Some Observations on the Pronominal System of Portuguese

Abstract I present a minimalist analysis of the pronominal system of Portuguese compatible with the claim that pronouns are underlying determiners (Postal 1969, Raposo 1973), including the nominative and dative forms. Whereas the form ois a «pure» determiner (a [+minimal, –maximal] projection), the nominative and dative pronouns will be analyzed as [+minimal, +maximal] projections with a nominal feature already «incorporated» in their grammatical feature matrix. I also argue that, under Postal’s theory, the ungrammaticality of sentences with an accusative clitic pronoun in Colloquial Brazilian Portuguese cannot be due to a loss of that lexical item, since the item is well and alive as an article (including as head of accusative DPs). Rather, the difference will be tracked down to the fact that the form ele lost its inherent (nominative) Case (which it maintains in European Portuguese), which puts it in direct competition with the form o. Considerations of economy, in the sense of Chomsky (1995), play an important role in this account. Key words: determiners, pronouns, Portuguese, syntax. Resum En aquest article presento una anàlisi minimista del sistema pronominal del portuguès que és compatible amb la proposta que els pronoms són determinants (Postal 1969, Raposo 1973), fins i tot les formes del nominatiu i les del datiu. Proposo que la forma oés un autèntic determinant (una projecció [+mínima, –màxima]), mentre que els pronoms nominatius i datius, pel contrari, s’han d’analitzar com projeccions [+mínimes, +màximes] que «incorporen» un tret nominal en la seva matriu de trets gramaticals. També considero que, d’acord amb l’anàlisi de Postal, la agramaticalitat de les oracions amb pronom clític acusatiu en el portuguès brasiler col·loquial no es pot atribuir a la pèrdua de l’element lèxic, ja que aquest element hi és present com a article * I would like to thank audiences at USC, University of Utrecht (Going Romance 95), Universidade Federal de Alagoas, PUC-Porto Alegre, Universidade Estadual de Campinas, Instituto Universitario Ortega y Gasset and Universidade Nova de Lisboa, where I presented parts of this work. I would like to thank in particular Charlotte Galves, Maximiliano Guimarães, Mary Kato, Sérgio Menuzzi, Jairo Nunes, Conceição Ramos and Juan Uriagereka for discussion and/or comments. All errors are mine. CatWPL 6, 1998 59-93 Some Observations on the Pronominal System of Portuguese* Eduardo P. Raposo University of California at Santa Barbara. Department of Spanish and Portuguese Santa Barbara, CA 93106. USA [email protected] Received: January 8th 1998 Accepted: December 12th 1998 CatWPL 6 059-093 21/10/99 12:47 Página 59 definit (fins i tot com a nucli d’un DP acusatiu). La diferència respecte al portuguès europeu es troba en el fet que la forma eleha perdut la seva marca inherent de Cas nominatiu, que en el portuguès europeu es manté, i això l’ha fet entrar en confrontació directa amb la forma o. Les qüestions que giren al voltant de la noció d’economia, en el sentit de Chomsky (1995), juguen un paper important en aquesta explicació. Paraules clau:determinants, pronoms, portuguès, sintaxi. 1. Introduction In 1973, I wrote a paper proposing that there was one and only one form o ‘the/him/it’ (Spanish lo) in the lexicon of Portuguese, and that different derivational histories were responsible for its use as an article (1a), as a «demonstrative» (1b), or as a clitic (1c) (see Raposo 1973):1 (1)a. já li [o livro] already I.read thebook] b.já li [o últimodo Blakee Mortimer] already I.read thelast of.theBlakeandMortimer] ‘I already read the last one of Blake and Mortimer.’ c. já o li already the.accread The proposal was based on Postal’s (1969) theory that pronouns are underlying determiners, which seemed appealing enough for the form o, for obvious morphophonological reasons.2While I had a plausible story on the derivational relations bet60 CatWPL 6, 1998 Eduardo P. Raposo 1. I will use the form oas representative of the whole paradigm of definite determiners and accusative clitics (o/a/os/as; Spanish lo/la/los/las). Consistently with the basic idea just proposed, I will henceforth gloss this form as ‘the’, independently of its function, signalling its pronominal use by giving an indication of its Case (for example, ‘the-acc’). I will also use ele‘he’ as representative of the nominative paradigm and lhe‘him-dat’ as representative of the dative paradigm. Henceforth, I use EP for European Portuguese, BP for Brazilian Portuguese, and simply «Portuguese» when general features of both dialects are under discussion. 2. A number of recent works has resurfaced the idea of «pronouns as determiners»; see Corver and Delfitto (1993) and Uriagereka (1995), who also cites work in progress by Esther Torrego. To the best of my knowledge, these works have not provided answers to some of the questions that I raise here; furthermore, my proposal of why determiners move to become «clitics» (see section 3.1) is Table of Contents CatWPL 6 059-093 21/10/99 12:47 Página 60 1. Introduction 2. Some Minimalist Assumptions 3. Cliticization as Move F 4. Lhe and Ele as Minimal-Maximal Determiners 5. The Lexical Characterization of Determiners 6. Deriving (62) From Economy Considerations 7. The Status of ele and o in Brazilian Portuguese 8. Conclusion References ween the three sentences of (1), I had much less to say about nominative and dative forms.3In Romance, these forms can be used as pronouns, but not as articles, raising obvious problems for a full-fledged theory of pronouns as determiners:4 (2) a. ele já resolveu o problema he already solved the problem b. *[ele professor] já resolveu o problema he professor already solved the problem Conversely, given that o, as an article, can head DPs with any one of nominative, accusative or dative Case (see section 5), it was not clear at the time why it is restricted in its pronominal use to the accusative Case, as shown by the ungrammatical (3), with oused as a nominative subject (with an intended meaning equivalent to (2a)): (3) *já o resolveu o problema already the.nom solved the problem Another important fact is that Colloquial Brazilian Portuguese is loosing the pronominal use of o((1c) is ungrammatical in that dialect), while maintaining the use of that form as an article.5This is prima facie a very strong argument against my 1973 analysis; rather, it seems to call for an analysis in which we have two different homophonous forms. In this work, I would like to come back to these problems, filling in some of the gaps, and quite likely leaving many others open for future research. I assume the Minimalist Program of Chomsky (1995), showing that it brings particularly simple solutions to some of the problems addressed; in particular, I show that economy considerations play a crucial role in accounting for phenomena involving lexical choice. One important assumption that I make here is that all arguments are DPs. This implies by necessity that all arguments are headed by D, which may be phonetically unrealized (see Stowell 1989, Longobardi 1994, Raposo 1998). Note that this assumption logically forces us to the idea that pronouns are Ds, or form part of a complex D element. The thrust of this paper, then, can be seen in providing empirical and conceptual arguments that support this «brute force» conclusion. Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 61 quite different from that of the works just cited. I will not comment any further on these papers here. I return to the issue in work in preparation. 3. I assume in this work that the form ele that appears in prepositional contexts («strong ele») is to be distinguished from nominative ele and accusative ele of BP (see Cardinaletti and Starke 1994, Galves 1988). I will thus leave it aside in this work. This may be a mistake, though, evaluation of which I leave for future research. 4. In this paper, I use the term «determiner» to refer to the syntactic category of the forms under discussion, independently of their function; and I use the terms «pronoun» and «article» to refer to the surface functions performed by the determiner. Care should be taken with the terms «pronoun» and «pronominal», because I also propose that there are items that do belong underlyingly to this category (such as the null pro that appears inside DPs, or English one, for example). The distinction is made explicitly in the text when necessary. 5. See, among others, Galves (1989), Cyrino (1996), Kato (1991) and Nunes (1992). CatWPL 6 059-093 21/10/99 12:47 Página 61 An important proviso is in order here. Although I will be discussing clitic pronouns and cliticization, I am not concerned in this work with the particular phenomenon of «enclisis» in EP.6Thus, in order to avoid unnecessary confusion, I will only use examples with contexts of proclisis. With such a restriction, I hope that many, if not most, of my conclusions, will apply to Spanish as well (and also to other Romance languages, although, of course, caution is in order). The paper is organized as follows. In section 2 I present the theoretical assumptions that are relevant here, especially those of checking theory and the role of the [±Intrinsic] distinction among formal features. In section 3, I discuss the hypothesis that pronouns are determiners based on the behavior of o‘the’, and I suggest that (pronominal) clitic placement in Romance may be viewed as a case of «generalized pied-piping», whereby Move F (for feature checking) pied-pipes the category D for PF convergence. In section 4, I extend the theory of pronouns as determiners to the nominative and dative forms ele ‘he’ and lhe ‘him-dat’. In section 5 I discuss why the form o‘the’ cannot be used as a nominative or dative pronoun. In section 6, I propose that the descriptive account of section 5 can be derived from «economy considerations», in the technical sense of Chomsky (1995). In section 7, I discuss the claim that the (accusative) clitic pronoun odisappeared from Colloquial BP, showing that it cannot be maintained under Postal’s theory of pronouns as determiners. I then show that the facts of BP can be explained from the properties of ele ‘he’ and considerations of economy. In section 8 I make some concluding remarks, and in the Appendix I discuss the concept of numeration, which is used in a central way in sections 6 and 7. 2. Some Minimalist Assumptions 2.1. Clausal Structure and Functional Categories I will assume the clausal structure proposed in Chomsky (1995: ch.4, section 4.10), shown in (4), for a typical transitive clause with a subject and an object: (4) TP T' T spec Subj Obj VP V v v' P v 62 CatWPL 6, 1998 Eduardo P. Raposo 6. For discussion, see Martins (1994), Uriagereka (1995) and Raposo (1995, to appear a, b). CatWPL 6 059-093 21/10/99 12:47 Página 62 In this system, there are no Agr projections, and functional categories are allowed to have multiple specifiers. There is a non-substantive category v, immediately c-commanding the VP, configurationally contributing the external θ-role to an argument in its spec position. vhas a strong [V] feature (I presume universally), forcing overt raising of the main verb to adjoin to v. Overt object shift (in languages such as Icelandic) is determined by a strong [D] feature of v(see Collins and Thráinsson 1996, Collins 1997) in very much the same way as movement of the external argument to [Spec, T] (to subject position) is determined by a strong [D] feature of T. I assume here that in Portuguese, vdoes not have a strong [D] feature, since there is no overt object shift in this language.7 2.2. Checking Theory and Types of Features I basically assume the Checking Theory of Chomsky (1995). Lexical items in general (i.e. both substantive and non-substantive categories) consist of three distinct sets of features: P(honological) features, which are relevant only for PF; S(emantic) features, which play no role in the computation and are relevant only for the symbolic systems that interface with LF; and F(ormal) features, which play a role in the computation:8 (5) {{P-f}, {S-f}, {F-f}} There are two crucial distinctions to be made within the set of formal features, that between [±Interpretable] features, and that between [±Intrinsic] features.9The two typologies cross-cut each other. See (6) and (7): (6) [+Interpretable] categorial and φ-features of arguments [–Interpretable] Case-features; φ-features of predicates (V, A); strong features (7) [+Intrinsic] categorial features; [assign-Case]; [gender] of Nouns; strong features (?) [–Intrinsic] [Case] of arguments; [number] of Nouns; φ-features of predicates (V, A); [αTense] of Verbs The role played by the [± Interpretable] distinction in the mechanics of derivations is well-known. A [+Interpretable] feature must be present at LF for Full Interpretation and therefore does not need to be checked; however, it may enter Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 63 7. But see Costa (1998) and Bos˘kovicˆ (1997), among others, for the contrary view. Maybe French, Italian and BP have some residual form of object shift at work in the syntax of some pronouns (see Cardinaletti and Starke 1994, among others). I omit from (4) a category C and a possible category F (or Σ) intermediate between C and T (see, e.g. Laka 1990, Raposo and Uriagereka 1996), since they play no role in this work. 8. The phonological set may be missing (as in pronominal empty categories), or the semantic set may be missing (as in expletives). 9. Chomsky (1995) calls the [–Intrinsic] features optional, a term which is misleading, since [–Intrinsic] features are clearly obligatory in derivations; for example, a Noun occurring in a particular derivation must have a [–Intrinsic] Case feature, otherwise the derivation crashes (because the [+Intrinsic] [assign-Case] feature of some functional category is not checked). CatWPL 6 059-093 21/10/99 12:47 Página 63 into a checking relation in order to check a corresponding [–Interpretable] feature. The [–Interpretable] feature is then erased; the [+Interpretable] feature, however, is not erased, and may enter in subsequent checking relations. The second distinction has been less prominent in current work. Briefly, [+Intrinsic] features are those that do not vary according to the particular derivation in which a given item enters. They are listed in the lexical entry of the item, or strictly determined by features listed in its entry. For example, in Nouns, the categorial feature N and the [gender] feature are intrinsic; T has an intrinsic [assignnom-Case]. See (7). [–Intrinsic] features, on the other hand, are those that vary according to particular occurrences of an item in particular derivations. Case and [number] are clearly [–Intrinsic] features for Nouns, since they may vary in different contexts and in different derivations, as shown in (8) for the Noun livro ‘book’. (8) a. eu comprei os livros [acc, pl] I bought the books b. o livro é meu [nom, sg] the book is mine [–Intrinsic] features are not listed in the lexical entry of an item. I will follow Chomsky (1995), and asssume that they are added to the set of formal features of a lexical item as the lexical item is taken from the lexicon into the numeration. Given the important role played by this operation in this work, I will capitalize it as ADD. Its function is pictured in (9), which informally represents the process by which the word livro ‘book’ is removed from the lexicon and included in the numeration underlying (8a). (9) Lexicon Numeration LIVRO →LIVRO(S) [N] ADD [pl] [N] [masc] [acc] [masc] [pl] [acc] Chomsky (1995: 277) suggests that the [±Intrinsic] distinction plays virtually no role in derivations. I will suggest, on the contrary, that it does. In particular, I propose that the operation ADD enters into considerations of economy, being in that respect crucial for the workings of the computational system. 2.3. The Distribution of Features I will now make the following preliminary assumptions with respect to the distribution of formal features in functional and lexical categories (with Portuguese in mind):10 64 CatWPL 6, 1998 Eduardo P. Raposo 10. The proposal in (10) is different from that of Chomsky (1995) in one respect: I assume that the [assign-acc-Case] is a property of v, rather than V (see Raposo and Uriagereka (1996) for discussion). However, nothing hinges on this particular point. CatWPL 6 059-093 21/10/99 12:47 Página 64 (10) V, A: [φ-features] ([–Interpretable, –Intrinsic]) v: [Assign-acc-Case] ([–Interpretable, +Intrinsic]) [Strong [V]] ([–Interpretable, +Intrinsic]) T: [Assign-nom-Case] ([–Interpretable, +Intrinsic] [strong [D]] ([–Interpretable, +Intrinsic]) [Strong [V]] ([–Interpretable, +Intrinsic]) Arguments are also endowed with formal features: (11) arguments: [φ: gender, number, person] ([+Interpretable]) [+Intr] [–Intr] [+Intr] [Case] ([–Interpretable, –Intrinsic]) Below, I make some specific proposals about the exact source of argumental features, and I also propose that certain pronouns are lexically marked for Case, i.e. that their Case feature is [+Intrinsic]. This proposal will be central in my account of the pronominal system of Portuguese. 2.4. Last Resort and Feature Checking One idea that plays an important role in this work is that Romance cliticization is nothing but an overt reflex of feature checking (see section 3). I will assume the theory of feature checking of Chomsky (1995), especially the proposals of section 4, including the following definition (Chomsky 1995: 297):11 (12) K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K. At LF, only the set of formal features of the head of an argument moves. Following Chomsky (1995), I will refer to this as FF[argument], for example FF[livros] in (8a).12 For concreteness, consider the main aspects of the derivation of a transitive sentence such as (13): (13) os operários construiram a casa the workers built the house The numeration underlying this sentence is given in (14):13 (14) {(os,1), (a,1) (operários,1), (v,1) (T,1) (casa,1), (construiram,1)} In (14), casa ‘house’ has the following formal features: {[acus, φ-O]}; operários ‘workers’ has the features {[nom, φ-S]} («φ-O» and «φ-S» stand for «object φ-features» and «subject φ-features», respectively); vhas {[assign-accus-Case]}; Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 65 11. I assume that K ranges over (functional) heads, and call it the target of movement. For the definition of «sublabel of K» and others which are relevant here, see Chomsky (1995: 268ff). 12. As pointed out to me by Jairo Nunes, in the covert component all phonological features will have been stripped away by Spell-Out, so that it is not clear whether a complete category is moving (minus its phonological set, by necessity) or just its formal features. 13. I omit the category C, which is not relevant in this work. CatWPL 6 059-093 21/10/99 12:47 Página 65 and construiram ‘built’ has {[φ-O, φ-S]} (both [–Interpretable]) and a Tense feature; finally, T has both a strong [V] feature and a strong [D] feature. Recall that some of these features (the [–Intrinsic] ones) are added to the items by ADD as they enter the numeration (see above). I now illustrate the derivation and checking relations of (13), starting at the point where vP is constructed, and what remains in the numeration is T. In (15), V has raised to v, checking its strong V feature (the result is represented as Vb; I use informal tree notation and labeling for expository purposes). (15) Next we Merge T to (15), deriving (16): (16) The next step is overt raising of Vb to T, where it checks the strong [V] feature of T: (17) TP T T VP Vv v' P DP os operarios DP a casa construiram tv Vb tVb v TP T VP Vv v' v DP os operarios DP a casa construiram tv Vb P VP Vv v' P DP os operarios DP a casa construiram tv Vb v 66 CatWPL 6, 1998 Eduardo P. Raposo CatWPL 6 059-093 21/10/99 12:47 Página 66 The external argument now moves overtly to [Spec, T], checking the strong [D] feature of T, the Case feature of T, the [–Interpretable] φ-S of the Verb and its own Case feature: (18) At this point, (18) enters into Spell-Out. Spell-Out creates a copy of (18) which enters into Morphology and Phonology, and strips away the set of phonological features from (18) itself, which continues into the covert component. Feature checking still requires raising of FF[casa], yielding (19), where the Case feature of v, the φ-O of the Verb and the object’s own accusative Case are checked:14 (19) TP VP v' P DP os operarios DP a casa tv tVb T' tsubj T T Vv construiram Vb T FF [casa] t (FF) v TP T T VP Vv v' P DP os operarios DP a casa construiram tv Vb tVb T' tsubj v Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 67 14. Note that in a language with overt verb movement, Move FF[object] for feature checking targets T, not V or v, or their traces (Chomsky 1995: 360). The latter is in fact precluded by the two following conditions (Chomsky 1995: 304): (i) Trace is immobile. (ii) Only the head of a chain CH enters into the operation of Attract/Move. The fact that the derivation of a transitive clause involves a step like (19) rather than a step that would adjoin FF[object] to the trace of vis crucial for an account that reduces cliticization to Move F (I return to this issue in work in preparation). CatWPL 6 059-093 21/10/99 12:47 Página 67 (34) The φ-features of a predicate are [–Interpretable]. Consider now the semantic status of Nouns. It is commonly assumed in semantic theory that Nouns are predicates, and that their contribution to the interpretation of a DP resides precisely in their predicative content. If (34) is the correct generalization, then the φ-features of Nouns are [–Interpretable].28 One empirical argument for this conclusion comes from the Caipira dialect of BP spoken in the interior of the state of São Paulo.29 In this dialect, the morphological manifestation of the φ-features of verbs and adjectives is quite impoverished, being mostly neutralized to 3rd person (in verbs) and to singular. One interesting fact is that within DP, the number feature of an argument is (minimally) required on D, but not on the Noun; in other words, if it occurs only once within DP, it must be manifested on D, as shown in (35): (35) a. osmenino cantou the.pl child.sg sang.3sg b. *o meninoscantou the.sg child.pl sang.3sg In (35), the noun is behaving like other verbal and adjectival predicates. It is plausible to assume that the principle governing φ-feature manifestation in the Caipira dialect is the following: (36) [–Interpretable] φ-features may be morphologically unrealized. If this is correct, we have one empirical argument for the conclusion that at least the number feature of Nouns is [–Interpretable]. I conclude, then, that the [+Interpretable] [number-person] features of an argument are located in D, not in N.30 If this conclusion is correct, T covertly attracts FF[D] of an object for feature checking. That still leaves us with the Case feature. I will not take any position here as to the source of the Case feature within DP. In section 5, I give a tentative argument suggesting that the source of the Case feature is N, but I am neutral as to whether this is also specified in D as D enters the numeration.31 What I want to show, howe74 CatWPL 6, 1998 Eduardo P. Raposo 28. Note that I am not counting Case among the φ-features. Semantically, the conclusion in the text seems to be plausible for [person] (see the discussion above) and for [number], a property which, if «referential», should be interpreted in the determiner. [Gender], however, seems to be an intrinsic property of Nouns. 29. The same facts occur in other dialects of BP. See Scherre and Naro (1997). 30. One alternative analysis might invoke some abstract Agr node inside DP. This is not in the spirit of a parsimonious theory of functional categories such as that of Chomsky (1995: 4.10). Notice that if the [–Interpretable] φ-features of the Noun are checked against the [+Interpretable] φ-features of D (through N-to-D raising), this involves a departure from Chomsky’s (1995: 282ff) claim that the features of the target are invariably [–Interpretable]. To the best of my knowledge, I can see no theoretical harm in that conclusion (note that what attracts FF[N] is the [–Interpretable] affixal feature of D). After checking of the N features , the [+Interpretable] features of D are still available for feature checking against T. 31. Although from a minimalist point of view, that should be barred, if the Case feature has its source in N. CatWPL 6 059-093 21/10/99 12:47 Página 74 ver, is that either solution is compatible not only with D being covertly attracted by T, but also with the stronger claim that only D is covertly attracted by T, once we assume that there is N raising to D at LF in Romance. Suppose then, as suggested by Chomsky (1995: 364), following work of Longobardi (1994), that D has an affixal feature [N-] that requires raising of the categorial feature [N] of the Noun to adjoin to D. Raising of [N] automatically carries the whole set of formal features FF[N], including the Case feature, leaving a trace of FF[N] inside the Noun. In the derivation of (13) above, for example, FF[casa] adjoins to the determiner, leaving a trace inside the Noun casa, as shown in (37): (37) Chomsky (1995: 365) proposes the following principle: (38) αcan be attracted by K only if it contains no trace. This implies that whatever features of N are relevant for checking against T must be pied-piped to T within the independently attracted Determiner. I thus conclude that, independently of the origin of the Case feature, the formal features of an argument that move covertly for feature checking are all within the category D, which is the only category attracted by T. The last step of the derivation of (13) is thus (39) rather than (19): (39) 3.3. Reducing Cliticization to Move F I have shown that the category that is covertly attracted for feature checking is D. We can now close our case, and conclude that cliticization of ocan be viewed as a TP VP v' P DP os operarios DP a casa tv tVb T' tsubj T T Vv construiram Vb T FF [a] t (FF) v N DP casa a tFF D FF [casa] D Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 75 CatWPL 6 059-093 21/10/99 12:47 Página 75 simple manifestation of Move F, with «generalized pied-piping» of the entire category, including its P(honological) features; in Chomsky’s (1995: 262) terms, carrying «just enough material for (PF) convergence». This is explicitly stated in (40): (40) Clitic placement (of o) is Move F pied-piping the whole category ofor PF convergence (i.e., to satisfy Property P). The derivation of (41b), with a clitic, is thus similar to the derivation of (41a), with a plain direct object. In both, the determiner raises to the checking domain of T, for feature checking with the verb. The only difference is that the whole category is pied-piped in (41b), whereas only its formal features are in (41a). Compare their respective structures in (39) and (42): (41) a. os operários (não) construíram acasa the workers not built the house b. os operários (não) aconstruíram the workers not the.acc built (42) Note that Property P is satisfied by the output of Move. The determiner lands at the left edge of T, where it right-attaches to the verb, also adjoined within T. One consequence of this analysis of clitic placement is that clitics in Romance do not target the verb directly, but rather T, which contains the sublabels for object checking (see section 2.4, especially fn. 14). The idea that the clitic targets an inflectional node was proposed by Kayne (1991) but, to the best of my knowledge, it had to be stipulated in that framework. The result is entirely derived in the present account. 3.4. Extension of the Proposal to the Clitic lhe This account can be extended to the dative pronoun lhe ‘him-dat’. That lhe is a clitic with a property similar to P (see (31)) is shown in (43): TP VP v' P DP os operarios DP tv tVb T' tsubj T T Vv construiram Vb T a tDpro v 76 CatWPL 6, 1998 Eduardo P. Raposo CatWPL 6 059-093 21/10/99 12:47 Página 76 (43) a. *eu não dei o livro (a) lhe I not gave the book to him.dat b. eu não lhe dei o livro I not him.dat gave the book ‘I did not give him the book.’ I assume that lhe is also an underlying determiner, with a dative Case feature. I also assume that this Case is assigned by the verb, perhaps as an inherent Case; in other words, verbs that select for indirect objects have a [+Intrinsic, –Interpretable] [assign-dative-Case] feature that must be checked. Whatever the precise mechanics of this checking are, they involve the «intermediate» dummy preposition a ‘to’ in the case of a full-fledged argument:32 (44) eu não dei o livro a[o estudante] I not gave the book to the student However, if lhe is a determiner with Property P, this particular derivation yields (43a), which crashes at PF; therefore, lhe cannot use the «prepositional astrategy» for Case checking; instead, it moves to the verb, where it checks directly its Case feature against that of the verb; and this movement must be overt, for the reasons just discussed (Property P). 4. Lhe and ele as Minimal-Maximal Determiners The idea that pronouns are underlying determiners is appealing for accusative clitics, since they clearly enter in paradigms where they function either as pronouns or as articles, and there is a plausible derivational connection between those paradigms. The idea, however, does not seem to extend easily to the nominative ele ‘he’ and the dative lhe ‘him-dat’. These do not present similar alternations, as shown in (45) and (46).33 (45) a. ele resolveu o problema he solved the problem b. eu não lhe dei o livro I not him.dat gave the book (46) a. *[ele professor] resolveu o problema he professor solved the problem b. *eu não dei o livro (a) [lhe professor] I not gave the book to him.dat professor Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 77 32. See Chomsky and Lasnik (1993) (Chomsky 1995: 113ff) for some observations concerning the largely unexplored issue of (inherent) Case checking through a «dummy» preposition. 33. Note that lhe in (46b) would satisfy Property P. The form ele, not being a clitic, is not subject to Property P. CatWPL 6 059-093 21/10/99 12:47 Página 77 The basic selectional property of determiners, that of subcategorizing for a nominal projection, does not hold of ele and lhe. In other words, a structure like (47) is not available for these forms (αa lexical Noun): (47) What I would like to propose is that ele and lhe are «complex» determiners, meaning by that «intransitive» Ds which satisfy their selectional feature via a [+N] feature in their grammatical matrix rather than by merging with a categorially independent Noun. In other words, in terms of Chomsky’s (1995) bare phrase structure theory, they are [+minimal, +maximal] projections rather than [+minimal,–maximal] projections like the «pure» determiners lo/o. Under this analysis, the structure of these forms is (48), where the superscripts are merely notational:34 (48) Thus, while ele or lhe are merged in argument positions as maximal projections, the «pure» determiner ois merged first with a nominal projection, which may be the empty category pro. The underlying structure for oas a pronoun is thus (49): (49) This analysis accounts in a straightforward way for the ungrammaticality of (46). Being intransitive [+minimal, +maximal] projections, the forms ele, lhe cannot be combined with an independent Noun: that derivation is simply impossible. Similarly, the following expressions, with modified pronouns, are also excluded, on the assumption that nominal modification requires a full categorial N and not just a feature (note that lhe in (50b) would satisfy Property P):35 pro DP D o D ele/lhe [....] [+N] [....] max/min NP *DP D ... N ... a ele/lhe 78 CatWPL 6, 1998 Eduardo P. Raposo 34. See Chomsky (1995: 249). Differently from Chomsky, though, I do not assume that this characterization is co-extensive with the notion of clitic. 35. A host of questions that I cannot address here arise, e.g., the impossibility as well of relative clause modification of these forms, and what this implies for the structure of relative clauses: (i) a. *[ele que eu encontrei ontem] resolveu o problema he.nom that I met yesterday solved the problem CatWPL 6 059-093 21/10/99 12:47 Página 78 (50) a. *[ele de matemática] resolveu o problema he of mathematics solved the problem b. *eu não dei o livro (a) [lhe de matemática] I not gave the book to him.dat of mathematics In sum, ele and lhe can only be used in their pronominal function, never in their article function (including as a «demonstrative»). In order to generate convergent expressions equivalent to (46) and (50), the system has to resort to the pure determiner o:36 (51) a. [oprofessor] resolveu o problema the professor solved the problem b. eu não dei o livro a [oprofessor] I not gave the book to the professor] (52) a. [o pro de matemática] resolveu o problema the of mathematics solved the problem b. eu não dei o livro a [o pro de matemática] I not gave the book to the of mathematics There is another property that ele and lhe have in common in EP, and which differentiates them from other determiners. Semantically, they favor a [+human] interpretation.37 However, a positive or negative specification for the feature [human] is not a typical property of determiners. For example, none of the following determiners is required to occur with a noun so specified, as shown in (60):38 (53) a. o/este/esse/aquele/um rapaz the/this/this-that/that/one boy b. o/este/esse/aquele/um livro the/this/this-that/that/one book These data suggest that «pure» determiners are not specified for the feature [human], even when they are clitics with a pro complement (pace Corver and Delfitto 1993); rather, the [human] specification of a DP has its source on the Noun.39 As to ele and lhe, we can speculate that their preferred reading as [+human] Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 79 b. *eu não dei o livro (a) [lhe que eu encontrei ontem] I not gave the book to him.dat that I met yesterday 36. pro allows modification unproblematically. I will return to these derivations from an economy perspective in section 6. 37. There is however a lot of idiolectal variation, especially concerning ele, as pointed out to me by Inês Duarte. The situation still seems to be different from BP, where ele has no semantic restrictions whatsoever along the [human] dimension (see Galves 1998). A similar lack of semantic restrictions applies to pro in EP. 38. I give ‘this-that’ as the gloss of esse. 39. Whatever process allows for the recovery of the content of pro in sentences like (25) and (27b) will also recover its [human] value. For example, even if discourse or pragmatic context is minimal, the selectional restrictions of the verb comprar ‘to buy’ will presumably impose a [–human, CatWPL 6 059-093 21/10/99 12:47 Página 79 has its source on the nominal feature that satisfies their selectional requirement (see (48) and preceding text; and fn. 37). 5. The Lexical Characterization of Determiners Let us now address another issue concerning the distributional properties of determiners, that of the grammatical functions that they can fulfill in a clause. In EP, ele is exclusively nominative, and lhe is exclusively dative:40 (54) a. ele saiu he left b. *eu vi ele I saw he c. *eu não dei o livro a ele I not gave the book to he (55) a. eu não lhe dei o livro I not him.dat gave the book b. *não lhe saiu not him.dat left c. *eu não lhe vi I not him.dat saw 80 CatWPL 6, 1998 Eduardo P. Raposo –abstract] reading on pro. This interpretive process presumably takes place at a level of representation that interfaces with LF. This is why Corver and Delfitto’s (1993) proposal seems to me to be on the wrong track. 40. I am ignoring here BP (and Spanish) dialects where lhe can be used as a direct object, as well as dialects where accusative pronouns can be used as indirect objects. These would require a different lexical characterization of the forms under discussion, but no major changes of the principles guiding the analysis would be needed, as far as I can see (Ramos in preparation). For the status of (54b) in BP, see section 7. Note that a sentence similar to (54c) with clitic doubling is grammatical in EP (clitic doubling is very restricted in EP, being possible only when the doubled element is itself a pronoun): (i) eu não lhe dei o livro a ele Inot to.him gave the book to him A reviewer asked how the phenomenon of clitic doubling fits in the present analysis. Although I have not given much thought to this matter, I assume Uriagereka’s (1995: 81) analysis (his structure (2a)), where the double is in [Spec, DP] headed by lhe. Notice that in my analysis the double (if it forms a single constituent with the clitic) has to be in [Spec, DP]; if it were a complement of D, Property P would be satisfied and there would be no need for cliticization of lhe. In addition, I assume that the preposition ain (i) is marking the doubled pronoun in [Spec, DP] rather than the whole DP; in this respect, note that the preposition must be present even if the doubled argument is an accusative direct object («Kayne’s generalization»): (ii) eu não o vi a ele Inot him.acc saw to him ‘I did not see him.’ CatWPL 6 059-093 21/10/99 12:47 Página 80 This is a lexical property of these items; as such, it should be encoded in their lexical entries. I thus claim that ele and lhe have a [+Intrinsic] [nominative] and [dative] Case feature, respectively. This is represented in (56) and (57):41 (56) ele [D, –P, φ, nom] (57) lhe [D, +P, φ, dat] The form o, however, can be used with any structural Case (and also dative Case), as shown in (58): (58) a. [oprofessor] não resolveu o problema [nom] the professor not solved the problem b. eu não dei o livro a [oprofessor] [dat] I not gave the book to the professor] c. eu não li [olivro] [acc] I not read the book Clearly, it would be a mistake to assign to oan entry with an inherent Case feature (say, accusative, as is often proposed in analyses that do not assume Postal’s theory). Rather, in light of (58), the correct lexical characterization of oseems to be (59), with no Case feature specified: (59) o[D, +P, φ] Notice that this state of affairs would follow trivially if «pure» determiners are simply not specified for a Case feature; in other words, if this feature is always contributed by the Noun which is the complement to the determiner. Under this view, ele and lhe have a Case feature because they already «incorporate» an abstract nominal feature, and that nominal feature is contributing a [+Intrinsic] Case feature. The form o, on the other hand, being a pure determiner, does not contribute a Case feature; and that feature is assigned to the Noun (or to pro) by ADD as it enters the numeration (see (9)). Given this, however, the following question immediately arises: if the form o can be used as an article in nominative, dative and accusative DPs, why is its pronominal use restricted to accusative Case, as in (60)? What prevents it from being syntactically combined with a pro specified as [nominative] or [dative], as in (61ab), which is intended to parallel (58a-b) in meaning?42 Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 81 41. For simplicity, I omit a codification of the phrase structure level status of these forms, and I encode the non-clitic status of ele as [–P]. 42. Concerning (61a), I put aside the complex question of whether an external argument may or may not be a syntactic clitic. See Chomsky (1995: 362) for some suggestive remarks that it may not. This would rule out (61a) independently of the text considerations. For the sake of the argument here, I assume that it can. CatWPL 6 059-093 21/10/99 12:47 Página 81 (60) eu não oli [acc] I not the.acc read ‘I did not read it.’ (61) a. *não oresolveu o problema [nom] not the.nom solved the problem ‘He did not solve the probem.’ b. *eu não odei o livro [dat] I not the.dat gave the book ‘I did not give the book to him.’ The Elsewhere Principle (Kiparsky 1973, Lumsden 1992, Halle and Marantz 1993, among many others) has been extensively used in the literature to account for precisely the type of problem under discussion here. This principle states that the availability of a given lexical form blocks lexical access to an otherwise quasiequivalent form, which, however, is not specified (or is underspecified) for a particular feature needed for the success of the derivation. The following version of the Elsewhere Principle is adapted from Lumsden (1992): (62) A form A that is specified for a relevant feature blocks a form B that is unspecified (or not-specified) for that feature. Descriptively, (62) has the desired effect. In EP, ele and lhe are lexically specified with a [+inherent] nominative and dative Case feature, respectively; thus, if the derivation calls for nominative and dative Case, respectively, these forms will be «selected», rather than o, which has no lexical specification for Case; instead of (61a-b), we have (63): (63) a. ele não resolveu o problema he.nom not solved the problem b. eu não lhe dei o livro I not him.dat gave the book Principle (62) (and the Elsewhere Principle in general), however, is somewhat stipulative. Furthermore, there is a redundancy in the way it applies to (61)/(63). Note that in addition to being inherently specified for a Case feature, the forms ele and lhe have a nominal feature that satisfies their selectional properties without the necessity of merging an independent Noun. A priori, thus, we do not know if (62) is sensitive to just one of these specifications, or to both; and in the former case, to which one. I would thus like to propose that (62) can be motivated on plausible theoretical grounds; in particular, I claim that it can be derived from economy considerations; it will then be seen that (62) is sensitive to the specification of a Case feature in ele and lhe rather than to their nominal feature, under plausible assumptions concerning the elements and processes that fall under the scope of economy. 82 CatWPL 6, 1998 Eduardo P. Raposo CatWPL 6 059-093 21/10/99 12:47 Página 82 6. Deriving (62) from Economy Considerations 6.1. Grounding the Proposal Let us consider more carefully the operation ADD (see section 2.1, especially (9) and its accompanying text); in particular, I am interested in the issue of whether it is subject to economy considerations. Chomsky (1995: 226) suggests that whether or not an operation actually falls within the domain of convergence and economy depends on its status with respect to derivations. Those operations that are a defining property of derivations are «costless». These include Merge and Select (from the numeration into the derivation). Insufficient application of Merge, or a non-exhausted numeration, fail to produce a derivation. They are like logical deductions where a rule has been misapplied: no questions of convergence arise, and thus no questions of economy arise either, given the way economy depends on convergence in Chomsky’s minimalist model. Chomsky’s (1995) proposal is that the only operations that enter into considerations of economy are those associated with convergence conditions, such as Move.43 From this perspective, forming the numeration appears to have the same status as Merge or Select: after all, if a numeration is not formed, we do not have a derivation at all: that seems to be the most basic of prerequisites. This in turn may follow from the core status of selectional properties in the «assemblage» of derivations (of Merge); it is their satisfaction that «creates» the derivation to start with, and if they are not satisfied, there simply is no derivation. Note, however, that the same is not true of the operation ADD. This operation is not a defining property of derivations. If it is not applied where it should be, i.e. if it fails to assign some [–Intrinsic] formal feature to a lexical item, the derivation may still take place and reach LF, but with an ill-formed object that either causes the derivation to crash or allows it to converge with a deviant interpretation. For example, if in the derivation of (13) the accusative Case feature is not added to the Noun casa ‘house’ as it enters the numeration, the derivation reaches stage (18); at that point, the feature in the object that is attracted cannot be [Case], since this is missing; but the object has φ-features, which are attracted by the [–Interpretable] object φ-features of the verb, deriving (39). At LF, this derivation crashes because the [+Intrinsic] [assign-acc-Case] feature vis not checked. In sum, from the point of view of derivations, ADD is associated with convergence, and should be included in the set of operations that are subject to economy conditions. Note that if this line of reasoning is on the right track, it forces us to include the formation of the numeration (especially ADD) within the set of derivational operations performed by CHL. This proposal about ADD can be used to place (62) on a less stipulative ground, and offers a principled solution to the problem of pronominal choice under discussion here. Suppose that the nominal feature that appears on ele and lhe is equivalent to pro, and that the items oand ele, and oand lhe, in (61) and (63), can be Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 83 43. For an extensive and illuminating discussion of these issues, see Nunes (to appear). CatWPL 6 059-093 21/10/99 12:47 Página 83 Appendix: The Concept of the Numeration Chomsky (1995) discusses the issue of how to characterize the set of derivations that are compared for the purpose of choosing the most economical one (the reference set). One of Chomsky’s proposals is the following: (77) The reference set is based on the same numeration. The notion of numeration, in turn, is defined as in (78) (from Chomsky 1995: 225, with a slight change): (78) Numeration A numeration is a set of pairs (LI, i), where LI is an item of the lexicon and iis its index, understood to be the number of times that LI is selected from the lexicon. Chomsky (1995: 236) also seems to (implicitly) assume that the [–Intrinsic] features of each LI are already part of the numeration, since he (explicitly) proposes that these features are «added [to each LI] by [the] step [that] form[s] a numeration» (p. 236) (material in square brackets is my responsibility). Notice, however, that twice in this work I have compared derivations which, strictly speaking, do not have the same lexical items, i.e. that are not based in the same numeration as defined in (78): the derivation of (61) with that of (63) (ovs. ele and ovs. lhe, respectively) and the derivation of (73a) (o) with that of (73b) (ele). Since the problems are the same, I will illustrate them here with (73a-b). The different numerations of these examples are partially represented in (79): (79) a. Numeration underlying (73a): {(o, 1), (pro, 1), ...} b. Numeration underlying (73b): {(ele, 1),...} Clearly, if our analysis is to be maintained, Chomsky’s view of the relationship between numerations and reference sets must be relaxed in some way (see Nakamura 1997 for a similar predicament). I will suggest one minimal adaptation of Chomsky’s ideas that is compatible with my analysis.52 The crucial observation is that the difference between (79a) and (79b) resides in the set of P(honological) features of the items in question. In particular, oand ele in (79) have the same set of formal and semantic features, including the accusative Case feature. One obvious way of relaxing the notion of Numeration is thus the following, which abstracts away from the P(honological) features:53 90 CatWPL 6, 1998 Eduardo P. Raposo 52. For a different (and less restrictive, in my view) solution, see Nakamura (1997). 53. Note that the different value for Property P of oand ele is correctly abstracted away under the assumption that this is a P-feature. CatWPL 6 059-093 21/10/99 12:47 Página 90 (80) Numeration A numeration is a set of pairs (LI, i), where LI is the set of {F-f, S-f} (see (5)) of an item of the lexicon and iis its index, understood to be the number of times that LI is selected from the lexicon. A second problem has to do with the categorial status of the two items. ois a [+minimal, –maximal] D requiring an additional categorial N in the numeration with which it can Merge, while ele is a [+minimal, +maximal] D already with a nominal feature. Can this difference also be abstracted away? I tentatively suggest that the answer is positive. As I briefly discussed in section 6.1, suppose that the nominal feature that appears on ele and lhe is the equivalent of pro; we might then propose that these two elements count as «the same» as far as the reference set is concerned. If something along these lines is correct, (79a) and (79b) are indeed equivalent from the perspective of the reference set.54 References Abney, S.P. (1987) The English Noun Phrase in its Sentential Aspect. MIT. Doctoral Dissertation. Barbosa, P. (1996). «Clitic Placement in European Portuguese and the Position of Subjects». In Halpern, A.L.; Zwicky, A.M. (eds.). Approaching Second: Second Position Clitics and Related Phenomena. CSLI Publications. Bos˘kovicˆ, Z. (1997). «Coordination, Object Shift, and V-Movement». Linguistic Inquiry 28: 357-365. Chomsky, N. (1981). Lectures on Government and Binding. The Hague: Foris. — (1995). The Minimalist Program. Cambridge, Massachusetts: MIT Press. Chomsky, N.; Lasnik, H. (1993). «The Theory of Principles and Parameters». In Jacobs, J.; Stechow, A. von; Sternefeld, W.; Venneman, T. (eds.). An International Handbook of Contemporary Research. Berlin: Walter de Gruyter. Reprinted in Chomsky (1995). Cardinaletti, A.; Starke, R. (1994). «The Typology of Structural Deficiency: On the Three Grammatical Classes». University of Venice and University of Geneva. Unpublished ms. Collins, C. (1997). Local Economy. Cambridge, Massachusetts: MIT Press. Collins, C.; Thráinsson, H. (1996). «VP-Internal Structure and Object Shift in Icelandic». Linguistic Inquiry 27: 391-444. Corver, N.; Delfitto, D. (1993). «Feature Asymmetry and the Nature of Pronoun Movement». University of Utrecht: OTS Publications. Costa, J. (1998). Word Order Variation: A Constraint-Based Approach. Holland Institute of Generative Linguistics Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 91 54. One can think of other problems, such as the fact that ADD is an operation that forms the numeration, which itself is the base for the reference set; this implies that we are comparing derivations even before the base of the reference set is established. I will address this issue in work in preparation. This also highlights in a dramatic way the global nature of economy conditions under the approach that I have assumed here. See Kitahara (1997) and Collins (1997) for two quite different approaches to this topic. CatWPL 6 059-093 21/10/99 12:47 Página 91 Cyrino, S. (1996). «O Objeto Nulo do Português Brasileiro». DELTA 12: 221-238. Galves, C. (1989). «L’Objet Nul et la Structure de la Proposition en Portugais du Brésil». Revue des Langues Romanes XCIII: 2. — (1998). «La Syntaxe Pronominale du Portugais Brésilien et la Typologie des Pronoms». In Zribi-Hertz, A. (ed.). Les Pronoms. Paris: Presses Universitaires de France. Halle, M.; Marantz, A. (1993). «Distributed Morphology and the Pieces of Inflection». In Hale, K.; Keyser, S.J. (eds.). The View From Building 20: Essays in honor of Sylvain Bromberger. Cambridge, Massachusetts: MIT Press. Kato, M. (1991). «Object Pronouns and Null Elements». In Ashby, W.; Mithun, M.; Perissinotto, G.; Raposo, E. (eds.). Linguistic Perspectives on the Romance Languages. Amsterdam: John Benjamins. Kayne, R. (1991). «Romance Clitics, Verb Movement, and PRO». Linguistic Inquiry 22: 647-686. Kiparsky, P. (1973). «“Elsewhere” in Phonology». In Anderson, S.R.; Kiparsky, P. (eds.). A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston, Inc. Kitahara, H. (1997). Elementary Operations and Optimal Derivations. Cambridge, Massachusetts: MIT Press. Laka, I. (1990). Negation in Syntax: On the Nature of Functional Categories and Projections. MIT. Doctoral Dissertation. Longobardi, G. (1994). «Reference and Proper Names». Linguistic Inquiry 25: 609-665. Lumsden, J. (1992). «Underspecification in Grammatical and Natural Gender. Linguistic Inquiry 23: 469-486. Martins, A-M. (1994). Clíticos na História do Português. Universidade de Lisboa. Doctoral Dissertation. Nakamura, M. (1997). «Object Extraction in Bantu Applicatives: Some Implications for Minimalism». Linguistic Inquiry 28: 252-280. Nunes, J. (1992). «Directionality of Cliticization, Distribution of Clitics and Null Objects in Brazilian Portuguese». University of Maryland. Unpublished ms. — (to appear). «Some Notes on Procrastinate and Other Matters». DELTA. Postal, P. (1969). «On So-called “Pronouns” in English». In Reibel, D.A.; Schane, S.A. (eds.). Modern Studies in English: Readings in Transformational Grammar. Englewood Cliffs, New Jersey: Prentice Hall. Ramos, C. (in preparation). O clítico de Terceira Pessoa: um Estudo Comparativo Português Brasileiro/Espanhol. Universidade Federal de Alagoas. Doctoral Dissertation. Raposo, E. (1973). «Sobre a Forma oem Português». Boletim de Filologia XXII: 361-415. — (1995). «Próclise, Enclise e a Posição do Verbo em Português Europeu». In Camilodos-Santos, J.; Williams, F.G. (eds.). O Amor das Letras e das Gentes. Center for Portuguese Studies, University of California, Santa Barbara, California. — (1998). «Definite/Zero Alternations in Portuguese (Towards a Unification of Topic Constructions»). In Schwegler, A.; Tranel, B.; Uribe-Etxebarria, M. (eds.). Romance Linguistics: Theoretical Perspectives. Amsterdam: John Benjamins. — (to appear, a). «Clitic Positions and Verb Movement». In Costa, J. (ed.). Portuguese Syntax: New Comparative Studies. Oxford University Press, Oxford. 92 CatWPL 6, 1998 Eduardo P. Raposo CatWPL 6 059-093 21/10/99 12:47 Página 92 — (to appear, b). «Directionality of Cliticization in European Portuguese: Left or Right?». In Franco, J.; Landa, A.; Martín, J. (eds.). Grammatical Analyses in Basque and Romance Linguistics. Amsterdam: John Benjamins. Raposo, E.; Uriagereka, J. (1996). «Indefinite SE». Natural Language and Linguistic Theory 14: 749-810. Scherre, M.; Naro, A. (1997). «A Concordância de Número no Português do Brasil: Um Caso de Típica Variação Inerente». In da Hora, D. (ed.). Diversidade Linguística no Brasil. João Pessoa: Ideia Editora. Stowell, T. (1989). «Subjects, Specifiers and X-bar Theory». In Baltin, M.; Kroch, A. (eds.). Alternative Conceptions of Phrase Structure. Chicago: The University of Chicago Press. Uriagereka, J. (1995). «Aspects of the Syntax of Clitic Placement in Western Romance». Linguistic Inquiry 26: 79-123. Some Observations on the Pronominal System of Portuguese CatWPL 6, 1998 93 CatWPL 6 059-093 21/10/99 12:47 Página 93