Decomposing Italian cliticsPaola Monachesi �Tilburg University/T�ubingen University1 IntroductionClitics have been the object of much discussion in the linguistics literature, however their de�nition is still atopic of debate. A crucial question is whether it is possible to formulate a concrete de�nition of what cliticsare, or if there is no uni�ed category of clitics.I will argue that the behavior of Italian clitics shows that they do not constitute a uniform class. There-fore, instead of assuming the existence of a special class clitics, whose elements exhibit variable behavior,I will try to decompose the properties of clitics and to assimilate them to those of other well establishedcategories.In this paper, I will consider Italian monosyllabic object clitics and the dative clitic loro. I will providemotivation for treating Italian object clitics as a�xes. Therefore, they will not be considered lexical items,but featural information which is provided in the lexicon and used in morphology and phonology for therealization of the cliticized verb form. In contrast, for the clitic loro I will show that this element exhibitsword-like properties and I will consider it a lexical item. In this way, it will be possible to account for thedi�erences in distribution between monosyllabic clitics and loro while also being able to derive their commonproperties.Italian clitics, both monosyllabic clitics and loro, can undergo clitic climbing, namely they can appearon a verbal head of which they are not an argument, if certain verbs are present. I will propose a lexicalanalysis of this construction which is based on the idea of argument composition, according to which thesubcategorization requirements of the embedded verb are passed up to the clitic climbing trigger verb. Iwill show that the proposed a�x vs. word distinction between monosyllabic clitics and loro, interacts in thedesired way with the argument composition analysis, accounting in an appropriate way for the similaritiesand di�erences of these clitics with respect to climbing.2 Italian clitics as a�xes: motivationsThe status of clitics is an issue which is still open for discussion; no de�nite answer has been provided in theliterature to the question of whether clitics should be treated as a�xes or as independent syntactic forms.There are arguments similar to the ones proposed by Miller (1992b) for French showing that Italian cliticsexhibit many properties that make them behave like lexically attached in ectional a�xes, as also pointedout in Monachesi (1993a). The arguments are mainly based on Zwicky & Pullum (1983) where criteria areproposed to distinguish between clitics and a�xes. According to these tests, Italian clitics exhibit propertiesof a�xes, as will be shown in the following sections where I present evidence in favor of this hypothesis.2.1 Degree of selection with respect to the hostSimilar to a�xes, Italian clitics are selective with respect to the host with which they combine, which isalways the verb, as can be seen in the following examples:(1) a. MartinaMartina locl.(acc) leggereads`Martina reads it'�This paper summarizes my dissertation and thus the acknowledments made there hold here as well. In addition, I wouldlike to thank Luca Dini and Graham Katz for the comments they made on an earlier draft of this paper.1

b. * UnAn articoloarticle glicl.(dat) incomprensibileuncomprehensible`An article uncomprehensible to him'c. UnAn articoloarticle incomprensibileuncomprehensible perfor luihim`An article uncomprehensible to him'In (1a), the sentence shows that Italian clitics attach to the verb, while in (1b) the sentence is ungrammaticalbecause the clitic combines with an adjective. In fact, if the host is di�erent from the verb a strong pronounshould be used instead, as shown in (1c). Furthermore, like in ectional a�xes, clitics do not a�ect the lexicalcategory of the host.2.2 Rigid orderingThe most crucial evidence for the a�xal nature of Italian clitics comes from their ordering. It is well knownthat their position within the clitic cluster is �xed, for example only the order in (2a) is correct, not that in(2b):(2) a. MartinaMartina mecl.(dat) locl.(acc) spediscesendsb. * MartinaMartina locl.(acc) micl.(dat) spediscesends`Martina sends it to me'In this case, the dative clitic, which represents the indirect object, precedes the accusative clitic, whichcorresponds to the direct object. This is not the usual order if a full phrase is present. Normally, the directobject precedes the indirect object:(3) MartinaMartina spediscesends lathe letteraletter ato VitoVitoGenerally, there is no clear relation between the order of the phrasal complements within a sentence andthat of the clitics which represent the same complements.As pointed out by Anderson (1992), there seems to be a similarity between clitics and a�xes whichalso exhibit a rigid ordering. In languages such as Swahili (Stump 1992) or Nimboran (Inkelas 1993), theidiosyncratic ordering of a�xes has been accounted for by assigning them to di�erent position classes andthe same seems to hold for Italian clitics which can be ordered according to the following template:(4) Template for the Italian clitic clusterPosition: I II III IV V VImi ci si lo si neti (ref) la (imp)gli lile leci (acc)viThis fact provides additional evidence for treating clitics as a�xes and for dealing with clitic orderingin morphology, as will be discussed in section 3.1. The same mechanism, namely template morphologycould be adopted to account for ordering in both cases. On the other hand, under the view that cliticsrepresent syntactically independent elements, their order should be dealt within syntax. However, given theidiosyncrasies which underlie clitic ordering and the fact that there is no clear correspondence between theorder of clitics and that of the related full complements, it is not possible to derive the sequencing of cliticsfrom general principles of syntax. 2

Another possibility would be to consider clitics as syntactic elements, but still postulate the existence ofa morphological module where clitic ordering and most of the morphophonological idiosyncrasies discussedin the next sections could be accounted for. An approach along these lines is advocated by Bonet (1991); inher analysis clitics are mapped into a structure which contains morphological features hierarchically ordered.In her account as well, morphology plays a crucial role in dealing with certain properties of clitics. However,the assumption that clitics are a�xes allows for a simpler account of the syntax-morphology interface. Infact, in the lexical analysis I will propose, clitics will not be considered lexical items which are located in aspeci�c position by rules of syntax, but featural information which is provided in the lexicon and used inmorphology and phonology for the realization of the cliticized verb form.2.3 CoordinationFurther evidence for the fact that Italian clitics behave like a�xes comes from coordination.1 As �rst noticedby Kayne (1975) for French, if there are two verbs which are coordinated and they share the same clitic, theclitic has to be repeated within each conjunct:(5) MartinaMartina locl.(acc) comprer�awill buy eand locl.(acc) indosser�awill wear`Martina will buy it and wear it'It cannot have wide scope over coordination, neither if it is a proclitic (6a) nor if it is an enclitic (6b):(6) a. * MartinaMartina locl.(acc) comprer�awill buy eand indosser�awill wear`Martina will buy and wear it'b. * PerTo comprarebuy eand indossarlowear cl.(acc)`To buy and wear it'This is also the case if two complex verbs are coordinated; the clitic has to be repeated in front of eachconjunct:(7) a. MartinaMartina locl.(acc) hahas compratobought eand locl.(acc) hahas indossatoworn`Martina has bought it and has worn it'b. * MartinaMartina locl.(acc) hahas compratobought eand hahas indossatoworn`Martina has bought and has worn it'Only if the second auxiliary is missing, namely if there is a coordination of two past participles, the secondclitic can be omitted:(8) MartinaMartina locl.(acc) hahas compratobought eand indossatoworn`Martina has bought and worn it'It seems, therefore, that coordination further supports the idea that Italian clitics should be considereda�xes.21Cf. Miller's discussion of French in Miller (1992a)2There is an exception to the fact that clitics must be repeated in front of each verb; namely with verbs of the X and re-Xtype, clitics can have wide scope over coordination. Given the marginality of these sentences they do not seem to a�ect thegeneral claim that clitics should be repeated in front of each conjunct.3

2.4 Arbitrary gapsZwicky & Pullum (1983) point out that in ectional paradigms allow for arbitrary gaps. In particular, thiscan be the case if a verb anomalously lacks certain forms, as can be seen also in Italian. Zanichelli (1988)reports that the verb involvere (to wrap) lacks the forms related to the past tense. Similarly, solere (to beused to) and sapere (to know) don't have the form for the present participle.Arbitrary gaps are also present in the case of clitics; they occur both when clitics combine with eachother and when they cliticize to the verb. An instance of the former case can be seen in the fact that Italian,like other Romance languages, doesn't allow the combination of a �rst or second person accusative clitictogether with a dative one:(9) a. * MartinaMartina glicl.(dat) micl.(acc) presentaintroduces`Martina introduces me to him'b. * MartinaMartina glicl.(dat) ticl.(acc) presentaintroduces`Martina introduces you to him'The sentences above become acceptable if the clitics are replaced by strong pronouns:(10) MartinaMartina presentaintroduces me/teme/you ato luihimSimilarly, the combination of a �rst and second person clitic doesn't produce acceptable results or at leastthere is variation in judgement among speakers:(11) Micl.(acc) ticl.(dat) raccomandarecommends`He recommends you to me/me to you'A sentence like the one above is not acceptable or marginally acceptable only in the interpretation Herecommends me to you which exhibits a di�erent order of clitics than the standard dative-accusative. Forother speakers the sentence is ambiguous between the two interpretations while for some other speakers itis simply not acceptable.Clitics present arbitrary gaps also when they attach to verbs; in Italian, this is the case when theycombine with a present participle. Beninc�a & Cinque (1991), point out that there are some limitations inthis combination, namely it is possible for a present participle to be followed by dative clitics:(12) IThe compensicompensation spettanti-ci/-mi/-vi/-gli/-ti/-lebelonging cl.(dat)`The compensation belonging to us/me/you/him/you/her'On the other hand, only �rst and second person accusative clitics can follow a present participle:(13) GliThe argomentitopics riguardanti-ci/-mi/-vi/-ti/concerning cl.(acc)`The topics concerning us/me/you'Third person clitics cannot combine with a present participle, even though the masculine plural form ismarginally acceptable with a plural present participle:(14) GliThe argomentitopics riguardanti*-lo/*-la/*-le/?-li/concerning cl.(acc)`The topics concerning him/her/them'However, the masculine plural form cannot follow a singular present participle:(15) L'The argomentotopic concernente*-li/concerning cl.(acc)`The topic concerning them' 4

If in the examples above, the clitics are replaced by strong pronouns which have the status of independentwords, the sentences become acceptable.These arbitrary gaps which occur both when clitics combine with each other and when they combinewith verbs seem to constitute evidence in favor of their a�xal status. Similar gaps do not seem to occur ifa pronoun or other independent words are present.2.5 Morphophonological idiosyncrasiesThe phonological shape of a�xes is often a�ected by the phonology of the stem or of other a�xes with whichthey combine. Zwicky & Pullum (1983) notice that morphophonological idiosyncrasies are rather commonof in ectional formations; their examples include forms like oxen or dice for plural a�x, slept or went for thepast tense a�x, best or worst for the superlative.Morphophonological idiosyncrasies occur also in the case of Italian cliticization. For example, if a cliticlike lo or la attaches as proclitic to a verb which begins with a vowel, a rule of vowel deletion applies, so thatthe clitic loses its vowel:(16) MartinaMartina l'cl.(acc) accettaaccepts`Martina accepts it'This applies also to mi, ti, ci, vi, si, even though mainly in spoken language.In Italian, the phonological shape of clitics can also be a�ected by the presence of other clitics; for exampleclitics ending in -i have their ending changed into -e if they are followed by another clitic which begins withl- or n-:(17) a. MartinaMartina *ti/tecl.(dat) locl.(acc) spedir�awill send`Martina will send it to you'b. MartinaMartina *mi/mecl.(dat) necl.(part) spedir�awill send moltimany`Martina will send me many of them'While Kaisse (1985) suggests a phonological analysis of the phenomenon by proposing a speci�c rule tohandle these cases; Anderson (1995) accounts for the change in shape of the clitics in terms of allomorphy.The [i/e] alternation mentioned above occurs only in the case of clitics.A change similar to the one discussed above is triggered by the third person dative feminine le; if itprecedes a clitic beginning with l- or n-, then the masculine dative form gli is used instead:(18) *le/glie-leCl.(dat) cl.(acc) hohave dategiven`I have given them to her/him'Another type of modi�cation occurs in the presence of two identical clitics where one of the two changes itsphonological shape. In Italian, if there is a sequence of two si, the �rst one surfaces as ci:(19) Ci/*sicl. sicl. vestedress pesantiheavy`One wears heavy cloths'A change which is apparently similar to the one discussed above, occurs in the sequence of two clitics whichboth surface as vi. In this case, the second vi is realized as ci:(20) ViCl.(acc) ci/*vicl.(loc) portatake`He takes you there'In Italian, the locative clitic can be either expressed as ci or vi, with vi used mainly in the literary language.The clitic ci will be used if the accusative clitic vi is present. A closer look at the data, shows that thepresence of vi is not only banned if another clitic with the same phonological realization is present, but moregenerally if an object clitic is present, as can be seen from the following examples:5

(21) a. micl.(acc) ci/*vicl.(loc) portatake`He takes me there'b. ticl.(acc) ci/*vicl.(loc) portatake`He takes you there'In example (21a), locative vi combines with the �rst person singular accusative clitic, while in (21b) itcombines with the second person accusative clitic and in both cases the result is ungrammatical. Thesequence is acceptable if locative ci is used instead.While changes occur if two identical clitics are present, this is not the case if two identical words occurone after the other, no changes occur, as can be seen in the following example:(22) a. RoccoRocco hahas seisix seisix inin pagellaschool report`Rocco has six six in his school report'b. VitoVito micl.(acc) hahas regalatogiven unaa rosarose rosapink`Vito has given me a pink rose'In the �rst example, the numeral sei is repeated twice, the second occurrence being interpreted as a gradein the school report; while in the second example, rosa is repeated twice, since the name of the ower andthat of the color are the same. In both examples, no modi�cations occur.It seems, therefore, that the morphophonological changes mentioned above, could be naturally interpretedif clitics are considered a�xes, while they would be unexpected if they are considered independent words.2.6 Verb left-detachmentFurther evidence for the a�xal status of Italian clitics comes from in�nitival left-detachment. In this con-struction, which is typical of the spoken language, a verb in the indicative can be doubled by the in�nitive.The two verbs must be the same and the latter will appear left-detached, as shown in the following sentence:(23) Dormire,To sleepinf , dormosleepind benissimovery well`As for sleeping, I sleep very well'Beninc�a (1988) notices that even though this construction resembles left dislocation it shouldn't be inter-preted as such. Indeed there are cases where it seems possible to assimilate it to hanging topic. There arecertain restrictions on which verbs can be duplicated, but what is crucial for this discussion is that only barein�nitives or cliticized in�nitives can undergo left-detachment:(24) Vedere/vederla,To see/to see cl.(acc), lacl.(acc) vedosee ognievery giornoday`As for seeing her, I see her everyday'It wouldn't be possible to have this kind of dislocation if the in�nitival is followed by complements:(25) * VedereTo see Martina,Martina, lacl.(acc) vedosee ognievery giornoday`As for seeing Martina, I see her everyday'The fact that bare in�nitives and cliticized ones have the same status seem to argue in favor of the a�xalnature of Italian clitics, as also proposed in Vallduv�� (1994) for Catalan clitics, on the basis of similarevidence.33It could be argued that these facts could be accommodated under a syntactic approach that views clitics as heads whichattach to a V. However, see Monachesi (1995b) for a discussion of why this cannot be the case.6

2.7 Direction of attachmentIn Italian, clitics can appear both as proclitic and enclitic; the direction of attachment is determined by themorphosyntactic features of the verb. They are proclitic if the verb is �nite (26a) and enclitic if the verb isnon-�nite (26b) or imperative (26c):(26) a. MartinaMartina locl.(acc) leggereads`Martina reads it'b. Vistolo,Seen cl.(acc), fuwas facileeasy decidereto decide`Having seen it, it was easy to decide'c. Leggilo!Read cl.(acc)`Read it !'It could be argued that such alternation is not typical of a�xes and, in particular, that in ectional a�xesare always enclitic in Standard Italian. However, there are languages in which such alternation can be foundwith a�xes; Fontana (1993) shows that in Arabic, the person marker is usually expressed in the imperfectby a pre�x while in the perfect by a su�x. Noyer (1994) discusses mobile a�xes in Huave; he notices thatthe a�x which expresses �rst person surfaces as a su�x in the past tense and as a pre�x otherwise.4Given the existence of languages with mobile position a�xes, it seems that the alternation exhibited byproclitics and enclitics in Italian doesn't constitute strong evidence against their a�xal status. It shouldalso be noticed, that while this alternation is still present in Standard Italian, it has been almost eliminatedin other Italian dialects where clitics show otherwise similar properties.Napoletano (Ba�le 1993) can be considered an example of a dialect where clitics usually appear beforethe verb.5 In fact, as in Italian, they are proclitic with �nite verbs (27a), but di�erently from Italian, theyprecede the verb also in the presence of an in�nitive (27b), furthermore with a negative imperative (27c) theclitics can only be proclitic, contrary to Italian where they also have the possibility of being enclitic:(27) a. V�Cl.(dat) r��ktell athe v�rit�atruth`I tell you the truth'b. pTo icl.(acc) kkur�acure`To cure them'c. n�uNot llcl.(acc) akkatt�abuy`Don't buy it'The only environment where clitics are enclitic is with imperative verbs, but in this case they have a di�erentphonological realization from proclitics.Additionally, in certain northern Italian dialects clitics usually appear as enclitic. Tuttle (1992) discussesa subset of Piedmontese dialects spoken between the Sesia and Ticino where enclisis is present in all envi-ronments. In contrast to Standard Italian, the clitic follows a �nite verb, as can be seen in the followingexample which belongs to the dialect spoken in Borgomanero, a town in the province of Novara:(28) iI serviviserve cl.(acc)`I serve you'4Afar (Fulmer 1990) also constitutes an example of a language where mobile position a�xes are attested. They appearas pre�xes on verbs beginning with [e,i,o,u] and as su�xes on verbs beginning with [a] or a consonant. Even though in thislanguage the mobility of the a�xes can receive a phonological explanation, it is still the case that the same a�x can appearboth in front and after the verb.5This is a dialect spoken in the city of Naples, in southern Italy.7

The clitic follows the complex verb also in the presence of an auxiliary verb (29a) and in the presence of acausative verb (29b):(29) a. iI �ohave truvallu�nd cl.(acc)`I have found it'b. aShe famakes vojmisee cl.(dat)`She makes me see'This is unlike Standard Italian where in the equivalent examples, the clitic would precede the complex verb.Therefore while Standard Italian maintains the alternation of clitics as proclitic and enclitic, there areItalian dialects where this possibility has been eliminated.2.8 Phonological status of Italian cliticsThe previous sections have shown that clitics share the behavior of a�xes from a morphosyntactic perspec-tive, the next question is whether they also share the phonological properties of a�xes.Previous literature (Nespor & Vogel (1986), Scalise (1986)) has denied that, from a phonological point ofview, Italian clitics behave as a�xes. Motivations are based on the fact that clitics behave di�erently froma�xes with respect to certain phonological rules such as Intervocalic s-Voicing (ISV) and RaddoppiamentoSintattico (RS) and most crucially with respect to stress. Therefore, in order to account for the particularphonological behavior of clitics, Nespor and Vogel have argued in favor of the existence of an appropriateconstituent within the prosodic hierarchy, namely the Clitic Group (cf. also Hayes (1989)). However, inMonachesi (1995a) and in Monachesi (1995b) I have shown that the behavior of Italian clitics is not verydi�erent from that of certain a�xes provided that a distinction is made between cases where one clitic ispresent and cases where two or more clitics are attached to the verb. Such a distinction follows as a naturalconsequence of Prosodic Minimality (McCarthy & Prince 1990), according to which a minimal word consistseither of two syllables or two moras. A clitic cannot constitute a prosodic word because it is monosyllabic,therefore it adjoins to the verb and it forms a prosodic word with it (cf. Booij (1995)). If two or moreclitics are present, they can merge together into a bi-/trisyllabic unit which combines with the verb as incompounds (cf. Inkelas (1993)). In this way, cliticization will be reduced in certain cases to a�xation andin other cases to compounding.3 The analysis of Italian monosyllabic cliticsGiven the evidence presented in the previous sections, I will assume that Italian clitics exhibit a�xal prop-erties; a lexical analysis of cliticization can be thus proposed. I will present a treatment, cast within theframework of Head-driven Phrase Structure Grammar (HPSG) (Pollard & Sag 1987, Pollard & Sag 1994),where clitics will not be considered lexical items which are located in a speci�c position by rules of syntax,but featural information which is provided in the lexicon and used in morphology and phonology for the real-ization of the cliticized verb form. However, clitics must also ful�ll the subcategorization requirements of theverb of which they are a semantic argument. In Italian, clitics are generally in complementary distributionwith full complements; a sentence like the following is not grammatical:6(30) *MartinaMartina locl.(acc) leggereads ilthe librobook`Martina it reads the book'Therefore, it is necessary to have a mechanism which can account for the complementary distribution ofclitics and full complements and which can provide the required featural representation. Within a lexicallybased theory of grammar such as HPSG, a lexical rule can be used to obtain this result. The ComplementCliticization Lexical Rule, proposed in Miller & Sag (1993) for French and adopted in Monachesi (1993b)and Monachesi (1993c) for Italian can serve this purpose:6If the full complement is left (or right) dislocated, its cooccurrence with a clitic pronoun is possible. For an analysis of thistype of construction which is compatible with the treatment of cliticization proposed here I refer to San�lippo (1996).8

(31) Italian Complement Cliticization Lexical Rule (ICCLR)2664verbCOMPS D..., 1 ,...ECLTS W 3775 =)264COMPS D...ECLTS W [n 1o375The rule applies to verbal forms that subcategorize for complements and its e�ect is that of removing onecomplement at a time from the COMPS list of the verb and adding this information as value of the CLTSfeature, which acts as interface to morphology. In addition, the rule has the side e�ect that the element inARG-S corresponding to the clitic will register the information associated with the clitic, as a consequence ofstructure sharing.7 This will be relevant to Binding Theory. This rule can account for cases of cliticizationlike the ones represented by sentences in simple tenses:(32) MartinaMartina locl.(acc) leggereads`Martina reads it'In this case the ICCLR applies to the verb leggere with the following COMPS list:(33) Example of the complement list of leggere�COMPS DNP[acc]E�The direct object will be removed from the COMPS list and it will be added as member of the CLTS feature:(34) Example of the entry for lo legge after ICCLR has applied�COMPS D E, CLTSnNP[acc][3sgm]o�The rule creates the featural information which will be used in morphology and phonology for the realizationof the cliticized verb form and also accounts for the complementary distribution of the clitic and the overtargument in a sentence like (30), repeated below:(30) *MartinaMartina locl.(acc) leggereads ilthe librobook`Martina it reads the book'As can be seen in (34), the COMPS list of the verb leggere is empty because the direct object has becomea member of the CLTS feature. It is, therefore, not possibile for a phrasal argument to be present, if thecorresponding clitic is also present. The ungrammaticality of sentences like (30) above, can be accountedfor in this way.Similarly, the rule can account for sentences where a ditransitive verb has one complement realized asclitic and the other one as an XP, as in the following case:(35) MartinaMartina locl.(acc) spediscesends ato RoccoRocco`Martina sends it to Rocco'Given the following COMPS list for the verb spedire:(36) Example of the complement list of spedire�COMPS DNP[acc], PP[a]E�The rule will apply once, eliminating the accusative argument from the COMPS list, while the other com-plement will remain in the COMPS list allowing for the presence of the overt PP a Rocco:7ARG-S stands for argument structure and it encodes information about the obliqueness hierarchy.9

(37) Example of the entry for lo spedisce�COMPSDPP[a]E, CLTSnNP[acc][3sgm]o�On the other hand, if both complements are cliticized, as in the following sentence:(38) MartinaMartina glielocl.(dat)(acc) spediscesends`Martina sends it to him'the lexical rule will apply twice. Given the COMPS list of the verb spedire in (36), the two complements willbe removed and added as value of the CLTS feature; since the COMPS list is empty, no overt complementcan appear:(39) Example of the entry for glielo spedisce�COMPS D E, CLTSnPP[a][3sgm] , NP[acc][3sgm]o�The CLTS feature contains the relevant information related to the clitic, which will be used in morphologyfor the realization of the cliticized verb form.The rule presented in (31) above, di�ers from a lexical rule with a similar function recently proposed byMiller & Sag (1995) for French.8 The rule which they suggest for French, uni�es the grammar of extractionand cliticization since it operates on a verb that has an element missing (thus encoded as value of the SLASHfeature) and gives as output an entry where that element has been removed from the SLASH feature andadded as value of the CLTS feature:9(40) French Complement Cliticization Lexical Rule(from Miller & Sag (1995))264verbSLASH S2CLTS S1 375 =)264SLASH S2n n 1oCLTS S1 [ n 1o375where [1] 2 S2Therefore, verbal forms �rst undergo the Complement Extraction Lexical Rule which creates slashed verbs.The rule reduces the valence of the verb by removing a given complement and adding this information asvalue of SLASH; in addition the position corresponding to the particular argument in ARG-S will be markedas a slashed position. The French Complement Cliticization Lexical Rule, presented in (40) above, will applyto this output to create cliticized verb forms.In this way, it is possible to give a uni�ed account of the fact that in French past participle agreement istriggered both by a cliticized or a wh-extracted direct object, but not by a following NP. Furthermore, thisapproach can also deal with the fact that both cliticization and extraction of the direct object allow for a oated quanti�er to occur between the auxiliary and the past participle, while this is not the case if an NPis present.On the other hand, Italian doesn't exhibit a similar correlation, since past participle agreement is triggeredby object clitics, but in general not by a wh-extracted direct object or by an object NP.10 Additionally, Italianpatterns di�erently from French regarding the oating quanti�er. In Italian, tutti, which is the equivalentof French tous, cannot intervene between an auxiliary and a past participle if an object NP is present; thesame acceptability holds if a cliticized or wh-extracted object is present.In contrast to French, in Italian, cliticization and extraction do not correlate with respect to past participleagreement and oating of the quanti�er tutti. The di�erent behavior of Italian and French with respect tothe phenomena mentioned above is re ected in the di�erent formulation of the Complement Cliticization8However, this is not their current position any more. See Abeill�e et al. (1996) discussed below.9The notation n stands for set di�erence, while [ stands for set union.10In Italian, third person object clitics trigger obligatory agreement with the past participle, while with �rst or second personclitics, agreement is optional. 10

Lexical Rule, namely the Italian version will not make reference to material contained in the SLASH feature.However, it would be desirable to unify the Italian and French lexical rules which deal with cliticization andto derive the di�erences with respect to past participle agreement and oating of the quanti�ers separately.In a more recent paper, Abeill�e et al. (1996) propose an alternative account for the French data discussedabove. They propose a cliticization lexical rule like the following:(41) Complement Cliticization Lexical Rule (CCLR)264verbCOMPS �1 . 1 . �2CLTS 2 375 =)24COMPS �1 . �2CLTS 2 [+ n 1 clitico35In the rule above, the notation � stands for list variables while dots refer both to appending and addingelements to lists. [+ stands for non-trivial set union, which will prevent 1 from being added to the set if itis already present.This rule is similar to the one adopted for Italian in (31), its e�ect is that of removing one complementat a time from the COMPS list of the verb; this information is then added as value of CLTS. The onlydi�erence is that in this approach SYNSEM objects will be distinguished between canonical (as in the caseof full phrases) and noncanonical ones, with clitic being a subsort of the latter. This rule has a relevantside e�ect: if the element which is removed from the COMPS list (to be added as value of CLTS) is also amember of the ARG-S list, the corresponding element in ARG-S will be of sort clitic. This information willbe relevant in the analysis of past participle agreement and quanti�er oating.On the other hand, verbal forms which are the output of the Complement Extraction Lexical Rulewill have the element in ARG-S, corresponding to the extracted one, of sort gap, which is also a subsortof noncanonical. Therefore, in French, both past participle agreement and the oating of quanti�ers willbe triggered if in the ARG-S list there is an element of sort noncanonical. As for Italian, past participleagreement will be triggered if an element of sort clitic is in the ARG-S list, in a way that will be made precisein section 5.3.In this manner, it is possible to have one lexical rule for cliticization which operates both in Italian andin French while still accounting for the di�erences with respect to past participle agreement and to quanti�er oating. The version of the lexical rule presented in (41) is the one which will be adopted in the rest of thiswork.3.1 The formation of the clitic clusterThe analysis which I have proposed for object clitics, relies on the assumption that clitics are not lexicalitems, but featural information which is present in the lexicon and used in morphology/phonology for thespell out of the cliticized verb form. The immediate question is, then, how the featural information will beused to produce the cliticized verb and in addition how the rigid ordering of the clitics can be derived.The issue of clitic ordering has been the topic of much discussion in the 70's; since the seminal workof Perlmutter (1971). Perlmutter claimed that the surface order of clitics cannot be determined by syntac-tic transformations and that linear order should be determined by �lters or templates (Surface StructureConstraints) that apply on the output of syntax.Much subsequent work has been concerned with criticisms and extensions of Perlmutter's proposal. Inhis attempt to account for Italian clitic ordering, Wanner (1977) argues that a Surface Structure Constraint(SSC) as proposed in Perlmutter (1971) would not be su�cient to account for the relevant data, but shouldbe supplemented by an additional negative constraint. Furthermore, a post-SSC reordering rule should beintroduced in order to account for the order of the clitic si which varies according to its interpretation.Therefore, he concludes that the Italian data constitute a serious challenge to the universal account of cliticordering advanced by Perlmutter.Additional criticism of Perlmutter's approach is provided by Seuren (1976). He points out some short-comings of the template approach and he proposes instead an alternative account which he develops forobject clitics while leaving aside the question of the ordering of adverbial clitics. His approach is based on acalculus where clitics are assigned a numerical value (functional load) and an index which determines theirposition within the cluster. Both the index and the functional load are calculated on the basis of a tableconsisiting of di�erent parameters which are based on morphological and semantic information. A maximumtotal load is �xed for di�erent languages and if the combination of clitics exceedes this limit, it is ruled11

out as not acceptable. Given the arbitrariness by which the numerical values are assigned in the di�erentlanguages, it is not clear in what sense this approach would be more insightful than Perlmutter's templates.More generally, it seems that purely syntactic approaches cannot provide the solution to the formationof the clitic cluster; additional mechanisms need to be introduced such as templates or numerical devices inorder to obtain the correct result. More recent syntactic analyses which address the issue of clitic ordering,such as Sportiche (1993) still rely on the idea of a template.The approaches previously discussed have mainly dealt with the ordering of clitics, but have generallyignored the problem of synthetic clusters and the cooccurrence restrictions in clitic combinations. If theseissues are taken into consideration, it becomes even more evident that a syntactic approach to the distri-bution of clitics is inadequate. On the other hand, they can be properly addressed within a morphologicaltreatment such as that of Bonet (1991). She views clitics as a hierarchical structure of morphological featureswhich is obtained through a mapping from S-structure to the Morphological component. This componentis characterized by speci�c morphological rules that can operate on the basic structure which is assignedto each clitic, by changing its features (cf. also Halle & Marantz (1993)). This system allows for an ac-count of synthetic clusters, namely non-transparent combinations of clitics, as in the case of the Spuriousse rule. Previous approaches have handled those cases in terms of phonological rules; in these accounts thefact that the output of the rule coincides with an independently existing clitic is a mere accident, whilethis is expected under her morphological approach. As for clitic ordering, Bonet maintains Perlmutter'stemplatic approach, namely clitics belong to di�erent position slots; linearization is then performed in themorphological component while spell out rules provide the input to phonology.The idea of accounting for the distribution of the clitics within morphology is not new, in fact Simpson &Withgott (1986) have argued that pronominal clitic clusters constitute a clear case of template morphology.They view template morphology as a word formation process that creates a structure which can be seen asa series of slots �lled by speci�c morphemes. However, it should be noticed that their notion of templatedi�ers from that proposed by Perlmutter (1971), in that the former creates a morphological structure whilethe latter acts as a �lter on syntactic structure. A crucial feature of template morphology is the notion ofposition classes, which also plays a relevant role in the analysis of languages where a�xes are rigidly ordered(cf. Inkelas (1993) for Nimboran), making the correlation between a�xes and clitics more evident.I would like to suggest that the notion of position class is the most appropriate to characterize thedistribution of Italian clitics. In fact, the investigation which I carried out on the Italian Reference Corpushas revealed that Italian clitics seem to be ordered in a way that doesn't have a straightforward connectionwith syntactic, semantic or phonological representation; their order appears as completely idiosyncratic.11Following Simpson & Withgott (1986) I will express the notion of position class by means of templatemorphology. If such an approach is adopted, the rigid ordering of Italian clitics can be accounted for. Theywill be ordered according to the following template:12(42) Template for the Italian clitic clusterPosition: I II III IV V VImi ci si lo si neti (ref) la (imp)gli lile leci (acc)vi11The corpus is being processed at the Institute for Computational Linguistics in Pisa. It consists of 13 million words fromtexts extracted from Italian newspapers, magazines, novels, short stories, handbooks, scienti�c texts and technical reports. Thesamples contained in the corpus belong to the period 1985-1989. I would like to thank Prof. Zampolli for letting me use thecorpus for this work and Remo Bindi for helping me with the extraction of the relevant data. I refer to Monachesi (1995b) fora detailed discussion of the output of the investigation carried out on the corpus, which includes a list of possible combinationsof clitics in Italian.12This chart re ects the order of clitics in Standard Italian. As for dialectal variation, it should be accounted for by thepostulation of di�erent templates. 12

Clitics are assigned to di�erent position classes and they will appear in the order prescribed by the template.Since no more than one clitic per position can be present, it will be possible to rule out ungrammaticalsequences like the following, simply because both gli and mi/ti belong to the same position class:(43) a. * MartinaMartina glicl.(dat) micl.(acc) presentaintroduce`Martina introduces me to him'b. * MartinaMartina glicl.(dat) ticl.(acc) presentaintroduce`Martina introduces you to him'On the other hand, whether the clitic(s) should be realized as proclitic or enclitic, is determined by themorphosyntactic features of the verb.4 The Italian clitic loroThe previous sections have provided an analysis for Italian object clitics together with arguments in favorof their a�xal status. In this section, I will consider another clitic, namely the Italian dative pronoun loro,which has never received much attention in the literature, the only exception being Cardinaletti (1991).Traditional grammars classify loro as a clitic even though one with an unusual distribution. Cardinaletti(1991) also points out the intriguing properties of loro which exhibits a behavior similar to that of theother Italian clitics since it occurs in a di�erent position from that of the related complement, it cannot becoordinated and it cannot be modi�ed, but on the other hand it di�ers from them in several respects.I will argue that the particular status of this element could be accounted for if it is considered a lexicalword. Therefore, a distinction will be drawn between Italian monosyllabic clitics, on the one hand and thebisyllabic clitic loro, on the other hand. The former exhibit a�x-like behavior and I have considered themnot as lexical items, but as featural information which is provided in the lexicon and used in morphologyand phonology for the realization of the cliticized verb form. The latter exhibits word-like behavior and itwill be considered a lexical item. In this way the di�erences in distribution between Italian monosyllabicclitics and loro can be easily derived.I would also like to suggest that the word-like behavior of loro could be related to its phonological shape,in fact loro is a bisyllabic clitic while all the other Italian clitics are monosyllabic.13 I will show that severalcases have been attested in the literature where the a�x vs. word distinction has been associated with thenotions of monosyllabicity and bisyllabicity, respectively.144.1 On the origin of loro: its developmentIn order to fully understand the idiosyncratic behavior of loro, it might prove useful to investigate its originsand its development.The dative plural clitic pronoun loro derives from the Latin genitive illorum. In Late Latin, there was aconfusion in the use of genitive and dative which originated from the fact that the two cases were realized inthe same way with nouns of the -a declension (e.g. puellae meaning `of the girl' and `to the girl'), as Rohlfs(1968) notices. Therefore, in a text of the 7th century, it is possible to �nd occurrences of genitive illorumused as dative, as in the following example from Tekav�ci�c (1980):(44) HocThis illorumto them dictumsaid estis`This was said to them'13However, in section 2.8, I have suggested that the combination of two clitics forms a bisyllabic unit; yet this should beseen as a morphological process which puts together two elements which are intrinsically monosyllabic. Unlike the case of loro,which is an intrinsically bisyllabic unit and which cannot be further decomposed.14The suggestion that the exceptional behavior of loro could be related to its phonological shape can be found also in Wanner(1987) and it is a possibility considered also by Nespor (1994). For a di�erent view, cf. Cardinaletti & Starke (1995) which arguethat the distinction between monosyllabic and bisyllabic entities is not crucial. They propose instead a division of pronounsinto three classes, namely weak pronouns, strong pronouns and clitics and they claim that it is the clitic/weak distinction whichis relevant for prosody and not the monosyllabic/bisyllabic distinction.13

In Old Italian loro, was a tonic pronoun whose behavior was similar to that of the singular forms lui, derivedfrom the dative illui and lei derived from the dative *illaei. These pronouns were usually preceded by apreposition, but they could also appear without a preposition in their genitive and dative use, as can be seenin the following examples from Rohlfs (1968):(45) ond'Because of that ioI risposianswered leito her (Dante, Purg. I, 62)`Because of that I answered to her'(46) quelThat che loroto them i'I volevawanted to diresay (Jacopone)`What I wanted to say to them'The examples above also show that these pronouns could both precede and follow the verb. In modernItalian this usage of the tonic pronouns without preposition has in general disappeared: in fact, the use ofgenitive and dative lui or lei is no longer attested. This is not the case of loro which still appears without apreposition in its genitive use:(47) lathe lorotheir casahouse`Their house'The use of loro without a preposition in its indirect object function is also disappearing from the spokenlanguage, being replaced by the singular masculine dative clitic gli. This is not the case in written languagewhere loro is widely used and still very productive, as an investigation of the Italian Reference Corpus hasrevealed. The study which will be presented in the following sections is based on data taken from this corpus.4.2 Italian monosyllabic clitics vs. loro: a comparisonEven though loro has always been considered a clitic in traditional Italian grammars, it di�ers from themonosyllabic clitics in several crucial respects, as pointed out in Cardinaletti (1991). In the following sections,a comparison will be drawn between loro and the monosyllabic clitics which will take their similarities anddi�erences into account.4.2.1 PositionDative loro like monosyllabic clitics cannot occur in the position of the corresponding full complement. Inparticular, loro doesn't occupy the canonical position for indirect objects, as the following ungrammaticalexample shows:(48) *MartinaMartina spediscesends lathe letteraletter loroto them`Martina sends them the letter'In fact loro, like monosyllabic clitics, must be adjacent to the verb:15(49) MartinaMartina spediscesends loroto them lathe letteraletter`Martina sends them the letter'The morphosyntactic features of the verb determine the direction of attachment for monosyllabic clitics.They are proclitic if the verb is �nite and enclitic if the verb is non-�nite or imperative. This is not the casefor loro, whose placement doesn't depend on the �niteness of the verb. Traditional grammars and previousstudies of this clitic have generally assumed that loro always follows the verb, as can also be seen in thefollowing examples:15It could be argued that this is an instance of double object construction, but see Cardinaletti (1991) for convincing evidenceagainst this possibility. In fact, the order in (49) is attested only with loro and is not productive for other pronouns and fullNP. Furthermore, loro is also found with intransitive verbs. 14

(50) a. DavaGave loroto them notizienews delleof the duetwo �gliedaughters sposatemarried`(She) gave them news about the two married daughters'b. UnaAn valangaavalanche diof lettereletters diof ammiratricifans chethat glicl.(dat) chiedonoask dito daregive loroto them una �gliochild`An avalanche of letters from fans that ask him to give them a child'Italian grammars do not mention the fact that loro can also precede the verb; Calabrese (1988) claims thatloro can never occur before the verb. An investigation of the corpus has shown that this is not always thecase since loro has the option of preceding the verb in certain contexts, namely within a relative clause andin the presence of an adnominal past participle, as can be seen in the following examples:(51) a. LaThe libert�afreedom diof fantasiafantasy chethat loroto them appartienebelongs`The freedom of fantasy that belongs to them'b. InnovazioniInnovations realizzateaccomplished dagliby the imprenditorientrepreneurs conwith ilthe poterepower d'of acquistopurchase loroto them fornitosupplieddalleby the banchebanks`Innovations which the entrepreneurs accomplished with the purchasing power supplied to themby the banks'These cases may be a remnant of the possibility that loro had in Old Italian to occur before the verb. Theydo not seem to relate to the more general possibility that monosyllabic clitics can be proclitic or enclitic,since monosyllabic clitics can never be proclitic on a past participle. A discussion of these cases will bepostponed until section 4.5.A further di�erence between monosyllabic clitics and loro is related to the fact that monosyllabic cliticscannot appear attached to an adjectival head, they can only appear attached to a verb, therefore a sentencelike the following is ungrammatical:(52) * UnA librobook glicl.(dat) incomprensibileincomprehensible`A book incomprehensible to him'On the other hand, the corresponding sentence where the monosyllabic clitic has been replaced by loro isgrammatical:(53) UnA librobook loroto them incomprensibileincomprehensible`A book incomprehensible to them'It should be noted that loro must precede the adjectival head; a sentence like the following, where it followsthe adjective, is not acceptable:(54) * UnA librobook incomprensibileincomprehensible loroto them`A book incomprehensible to them'In this respect, there is a di�erence with the cases mentioned above since in the presence of a relative clauseor of an adnominal past participle, loro can either precede or follow the verb, but if an adjective is present,loro can only precede it.Italian monosyllabic clitics must attach to a verb; no other constituent can intervene between them andthe verb. Loro, however, can be separated from the verb by negative adverbs such as mai (never) or bypositive ones like sempre (always), as also noted by Cardinaletti (1991). The following is an example of sucha case: 15

(55) IlThe mercatomarket nonwill not dar�agive mainever loroto them lethe sommeamounts necessarienecessary perto e�ettuaremake ilthe pagamentopayment`The market will never give them the amount necessary to make the payment'Therefore, in the case of loro, it seems more appropriate to talk about verb adjacency than of proclisis andenclisis.To summarize: the data presented above have shown that both monosyllabic clitics and loro occur ina special position within the sentence. While monosyllabic clitics attach to the verb which acts as a host,this is not the case of loro which can be separated from the verb by an adverb, behaving therefore more asan independent word. Furthermore, both loro and monosyllabic clitics can precede or follow the verb, butdi�erent conditions determine the direction of attachment.4.2.2 OrderingAs already discussed, Italian monosyllabic clitics are rigidly ordered; if there are two clitics which representa direct and an indirect object, they will follow the order dative-accusative, as can be seen in the followingsentence:(56) MartinaMartina devemust spedirglielasend cl.(dat) cl.(acc)`Martina must send it to him'On the other hand, if the indirect object is represented by the clitic loro, the order will be accusative-dative:(57) MartinaMartina devemust spedirlasend cl.(acc) loroto them`Martina must send it to them'The generalization seems to be that loro is always most external than the monosyllabic clitics.It seems, therefore, that this di�erence with respect to ordering argues in favor of a di�erent status forloro vs. monosyllabic clitics.4.2.3 CoordinationMonosyllabic clitics and loro behave di�erently also with respect to coordination. As already mentioned,monosyllabic clitics do not have wide scope over a coordination of verbs. If two coordinated verbs subcate-gorize for the same clitic, the clitic must be repeated with each verb, therefore a sentence like the followingis not grammatical:(58) * PresentatoShown eand ripresentatoglireshown cl.(dat)`Shown and shown again to him'This is not the case of loro which can have wide scope over a coordination of verbs:(59) PresentatoShown eand ripresentatoreshown loroto them`Shown and shown again to them'These examples constitute additional evidence for the di�erent status of lorowith respect to the monosyllabicclitics.4.2.4 Clitic ClimbingItalian monosyllabic clitics undergo clitic climbing. I will consider two cases where clitic climbing is triggered,namely if an auxiliary verb or if a restructuring verb (Rizzi 1982) is present.Monosyllabic clitics must climb if an auxiliary verb is present; it is not possible to have the clitic attachedto the lower verb: 16

(60) a. MartinaMartina l'cl.(acc) hahas mangiataeaten`Martina has eaten it'b. * MartinaMartina hahas mangiatolaeaten cl.(acc)`Martina has eaten it'On the other hand, climbing of monosyllabic clitics is optional with a restructuring verb, namely a verbwhich belongs to one of the following classes: modals, aspectuals, motion verbs, as the examples below show:(61) a. MartinaMartina lacl.(acc) volevawanted to spediresend`Martina wanted to send it'b. MartinaMartina volevawanted to spedirlasend cl.(acc)`Martina wanted to send it'The behavior of loro is di�erent, in fact if an auxiliary verb is present, climbing of loro is optional:(62) a. IThe sognidreams chethat glithe altriother pazientipatients hannohave loroto them raccontatotold`The dreams that the other patients have told them'b. InIn mezz'half an orahour DeDe BenedettiBenedetti avevahad spiegatoexplained loroto them chethat nonnot erawas pi�umore cos��so`In half an hour De Benedetti had explained them that it wasn't the case any more'If a restructuring verb is present, climbing of loro is also optional, as with monosyllabic clitics. It should benoted that loro doesn't appear before the auxiliary or restructuring verb as in standard clitic climbing casesbecause, as already mentioned, this clitic usually follows the verb, also in the case of �nite verbs:(63) a. Perch�eBecause eranowere tempitimes diof sospetti,suspicions eand nonnot sapevanoknow cosawhat potessecould loroto them capitarehappen`Because those were suspicious times and they didn't know what could happen to them'b. SoloOnly quandowhen ilthe rumorenoise delleof the campanebells cessa,stops, possocan rivolgereturn loroto them lathe parola.word`Only when the noise of the bells stops, can I talk to them'Clitic climbing in Italian has an interesting property. If there are two monosyllabic clitics which semanticallybelong to the same verb, they cannot be separated; they must climb together. Therefore the followingsentence where the clitics are split is not grammatical:(64) *MartinaMartina lacl.(acc) volevawanted to spedirglisend cl.(dat)`Martina wanted to send it to him'In fact, both clitics must attach either to the embedded verb or to the main verb:(65) a. MartinaMartina volevawanted to spedirglielasend cl.(dat) cl.(acc)b. MartinaMartina glielacl.(dat) cl.(acc) volevawanted to spediresend`Martina wanted to send it to him'This is not the case of loro which doesn't have to cluster with the other clitics, showing again anotherinstance of its idiosyncratic behavior: 17

(66) LaThe scintilla,spark, lacl.(acc) pu�ocan fornireprovide loroto them unaa dichiarazionedeclaration delfrom the ministeroMinistry dell'ofInternoInternal A�airs`A declaration of the ministry of Internal A�airs can provide them with a spark'To summarize: the data above have shown that loro, like the monosyllabic clitics can undergo clitic climbing,still the fact that climbing of loro is optional with auxiliary verbs and that it doesn't cluster with other cliticsshows that it has a di�erent status.4.2.5 PhonologyThe main di�erence between loro and the monosyllabic clitics is with respect to stress.Nespor (1994) argues that loro, as opposed to monosyllabic clitics, bears primary word stress. She pointsout that the rhythmic contour of a sentence containing this clitic is not di�erent from the contour of othersentences where full words appear in the same position as loro. This can be shown in the following examplesfrom Nespor (1994):(67) a. TelefonavaCalled loroto them tuttiall ithe giornidays`He called them every day'b. TelefonavaCalled pocolittle tuttiall ithe giornidays`He called little every day'Since loro is a bisyllabic unit, it ful�lls the minimal requirement on prosodic words, namely that they mustbe at least bimoraic or bisyllabic. Therefore, at a phonological level, the following representation can beproposed:(68) Representation of loro PPh����� HHHHHPWTelefonava PWloroLoro constitutes a prosodic word which attaches at the level of the Phonological Phrase; in this way thebehavior mentioned above is expected. Furthermore, this representation allows for the presence of certainadverbs to intervene between the verb and loro.16Nespor (1994) mentions also that loro di�ers from monosyllabic clitics in another respect, namely whilethe former undergo obligatorily the rule of troncamento, the application of this rule in the presence of lorois optional. However, I have shown in Monachesi (1995b) that there are cases where the rule doesn't applyin the presence of clitics:(69) mangiamo+celo/*mangiam+celoEat cl.(dat) cl.(acc)`Let's eat it'Given the unclear status of this rule, I will not consider it here.16For a discussion of possible representations in the case of monosyllabic clitics, cf. Monachesi (1995b).18

4.3 Bisyllabic units as independent wordsOn the basis of the di�erences shown above, a distinction will be drawn between Italian monosyllabic cliticsand the bisyllabic loro, the former are considered a�xes while the latter exhibits word-like behavior.This distinction between monosyllabic and bisyllabic units has various precedents in the literature.Zwicky (1977) discusses bisyllabic clitics in Bikol; he points out that they exhibit properties of indepen-dent words since they bear stress and they don't trigger a rule lengthening certain preceding vowels, whilemonosyllabic clitics do.Within the a�x system of Yidiny, Dixon (1977) makes a distinction between monosyllabic and bisyllabicsu�xes. The former are cohering a�xes that attach to a stem and form a unit with it while the latter arenoncohering a�xes that behave as independent words. Therefore a monosyllabic a�x coheres with a stemand forms a phonological word with it which constitutes the domain of application of certain phonologicalrules. On the other hand, a bisyllabic a�x does not cohere with the stem and begins a new phonologicalword; the relevant phonological rules apply separately to each phonological word.Similarly within the a�x system of Dutch, Van Oostendorp (1994) makes a distinction between mono-syllabic a�xes and bisyllabic ones. Dutch su�xes trigger two processes which are related to syllabi�cation:resyllabi�cation of base-�nal consonants and deletion of base-�nal schwa, while these generalizations holdfor monosyllabic su�xes, -achtig seems to be an exception. Van Oostendorp argues that it is not necessaryto postulate a morphological distinction between two classes of su�xes in order to explain the exceptionalbehavior of -achtig. This di�erence should follow from the di�erent phonological shape, namely -achtig isbisyllabic while the other su�xes are monosyllabic. Therefore -achtig acts more as an independent unit andforms su�xed with -achtig behave like compounds.The cases mentioned above seem indeed to show that there is a correlation between phonological shape- between the monosyllabic vs. bisyllabic distinction - and the a�x vs. word behavior.4.4 The distribution of the bisyllabic clitic loroAs already mentioned, I will assume that loro has a word-like status and in particular, I will consider it anelement of category XP[+CL].17 As pointed out in Cardinaletti (1991), loro can be marginally doubled by amonosyllabic clitic, behaving in this respect as a maximal projection:(70) ?? Gliel'Cl.(dat) cl.(acc) hohave datogiven loroto them`I have given it to them'Furthermore, if loro is considered a saturated XP, it is possible to account for the fact that it cannot havecomplements.Since loro is an XP, its position is determined by the general mechanism responsible for constituentordering in HPSG, namely Linear Precedence (LP) constraints. It should be recalled that loro is generallyadjacent to a verb, usually following it and preceding therefore all the other complements. This order can bederived by the interaction of two LP constraints; one, which is independently motivated, orders complementsafter the head:(71) Linear Precedence Constraint 1 (LP 1) HEAD � XPThis will account for the fact that loro appears after the verb. On the other hand, loro doesn't appear inthe usual position of an indirect object. The following constraint will order an XP[+CL] before all the otherphrasal complements:(72) Linear Precedence Constraint 2 (LP 2)COMPLEMENT[+CL] � COMPLEMENT[�LEX]In this way a sentence like (50b), repeated below, where the clitic precedes an object NP, will be accountedfor:17Alternatively, loro could be considered a full phrase, whose SYNSEM, unlike that of other full phrases, is of sort clitic.19

(50b) UnaAn valangaavalanche diof lettereletters diof ammiratricifans chethat glicl.(dat) chiedonoask dito daregive loroto them una �gliochild`An avalanche of letters from fans that ask him to give them a child'It will have the following structure:(73) Structure of dare loro un �glioVP �COMPS D E�V �COMPSD 1 , 2E� 2 XPh+CLi 1 NP h-LEXidare loro un �glioOn the other hand, the fact that loro follows a direct object monosyllabic clitic as in sentence (57), repeatedbelow, is a direct consequence of the a�xal status of monosyllabic clitics:(57) MartinaMartina devemust spedirlasend cl.(acc) loroto them`Martina must send it to them'Recall that a lexical rule has been proposed to account for simple cases of cliticization both in Italian andin French, namely the Complement Cliticization Lexical Rule presented in (41):(41) Complement Cliticization Lexical Rule (CCLR)264verbCOMPS �1 . 1 . �2CLTS 2 375 =)24COMPS �1 . �2CLTS 2 [+ n 1 clitico35The e�ect of the rule is that of removing a complement from the COMPS list of a verb; this information willbe added to the CLTS feature which acts as interface to morphology. Therefore, the lexical rule providesthe featural information which is necessary for the spell out of the cliticized verb form.Under this analysis, in sentence (57) above, the Complement Cliticization Lexical Rule applies to spedire,removing the complement from the COMPS list and adding the relevant information to the CLTS feature.The feature will trigger the construction of the cliticized verb form in morphology. Since loro is not ana�x, but an independent word, it is ordered after the verb (in this case the cliticized verb form) by LP1.Therefore, the sentence above will have the following representation:(74) Structure of deve spedirla loro20

VPV VP�COMPSD E�deve V2664COMPS D 2ECLTS�NPhacc,3fsgi�3775 2 XPh+CLispedirla loroIn contrast, monosyllabic clitics follow the dative-accusative order, as shown in sentence (56), repeated below:(56) MartinaMartina devemust spedirglielasend cl.(dat) cl.(acc)`Martina must send it to him'Recall that they are considered a�xes which belong to di�erent position classes. Therefore, the cliticizedverb form will be spelled out in morphology while the dative-accusative order is prescribed by the template,thus accounting correctly for the sentence above.4.5 Problematic dataThe data presented in (51), which I repeat below, where loro has the option of appearing before the verbwithin a relative clause and in the presence of an adnominal past participle are rather problematic:(51) a. LaThe libert�afreedom diof fantasiafantasy chethat loroto them appartienebelongs`The freedom of fantasy that belongs to them'b. InnovazioniInnovations realizzateaccomplished dagliby the imprenditorientrepreneurs conwith ilthe poterepower d'of acquistopurchase loroto them fornitosupplieddalleby the banchebanks`Innovations which the entrepreneurs accomplished with the purchasing power supplied to themby the banks'I think they should be seen as a remnant of the possibility that loro had in the past to precede the verb.Still, further research would be necessary to establish if loro could indi�erently appear before or after theverb in any context or if it had a preference for one or the other position in speci�c contexts.In particular, the option that loro has to appear before the verb within a relative clause could be relatedto the fact that in Old Italian clitics appeared in second position (Tobler-Mussa�a law) (Wanner 1987).Therefore, within a relative clause, they would follow a relative pronoun, and precede the verb. This isstill the case in European Portuguese. On the other hand, European Portuguese has clitics appearing asproclitics also in other contexts, (in clauses containing a WH question word, in negative clauses, when thesubject is a quanti�ed NP and when the verb is in the in�nitive governed by a preposition) where it wouldn'tbe possible for loro to occur before the verb. Since an adnominal past participle can be seen as an implicitrelative, this could explain why the possibility that loro has to precede the verb can be extended also tothose cases in (51b).Further research on the development of this clitic is needed in order to assess whether this could be apossible line of explanation of the data above. 21

5 Clitic climbing and argument compositionAs alreday mentioned, Italian clitics have a peculiar property. In certain cases they may appear on a verbalhead of which they are not an argument and this is the case both for monosyllabic clitics and for loro. Thisphenomenon goes under the name of clitic climbing.Sportiche (1993) claims that clitic climbing is incompatible with a lexical analysis of cliticization becausein this construction the clitic bears no lexical relationship to the main verb it is attached to, but insteadit is lexically related to the embedded verb. However, in Monachesi (1993c) and in Monachesi (1993b), Ihave argued that an analysis in terms of argument composition can provide an adequate account of Italianclitic climbing, while being compatible with a lexical approach to cliticization, such as the one proposed herefor monosyllabic clitics. The argument composition mechanism crucially relies on the notion of structuresharing which is widely used within HPSG. In the case of clitic climbing, the arguments of a verb governedby a clitic climbing trigger verb are raised to become arguments of the governor; therefore there will bestructure sharing between the complements of the embedded verb and those of the main verb.18The mechanism of argument composition was �rst introduced in HPSG by Hinrichs & Nakazawa (1990)for the treatment of the German verb cluster and it is currently employed in the analysis of Dutch (Rentier(1994) and Van Noord & Bouma (1994)) and French ( Miller & Sag (1993), Abeill�e & Godard (1994), Miller& Sag (1995), Abeill�e et al. (1996)). Its e�ect is similar to that of division in Categorial Grammar (Moortgat1988), which has also been used to account for Spanish clitic climbing (Nishida 1991).In the rest of this paper, I will consider two environments where clitic climbing is triggered, namely in thepresence of restructuring verbs and of auxiliary verbs. I will show that the argument composition approachcan provide an elegant account of clitic climbing both of monosyllabic clitics and of loro.5.1 Restructuring verbs and clitic climbingIn Italian, there are di�erent environments where clitic climbing is triggered, in particular it occurs withthree classes of verbs:19(75) Restructuring verbs� modal verbs(e.g. potere (can), dovere (must), volere (want))� aspectual verbs(e.g. cominciare (to begin), �nire (to �nish), continuare (to continue))� motion verbs(e.g. venire (to come), andare (to go), tornare (to come back))As already mentioned, if one of the verbs mentioned in (75) is present, the clitic may attach to the mainverb:(76) MartinaMartina locl.(acc) vuolewants to leggereread`Martina wants to read it'Clitic climbing is optional with these verbs; the following sentence where the clitic stays within the lowerclause is also grammatical:20(77) MartinaMartina vuolewants to leggerloread cl.(acc)`Martina wants to read it'18This analysis captures intuitions similar to those underlying Rizzi's restructuring rule (Rizzi 1982) and the clause reductionanalysis of Aissen & Perlmutter (1983).19Clitic climbing in Italian occurs also with auxiliary verbs, causative and perceptual verbs. In this paper, I will be onlyconcerned with restructuring verbs and auxiliary verbs. However, Abeill�e et al. (1996) show that the argument compositionanalysis can be easily extended to account for the behavior of causative and perceptual verbs in French. I believe that theiranalysis could be carried over to Italian.20In general there is no di�erence in meaning between the two sentences. However, see Napoli (1981), for a potentialcounterexample. 22

Following Rizzi (1982), I will assume that while there is motivation for having the embedded verb and itscomplements to form a constituent if no clitic climbing occurs, this is not the case when clitic climbingis triggered. Therefore, in this con�guration, the possibility of having a partial VP should be excluded,instead a at structure should be created, where the in�nitival verb and its complements are sisters of therestructuring verb. On the other hand, if no clitic climbing occurs, the in�nitival verb and his complementsform a VP constituent. Both structures are illustrated below:21(78) Di�erent structures associated with restructuring verbsVPV V C VPV VPV CGiven the di�erent structures presented above, two di�erent lexical entries should be associated to restruc-turing verbs; they are related by the following lexical rule:(79) Argument Composition Lexical Rule (ACLR)2666666664restr-verbSUBJ DNPECOMPS*VP264SUBJ DNPECOMPSD E375+3777777775 )2666666666664SUBJDNPECOMPS 1 �*V26664CLTSn oSUBJDNPECOMPS 1 37775+ARG-SDNP, VE� 1 3777777777775The lexical rule takes a restructuring verb which subcategorizes for a saturated VP as input and produces asoutput an alternative entry where the restructuring verb subcategorizes for a verbal complement and for thearguments of the latter.22 This operation of argument composition is indicated by the presence of the tag1 both within the COMPS list of the in�nitival and that of the restructuring verb and it indicates that thecomplements of the embedded verb become complements of the restructuring verb. Complements will beordered according to Linear Precedence constraints which will be formulated on the basis of the obliquenesshierarchy expressed in ARG-S. In particular, the following constraints would be relevant in this context:� A lexical head precedes all its complements� Verbal complements are ordered according to the obliqueness hierarchy� A verbal complement precedes a phrasal complementGiven the Argument Composition Lexical Rule proposed above, it is possible to account for cases whereclitic climbing occurs as in (76), repeated below:(76) MartinaMartina locl.(acc) vuolewants to leggereread`Martina wants to read it'In fact, the ACLR applies to the lexical entry for vuole, which has the following COMPS list:(80) Example of the complement list of vuole24COMPS*VP �COMPSD E�+3521Cf. Monachesi (1995b) for a detailed discussion of the syntactic structure of restructuring verbs.22Given the lexical rule, two structures will be created if the subcategorization slots are �lled by complements instead ofclitics: a at one and one where the lower verb and its complements form a constituent. However, data from long NP-movement,tough constructions and auxiliary selection show that they are both necessary also when clitics are not present.23

The output of the lexical rule will produce a schematic lexical entry which can be instantiated in variousway, the following is a possible instantiation:(81) Possible instantiation of the complement list of vuole after the application of the ACLR24COMPS* 1NP[acc] , V �COMPSD 1E�+35This output feeds the Complement Cliticization Lexical Rule, which I repeat below:(82) Complement Cliticization Lexical Rule (CCLR)264verbCOMPS �1 . 1 . �2CLTS 2 375 =)24COMPS �1 . �2CLTS 2 [+ n 1 clitico35Recall that the rule removes the relevant complement from the COMPS list and adds it as value of theCLTS feature; the cliticized verb form lo vuole will eventually be spelled out. In this case, there will be a at structure, as illustrated in the following example:(83) Representation for lo vuole leggereVP�COMPSD E�V26664COMPS* 2V �COMPSD 1E�+CLTS� 1NP hacc, 3Msgi� 37775 2Vlo vuole leggereCases where the clitic doesn't climb as in (77), repeated below:(77) MartinaMartina vuolewants to leggerloread cl.(acc)`Martina wants to read it'are parallel to the simple cases of cliticization presented in section 3. The Complement Cliticization LexicalRule applies to leggere and will move the relevant information into the CLTS feature; the cliticized verb formleggerlo will eventually be triggered. In this case, there will be a structure like the following, in which themain verb subcategorizes for a VP:(84) Representation for Vuole leggerlo24

VP�COMPSD E�V�COMPSDVPE� VPV2664COMPS D ECLTS�NP hacc, 3Msgi�3775vuole leggerloIt should be noticed that only a speci�c set of verbs trigger clitic climbing. An advantage of the argu-ment composition approach is that it is lexically constrained, therefore the mechanism is triggered only byrestructuring verbs. A sentence like (85):(85) *MartinaMartina locl.(acc) promettepromises dito leggereread`Martina promises to read it'is ungrammatical because promettere doesn't belong to the class of restructuring verbs, therefore it can onlysubcategorize for a saturated VP and not for the arguments of the verbal complement.In Italian it is possible to have several restructuring verbs embedded one under the other; in this casethe clitic can attach to any of the verbs:(86) a. MartinaMartina locl.(acc) vuolewants to potercan leggerereadb. MartinaMartina vuolewants to poterlocan cl.(acc) leggerereadc. MartinaMartina vuolewants to potercan leggerloread cl.(acc)`Martina wants to be able to read it'These data receive a straightforward interpretation under the argument composition analysis. In (86a), theACLR is triggered twice, applying to the two restructuring verbs. Therefore, the �rst restructuring verbinherits the complements of the verb that follows, which in turn inherits the complements of the in�nitivalverb. It is through the interaction of the schematic lexical entries, which are the output of the ACLR, thatthe desired e�ect is obtained, namely that the subcategorization requirements of the last verb are passed upto the main verb. The result will be a completely at structure. As for (86b), the ACLR will be triggeredonce; the middle verb subcategorizes for the complement of the embedded verb which is realized as a clitic.The middle verb and the lower one form a constituent, namely a VP which is the complement of the higherone. The last example is similar to the one presented in (77). In particular, the Complement CliticizationLexical Rule applies to the lower verb in order to obtain the featural information necessary for the spell outof the cliticized verb form.Clitic climbing in Italian has an interesting property. If there are two clitics which are subcategorized bythe same verb, they cannot be separated; they must climb together:25

(87) a. * VitoVito locl.(acc) volevawanted to spedirglisend cl.(dat)`Vito wanted to send it to him'b. VitoVito glielocl.(dat) volevacl.(acc) spedirewanted to send`Vito wanted to send it to him'The ungrammaticality of (87a) is accounted for by the requirement that the CLTS feature must be empty,as stated in the output of the lexical rule in (79). The subcategorization requirements of the embedded verbcan be passed up to the main verb only if the verbal complement doesn't constitute a cliticized verb form.This implies that the CLTS feature of the lower verb must be empty. This is not the case in (87a) wherethe CLTS feature contains the information related to the dative clitic. This is because the ComplementCliticization Lexical Rule applies to the lower verb, removing the complement from the COMPS list andadding the information to the CLTS feature:(88) Example of the entry for spedirgli after CCLR has applied�COMPS DNPE, CLTSnPP[dat][3sgm]o�Therefore, it will not be possible for the higher verb to subcategorize for the object since the subcategorizationrequirements of the lower verb cannot be inherited and the sentence is correctly ruled out.5.2 Restructuring verbs and climbing of loroRestructuring verbs allow not only for the climbing of monosyllabic clitics, but also for the climbing of thebisyllabic clitic loro. As in the case of monosyllabic clitics, loro can undergo optional clitic climbing withrestructuring verbs:(89) a. Perch�eBecause eranowere tempitimes diof sospetti,suspicions eand nonnot sapevanoknow cosawhat potessecould loroto them capitarehappen`Because those were suspicious times and they didn't know what could happen to them'b. SoloOnly quandowhen ilthe rumorenoise delleof the campanebells cessa,stops, possocan rivolgereturn loroto them lathe parola.word`Only when the noise of the bells stops, can I talk to them'Recall that I have considered loro an element of category XP[+CL], therefore its distribution is determined bythe Linear Precedence constraints. LP1, which states that a phrasal complement follows the head, accountsfor the fact that loro comes after the verb:(90) Linear Precedence Constraint 1 (LP 1) HEAD � XPOn the other hand, LP2 which says that a complement marked [+CL] should precede a phrasal complementaccounts for the fact that loro precedes all the other complements:(91) Linear Precedence Constraint 2 (LP 2)COMPLEMENT[+CL] � COMPLEMENT[�LEX]In this section, I will show that the analysis suggested for loro interacts in the desired way with the accountproposed for restructuring verbs. As previously explained, the Argument Composition Lexical Rule takes arestructuring verb, which subcategorizes for a saturated VP, as input:(92) Example of the complement list of a restructuring verb24COMPS*VP �COMPSD E�+3526

in order to produce a new lexical entry where the verb subcategorizes for the verbal complement and for thearguments of the verbal complement:(93) Example of the complement list of a restructuring verb after the application of ACLR"COMPS 1 ��VhCOMPS 1 i�#Given the fact that loro is considered an XP, if climbing occurs, as in (89a) above, its ordering will bedetermined by the LP constraints. LP1 orders loro after the main verb. Since capitare is a V, (it is marked[+LEX]) and it doesn't represent a phrasal complement, LP2 will not apply; therefore two orders will beallowed which are both grammatical. The example below illustrates the order shown in (89a):(94) Representation of potesse loro capitareVP�COMPS D E�V�COMPS D 1 , 2E� 1 XPh+CLi 2V24+LEXCOMPS D 1E35potesse loro capitareHowever, another order will be allowed, where loro follows the verbal complement as in (89b).23However, a normal XP cannot occur in the same position as loro; a sentence like the following is notgrammatical:(95) * RoccoRocco nonnot sapevaknow cosawhat potessecan ato GiovanniGiovanni capitarehappen`Rocco didn't know what could happen to Giovanni'This sentence is ruled out by the Linear Precedence constraint already discussed, which states that verbalcomplements should precede phrasal complements. However a further speci�cation should be added: itshould be speci�ed that this applies only to phrasal complements which are marked [-CL]. In this way,sentence (95) above will be ruled out since a Giovanni will be marked [-CL], while a sentence like (89a)where the verbal complement follows the XP loro will be allowed. This is because loro is marked [+CL],therefore it escapes the constraint mentioned above.24 Climbing of loro is forbidden with verbs which arenot members of the restructuring class; as can be seen from the following sentence:23This treatment produces spurious ambiguity since a sentence like (89b) will be also licensed by the entry corresponding tothe input of the ACLR.24Incidentally, it should be noticed that the Linear Precedence constraint mentioned above together with the di�erent struc-tures associated with restructuring verbs make certain predictions, as suggested to me by Luca Dini (p.c.). In particular, thefollowing sentences should be grammatical:(i) RoccoRocco �eis andatogone ato NapoliNapoli ato comprarlobuy cl.(acc)`Rocco has gone to Napoli to buy it'(ii) RoccoRocco locl.(acc) �eis andatogone ato comprarebuy ato NapoliNapoli`Rocco has gone to Napoli to buy it'While the following one should be ungrammatical:(iii) * RoccoRocco locl.(acc) �eis andatogone ato NapoliNapoli ato comprarebuy 27

(96) * RoccoRocco decidedecides loroto them dito spediresend lathe letteraletter`Rocco decides to send the letter to them'Sentences like the one above are ruled out because the argument composition mechanism is lexically con-strained. As already discussed, this implies that only restructuring verbs can subcategorize for the comple-ments of the embedded verb; since decidere is not a restructuring verb, it cannot inherit the complementsof spedire, therefore the presence of loro in that position is not allowed.I have already shown that monosyllabic clitics cannot be split when they climb, therefore a sentence like(87a), repeated below:(87a) * VitoVito locl.(acc) volevawanted to spedirglisend cl.(dat)`Vito wanted to send it to him'is ungrammatical. Loro di�ers from monosyllabic clitics in this respect because it doesn't cluster with theother clitics:(97) LaThe scintilla,spark, lacl.(acc) pu�ocan fornireprovide loroto them unaa dichiarazionedeclaration delfrom the ministeroMinistry dell'ofInternoInternal A�airs`A declaration of the ministry of Internal A�airs can provide them with a spark'This di�erent distribution provides further support for the analysis proposed for loro. Since loro is anXP, it doesn't combine with the verb in morphology; this implies that the CLTS feature will be empty,therefore argument composition will be possible. In this way, the restructuring verb will subcategorize forthe complements of the embedded verb:(98) Example of the entry for pu�o after ACLR has applied24COMPS* 1NP, 2PP, V �COMPSD 1 , 2E�+35This will create the input for the application of the Complement Cliticization Lexical Rule which will removethe object NP from the COMPS list and add it as value of the CLTS feature. The cliticized verb form la pu�owill eventually be spelled out. On the other hand, since loro is an XP, it will be ordered according to theLinear Precedence constraints. In this particular case, it complies with LP1 which states that a complementshould follow the head, accounting in this way for the fact that it doesn't cluster with the monosyllabicclitics. The following representation shows this:(99) Representation of La pu�o fornire loro`Rocco has gone to Napoli to buy it'These predictions are indeed borne out. In the case of example (i), there is no clitic climbing, therefore there will be a VPwhich is a complement of the main verb, thus the NP complement can precede it. While in sentence (ii) there is a at structurewhere the NP follows the in�nitival verbal complement in accordance with the Linear Precedence constraint. On the otherhand, example (iii) is ungrammatical and in fact it is not in accordance with the Linear Precedence constraint since a phrasalcomplement precedes a verbal complement.28

VP�COMPS D E�V2664COMPS* 2PP, 3V �COMPS D 1 , 2E�+CLTSn 1NP[acc][3sgf ]o 3775 3V�COMPS D 1 , 2E� 2PPh+CLila pu�o fornire loroThis analysis makes further predictions. In particular, it predicts the fact that a sentence like the followingis possible:(100) LaThe scintilla,spark, lacl.(acc) pu�ocan loroto them fornireprovide unaa dichiarazionedeclaration delfrom the ministeroMinistry dell'of theInternoInternal A�airs`A declaration of the ministry of Internal A�airs can provide them with a spark'In this case, as well, loro complies with LP1 since it follows the verbal head. Additionally, the analysisproposed predicts that a sentence like the following is not grammatical:(101) * LaThe scintilla,spark, pu�ocan loroto them fornirlaprovide cl.(acc) unaa dichiarazionedeclaration delfrom the ministeroMinistry dell'of theInternoInternal A�airs`A declaration of the ministry of Internal A�airs can provide them with a spark'Since the Complement Cliticization Lexical Rule applies to the embedded verb in order to produce thecliticized verb form fornirla, the CLTS feature will not be empty. The argument composition mechanismwill not be triggered, so loro cannot become a complement of the restructuring verb; this implies that itcannot occur in the position given above.5.3 Auxiliary verbs and clitic climbingClitic climbing can be triggered also in another environment, namely in the presence of auxiliary verbs.However, while with restructuring verbs climbing of monosyllabic clitics is optional, with auxiliary verbs itis obligatory.A clitic which originates as complement of the past participle must climb (102a); it is not possible tohave the clitic attached to the embedded verb (102b):(102) a. RoccoRocco locl.(acc) hahas lettoread`Rocco has read it'b. * RoccoRocco hahas lettoloread cl.(acc)`Rocco has read it'Following Napoli (1981), I will assume that auxiliary verbs receive the same representation as restructuringverbs which have undergone clitic climbing. Therefore in the case of a sentence like:(103) GiovanniGiovanni hahas mangiatoeaten glithe spaghettispaghetti`Giovanni has eaten spaghetti' 29

the following at structure will be associated with it:25(104) Representation of Ha mangiato gli spaghettiVPV V NPHa mangiato gli spaghettiAs in the case of restructuring verbs which have undergone clitic climbing, I will propose that auxiliaryverbs subcategorize for the past participle and for the complements of the latter. The argument compositionmechanism which I have adopted for restructuring verbs is also employed in this case.The following lexical entry can be proposed for auxiliary verbs; it corresponds to the output of theArgument Composition Lexical Rule which I have proposed for restructuring verbs:26(105) 266666666664verbSUBJ D 1NPECOMPS 2 �*V26664CLTSn oSUBJ D 1NPECOMPS 2 37775+377777777775Auxiliary verbs are considered raising verbs, therefore there is structure sharing between the subject of theauxiliary verb and that of the past participle, as indicated by the tag 1 . Furthermore, the complements ofthe past participle are inherited by the auxiliary, as shown by the tag 2 . Given the representation proposedfor auxiliary verbs, the analysis of clitic climbing becomes straightforward. In a sentence like (102a), repeatedbelow:(102a) RoccoRocco locl.(acc) hahas lettoread`Rocco has read it'the auxiliary subcategorizes for the past participle and for the NP object which is a complement of the pastparticiple:(106) Example of the complement list of aux24COMPS 2 �*V �COMPS D 2E�+35An application of the Complement Cliticization Lexical Rule will remove the NP complement from theCOMPS list and add it as value of the CLTS feature:(107) Example of the entry for ha letto after the application of CCLR24CLTSn 2NP[acc]3msgo, COMPS*V �COMPS D 2E�+3525A at structure has been proposed for French auxiliaries by Emonds (1995) and Abeill�e & Godard (1994).26In Monachesi (1993c) I had assumed that auxiliary verbs subcategorize for a VP. In this work I have adopted a at structureas in Abeill�e & Godard (1994) for French which allows for a more appropriate treatment of auxiliary verbs. Cf. also Miller &Sag (1995) . 30

The cliticized verb form is then spelled out in morphology on the basis of the information contained in theCLTS feature. In this case the information CLTSn 2 NP[acc]3msgowill be associated with the clitic lo. Thefollowing structure exempli�es this:(108) Structure for lo ha letto VPV2664COMPS* 2V �COMPS D 1E�+CLTSn 1NP[acc]3msgo 3775 2Vlo ha lettoIt should be noted that in the lexical entry presented in (105), the auxiliary subcategorizes for a pastparticiple which has not combined with clitics; in fact the CLTS feature of the past participle must beempty. This restriction accounts for obligatory climbing in the case of auxiliary verbs. Therefore, a sentencelike (102b) where the clitic has not climbed is ruled out:(102b) * RoccoRocco hahas lettoloread cl.(acc)`Rocco has read it'In this case the CLTS feature on the past participle is not empty. In fact, the verb undergoes the ComplementCliticization Lexical Rule, therefore the CLTS feature will contain the information related to the object clitic,necessary for the spell out of the cliticized verb form:(109) Lexical entry for lettolo after the application of CCLR�CLTSn 2NP[acc]3msgo, COMPS D E�It is clear that in this way the past participle doesn't meet the constraints imposed by the lexical entry in(105) above, where CLTS must be empty.In Italian, the combination of a past participle with a clitic cannot be ruled out at the morphologicallevel, since it is grammatical in absolutive past participle constructions:(110) Lettolo,Read cl.(acc), fuwas facileeasy decidereto decide`Having read it, it was easy to decide'It is only when the past participle combines with an auxiliary that the clitic must attach to the latter, theconstraint imposed on the CLTS feature, guarantees this result. Given the fact that in French clitics arenever realized on a past participle, this combination can be ruled out at the morphological level.As already mentioned in section 3, object clitics trigger past participle agreement in Italian:27(111) RoccoRocco licl.(acc)Msplu hahas lettireadMsplu`Rocco has read them'27It should be noticed that in the absence of object clitics, the past participle agrees with the subject if there is an intransitiveverb which takes essere as auxiliary or if there is a passive. If there is an intransitive verb which takes avere as auxiliary or atransitive verb, the past participle takes the masculine singular in ectional ending -o. For a treatment of this type of agreementwithin HPSG I refer to Dini (1995). In addition, it should be mentioned that past participle agreement is also triggered by theclitic ne. Further research would be necessary to assess if the treatment of agreement suggested in the case of object cliticscould be extended to account also for the agreement triggered by ne.31

Recall that the e�ect of the Complement Cliticization Lexical Rule is that of removing an element at a timefrom the COMPS list of the verb and add this information into the CLTS feature. Cliticized verb forms willhave in their ARG-S and in their CLTS an element whose SYNSEM is of sort clitic. Therefore, agreeingparticiple forms will require that their �rst element in the COMPS list, namely the object, should be of sortclitic. This will be true if the information related to this element is structure shared with that containedin the CLTS feature of an auxiliary verb.28 Lexical rules in ecting past participles will be sensitive to thisinformation.Before concluding this section, I want to point out that auxiliary verbs can also trigger the climbing ofthe clitic loro. While climbing of the monosyllabic clitics is obligatory if an auxiliary verb is present, climbingof loro is optional, as can be seen in the following examples:(112) a. IThe sognidreams chethat glithe altriother pazientipatients hannohave loroto them raccontatotold`The dreams that the other patients have told them'b. InIn mezz'half an orahour DeDe BenedettiBenedetti avevahad spiegatoexplained loroto them chethat nonnot erawas pi�umore cos��so`In half an hour De Benedetti had explained them that it wasn't the case any more'Since loro is an element of category XP[+CL], in the examples above, it will be just like an ordinarycomplement of the past participle which is inherited by the auxiliary. Two possible orders of loro will beallowed, accounting in this way for the fact that it can optionally climb if an auxiliary verb is present. Infact, it will be possible for loro both to precede and to follow the past participle; this is because the pastparticiple is not a phrasal complement therefore only LP1, repeated below:(113) Linear Precedence Constraint 1 (LP 1) HEAD � XPwill be relevant in this context and this constraint is compatible with both orders.6 ConclusionIn this paper I have taken Italian object clitics and the dative clitic loro into consideration. I have arguedthat Italian object clitics exhibit a�x behavior and I have proposed a lexical analysis to account for theirspecial position and for their distribution. On the other hand, loro has word-like status and I have consideredit a lexical item. In this way it is possible to account for the di�erences in distribution between loro andmonosyllabic clitics while being able to derive the common properties.I have also discussed clitic climbing, showing how an analysis in terms of argument composition canaccount for the di�erences and similarities with respect to the climbing of monosyllabic clitics and of loro.Two environments where clitic climbing occurs have been considered, namely in the presence of auxiliaryverbs and in the presence of restructuring verbs. Furthermore, as shown in Monachesi (1993b) and inMonachesi (1995b), the analysis in terms of argument composition which has been proposed for restructuringverbs, can also provide a uniform account of the other properties of these verbs, namely the possibility oftriggering long NP-Movement, of allowing "unbounded" tough constructions and of auxiliary selection.7 ReferencesAbeill�e, A. & D. Godard (1994): \The complementation of French auxiliaries." In: Proceedings of the Thir-teenth West Coast Conference on Formal Linguistics. CSLI, StanfordAbeill�e, A., D. Godard, P. Miller & I. Sag (1996): \French bounded dependencies." In: S. Balari & L. Dini(eds.) HPSG in Romance, Stanford. (To appear)Aissen, J. & D. Perlmutter (1983): \Clause reduction in Spanish." In: D. Perlmutter (ed.) Studies in Rela-tional Grammar . Chicago University Press, Chicago, pp. 360{403Anderson, S. (1992): A-Morphous Morphology. Cambridge University Press, Cambridge28Thanks to Ivan Sag for suggesting this to me. 32

Anderson, S. (1995): \Rules and constraints in describing the morphology of phrases." In: Proceedings of theParasession of the 31th Regional Meeting of the Chicago Linguistic Society. (To appear)Ba�le, L. (1993): \Fonologia prosodica e teoria metrica: accento, cliticizzazione e innalzamento vocalico innapoletano." Universit�a di Firenze, PhD ThesisBeninc�a, P. (1988): \L' ordine degli elementi nella frase e le costruzioni marcate." In: L. Renzi (ed.) GrandeGrammatica Italiana di Consultazione. Vol.1 . Il Mulino, Bologna, pp. 153{194Beninc�a, P. & G. Cinque (1991): \Frasi subordinate al participio: participio presente." In: L. Renzi & G.Salvi (eds.) Grande Grammatica Italiana di Consultazione. Vol.2 . Il Mulino, Bologna, pp. 604{609Bonet, E. (1991): \Morphology after Syntax: pronominal clitics in Romance." MIT, PhD ThesisBooij, G. (1995): The Phonology of Dutch. Oxford University Press, OxfordCalabrese, A. (1988): \I pronomi clitici." In: L. Renzi (ed.) Grande Grammatica Italiana di Consultazione.Vol.1 . Il Mulino, Bologna, pp. 549{592Cardinaletti, A. (1991): \On pronoun movement: the Italian dative loro." Probus 3.2:127{153Cardinaletti, A. & M. Starke (1995): \The typology of structural de�ciency. On the three grammaticalclasses." In: H. Van Riemsdijk (ed.) Clitics in the languages of Europe. Language Typology. Vol. 8.Mouton de Gruyter, Berlin. (To appear)Dini, L. (1995): \Unaccusative behaviors." Quaderni del laboratorio di linguistica 9Dixon, R. (1977): A grammar of Yidiny. Cambridge University Press, CambridgeEmonds, J. (1995): \How clitics license null phrases: a theory of the lexical interface." In: H. Van Riemsdijk(ed.) Clitics in the languages of Europe. Language Typology. Vol. 8. Mouton de Gruyter, Berlin. (Toappear)Fontana, J. (1993): \Phrase structure and the syntax of clitics in the history of Spanish." University ofPennsylvania, PhD ThesisFulmer, S. (1990): \Dual-position a�xes in Afar: an argument for phonologically-driven morphology." In: A.Halpern (ed.) Proceedings of the Ninth West Coast Conference on Formal Linguistics. Stanford, pp. 189{203Halle, M. & A. Marantz (1993): \Distributed morphology and the pieces of in ection." In: K. Hale & S.Keyser (eds.) The view from Building 20 . MIT Press, Cambridge, MA, pp. 111{176Hayes, B. (1989): \The prosodic hierarchy in meter." In: P. Kiparsky & G. Youmans (eds.) Phonetics andPhonology. Rhythm and meter . Academic Press, New York, pp. 201{260Hinrichs, E. & T. Nakazawa (1990): \Subcategorization and VP structure in German." In: Hughes, Shaun& Salmons (eds.) Proceedings of the Third Symposium on Germanic Linguistics. Benjamins, AmsterdamInkelas, S. (1993): \Nimboran position class morphology."Natural Language and Linguistic Theory 11:559{624Kaisse, E. (1985): Connected speech. The interaction of Syntax and Phonology. Academic Press, OrlandoKayne, R. (1975): French syntax: the transformational cycle. MIT Press, CambridgeMcCarthy, J. & A. Prince (1990): \Foot and word in prosodic morphology: the Arabic broken plural."Natural Language and Linguistic Theory 8:209{283Miller, P. (1992a): \Postlexical cliticization vs. a�xation: coordination criteria." In: C. Canakis, G. Chan &J. Denton (eds.) Proceedings of the 30th Regional Meeting of the Chicago Linguistic Society . Chicago,pp. 382{396Miller, P. (1992b): Clitics and constituents in Phrase Structure Grammar. Garland, New YorkMiller, P. & I. Sag (1993): \French clitic movement without clitics or movement." Paper presented at theLSA annual meeting and ms., Los AngelesMiller, P. & I. Sag (1995): \French clitic movement without clitics or movement." Lille & Stanford. Ms..(French version to appear in Revue Quebecoise de Linguistique)Monachesi, P. (1993a): \The use of nonlocal features in the analysis of Italian object clitics and cliticclimbing." In: W. Sijtsma & O. Zweekhorst (eds.) Proceedings of the third CLIN meeting . Tilburg, pp. 71{82Monachesi, P. (1993b): \Restructuring verbs in Italian HPSG grammar." In: K. Beals, G. Cooke, D. Kath-man, S. Kita, K. McCullough & D. Testen (eds.) Proceedings of the 29th Regional Meeting of the ChicagoLinguistic Society . Chicago, pp. 281{295Monachesi, P. (1993c): \Object clitics and clitic climbing in Italian HPSG grammar." In: Proceedings ofthe sixth conference of the European Chapter of the Association for Computational Linguistics. Utrecht,pp. 437{442 33

Monachesi, P. (1995a): \On the status of the Clitic Group." Paper presented at the 18th GLOW colloquium,Tr�omso. (GLOW Newsletter 34:36-37)Monachesi, P. (1995b): \A grammar of Italian clitics." Tilburg University, ITK Dissertation Series 1995-3and TILDIL Dissertation Series 1995-3. (Available from: http://www.sfs.nphil.uni-tuebingen.de/�paola)Moortgat, M. (1988): Categorial investigations: logical and linguistic aspects of the Lambek calculus. Foris,DordrechtNapoli, D. (1981): \Semantic interpretation vs. lexical governance: clitic climbing in Italian." Language57.4:841{887Nespor, M. (1994): \The phonology of Clitic Groups." In: L. Hellan & H. Van Riemsdijk (eds.) Clitic Doublingand Clitic Group. EUROTYP working papers, pp. 67{90Nespor, M. & I. Vogel (1986): Prosodic phonology. Foris, DordrechtNishida, C. (1991): \A non-transformational analysis of clitic climbing in Spanish." In: A. Halpern (ed.)Proceedings of the Ninth West Coast Conference on Formal Linguistics. Stanford, pp. 395{409Noyer, R. (1994): \Mobile a�xes in Huave: optimality and morphological well-formedness." In: Proceedingsof the Twelfth West Coast Conference on Formal Linguistics. University of California, Santa CruzPerlmutter, D. (1971): Deep and surface structure constraints in Syntax. Holt, Rinehart and Winston, Inc.,New YorkPollard, C. & I. Sag (1987): Information-based Syntax and Semantics. Fundamentals. Vol.1. CSLI, StanfordPollard, C. & I. Sag (1994): Head-Driven Phrase Structure Grammar. University of Chicago Press and CSLI,StanfordRentier, G. (1994): \A lexicalist approach to Dutch cross serial dependencies." In: K. Beals, J. Denton,R. Knippen, L. Melnar, H. Suzuki & E. Zeinfeld (eds.) Proceedings of the 30th Regional Meeting of theChicago Linguistic Society . Chicago, pp. 376{390Rizzi, L. (1982): Issues in Italian Syntax. Foris, DordrechtRohlfs, G. (1968): Grammatica storica della lingua italiana e dei suoi dialetti. Einaudi, TorinoSan�lippo, A. (1996): \Thematically bound adjuncts." In: S. Balari & L. Dini (eds.) HPSG in Romance,Stanford. (To appear)Scalise, S. (1986): Generative morphology. Foris, DordrechtSeuren, P. (1976): \Clitic pronoun clusters." Italian Linguistics 2:7{35Simpson, J. & M. Withgott (1986): \Pronominal clitic clusters and templates." In: H. Borer (ed.) The syntaxof pronominal clitics. Syntax and Semantics. Vol.19 . Academic Press, Orlando, pp. 149{174Sportiche, D. (1993): \Clitic constructions." UCLA. Ms.Stump, G. (1992): \On the theoretical status of position class restrictions on In ectional a�xes." In: G.Booij & J. V. Marle (eds.) Yearbook of Morphology . Kluwer Academic Press, Dordrecht, pp. 211{241Tekav�ci�c, P. (1980): Grammatica storica dell' italiano. Il Mulino, BolognaTuttle, E. (1992): \Del pronome d' oggetto su�sso al sintagma verbale. In calce ad una nota salvoniana del1903." L' Italia Dialettale LV:13{63Vallduv��, E. (1994): \On Catalan verb left-detachment." Paper presented at Workshop on Left Dislocation,TilburgVan Noord, G. & G. Bouma (1994): \Adjuncts and the processing of lexical rules." In: Proceedings ofCOLING 94 . Kyoto, pp. 250{256Van Oostendorp, M. (1994): \A�xation and integrity of syllable structure in Dutch." In: R. Bok-Bennema& C. Cremers (eds.) Linguistics in the Netherlands . John Benjamins Publishing Company, Amsterdam,pp. 151{162Wanner, D. (1977): \On the order of clitics in Italian." Lingua 43:101{128Wanner, D. (1987): The development of Romance clitic pronouns. Mouton de Gruyter, BerlinZanichelli (1988): \Flessioni, rime, anagrammi: l' italiano in scatola di montaggio, BolognaZwicky, A. (1977): On clitics. IULC, BloomingtonZwicky, A. & G. Pullum (1983): \Cliticization vs. in ection: English n't." Language 59.3:502{51334