Syntax and Morphology in the Placement of Clitics

This article proposes an analysis of clitic placement in European Portuguese and Brazilian Portuguese (henceforth, respectively EP and BP), from a comparative perspective. In the first part of the article, we use as a comparative corpus the original text of Paulo Coelho’s novel O Alquimista and the adapted version of the Portuguese edition in order to illustrate the differences between the two varieties of Portuguese. The second part of the article presents a review of some recent proposals dealing with enclisis in tensed clauses in EP and of the few studies that explicitly raise the question of the difference between BP and EP clitic-placement. In the third part we present a new comparative analysis for EP and BP cliticplacement. Our proposal is that the different behavior of clitics in EP and BP follows from the interaction of two different properties: i) a syntactic property: EP clitics are Infl-clitics and BP clitics are V-clitics. This analysis allows us to tighten the correlation between the placement of clitics in BP with the reduction of the paradigm of the accusative/dative clitics. We adopt Galves’s (2002) idea that a pure accusative marking that is structural requires an Agreement node. Since Infl no longer contains Agr in BP, the only alternative left is the inherent case marking/checking of the clitic by the verb; ii) a morpho-phonological property: EP clitics, but not BP clitics, are required to be in a non-initial position with respect to some boundary. This part of the analysis is based in great part on Galves and Sândalo (2004) who consider clitics as phrasal affixes that are subject to word formation rules like any other affixes. This line of thinking is very much in accordance with recent proposals in the framework of Distributive Morphology. Finally, in the fourth part of this article, we argue that our analysis is preferable to others for both empirical and theoretical reasons. We show that it is able to explain EP clitic-placement in both tensed and infinitival clauses and to account for the variation observed in some contexts. We also bring historical data into the discussion, which we argue can be harmoniously integrated into our explanation of the synchronic facts.


Introduction
This article proposes an analysis of clitic placement in European Portuguese and Brazilian Portuguese (henceforth, respectively EP and BP), from a comparative perspective.In the first part of the article, we use as a comparative corpus the original text of Paulo Coelho's novel O Alquimista and the adapted version of the Portuguese edition in order to illustrate the differences between the two varieties of Portuguese.
The second part of the article presents a review of some recent proposals dealing with enclisis in tensed clauses in EP and of the few studies that explicitly raise the question of the difference between BP and EP clitic-placement.
In the third part we present a new comparative analysis for EP and BP cliticplacement.Our proposal is that the different behavior of clitics in EP and BP follows from the interaction of two different properties: i) a syntactic property: EP clitics are Infl-clitics and BP clitics are V-clitics.This analysis allows us to tighten the correlation between the placement of clitics in BP with the reduction of the paradigm of the accusative/dative clitics.We adopt Galves's (2002) idea that a pure accusative marking that is structural requires an Agreement node.Since Infl no longer contains Agr in BP, the only alternative left is the inherent case marking/checking of the clitic by the verb; ii) a morpho-phonological property: EP clitics, but not BP clitics, are required to be in a non-initial position with respect to some boundary.This part of the analysis is based in great part on Galves and Sândalo (2004) who consider clitics as phrasal affixes that are subject to word formation rules like any other affixes.This line of thinking is very much in accordance with recent proposals in the framework of Distributive Morphology.
Finally, in the fourth part of this article, we argue that our analysis is preferable to others for both empirical and theoretical reasons.We show that it is able to explain EP clitic-placement in both tensed and infinitival clauses and to account for the variation observed in some contexts.We also bring historical data into the discussion, which we argue can be harmoniously integrated into our explanation of the synchronic facts.

I. Clitic placement in EP and BP: a comparative description
To illustrate the differences in clitic placement between EP and BP we shall use as a comparative corpus the original text of Paulo Coelho's novel O Alquimista and the adapted version of the Portuguese edition1 .In the remainder of this section, we shall refer to the sentences of the original text as BP and to the revised sentences of the adapted version as EP 2 .

The contexts for obligatory enclisis in EP
As it is well-known, EP is characterized by the fact that enclisis is obligatory in certain finite contexts.We can divide these contexts into two main classes, according to the position of the verb in the clause.We shall thus distinguish V1 and V2 contexts 3 .
1.1 V1 contexts V1 contexts are the ones in which either there is no phrase at all preceding the finite verb inside the sentence (absolute V1) or the preceding material is not grammatically connected to the verb.This is the case at the left boundary of parentheticals, or just after parentheticals, or in coordinates not introduced by a connector.In all these cases, we find proclisis, with few exceptions, in BP, and enclisis without exception in EP.
It is worth noting that proclisis in absolute V1 contexts is one of the great innovations of BP syntax, since no examples of this placement are found in the history of European Portuguese (cf.Martins 1994, for Old  The same variation is found in Coordinate clauses in which the verb immediately follows the coordination conjunction "e", and, or "mas", but:

V≥2 contexts
Enclisis is categorical in V2 contexts in EP in root affirmative clauses unless the verb is in the scope of some focalized or quantified phrase5 or adverbs of certain classes.The pre-verbal phrase can be the subject of the verb or some other fronted phrase, PP or adverb.Accordingly, we systematically find enclisis in these cases in the Portuguese version, when the original text is proclitic.
(6) a. Ele me parece mais velho e mais sábio.(BP) He CL1sg seems more old and more wise b.Ele parece-me mais velho e mais sábio.(EP) He seems CL1sg more old and more wise

Variation in BP
Enclisis is also found in these contexts in BP.This reflects the written norm, which is specially strong in the absolute first position since proclisis in this position is an innovation of BP 6 , not yet recognized as a legitimate construction.We also find some variation in V2 contexts, after subjects, PPs and adverbs, as well as in V1 coordinate clauses.We interpret this variation as the effect of grammar competition, in the sense of Kroch (2001).This means that, although they occur in the original Brazilian version of the text, the following sentences are not produced by the BP grammar, but correspond to pieces of other grammars, learned by the speakers by means of a special training: (10) Procurei-o a manhã inteira, disse.(BP-EP) Looked for CL3sg the morning whole said 'I have looked for him during the morning, he said.' 6But, as Coelho's style is very close to the Brazilian colloquial speech, such a variation is relatively marginal.In other kinds of written texts, we find much less proclisis in V1 contexts.Cf.Duarte, Matos and Gonçalves (2002), who report that in 37 occurrences of V1 sentences, they only find 4 cases of proclisis.They also comment that in imperative sentences, they only find enclisis.Example (i)  It is interesting to note that the variation observed in ( 6)-( 9) vs. ( 10)-( 14) is also found in Classical Portuguese (see Section IV).This raises the question of whether the competition found in BP written texts is a competition with the EP grammar or with the Classical grammar or both.We won't go further into this issue in this paper.

Clitic placement in verbal clusters
As already mentioned by Teyssier (1976), one important aspect of the difference between EP and BP clitic-placement concerns the position of the clitic in verbal sequences.Two main cases should be considered:

Auxiliary+participle
In many cases, it seems, at a first glance, that the only difference has to do with hyphenization.
(15) a. E tinham se entendido perfeitamente.(BP) And had CL3sg understood perfectly b.E tinham-se entendido perfeitamente.(EP) And had CL3sg understood perfectly 'And they had understood each other perfectly.'But a closer look shows that things are more complicated than they seem.First, the absence of hyphen is not a general rule of the Brazilian orthography, since it is found in many other cases when the clitic is post-verbal.Second, we saw above that the general tendency of BP, which is clearly represented in the Brazilian text, is not enclisis but proclisis.So we can interpret the superficial order AUX CL V not as the reflex of the encliticization of the pronoun to AUX, as in the Portuguese version, but as deriving from its procliticization to V.This is yet another Brazilian innovation with respect to Classical Portuguese which, like EP, does not allow the clitic to attach to participles (cf.Pagotto (1992) and Cyrino (1994) for the evolution of this construction in BP).In EP, by contrast, we have additional evidence that the clitic is enclitic to the auxiliary in examples in which a preposition appears before the verb: (  19) show that clitic-placement in BP is not sensitive to the presence of subordinators like "como" (as) and "onde" (where) , or quantified expressions like "cada vez mais" (more and more) which yield proclisis to the inflected verb in EP.In BP, the pronoun keeps attached to the thematic verb, independently of what happens in the left periphery of the clause7 .

Infinitival clauses
When the nonfinite form of the verbal cluster is infinitival, the general rule of proclisis to the thematic verb continues to apply in BP.In EP, the clitic can be either enclitic to the infinitival verb (20-21) or adjoined to the inflected verb (22)(23).In the latter case, its being enclitic or proclitic follows from the general rule for tensed clauses.Like in other Romance languages, clitic-climbing to the inflected verb is dependent on the class of this verb.But it is not obligatory (see the contrast between 20b and 22b, with the verb "querer" -to want-or between 21b and 23b with the verb "ir" -to be going to-)8 .If the inflected verb does not allow restructuring, the clitic invariably remains enclitic to the infinitive.Again, in BP, the typical position for the clitic in this context is proclisis to the thematic verb, which is not available in EP.As in the case of the participles and gerunds above, the position of the clitic is completely independent of the kind of elements which precedes the verb and the class of the verb itself.( 20

The case of prepositional clauses
The case of prepositional clauses is the one in which we find less apparent difference between EP and BP.This is due to the fact that in this context we find a great variation in both EP and BP.We see proclisis and enclisis with "para" (to) in both variants (24 and 25), and enclisis and proclisis with "de" (to, of) in EP (26b and 27b.).BP has enclisis and proclisis with "a" (to) (28a.and 29a.).In BP, this is the context which favors enclisis, in particular when the pronoun is the third person accusative clitic "o/a" (it/him/her).In the examples below, we see several cases in which the Portuguese translator leaves the clitic where it is in the original.The changes occur with the preposition "de" (enclisis>proclisis, ex.26b.)(but enclisis is maintained with "gostar de" -to like-, ex.27b.), the preposition "a"9 (proclisis>enclisis, ex.29b.), and when the negation intervenes between the preposition and the verb (enclisis>proclisis, ex.30b.).( 24 Again, we have evidence that the variation between enclisis and proclisis in this context in BP is a matter of grammar competition.Enclisis results from late acquisition in both tensed and infinitival sentences, and it is favoured by the clitic "o/a", which also originates in late acquisition. As for the alternation enclisis/proclisis in EP infinitival clauses, apart from few exceptions, it has not been integrated into the discussion of clitic placement in Portuguese since it poses a challenge for any theory of clitics.Below we shall propose an alternative analysis to the analysis proposed by Duarte, Matos and Gonçalves (2002).We return to this matter in Section IV.

The 3 rd person clitics in BP
Beyond the differences in placement, it is well-known that EP and BP differ in their use of clitics (cf.Galves 2001a, Monteiro 1992, Abaurre and Galves 1996, Kato 1999, Raposo 1999 among others).In particular, third person clitics have been argued to no longer belong to the BP paradigm (cf. Correa 1991, Galves 2002, among others).In written texts, we do find both the accusative and the dative 3rd person clitics "o" (it/him/her) and "lhe" (to it/him/her).But, when we compare the BP and the EP version of O Alquimista, there is evidence that the former tends to avoid the use of these forms, in accordance with what we find in spoken language.

Null objects
In many cases, a 3 rd person clitic in the Portuguese version corresponds to a null argument in the Brazilian counterpart10 .( 31)-( 32) show null accusative arguments, and ( 33)-( 34) show null dative arguments.Note that the EP version of (34) contains a possessive dative clitic, which is completely absent from spoken BP.

Use of strong pronouns
Another strategy to avoid clitics in BP is the use of strong pronouns.We see in the examples below that the dative clitics "lhe" (to it/him/her/you), "te" (to you), "nos" (to us), "a" ( her) correspond in the EP sentences to the pronouns "ele" (it/him) or "você" (you) governed by the prepositions "para" or "a" (to)11 .The preposition in (39a) allows the writer to avoid the use of the tonic pronoun in direct object position.This is very frequent in speech but stigmatized by the written norm.The Portuguese version keeps the accusative clitic, since the verb is transitive.We shall see in the next section that the variation in BP between dative and accusative can be observed also in the use of the clitics.

2 nd person lhe
In BP, the clitic pronoun lhe tends to disappear as a third person pronoun, and is normally used to refer to the second person of the discourse, corresponding to the address form "você" (you).This is illustrated by the sentences below where "lhe" is systematically translated in EP by the second person pronoun "te": In its canonical use, "lhe" differs from "te" in that it is only dative, while "te", like "me" can be either dative or accusative.But in colloquial speech, we can find lhe as the direct object of a transitive verb.In the following examples, it is interesting to note that accusative "lhe" can correspond both to the 2 nd person of the discourse, translated by "te" and the 3 rd person translated by "a/o".( 42 In conclusion, the comparison between the original version of O Alquimista and its adaptation to EP interestingly confirms many grammatical aspects that had been already described for the two variants independently, on the basis of other corpora or the intuition of native speakers.

Enclisis in tensed sentences in EP
Enclisis in tensed sentences in EP has been described and analyzed in many papers since the eighties.It is outside the scope of this paper to fully review the great number of analyses made.We shall only bring here some of the recent proposals, which we shall divide in three great groups, according to the theory of enclisis they put forward 12 .

The syntactic computation only generates proclisis.
From this point of view, enclisis is the result of a special operation, which is sensitive to morpho-phonological properties of the categories involved.This operation can still be syntactic, as in Raposo (2000), or post-syntactic, as in Barbosa (1991Barbosa ( , 1993) ) and Costa & Martins (2003).Raposo (2000) 12 All the analyses presented here can be characterized as a 2 nd generation of analyses of clitic-placement in the sense that they do not assign to Syntax the unique responsibility of the respective order between the verb and the clitic.The purely syntactic analyses were essentially based on Kayne's (1991) proposal that adjunction to a head is invariably to its left.From this point of view, if the clitic raises to the category hosting the verb, the result will be invariably proclisis, while enclisis will not be derived unless the verb is attracted to a higher position.Alternatively, the clitic can raise first and the verb adjoins to its left, yielding enclisis.Both proposals were made to derive the characteristic enclisis of EP.
Raposo, based on Uriagereka (1995), assumes the existence of a category higher than Infl and lower than C, called F. According to his analysis, the crucial property of F in EP is that it is enclitic; as a consequence, when nothing lies on its left, the verb must raise to Spec/F, to provide F with a phonological host.According to this analysis, "Verb movement to Spec-F is a Last Resort operation that applies to satisfy PF properties of the functional category" (Op.cit.p. 280).It is the movement of the verb to a position higher than the one occupied by the clitic that creates the order V-CL 13 .
In order for this analysis to account for the distribution of enclisis in EP, as described in Section I, it is necessary to claim that when some phrase is on the left of the verb, and enclisis is required, this is due to the fact that this phrase is not in a position which counts as a possible host for F. This is straightforward for the fronted NPs, PPs and adverbs, but it needs a special claim for pre-verbal subjects.Following several papers by Pilar Barbosa, Raposo assumes that pre-verbal subjects occupy a peripheral position in EP, namely spec/Top.Barbosa (1991Barbosa ( , 1993) ) According to Barbosa, there is a phonological restriction active in EP which blocks non-stressed elements at the left edge of the Intonational Phrase.In the line of works like Salvi (1990), andBenincà (1995), she thus attributes the obligatory enclisis in certain contexts in EP from the application of the so-called Tobler-Mussafia Law, which bans unstressed words at the absolute beginning of sentences.Barbosa (1991Barbosa ( ,1993) ) assigns to the morphological component the task of inverting the order between the verb and the clitic, in the framework of Distributed Morphology.
Again, the question of why enclisis is obligatory even when a subject precedes the verb needs special attention.Barbosa, following the analysis of the position of subjects in null subject languages, argues that pre-verbal subjects in EP do not occupy a position internal to the clause, but are dislocated, like topics.According to this line of argumentation, the A-position for subjects in NSLs is the post-verbal position and pre-verbal subjects occupy a A' -position.Costa and Martins (2003) Costa and Martins propose that the clause contains a functional head called Σ, similar to Raposo's F in that it is intermediary between Infl and C. According to them, this functional category is morphologically strong in EP.If no phrase raises to its specifier, it must be licensed by morphological merger or V-raising.Since, according to the authors, there is no evidence of V-raising to Σ in EP, the only possibility is merger, which is a morphological operation that puts together Σ and the verb.But since this operation requires adjacency between the elements to be merged, pre-verbal clitics block it.Thus, inversion must apply in order for the merger operation to be possible.As in Raposo (2000), this operation is characterized as a Last Resort process.
All these analyses share crucial properties, which can be summarized as follows: (i) the property that provokes enclisis is not syntactic but morphological or phonological; (ii) the order V-CL is created by a late syntactic or a postsyntactic process that rearranges the order produced by the syntactic component in such a way that the phonological or morphological property involved is satisfied; (iii) the special feature that is responsible for this rearrangement, and, ultimately, accounts for the difference between EP and the other Romance languages, is not on the clitic but on the category that initiates the clause (F or Σ) or on the left boundary of the Intonational Phrase; (iv) in EP, pre-verbal subjects are peripheral.They occupy the same position as dislocated topics: they are either in Spec/Top or adjoined to the clause.

The syntactic component generates enclisis
An alternative conception of enclisis is proposed by Duarte and Mattos (2000), Frota and Vigário (1996) and Duarte, Matos and Gonçalves (2002).From their point of view, the relationship between syntax and phonology in the derivation of clitic-placement is somehow the reverse of what is proposed in the analyses presented above.In effect, the basic syntactic derivation yields enclisis.Proclisis only occurs in presence of a heavy functional category.Enclisis is thus characterized as the basic, non-marked, pattern of EP clitic-placement.
Enclisis is derived by the movement of the clitic to a functional position and the left adjunction of the verb to the left of the category that hosts the clitic 14 .As for proclisis, it is conditioned by phonological phrasing when heavy functional words c-command and precede the clitic.In this case, the clitic is attracted to a higher position, yielding the order clitic-verb.It is important to note that this attraction is dependent on the phonological phrasing.If the heavy functional head is outside the Intonational Phrase which contains the clitic, it ceases to work as a proclisis trigger.
The above claim allows Frota and Vigário (1996) to explain the variation between enclisis and proclisis when some phrase occurs between the complementizer and the verb 15 .If the functional category which yields proclisis is located inside the Intonational Phrase which contains the clitic, proclisis obtains.If it is outside, enclisis obtains, instead, as illustrated in ( 44)-( 45): (44) a. Acho [Int P que ao João, 14 This category is Spec/AgrO in Duarte and Matos (2000), and Asp or T, according to the argumental or non-argumental status of the clitic in Duarte, Matos and Gonçalves (2002).
The claim that enclisis derives from the movement of the verb to a position already occupied by the clitic is also found in Madeira (1992), Manzini (1994), andRouveret (1992). 15There is some controversy with respect to the restrictions on the occurrence of enclisis in this context.For Frota and Vigário (1996), the pre-verbal phrase must be heavy (cf.Section IV).For Raposo (1994), there is no such restriction.The answer the authors give is that, in EP, the clitics have entered into a process of reanalysis that confers to them a status of quasi-suffixes.This is the reason why the presence of the clitic on the right of the functional category does not prevent the checking of the inflectional features of the verb.The locus of the difference is therefore no more a property of the clause, as in the analyses presented in 1.2, but a property of the clitics themselves 16 .
1.3 Clitic-placement is not derived at all by syntactic processes.Galves and Sândalo (2004), based on Anderson (2000) and Legendre (2000), assume a model of clitic placement in which the position of clitics in the sentence is not defined by syntactic rules but by morphological rules.In this model, clitics are considered as phrasal affixes and, like word affixes, are subject to alignment constraints.The interaction of these constraints is put in optimalistic terms: they are ranked in such a way that the satisfaction of the highest one leads to violations of the lowest one.Galves and Sândalo argue that the complex pattern of clitic 16 An intermediate position is held by Pilar Barbosa in her most recent works (Barbosa, 1996(Barbosa, , 2000)).She adopts Duarte and Mattos's analysis of enclisis and proposes that when the movement of the verb to Infl (Tense) creates a configuration in which the clitic is at the left edge of the Intonational Phrase, the alternate derivation, with the clitic in a lower position (Asp or AgrO) and the verb adjoined to it, is chosen.placement in EP simply derives from the interaction of the two following constraints: Edgemost (L, I-bar): a clitic is aligned with the left edge of I-bar Non-initial (1 X-bar): a clitic cannot be the first element of the first X-bar of the clause.
Non-initial being higher ranked than Edgemost, the model predicts that we find enclisis every time I-bar is the first X-bar of the clause.It is easy to see that the EP paradigm fits exactly within this prediction, since all the cases of obligatory enclisis are cases in which no functional category is projected above IP (matrix clauses with no operators), and all the cases of obligatory proclisis are the cases in which there is some functional category projected above IP (CP in the case of interrogative and subordinate clauses, some intermediate category like ΣP in the case of focalized or quantified phrase preceding the verb, NegP in negative clauses) 17 .
Note, however, that this analysis also articulates a syntactic component and a prosodic component.The syntactic component has to do with the reference to I-bar as the locus of the realization of the clitic, and to X-bar as the domain in which the clitic cannot be initial.The prosodic component concerns the prohibition of the clitic at the initial position of some domain.However this approach differs from the proposals in 1.1 in that the initial position is defined in such a way that it does not require that the verb be in absolute first position in the clause.Namely, if the specifier of IP is filled by a subject, and no category higher than IP is projected, Ibar is still the first X-bar of the clause, and the clitic cannot be initial.This is an important difference with the other approaches that derive enclisis from a prohibition of the clitic in first position, since it does not force one to attribute a peripheral position to the subject.Another important difference regarding the other analyses presented above is that it does not refer to the Intonational Phrase.We come back to both points in Section IV.

The contrast BP/EP
In this section, we shall briefly review the few studies that explicitly raise the question of the difference between BP and EP clitic-placement.They all share the basic idea that what crucially differs in the two variants is the nature of the pronoun itself.They do, however, differ in the way they formulate this idea, and how they articulate the special property of the clitics with the syntax of the language.According to Duarte, Matos and Gonçalves (2002), the reason why BP does not display enclisis like EP is that clitics in the former are no more reanalyzed as quasi suffixes.Interestingly, the loss of this reanalysis is correlated with the general weakening of inflectional morphology in BP, which opens the way to correlating clitic-placement with other syntactic phenomena.However, underlying this analysis is the claim that BP evolved from EP, which is problematic.We'll come back to this question in Section IV.Abaurre and Galves (1996) and Galves (2002) also argue that BP clitic pronouns are somehow stronger than EP clitic pronouns.According to these authors, while EP clitics behave like heads, BP clitics behave like phrases 18 .Additionally, they claim that in BP clitics do not adjoin to Infl, but to V, because of the lack of Agr in Infl.This again draws a correlation between the behavior of clitics and other syntactic phenomena typical of BP.Costa and Martins (2003) also propose that the difference between EP and BP clitic-placement is that clitics attach to Infl in the former and to V in the latter.According to them, this is why there is no enclisis in BP although they argue that BP shares with EP the property of having a morphologically strong F.

III. A new comparative analysis for EP and BP clitic-placement
In this paper, we shall maintain some of the basic ideas of the previous analyses, but in a slightly different framework.Our proposal is that the different behavior of clitics in EP and BP derives from the interaction of two different properties: a.The category to which clitics syntactically attach: EP clitics are Infl-clitics and BP clitics are V-clitics.b.The phonological requirement that forces clitics to be in a noninitial position with respect to some boundary: This requirement is active in EP but not in BP.
The first part of the analysis will allow us to keep the correlation between the placement of clitics in BP with the loss of the third person accusative clitic.We shall adopt Galves's (2002) idea that accusative marking, which is structural, requires an Agreement node.Since Infl no more contains Agr in BP, the only alternative left is the inherent case marking/checking of the clitic by the verb.
The second part of the analysis is in great part drawn from Galves and Sândalo (2004) who rely on Anderson's (2000) and Legendre's (2000) account of clitic placement.For these authors, clitic placement is not governed by syntax, but by morphology.They consider clitics as phrasal affixes which are subject to word formation rules like any other affixes.This line of thinking is very much similar to the one advocated in Distributed Morphology (cf.for instance Harley and Noyer, 1999).In this framework, the syntactic computation does not manipulate words but roots and features.Words are formed post-syntactically by the general principles of morphology.The difference between the two approaches is that the former works with representations in the framework of Optimality Theory, and the latter is derivational.What mainly interests us here, nevertheless, is that they share the conception of Morphology as a post-syntactic component.This conception has strong consequences for the analysis of clitic-placement, as already shown by Anderson (2000) and Legendre (2000).
In the remainder of this article, we shall argue that the division of work between Syntax and Morphology allows us to better understand the idiosyncrasies of clitic-placement in both EP and BP.Instead of studying each variety independently, we shall adopt a comparative approach on both levels of analysis.
1.The syntactic level, or where to attach your clitics It has been argued that in Romance Languages clitics adjoin to Infl.Leaving aside the question of whether the clitic attaches onto the right or left of the verb, EP displays the basic properties which have been taken as evidence for this claim, mainly the fact that in tensed sentences, clitics always attach to the inflected auxiliary, and never to the gerundive or participial form of the thematic verb.Additionally, EP behaves like other Romance null subject languages in allowing climbing of the clitic up to the inflected verb, as in ( 46) in constructions with some classes of verbs19 .The attachment of the clitic to the infinitival form, as in (46b) which varies with (46a), cannot be taken as an argument that the clitic attaches to V, since, following Stowell (1982) it has long been assumed that infinitival clauses can also be projections of Infl.In both sentences, the clitic is adjoined to Infl, tensed Infl in (46a) and Infinitival Infl in (46b) The difference between (46a) and (46b) simply derives from the restructuring of the clauses, as argued by Rizzi (1982) and others.In (46a), there is only one Infl, the tensed one.These arguments do not apply to the core case of clitic placement in BP.In effect, as we saw above, there is evidence that clitics do not raise to Infl.In any circumstances, they are attached to V. As we showed in Section I, the natural way to utter the sentences above is neither (46a) nor (46b) but ( 47): (47) Não quero te ver.Not want CL2sg see 'I don´t want to see you.' Furthermore, the fact that the clitic is proclitic to the infinitival verb, and not enclitic to the tensed verb is supported by sentences like (48), where an adverb occurs between the auxiliary and the clitic. 20: (48) não posso no momento lhe dar.not can at-the moment Cl3sg give 'I can't at the moment give it to you.' (48) could be taken as evidence that clitic placement in infinitival sentences is like in French, i.e., with proclisis to the infinitival verb.But as we have seen above, BP differs from French in having proclisis also to participles and gerunds.( 49) clearly shows that the clitic is proclitic to the thematic verb and not enclitic to the auxiliary.The reason why we find (50b) instead of (50a) in BP is not completely clear.It could be that the EP pattern is acquired altogether with the 3 rd person pronoun.But alternatively, we can think that the different pattern of placement for the accusative clitic has to do with its morphological features, and the fact that it needs an inflectional head to be licensed.This alternative hypothesis is interesting because it is coherent with the idea that there exists a relationship between the peculiarities of the pronominal paradigm in BP and the category to which clitics attach.As we saw above, the differences between EP and BP concerning the use of clitics cannot be reduced to differences in clitic-placement.It also concerns the pronominal paradigm itself.As we showed in the first section from O Alquimista, there is a strong tendency to avoid the third person clitic, and to replace it with null objects, strong pronouns, repetition of a noun phrase, and even with the dative pronoun "lhe".Taken together, the loss of the third person accusative clitic "o/a" and the use of the third person dative clitic "lhe" for direct objects allow us to understand what happened with the dative/accusative first and second person clitics "me" and "te" and the dative/accusative clitic "se": they have lost their accusative features, and become pure dative forms.This change is not morphologically visible, since the the five greatest towns of Brazil.
form has always been the same, but it is coherent with what happens at the third person where different case-marking correspond to different forms This analysis also gives us the key of the peculiar placement of the clitics in BP.As already suggested by Galves (2002), the relationship between the dative marking of the clitics in BP and their invariable attachment to the verb which assigns them their θ-role lies in the notion of inherent case.The clitics are inseparable from their thematic verb because it is the source of their casual licensing.This is compatible with the dative forms but excludes the accusative ones, since Accusative is the structural case by excellence.We shall not go further into why BP clitics are no more Infl clitics.Many researchers have suggested that this is due to the weakening of Infl in this language.We shall adopt this view here and suggest without discussion that Infl in BP lacks the property that structurally licenses clitics, forcing their licensing in association with their θ-marking.

The morphological level, or how to attach your clitics
As we have seen in Section II, most of the analyses proposed to account for clitic placement have a syntactic basis.Syntactic devices derive one of the two orders, proclisis or enclisis, as the basic order.Following Kayne's (1991) theory of clitics, the adjoined head is located on the left of its host.This derives proclisis if the clitic raises to the category which already hosts the verb, and enclisis if the verb raises to a category which is headed by the clitic.
Here, in the spirit of Galves and Sândalo (2004), we propose a radically different conception of the way clitics surface on their verbal hosts.We assume that clitics are not syntactically autonomous elements, but the morphological spell out of functional features, which, therefore, are subject to word formation rules, as any other affix.The difference is only that clitics are instances of phrasal morphology This view is compatible with both the Optimality Theory framework, as in Galves and Sândalo (2004) and the Distributed Morphology framework.In the latter, there is a derivational component, in which the functional features corresponding to the clitics move to some category.According to the previous section, this category is Infl in EP and V in BP.The important point is that this movement has no effect on the pre-verbal or post-verbal position of the clitic.It is at the Morphological level that this position will be defined.
As mentioned above, the two constraints considered by Galves and Sândalo (2004) are: (a) Edgemost: a clitic is aligned with the left edge of I-bar (b) Noninitial: a clitic cannot be the first element of the first X-bar of the clause.
If we assume a derivational component, we can simplify the formulation of Edgemost.First, we do not need to mention I-bar, since it is defined on the grounds of syntactic properties.Second, we can assume that the realization of the functional features at the left edge of the I-bar boundary is the default linearization of features adjoined to Infl.Edgemost is therefore given for free by the model.The crucial point is Non-Initial.If this constraint is active in the morphology of a given language, it forces the realization of the clitic as a suffix on the verb.
Note that given the formulation of Non-Initial, enclisis continues to be dependent on the structure of the clause, since it depends on the non-embedding of I-bar in the clause.However, it does not derive from syntactic processes.Another important difference with all the analyses reviewed is that enclisis is not dependent on V1.Given its definition above, Non-Initial is not affected by the presence of a subject in the specifier of Infl.This means that we do not need to put the preverbal subject outside the boundaries of IP to derive enclisis.We shall come back to this important point below.
In the next section, we shall discuss the advantages of this analysis, and its empirical superiority over the others.

IV. Discussion
In this section, we shall argue that the analysis we have proposed is preferable to the others for the following reasons: (i) it avoids some empirical and theoretical problems encountered by the other proposals; (ii) it captures new correlations that are not grasped by the others; (iii) it does not discard the results obtained by the others.

The position of pre-verbal subjects in EP
The first important result we obtain has to do with the position of pre-verbal subjects in EP.We obtain the distribution of enclisis and proclisis in tensed sentences without assigning the pre-verbal subject an external position, contrary to all the analyses reviewed, which depend on the externality of the subject.
The claim that subjects occupy a topic position in EP is problematic from both a synchronic and a diachronic point of view.As for synchrony, we refer the interested reader to Costa (1998Costa ( , 2000) ) who brings evidence that subjects and dislocated topics behave differently in EP.As for diachrony, Galves, Britto and Paixão de Sousa (2005) show that enclisis in Classical Portuguese (henceforth ClP) was indeed a V1 phenomenon 21 .16 th and 17 th texts instantiate the following pattern: a) Enclisis is obligatory when the verb is in absolute first position b) Proclisis is obligatory in the same contexts as in EP c) In the other cases, there is variation between enclisis and proclisis, the latter being highly dominant Case c), which crucially differentiates ClP and EP, concerns the case in which enclisis is obligatory nowadays: matrix affirmative clauses in which the verb is preceded by either a referential subject, an adverb of a certain kind or a preposed phrase.In this case, we find proclisis in 90 to 100% of the cases22 .This pattern begins to change in the texts of the authors born in the first half of the 18 th century, and the change to the modern pattern almost gets to completion in authors born in the middle of the 19 th century.The important point is that the change in clitic-placement is accompanied by a change in subject position, as shown by Paixão de Sousa ( 2004).Paixão de Sousa argues that in ClP, the Aposition for subjects was the post-verbal position.From the 18 th century on, all the syntactic phenomena associated with this grammar change together.The rate of post-verbal subjects goes from 20% in the 16 th /17 th centuries to 10% in the 18 th /19 th centuries.At the same time, we observe that the frequency of pre-verbal subjects with enclisis increases much faster than other pre-verbal phrases with enclisis.Finally, Galves and Paixão de Sousa (2003) observe that in V3 sentences, the sequence XSV-cl, which was absent from the texts in the classical period, becomes frequent in the texts of the authors born in the 18 th and 19 th centuries.All these observations taken together strongly suggest that in the new grammar, there is a position for pre-verbal subjects, distinct from the position of topics and fronted constituents.This conclusion is consistent with our analysis but not with the ones that need to assign a topic position to pre-verbal subjects in EP.

The evolution of BP clitic-placement
As we saw in Section II, Duarte, Matos and Gonçalves (2002) argue that enclisis in EP is due to the fact that clitics have entered into a process of reanalysis that confers to them a status of quasi-suffixes.Accordingly, they derive proclisis in BP from the loss of this process of reanalysis, which they associate with the general weakening of morphology in BP.
The problem with this analysis is that it implies that BP evolved from EP.This is a very dubious claim.In effect, as we mentioned above, the grammatical change that led to EP appears in texts of authors born in the first half of the 18 th century.This change would have to be taken to Brazil by the Portuguese emigrants, and followed by another change, in the opposite direction.Since the phenomena typical of the modern Brazilian syntax begin to appear at the end of the 18 th century (cf.Cyrino 1994), this picture seems difficult to sustain.It seems much more realistic to hypothesize that both EP and BP independently evolved from ClP.The generalized enclisis is a EP innovation.Therefore, there cannot be a loss of enclisis associated with the weakening of morphology in BP.The alternative explanation we propose here is that what changed in BP, in connection with the weakening of Infl, is that clitics no more adjoin to Infl, but to V.This change is not directly related to whether clitics are pre or post-verbal but it creates a new position for clitics, which exists neither in ClP nor in EP.This position is proclisis to non-finite forms as illustrated in (15a)-(24a) above.Duarte et al.'s analysis says nothing about this Brazilian innovation.

Theoretical problems
Finally, we shall briefly comment on some theoretical problems encountered by the analyses presented in Section II.
First, in many analyses we find stipulations about the categories involved which have no independent motivation.For instance, Raposo (2000) derive enclisis from the claim that the functional category F is enclitic.Since F has no specific lexical realization in EP, we cannot say that this is a phonological property of the word that realizes F. It must be the property of the abstract category itself.Besides the issue of whether abstract categories have phonological properties independent of their morphological realization, there is no way to find independent evidence for this claim, and the analysis based on it becomes rather circular.
Another theoretical problem arises in connection with the effects of the special properties of the categories F or Σ.Since they have special requirements, they force movement of the verb to their specifier (Raposo, 2000) or merger with the verb (Costa and Martins, 2003).In both cases, these operations are defined as "last resort" operations.This is a problematic extension of the restrictive notion of Last Resort proposed by Chomsky in the Minimalist program.In particular, it is not limited by Greed (in Raposo's analysis the verb moves to solve the problem of F), and it is not involved in checking relations, since, crucially, there is no feature checking at stake in the derivation of enclisis.The use of the notion of strong category in Costa and Martins' analysis suffers from the same problem.This strength is not defined in terms of strength of features, which must be checked, but as a morphological property, which can be satisfied by a morphological operation like merger.
The analysis we propose here definitely assumes that the pre-verbal or postverbal position of the clitics is a matter of Morphology.This means that it derives from conditions on the well-formedness of words, and not from the properties of functional categories, which is a matter of syntax.If clitics are affixes, they are subject to alignment constraints, like word-affixes.The constraint Non-Initial is such a condition.Its articulation with syntax lies in the definition of the relevant domain in which it applies, and we shall argue in the next section that this domain has changed during the history of Portuguese, and may be changing again.
2. New correlations that do not derive from the other analyses

The interface prosody/syntax and the history of Portuguese
Our analysis also differs from many others (see the references above) in that we do not derive the presence of enclisis in EP and its absence in BP from the properties of the clitic itself, but from the effect of the morphological constraint Non-Initial, which is active in EP but not in BP.
Non-Initial is clearly linked to prosody, since it defines a position in which a non-stressed element cannot occur.In this sense, it is similar to the Tobler-Mussafia Law.In fact, the Tobler-Mussafia law is a Non-Initial Constraint that defines its domain as the whole sentence (i-e the first XP).We have given evidence above that this constraint was active in ClP.But we have argued against Barbosa's claim that it is also active in EP.We have adopted Galves and Sândalo's claim that the domain of Non-Initial in EP is the first X-bar of the clause.We conclude that the change in clitic-placement from ClP to Modern European Portuguese is correlated with a change in the domain of Non-Initial.This is a satisfactory conclusion since we know that an important change occurred in the Portuguese prosody after the 16 th century, which affected the pronunciation of the pre-tonic vowels 23 .It is beyond the limits of this paper to go further into this matter.
From the same point of view, we can say that from ClP to BP, Non-Initial was lost 24 .But this difference does not suffice to explain all the differences between BP and EP.It explains proclisis in absolute first position, but it does not explain the attachment of the clitic to the non-finite verbal form in verbal clusters.The latter innovation is not due to the prosodic evolution of BP, but to the weakening of the agreement system, which changed the properties of Infl in this language.This is why the change in clitic-placement from ClP to BP is much more dramatic than from ClP to EP.
In the next section, we shall see how our analysis can account for the new tendencies of clitic-placement in EP.

The recent evolution of EP
Several studies (cf.Duarte and Faria 1994, among others) show from acquisition data and pieces of written and oral colloquial speech, that there is a tendency in European Portuguese to extend the contexts of enclisis to the detriment of proclisis.Duarte et al.'s approach does not explain this fact straightforwardly.In effect, according to them, the basic property of the syntax of Standard EP is the fact that clitics are quasi-suffix.This explains the possibility of enclisis in tensed sentences.However, the increase of the use of enclisis cannot be explained by this same property, which already concerns all the occurrences of clitics, but only by a blocking of proclisis in certain contexts.In the framework proposed, the only reason for the increase of this blocking would be either a tendency to put the strong functional words that yield proclisis (the "proclisis triggers") outside of the Intonational Phrase that contains the clitic, or a weakening of the strong functional words.This would be a new fact, which has nothing to do with the basic property that allows, and forces, enclisis in certain contexts, that is, the nature of quasiaffixes of the clitic pronouns in EP.
In our approach, it is possible to link the current tendency of the "advanced" language with the basic property that is responsible for enclisis.A simple explanation would be that what is changing, again, is the domain of Non-Initial.If the domain for the constraint is any X-bar (i-e, if we drop 1 rst from the definition) we shall obtain a generalized enclisis.The difference between the distribution of enclisis and proclisis in Standard EP and in the speech of children as well as in "advanced" colloquial speech is straightforwardly explained if NonInitial no longer makes reference to the first X-bar.We therefore expect to find a general pattern of enclisis.If this is true, we are currently observing the end of a process of generalization of enclisis, which corresponds to successive modifications of the domain of Non-Initial that have the effect of extending the contexts in which enclisis is produced.

Other advantages of the analysis
The analysis proposed also allows us to account for two other aspects of clitic placement in EP: the weight effects described by Frota andVigário (1996, 2002) in finite subordinate clauses, and the complicated pattern of clitic-placement in infinitival clauses.These two contexts have in common the fact that they allow variation between proclisis and enclisis25 .

Weight effects in EP
One of the few contexts of variation between enclisis and proclisis in EP is illustrated in (51) below (cf.Frota and Vigário 1996).Also, according to Frota andVigário (1996, 2000) Frota and Vigário claim that enclisis is possible in a subordinate clause when either a pause, or a heavy constituent, or a parenthetical, or a topic phrase, intervenes between the element which triggers proclisis and the clitic.
An empirical study is still to be done on this variation.As already mentioned, the exact characterization of the contexts in which enclisis may occur in embedded tensed sentences is not very clear.Raposo (1994) for instance claims that in sentences like (51b) the clitic can be post-verbal.
What is at stake is the definition of "heavy", and its relationship with the notion of Intonational Phrase.The use of this notion by the authors reviewed must be rather abstract, since it is not the case that enclisis is always associated with an intonational boundary at the immediate left of the verb.This is a problem in particular for Barbosa's approach since subjects in EP can be in the same Intonational Phrase as the verb, and generally are.This is also a problem for structures like ( 44) and ( 45) above, since enclisis is not dependent on an intonational break between "que" and "o João" in the former and between "que" and "embora" in the latter.Finally, if Raposo is right in claiming that enclisis is possible in sentences like (51b), it is very unlikely that the embedded subject in this sentence has an intonational contour of its own, or that enclisis is dependent on an intonational break between the complementizer and the subject.
Our analysis straightforwardly accounts for the facts if we understand "1 st Xbar in the clause" in the definition of Non-Initial as "1 st X-bar in CP".All the cases of enclisis in embedded contexts would be therefore cases of CP recursion.We thus replace the representations in ( 44) and ( 45) by the representations in (52a)-( 52b) and (53a)-(53b).Sentence (53b) with enclisis can be represented as ( 54 As already mentioned above, we lack a description of all the discursive and prosodic aspects of this variation, which could empirically support this analysis.As far as subjects are concerned, we would expect this construction to be somehow marked, and associated to special discourse functions like contrastiveness or emphasis.This is a common point with the approach based on the Intonational Phrase.But the advantage of our analysis is that we do not predict that there are obligatorily intonational breaks or special intonational contours associated with enclisis, and that such breaks or contours are impossible when the clitic is preverbal. It must be noted additionally that all the examples of enclisis in embedded contexts involve bridge verbs like "achar" (to think), or "dizer" (to say).This is predicted by our analysis, since the recursion of CP is a phenomenon limited to this kind of verbs26 .

Clitic-placement in infinitives
Finally, our analysis allows us to suggest an explanation for the other context in which we find variation in clitic placement in EP: infinitival clauses introduced by a preposition.
First, it is important to note that in this case an explanation based on prosodic considerations is not easy to defend, since there are clearly no prosodic correlates associated with the variation described in ( 24)-( 30).An analysis based on this kind of explanation for tensed sentences cannot therefore be used to explain the variation in infinitival clauses.
(55) a.  55) -(56) show that the variation between enclisis and proclisis observed in noninflected infinitival clauses is not found in inflected infinitival constructions, in which enclisis is impossible, except with the preposition "a", which invariably requires enclisis.According to Duarte et al., all the prepositions except "a" are heavy functional words, which are proclisis triggers.(55a) is therefore unexpected.They solve this problem by assigning different structures to (55a).and (55b).According to them, the former is an "elliptical structure" that contains a null verb with a modal or temporal value.No such verb is present in the latter.Because of the null verb in (57a), the clitic is no more in the scope of the preposition, and enclisis is derived.According to the same analysis, enclisis is blocked with inflected infinitival sentences because the modal or temporal null verbs do not select inflected infinitive.Additionally, this analysis straightforwardly derives the absence of proclisis with "a" from the claim that it is not a strong functional category.
We should now capitalize on how our analysis can account for the double contrast observed: inflected infinitive vs. non inflected infinitive and "a" versus the other prepositions.
First, we shall recall that enclisis is obligatory in infinitival sentences complement of verbs.This is expected in our framework if all infinitival clauses are CPs (cf.Boskovic, 1997), in which the first X-bar contains the verb.On this basis, we can derive the variation between enclisis and proclisis with prepositions from their ambiguous status.They can be either the category that selects CP, or the complementizer itself.Assuming Non-Initial, we obtain enclisis in the first case and proclisis in the second case, as represented below: (58) a. Saímos [PP para [CP vê-lo b. Saímos [CP para o ver] As for inflected infinitival constructions, we must assume that they only correspond to the structure.This is coherent with Raposo's (1987) analysis, since only in C is the preposition able to govern the agreement on the verb 27 .Outside CP, the preposition is not able to license inflected infinitive.We therefore straightforwardly derive the obligatoriness of proclisis in this context.
As for "a", our analysis leads us to propose that it is never in COMP.This claim is supported by another important difference between "a" and the other prepositions.In clauses introduced by "a", we never find a pre-verbal subject.Since we do find inflected infinitive clauses introduced by "a", we can rely again on Raposos's (1987) analysis to propose that in these constructions the verb raises to COMP in order to have its agreement features governed by the preposition.This explains why only the post-verbal position is available for a subject.
We conclude that our analysis of the distribution of enclisis and proclisis in tensed sentences based on Non-Initial is also able to account for the distribution of enclisis and proclisis in infinitival sentences.

V. Concluding remarks
The main conclusions of this article are the following: -In BP, clitics are not licensed by Infl but by the verb that assigns their thematic roles to them.This explains the fact that they appear attached on the thematic verb even if it is a participle.Proclisis is the normal pattern since Non-Initial is not active in this language.Enclisis continues to appear more or less marginally according to the discourse and syntactic context.This is the effect of competition between the grammar people acquire in natural acquisition, and pieces of grammars learned at school or in contact with formal written and spoken language.The third person accusative clitic "o/a" is also reminiscent of old grammars.Its obsolescence is due to the fact that in this system the clitics that survive are the ones that have dative features compatible with the inherent case marking by the verb.
-In EP, the distribution of enclisis and proclisis both in tensed and infinitival sentences derives from the application of Non-Initial, which bans clitics at the beginning of the first X-bar of CP.This analysis derives both the cases of categorical enclisis and the cases of variation, which, contrarily to BP, are produced by the grammar itself.
Portuguese; and Galves, Britto and Paixão de Sousa 2005, for Classical Portuguese) 4 .(1)a. Me chamo Fátima -disse a moça... (BP) CL1sg call Fatima said the girl b.Chamo-me Fátima -disse a moça... (EP) Call CL1sg Fatima said the girl 'My name is Fátima said the maid...' (2) a. Não procurem entrar na vida do oásis, concluiu, not seek enter in-the life of-the oasis concluded se afastando.(BP) CL3sg going away b.Não procurem entrar na vida do oásis, concluiu, not seek enter in-the life of-the oasis concluded afastando-se.(EP) going away CL3sg 'Do not seek to come into the life of the oasis, he concludes, going away.' (3) a. Tem sonhos, se emociona, e está apaixonado...(BP) Has dreams CL3sggets emotional and is in love b.Tem sonhos, emociona-se, e está apaixonado...(EP) Has dreams gets emotional CL3sg and is in love 'He has dreams, gets emotional and is in love …' CL3pl in-the bag 'And afterwards he took the two stones from the earth and put them back in the bag.' accusative (31) a. O rapaz acordou --uma a uma (BP) (as ovelhas) The boy woke up --one by one (the sheep) b.O rapaz acordou-as uma a uma The boy woke up CL3pl one by one (the sheep) (EP) 'The boy woke them up one by one' in-the pocket (the stones) 'He wrapped them up in a handkerchief and put them back in his pocket.'dative 3rd person (33) a. Um certo amigo tinha--indicado a loja... (BP) A certain friend had referred the store b.Um certo amigo tinha-lhe indicado a loja... (PE) A certain friend had CL3sg referred the store 'A certain friend had referred him to the store...' (34) a. a velha segurava suas mãos.Repetition of the referential noun We also find cases in which the EP clitic corresponds to a repetition of an argument in two successive sentences or in a coordinated clause.(35) a. Então começaram a bater no rapaz.they began to beat the boy.They beat on the boy/him…'

(
36) a. Vamos, pergunta a ela! (BP) Come on ask to her b.Vamos, pergunta-lhe!(EP) Come on ask CL3sg 'Come on, ask her!' (37) a. Para mostrar a você uma simples lei do mundo (PB) To show to you a simple law of-the world b.Para te mostrar uma simples lei do mundo (PE) To CL2sg show a simple law of-the world 'In order to show you a simple rule of the world' (38) a. ...mas na verdade estão ensinando a você...I look at your sand I contemplate her also' I want to ask you a favor.' 't want to see you' always seeing you' It is worth emphasizing that this pattern concerns the first and second person clitics.Third person accusative clitics o/a, which are acquired later (cf.Correa 1991) normally follows the Portuguese pattern, i.e., attaches to Infl.(50b) is therefore more natural than (50a) (57) a. Saímos [PP para [TP [V e] vê-lo b.Saímos [PP para [TP o ver] Additionally, when some pre-verbal element prevents enclisis, we see that the clitic changes its position in EP, while it remains at the same place in BP.

vão acostumar-se... (
They told me that, although it was difficult, they gave the grant to him.'Finally, a question remains to be answered: from this point of view, what prevents enclisis from being derived in the other Romance languages?In other words, what is the special property of EP with respect to these other languages?