chicago Linguistic Society, edited by B. J. Darden, C.-J.. Bailey, and A. Davidson, pp. 71 80. University o Chicago, Chicago. Muriungi, Peter. 2008. Phrasal oveent inside Bantu verbs. Ph.D. thesis, CASTL Trosø. eelean, Ad and Krista Szendröi. 2007. Radical Pro Drop and the Morphology o Pronouns. Linguistic Inquiry 38 4: 671 714. Rachand, Gillian. 2008. Verb Meaning and the Lexicon: A First Phase Syntax. Cabridge University Press, ew York. Taraldsen, Knut Tarald. to appear. The nanosyntax o guni noun class preixes and concords. Lingua. Weeran, Fred and Jacqueline Evers-Vereul. 2002. Pronouns and case. Lingua 112: 301 338. 127