Ensaio teórico BRAZILIAN PORTUGUESE AND FINNISH REFERENTIAL NULL SUBJECTS* RODRIGUES Cilene revistadaabralin@gmail.com Universidade de Brasília 3 1/2 BRAZILIAN PORTUGUESE AND FINNISH REFERENTIAL NULL SUBJECTS 75 119 75-119 http://creativecommons.org/licenses/by/4.0/

This paper focuses on the syntax of referential null subjects in Modern Brazilian Portuguese and Finnish. It will be argued that these language art not null subject grammars; that is, their referential null subjects are the residue of movement, rather than lexical pronominal null categories. This is to be formally correlated with the weak verbal agreement morphology of the two grammars under consideration.

 Resumo

Este artigo trata das propriedades sintáticas dos sujeitos nulos referenciais do português do Brasil moderno e do finlandês. Na análise aqui apresentada, essas não são linguas de sujeito nulo. Os sujeitos nulos que nelas observamos são resultados de movimentos, e não pronomes nulos. Essa característica do português do Brasil e do finlandês deve ser fomalmente correlacionada com o fraco sistema de concordância verbal dessas gramáticas.

 Sujeitos nulos movimento morfologia verbal Introduction

This paper focuses on null subjects in two of the so-called partial pro-drop languages, Modern Colloquial Brazilian Portuguese (BP) and Finnish, seeking an answer to (1):

.

(1) What is the syntactic nature of referential null subjects in BP and Finnish?

.

The possibility of dropping any pronominal subject is taken to be a manifestation of the pro-drop parameter (e.g. Chomsky 1981, Rizzi 1982, Jaeggli and Safir 1989). 'Hius, Italian and European Portuguese, in which expletives as well as referential subjects can be null are bona fide examples of a positive setting of this parameter.

Partial pro-drop languages allow null expletives (2).1 But, differently from full pro-drop grammars, in these grammars, 3rdPerson (3rdP) referential null subjects occur only in embedded clauses (3)(4), and behave like anaphors, requiring a sentendal antecedent.

In short, the state of affairs is the following: in full pro-drop grammars, referential null subjects are null pronouns (pro), whereas, in partial pro-drop languages, they are null anaphors. This raises a concern for the pro-drop parameter. Shall we take BP and Finnish to be pro-drop grammars? If the answer is positive, then we need to understand how a fixed parameter can have different outcomes: while in Italian and European Portuguese the fixation of the pro-drop parameter opens up the availability of pronominal null subjects, in BP and Finnish it results in anaphoric null subjects. On the other hand, in analyzing the grammars under considerarion as non pro-drop, one needs to account for (4), explaining the nature of the empty category observed there. Hence, question (1) is inevitable, and any theory about the pro-drop nature of BP and Finnish has to be built upon an answer for that question.

This paper answers (1) by suggesting that Modern BP and Finnish are not null subject languages. In these grammars, referential null subjects arc traces rather than null pronouns.2 It will be suggested that the availability of a trace in the subject position of a finite clause is the result of weak verbal agreement morphology.

The sections are organized as follows: section 1 introduces the loss of verbal agreement morphology in BP and Finnish. Section 2 examines the licensing of null subjects in these languages. Section 3 proposes that instances of null subjects found in matrix clauses of BP and Finnish are cases of topic deletion. Section 4 takes on 3rdP null subjects, presenting evidence thar these subjects result from A-movement out of finite clauses. Section 5 offers a technical implementation for a movement analysis, correlating the possibility of having A-movemcnt out of finite domains with loss of verbal agreement morphology. Section 6 summarizes the discussion.

 1. Verbal Agreement Morphology and Licensing of Null Subjects

In the recorded history from 1845 to 1992, BP underwent a simplification in its verbal inflection (cf. Fig 1). A paradigm with six forms (paradigm 1) was reduced to a paradigm with four forms (paradigm 2), and dien to a paradigm with only three forms (paradigm 3). From Stage 1 to Stage 2, the indirect 2ndPSg replaced the direct 2ndPSg agreement, which is phonologically identical to 3tdPSg. From Stage 2 to Stage 3, the 1stPPI was replaced by 3r,,PSg. This latter change occurred since the pronoun a gente (‘we’, literally-, ‘the people*) replaced the 1“P1 pronoun nós 'we'. A gente triggers 3rdPSg agreement.

Fig. 1. Falar Finlandês 'to speak Finnish'

Source: Duarte (1996)

Person and Number Paradigm 1 Paradigm 2 Paradigm 3
1stSg Fal-o Finlandês Fal-o Finlandês Fal-o Finlandês
2ndSg direct Fala-s ------------------------ ------------------------
2ndSg indirect Fala-Ø Fala-Ø Fala-Ø
3rdSg Fala-Ø Fala-Ø Fala-Ø
1stPl Fala-mos Fala-mos Fala-Ø
2ndPl direct Fala-is ------------------------ ------------------------
2ndPl indirect Fala-m Fala-m Fala-m
3rdPl Fala-m Fala-m Fala-m

The verbal agreement morphology of Standard Finnish (SF) seems to be quite rich when compared to diat of BP. For every combination of person and number, there is a different ending on the verb, as figure 2 illustrates:

Fig. 2. Puhu Portugalia 'to speak Portuguese'

Person and Number Standard Finnish
1stSg Puhu-n Portugalia
2ndSg Puhu-t
3rdSg Puhu-u
1stPl Puhu-mme
2ndPl Puhu-tte
3rdPl Puhu-vat

However, rhe richness of diis paradigm becomes questionable once we consider the 3n,P in detail. As apparent in figure 2, the l*PSg/Pl, 2ndPSg/Pl and 3rdPPl verbal inflection contains a consonant, but the 3nlPSg consists of lengthening of the last vowel of die stem. This lengdiening process occurs in the present tense only. In the past and conditional tense (5), die process is neutralized and there is no overt agreement suffix for the S^PSg (cf. Vainikka, 1989 and Holm berg and Nikannc, 1993).3

.

(5) a. Hän Puhui Portugalia

s/he spoke -3Sg Portuguese

'S/he spoke Portuguese'

b. Hän puhusi Portugalia

s/he speak-3Sg-Cond Portuguese

'S/he would speak Portuguese'

.

Vainikka and Levi (1999), citing work done by Hakulinen (1979), present furdier evidence that 3rdP verbal agreement morpheme is weak in SF. All rhe agreement suffixes are phonologically related to the corresponding pronoun, except for die 3rdP suffixes (cf. figure 3). The 2ndPSg agreement suffix t is diachronically related to die pronoun sina, since this pronoun is reconstructed as *lina.A But between the 3rdPSg/PI agreement suffixes and the corresponding pronouns there is no such historical connection.

Fig. 3

Source: Vainikka and Levy (1999)

1stSg 2ndSg 3rdSg 1stPl 2ndPl 3rdPl
Agreement Suffix -n -t -V -mme -tte -vat
Pronoun mina sinä hän me te he

This leads us to consider that the verbal inflection paradigm of SF is not uniform under Jaeggli and SafiFs (1989) definition of a morphologically uniform verbal inflection paradigm. It is a mixed paradigm, containing an undcrived inflectional form, namely the form 3rdSg.5 Hence, under Jaeggli and SafiFs theory of null subjects, SF is not pro-drop language.

Having examined Standard Finnish, we now turn to Colloquial Finnish. Similarly to BP, Colloquial Finnish (CF) has a simplified verbal inflection paradigm.6 The 3rdPSg replaced the 3ft,PPl and the 1stPPl was substituted by a new form that is phonologically identical to die impersonal passive morpheme (6).

Fig. 4. <italic id="italic-1d33805986c3dcdd1c8126eb816e1398">Puhu Portugalia</italic> 'to speak Portuguese'

Person and Number Colloqual Finnish
1stSg Puhu-n Portugalia
2ndSg Puhu-t
3rdSg Puhu-u
1stPl Puhu-taan
2ndPl Puhu-tte
3rdPl Puhu-u

(6) Koulussa opetetaan lapsia

school-Ine teach-Pass children-Par

'Children are taught at school'

.

In conclusion, the grammars under investigation share the property of having a weak 3rdP verbal morphology. Thus, given the fruitful literature on the correlation between rich morphology and licensing oi' null subjects (cf. Taraldsen 1980, Chomsky, 1981, 1982, Rizzi, 1982, Borer 1986, 1989, Jaeggli and Safir 1989, among others), the content of the next section should he predictable: BP and Finnish are not fully-fledged pro-drop languages.

 2. Partial Null Subjects

As shown below, null expletive and generic null subjects arc licensed in BP and Finnish:7

For the time being, I will put generic null subjects aside. As I discussed in Rodrigues (2004a), these subjects are unable to bind ana- phors, unable to control inside infinitival clauses and unable to be modified by subject-oriented adverbs. Thus, since these properties arc usually observed when the subject position is occupied by a null pronoun, it is unclear whether the missing subject of (8) is syntactically realized or not.9

CF differs from SF and BP in diat it has overt expletives, even though they are optionally inserted, as shown below (Cf. Holmberg and Nikannc 1994, 2002 and Holmberg 2003).10

Interestingly, though, none of these three grammars allows 3rdP referential null subjects in matrix clauses:

Another interestingly property of present-day BP and Finnish is that they do not pattern with Romance null subject languages with respect to die Avoid Pronoun principle (11). While Romance pro-drop languages accept overt subject pronouns only when they are emphatic, BP and Finnish allow overt subject pronouns that do not carry any emphatic force. Hence, BP and Finnish arc insensitive to the Avoid Pronoun principle.

.

(11) Avoid pronoun

Empty categories have preference over overt pronoun (cf. Chomsky, 1981:65)

.

In fact. Modern BP and CF seem to prefer overt subject pronouns rather than null subjects.11 Duarte (1995, 1996) presents historical facts indicating that in BP the use of referential null subjects lias decreased considerably. Until 1918, there was a preference for null subjects, bur around 1937 this preference started falling, reaching less than 30% of the occurrence in 1992 Similarly, lleinonen’s (1995) study of CF shows that the l”PSg subject pronoun minaimã T is omitted only 12% of the rime.

Related to the inapplicability of (11) is the fact that, contrary to what happens in the Romance pro-drop languages, in BP and Finnish there is no preference for a non coreferenrial reading of embedded overt subject pronouns.

The facts presented so far suggest that neither BP nor Finnish are null subject grammars. I will argue that the incapability of licensing null subjects is to be correlated to the fact that these grammars have weak verbal morphology at least for the 3rdP. But, assuming this to be the right conclusion, referential null subjects should be disallowed altogether. However, BP allows ll'P null subjects in main clauses and Finnish allows 1stP and 2ndP (13). Moreover, 3rdP referential null subjects arc allowed in embedded clauses (14).

As I discuss below, rhese cases do not constitute real evidence that Finnish and BP still license referential null subjects. These empty subjects are rraces rather dian null pronouns, as we will presently discuss.

 3. Matrix Null Subjects: Cases of Topic Deletion

Rodrigues (2002, 2004a) has shown that BP 1”P null subjects behave as if they were topics. They are excluded from clauses with a fronted wh-phrase (15a) or a topicalized constituent (15b), and cannot occur inside relative clauses (15c).

.

(15) a. O que *e/eu fiz

what i did-1Sg

'What did i do'

b. O João, *e/eu acho que vai ser promovido

the João I think-1Sg that will be-Inf promoted

'As for João, I think he is going to be promoted'

c. Eu comprei aquele vestido [que *e/eu vi ontem]

I buy-1SgPast that dress that saw-1Sg yesterday

'I bought that dress that i saw yestersay'

.

This is characteristic of Modern BP. As (16) illustrates, the 19th- century BP did not present these restrictions. (These sentences were extracted from O Judas no Sábado de Aleluia, Martina Pena, 1844.)

.

(16) a. E o que e direi da menina

and what say-1SgFut of-the girl

‘And what am I going to say about the girl’

b. Mas, olha, o meu vestido está quase pronto, e o teu, e não sei quando estará

but took the my dress is-3Sg almost dont and the yours not know-1Sg when be-3SgFut

‘But, look, my dress is almost done, and yours, I don’t know when it will be done’ (Martins Pena 1844. O Judas no Sábado de Aleluia)

c. Que se case, e quanto antes, com a noiva [que e lhe dou]

that SE marry-3Sg and as soon as possible with the fiancée that you give-1Sg

'As soon as possible, get married with the fiancée that I give you’

.

Assuming that the 19-ccntury BP was pro-drop language, the contrast between (15) and (16) suggests that 1stP null subjects in Modern BP arc not null pronouns. In fact, these subjects behave like dropped ich 'I' in German, which occurs in the following situations: (a) it occupies the scnrcncc-inilial position (i. e., when it is a topic (17a)); (b) when spec of CP is not filled (17b); and (c) when it is not inside a relative clause (17c).

.

(17) a. (Ich) hab' *(ich) ihn schon gesehen

'I saw him already

b. * Was machte

'What did I make

c. (*Ich) kenne das mädchen, daß gestern getroffen habe

'I know the girl that I met yesterday'

.

German is not pro-drop language, thus the similarities between (17) and (15) suggest that 1stPerson null subject in Modern BP may not be pro. In fact, in (2002, 2004a) I argued that these subjects are deleted topics à la Ross (1982) and Huang (1984). That is, they are topicalized overt pronouns that turn out to be targets for deletion. Thus, they are variables bound by zero topics, and, as such, they block A’-movement of another constituent.12

I will not work out the technical details of the topic-deletion operation (see Huang 1984). The main point is that modem BP 1stP null subjects are not null pronouns. They do not have the freedom of null pronouns, being disallowed in structures with either a filled spec CP or topicalizcd phrase. This indicates that they are deleted topics.

Matrix referential null subjects in Finnish might also be instances of topic deletion. As shown in (18a), they are not allowed in structures with a topicalized constituent (cf. Vainikka and Levi, 1999), even though topicalization occurs without difficulty when the subject is a pronoun, as in (18b).

Finnish allows only one fronted topic per sentence (cf. Holmberg and Nikanne 2002). it follows then that in (18a), topicalization of Palkankorotusta ‘raise' bars topicalization of the subject pronoun mina T, which, as a consequence, is not targeted by deletion.

In addition to this, Holmberg and Nikannc (2002) and Holmberg (2003) observe that 1stP and 2ndP null subjects cannot co-occur with an expletive.

.

(19)* Sitä uskon . . . . . . . . vallankomousksen

Expl believe1stSg . revolutions-Ill

'I believe in Revolution'

.

This has led Holmberg (2003) to propose the very same analysis suggested by Rodrigues in (2002 and 2004a): null subjects in Finnish involve pronoun deletion. The idea here is that the expletive occurs in the topic position (spec of FP — Finite Phrase — in Holmberg and Nikanne’s analysis), thus blocking deletion of the subject pronoun.

One could object to extending the topic deletion analysis to Finnish since in this language 1stP and 2ndP null subjects can co-occur with a wh-phrase in spec of CP:

.

(20) Missä e voimme . . . . lukea . . . . . sanomalethiä

. . . where . . . can-1Pl . . . read-Inf . . . Newsapers-Par

'Where can we read newspapers'

.

If (20) involves topicalhation followed by deletion of the subject pronouns, we expect this process to interfere with the wh-movement. Hence, (20) should be ungrammatical. Crucially, however, Finnish does not display minimality effects in structures with both wh-move- ment and topicalhation (Holmberg and Nikanne (2002):

.

(21) Kuka tämän kirjan . . . . on . . . . . . . . . . . kirjoittanut

who . . . . this book-Acc . . have-3SgPres write-PastPart

'(What about this book:) Who has written this book'

.

I will not offer an analysis for this lack of minimality effects.13 For the purpose of the present discussion, it is sufficient to observe that whatever explains (21) also explains the possibility of topicalhing (and deleting) the subject pronoun in (20). 'therefore, 1 conclude that matrix null subjects in Finnish are also instances of deleted topics.

 4. Referential 3<sup id="superscript-14c92d80139793d3d3bcf46c2b31ff02">rd</sup>P Null Subjects: Evidence for a Movement Analysis

This section provides several sources of evidence that 3rdP referential null subjects in BP and Finnish are residues of movement rather than null pronominal categories.

4.1. Anaphoric Behavior

The only difference between the sentences below is that the null subjects are provided with a proper antecedent in (22), but not in (23). Note that in (23), the empty subjects must refer back to the matrix subject. It docs not take discourse referents. Thus, BP and Finnish 3rdP referential null subjects behave as anaphors.

In pro-drop languages, represented here by European Portuguese and 19th-century BP, null subjects are free in reference. Thus, they can appear in matrix clauses, referring to an entity mentioned in the discourse. When they appear in embedded clauses (24), they are ambiguous in reference, being either coreferential with a DP in a higher clause or related to some person identifiable in the discourse. ((24a) is from Modesto (2000), and (24b) was extracted from O Judas no Sábado de Aleluia, Martins Pena (1844).)

Chomsky and Lasnik (1993) remark that a [+anaphor] empty category, i.e. NP-trace, gets its reference from an antecedent, while a [+pronoun) empty category, pro/PRO, can have reference on its own, although its reference may also be determined by an antecedent Thus, the null subjects in (24) are [+pronoun] empty categories (pro), whereas the subjects in (22) are identified as [+anaphor] empty categories, therefore as NP-traces, according to Chomsky and Lasnik’s classification.

 4.2. Locality

Chomsky (1995:311) suggests that the definition of Move incorporates the property in (25), where closeness is defined, as in (26):

.

(25) Minimal Link Condition (MLC)

K attracts α only if there is no β, β closer to K then α, such that k attracts β.

.

(26) Closeness

β is closer to K than α if β c-commands α, and β is not in the same minimal domain as τ or α, where τ is the target of raising. (Chomsky 1995: 335-6)

.

In BP and Finnish, the antecedence relationship between a 3rdP null subject and a DP obeys the MLC, the antecedent being the closest c-commanding DP. In (27a,b), for instance, the DPs with index 1 cannot be the antecedents because they do not c-command the null subjects. In (28a,b), the matrix subjects fail to be rhe antecedents because the subject of the intermediate clause is closer to the null subjects:

Again, in the pro-drop grammars, when a null subject refers to a DP in the sentence, the relation is not subject to the MLC. Witness the acceptability of (29) and (30). ( (29) is European Portuguese (Modesto, 2000),14 and (30) is 19th-century BP, extracted from O Judas no Sábado de Aleluia, Martins Pena, 1984.)

(27) and (28) follow straightforwardly if BP and Finnish 3rdP referential null subjects are formed by movement. The antecedent of these subjects must be the closest c-commanding DP because MLC is a condition on movement (28) is parallel to super-raising structures like (31). In (28) , o João/Jukka cannot be the antecedent because movement to the matrix subject position would cross over the subjects of the intermediate clauses. Similarly, in (31), the subject of the intermediate clause intervenes, blocking the movement of John to the matrix clause.

.

(31) *John1 seems that it is likely t1 to win

 4.3 The resumption test

Relative clauses are strong Islands in Romance. But, as observed in Rizzi (1982), Chomsky (1981) and Jaeggli and Safir (1989), these languages have a resumption strategy that salvages subject extraction from relative clauses via insertion of a resumptive pro, as the Spanish sentence in (32) illustrates.

.

(32) Ese es el tio que1 María conoce a la mujer [con quien pro1 se casó]]

this is the guy that Maria know-3Sg to the woman with whomSE married-3Sg

'That is the guy that Maria knows the woman who he married'

.

By comparing Spanish with BP, we can find additional suggestive evidence that null subjects in modern BP do not behave like null subject pronouns. In (33), the empty subject fails the resumption test. As in English, only overt pronouns can be used in a resumption strategy in BP:

.

(33) Esse é o rapaz que1 Maria conhece a garota2 que2 ele1/*e1 beijou t2

this is the guy that the Maria know-3Sg the girl that he kissed-3Sg

'This is the guy that Maria knows the girl that he kissed'

.

Moreover, the contrast below show's that Spanish non-resumptive null pronouns can freely occur inside relative clauses, whereas in BP, they cannot

Thus, clearly embedded null subjects in modem BP pattern like wh-traces:

.

(35)* Quem1 que o João viu a garota que t1 beijou na noite passada

who that the João saw-3Sg the girl that kissed-3Sg in the night last

'Who is the x such that João saw the girl that x kissed last night'

.

The traditional view of restrictive relative clauses involves adjunction: the embedded CP is an adjunct to the maximal projection of the NP, which is taken as the nominal head of the relative clause (Cf. Safir 1986, Browning 1987 and Fabb 1990). However, following Vergnaud (1974), Kaync (1994) argues for a raising analysis of relative clauses, according to which a relative CP is a complement of D° and the gap inside die relative clauses is formed by movement as illustrated in (36):

(36) [DP the [CP [DP picture2 [D' which t2]1[C' [Bill saw t1 ]]]

Due to space limitations, I will not discuss these two analyses, rather I will assume the raising analysis, suggesting that BP 3rdP referential null subjects occur inside relative clauses for locality reasons.15 According to the raising analysis, (37a) lias the structure sketched in (37b).

In (37b), the movement of the DP o João towards the matrix clause has to cross over the relativized NP. Therefore, given that movement obeys locality, (37) is ungrammatical.

Finnish, differendy from BP, allows null subjects inside restrictive relative clauses.

Interestingly, Finnish is similar to BP in blocking A’-movement out of relative clauses. This is unexpected given the analysis I am proposing for Finnish null subjects.

Importantly, however, Finnish allows topicalization to co-occur with wh-movement in both free and relative clauses.

Therefore, it might be that in this language, a restrictive relative clause has the structure in (41), where die NP heading the relative clause moves to spec of CP, and the subject also moves to the left periphery, arguably to a topic position (call it TopP) between TP and CP.

(41) [CP NP1 [TopP NP2 [TP t2 V t1]]]

If this is right, (38a) is compatible with the topic deletion operation 1 have proposed in section 2 for 1stP null subjects. During rhe derivation of tliis sentence, the 1stP overt pronoun minä moves to spec of TopP, becoming thus the target of deletion.

The fact that a null subject is allowed in (38b) would seem to suggest that in Finnish, the antecedence relationship between a null subject and a DP does not obey the MLC. However, this conclusion is unlikely to be correct because as we have seen above this relationship is independendy restricted by locality, the antecedent being the closest c-commanding DP. Therefore, it must be the case that in (38b) the nominal head of the relative clause is not actually intervening between the null subject and the DP Pekka.

One interesting aspect of restrictive relative clauses in Finnish is that the relative pronoun and the noun that heads the relative clause are morphologically marked with different Cases. In (38b), for example, the relative pronoun is marked with the Case checked by the embedded verb (partitive) and the nominal head realizes the Case checked by the matrix verb (accusative). In addition, the head of the relative clause might surface separated from the relative CP (Cf. Sulkala and Karjalainen 1992 and Helasvuo 1994).

Thus, it is plausible that in Finnish relative clauses the nominal head moves from the relative CP to a Case position in the matrix VP shell, as sketched in (44).

The question is about the empty subject inside the relative clause. Can it be a trace of the DP Pekka? (44) differs from (37) in that in (44) the movement of Pekka towards the matrix crosses over a trace, whereas in (37) movement o João crosses over an NP. Thus, the conclusion seems to be that traces do not block extraction of a lower constituent. This raises an issue about the MLC. If our analysis of (44) is right, traces are not counted in defining closeness. This accords with Chomsky’s (2001) idea that only the head of an A-chain (i.e., the whole chain) blocks movement under the MLC. Note that movement of Pekka to die matrix clause might happen before the movement tytön. Therefore, it is possible that Pekka’s movement docs violate the MLC derivationally, though not representationally. This leads to the conclusion that this condition is a condition on representations, as already suggested by Chomsky (2001), among others.

 4.4. Obligatory Control Properties

Hornstein (1999, 2001) proposes that obligatory control configurations (OC) are formed by movement. Thus, (45a) has the structure in (45b). The DP John is first merged as the external argument of the verb win. After that, John moves to the spec of the embedded infinitival TP to check the relevant EPP feature. When the matrix VP is built, the same DP moves to spec of the matrix clause, being, thus, interpreted as both the winner and the hoper. Finally, a TP is projected on the top of the structure and John moves to spec of TP, where it checks its Case feature.

.

(45) a. John hopes to win

b. [TP John1 [T' [VP t1 [hopes [TP [t1 to [T' [VP t1 win]]]]]]]]

.

It is worth recapitulating Horsntein’s arguments for a movement analysis for obligatory control. First, the controlled element is anaphoric, taking the closest c-commanding NP as its antecedent; second, split antecedents are not allowed, third, a de se reading is forced, and fourth, if the closest c-commanding NP has die format then only-NP, then only a covariant interpretation is possible.

Interestingly, Modern BP and Finnish 3rdP referential null subjects display the OC properties listed above. First, as we already saw, BP and Finnish referential 3rdP null subjects require a sentential antecedent, which must be the closest c-commanding DP (Cf. (22)-(23) and (27)-(28)). Second, Figueiredo Silva (1996) reports a prohibition against split antecedents in modern BP (46a), and Gutman (1999) and Vainikka and Levi (1999) report the same prohibition in Finnish (46b). Note that the sentences are acceptable if the embedded subjects are overt pronouns. Hence, overt pronouns allow split antecedents, but null subjects do not:

3rdP null subjects behave as OC PRO also with respect to VP ellipsis. As discussed in Negrão (1999), only a sloppy reading is available for the null subjects in (47a). (47b) shows that the same interpretative restriction holds in Finnish, and (48) illustrates that overt subject pronouns allow a strict reading.

With respect to de se interpretation, the embedded null subjects of BP behave exactly like an OC empty category.17 For instance, in the context given in (49), the BP statement in (50) is false. For (50) to be true, it demands a de se belief (cf. Castaneda 1966, Salmon 1986 and Chierchia 1990). That is, Ronald Reagan must be aware that it is himself that he believes to be the fortieth president of the United States. Using Cole’s et al. (2001) way to put it, we say that under a de se reading the protagonist of the event actually ascribes, or is disposed to ascribe, to himself/hcrself the property denoted by the predicate containing the empty category or anaphoric element.

.

(49) Because of his Alzheimer’s disease, Ronald Reagan does not remember who he was. One day, reading the newspaper, he reads the headline ‘Reagan was the fortieth President of the United States’. After having finished reading the article, Ronald Reagan comes to know that the person called Reagan was the fortieth President of the United States.

.

(50) O Ronald Reagan1 sabe que e1 foi o quadragésimo presidente dos EUA

the Ronald Reagan know-3Sg that was-3Sg the fortieth president of the USA

‘Ronald Reagan knows that he himself was the fortieth president of the United States’

.

Although subtle, there is a contrast between (50) and (51) below. (51) is ambiguous. The pronoun ele ‘he’ might refer back to Ronald Reagan, forcing us to attribute a de se knowledge to this entity, and judge die statement as false given the context in (49). However, the pronoun might refer not to Ronald Reagan, but to Reagan, the name in the headline. Following Higginbotham (1992a), let me indicate the concept of Reagan as [1/2the person whom Ronald Reagan is reading about1/2]. If ele refers to Reagan, a de se reading does not emerge.

.

(51) O Ronald Reagan sabe que que ele foi o quadragésimo presidente dos EUA

the Rond Reagan know-3Sg that he was-3Sg the fortieth president of.the USA

'Ronald Reagan knows that he was the fortieth president of the United Stated’

.

We might use Chierchia’s (1990) contradiction test to verify the difference between null and an overt subject pronouns. Take (52) as an example. (52a) yields a contradiction, whereas (52b) does not. In (52a) the null subject induces a de se reading, which is contradicted by the content of the conjoined sentence. The contradiction disappears in (52b) because the overt subject pronoun does not force a de se reading.

Traditionally, attitude report verbs are taken to express relations between agents and propositions (cf. Salmon, 1986). Hence, in the sentence above, the matrix verb saber ‘know’ expresses a knowing relation between Ronald Reagan and the proposition ‘x was the fortieth president of USA’. Thus, if the embedded subject is an overt pronoun, the x in the proposition can be valued either as Ronald Reagan or as Reagan. If it is valued as Ronald Reagan, then Ronald Reagan is in a knowing relation with a proposition about himself; therefore, this entity is in a de se knowing relation. Conversely, if x is value as Reagan, Ronald Reagan is not in a de se knowing relation, for he knows a proposition that is about somebody else, mainly die person whom he read about.

Having said that, the question is about null embedded subjects. Why does their presence force a de se reading? The movement analysis proposed here gives us an answer. In the structure of (50), the embedded sentence is [t foi o quadragésimo presidente dos EUA] ‘[t was the fortieth president of USA]’, where t is a trace of Ronald Reagan, the subject of the matrix clause. Therefore, (50) asserts that Ronald Reagan is a knowing reladon with the proposition ‘Ronald Reagan was the fortieth president of the United States’. In other words, Ronald Reagan is in a knowing relation with a proposition about himself.

Consider now another property of OC: the absence of an invariant for cw/y-NPs. The pronoun A in (53) can take as its antecedent cither John (54a) or the antecedent is die whole DP only John (54b). 'lTicsc two co-indexations lead to two different readings: (54a) is assigned the covariant interpretation in (55a) and (54b) gets rhe covar- iant interpretadon in (55b) (cf. Higginbodiam, 1992b).

.

(53) Only John expects [that he will win

.

(54)a. [Only [John]1]2 expects [that he2 will win]

b. [Only [John]1]]2 expects [that he1 will win]

.

(55)a. Only John is an x such that x expects x will win (Covariant interpretation)

b. Only John is an x such that x expects he, John, will win (Invariant interpretation)

.

As pointed out in Negrão (1999), the BP sentences in (56) also receive different interpretations. (56a) is assigned the covariant interpretation in (57a), and (56b) is assigned the invariant interpretation in (57b):

(57) Only Maluf is an x such that x thinks that x will win the elections

b Only Maluf is an x such that x thinks that he, Maluf, will win the elections

.

Thus, BP 3rdP null subjects cannot be co-indexed with the NP contained in the only-NP phrase. It must be co-indexed with the whole only-NP phrase, as represented in (58a). Conversely, an overt pronoun must be co-indexed with the NP contained in the only-NP phrase, as in (58b). These restrained co-indexations are responsible for the readings in (58): (58a) yields (57a), whereas (58b) yields (57b).

As pointed our in note 11, BP grammars generally obey Montalbetti’s constraint, prohibiting co-indexation between an overt pronoun and a quantifier-like expression. Therefore, if the phrase only-NP is a quantified NP, the impossible co-indexation in (58b) might fall within the realm of Montalbetti’s constraint This interaction can be tested. As I mentioned earlier (note 11), in some dialects of BP the quantifier expression lodo-NP ‘every-NP’ allows co-indexation with an overt pronoun. Therefore, if todo-NP can violate Montalbetti’s constraint, it can be co-indexed with an overt pronoun, triggering, thus, a covariant interpretation, this is shown by the ambiguity of (59), which, similarly to the English sentence in (53), can receive either the covariant (60a) or invariant reading (60b).

.

(59) [toda fã da [Carla Peres]1]2 acha que ela1/2 deve agir como esposa do Xande

every fan of the Carla Peres think-3Sg that she have-3Sg act-Inf as wife of Xande

'Every fan of Carla Peres thinks that she has to behave as Xande’s wife’

.

(60) a. x(x=a fan of Carla Peres) x thinks that x has to behave like Xande’s wife

b. x(x=a fan of Carla Peres) x thinks that Carla Peres has to behave as Xande’s wife

.

In sum, the co-indexing relations between an overt subject pronoun and the terms of an only-NP, BP is different from English only in that the former, but not the latter, is subject to Montalbetti’s constraint.

We should now ask ourselves about the impossibility of an invariant interpretation for (56a). According to the analysis suggested here, the gap in the embedded subject position of (56a) is a trace of the matrix subject. Thus, the antecedent cannot be o Maluf because this DP does not c-command anything outside the containing phrase só o Maluf. Thus, the conclusion is that the antecedent (the movement element) in (56a) must be the whole quantifier phrase só o Maluf. This explains why (56a) accepts only a covariant reading.

Finnish is not subject to Montalbetti’s constraint, as shown in (61).

.

.

Hence, (62a) receives cither an invariant or a covariant reading; ban/se ’he’ being co-indexed either with Jukka or with min Jukka, in (62b), as predicted by a movement analysis, only a covariant interpretation is possible; the gap in die embedded subject position must take the whole quantifier phrase vain Jukka as its antecedent.

Treating BP and Finnish 3rdP referential null subjects as formed by movement has the advantage of explaining the OC properties discussed above. It also explains why these null subjects are disallowed with relative clauses and why dieir antecedents must be the closest c-commanding DP in accordance with the MLC.18

The next section offers a way of implementing the analysis proposed here, correlating subject-to-subject movement with loss of verbal agreement morphology'.

 5. Technical Implementation 5.1. Verbal Agreement Morphology and Null Subject Pronouns

Since Taraldsen (1980) there has been the intuitive idea that the licensing of referential null subjects is somehow related to verbal agreement morphology. Recently it has beet» proposed that in null subject languages with rich agreement morphology, spec of TP/AgrP might not be realized by pro or any other material because the verbal agreement morphology carries a D feature, being, thus, able to satisfy the EPP features of these non-substantive categories. For developments of this idea, see Barbosa (1995), Alexiadou and Auagnostopoulou (1998), and Kato (1999, 2000), among others.

Alexiadou and Anagnostopoulou distinguish two types of verbal agreement systems: a weak and a strong one. In weak system, verbal agreement affixes (Agr, henceforth) are not independent units of the computational component, entering derivations already attached to their hosts. A weak Agr, therefore, is arguably not manipulablc by syntactic operations. In a strong Agr system, insrances of Agr are listed in the lexicon as separate lexical items, and enter the computational component as independent syntactic units that can be used by syntactic operations. Tn particular, it is suggested that a strong Agr enters a derivation furnished with a D-featurc, f-featurcs and perhaps with Case feature. Hence, syntactically it is expected to function as a pronoun. The authors were concerned mainly with null expletive constructions, and their suggestion is that in grammars with a strong agreement system, a null expletive pronoun is Agr itself. 'Ihc assumption is that a strong Agr might be merged directly on die head of die Agreement phrase (AGKP), sadsfying, thus, the EPP feature of AGR. Hence, when the verb adjoins to AGR, AGR projects, but since its EPP feature had already being sarisfied, it docs not project a specifier. Though they did not explore it, Alexiadou and Anagnostopoulou raise the possibility of replacing referendai pro by Agr, presupposing that Agr counts as a theta bearing argument in null subject grammars. Kato (1999, 2000) puts this possibility forward, suggesting that in Romance pro-drop languages, a sentence like (63a) is derived as in (63b), Agr enters die derivadon as the external argument of VP and, being a maximal minimal projection, it adjoins to T°, where it checks the EPP features of T and its own Case feature. As a result, spec of TP is not projected.

(63)a. viajó

traveled-3Sg

'S/he traveled'

b. [TP [T [T viaj2-[Agr o]1 + T][VP t1 t2 ]]]

As Karo observes, this analysis presupposes that DPs and overt pronouns in subject positions are instances of doubled subjects in the sense that they double Agr, being generated at the left periphery of the clause, possibly receiving a default Case.

Assuming this analysis to be on the right track and considering the correlation between loss of referential null subjects and weak verbal agreement morphology' in BP, I hypothesize that in these grammars Agr became f-defecrivc, being unable to instantiate person & number distinctions (for discussion, see Galves 1996 and Kato 1999) and, as a consequence, was reanalyzed as part of the verb, losing its syntactic independence. However, 1 presuppose that BP and Finnish Agrs still have a D-fearurc. Hence, in these grammars, when V adjoins to T°, carrying Agr, Agr satisfies the F.PP feature of T, but, being f-dcfective, is unable to delete the f- features of T.19

The assumption that Agr is able to satisfy the F.PP feature of T is motivated by the fact that null expletives are still allowed in present- day BP and in Finnish. Thus, if there is no pro, in (64), the EPP feature of T is satisfied by Agr.

If this is right, it follows that in a finite clause with a referential null subject, a f-complcte item must be inserted in the complement domain of T, such that the theta-role of the verb and the f-features of T can be checked. But, the [spec, TP] need not to be projected, since Agr is able to satisfy the EPP feature. Thus, if this is right, in the derivation of sentences like (65), the DP o João/Jukka needs to be first merged within the embedded clause, and then moved to the matrix clause. Moreover, this movement may not involve [spec, TP|.

5.2. Deriving Embedded Null Subjects by Movement

If the 3rdP null subjects in (65) are the result of movement, then two questions arise: (i) How can a DP receive more than one theta- role? (ii) How can a DP check more than one structural Case? The first question can be answered by assuming with Boskovic (1994), Boskovic and Takahashi (1998), Lasnik (1995), Ilornstein (1999, 2001) and Rodrigues (2004b), among others, that movement into theta-positions is allowed. The second reason is more complex since it is standardly assumed that a DP checks structural Case only once.

I provide an answer to this question without overgencrating. I assume that Case-checking accords to (66):

.

(61) a. A structural Case feature is checked in a spec-head relation

b. Agreement in f-features prompts movement to check Case

.

(66a) revamps Chomsky’s (1993) proposal, according to which a DP can check its Nominative Case against a functional Category only if it had moved to the spec of that functional category, checking thus the EPP feature related to the functional head. For example, a DP checks Nominative Case against T only after having moved to spec of TP, checking die EPP feature of T In his recent proposals (2000 and thereafter) however, Chomsky dissociates Case checking from EPP checking, and correlates Case with agreement in f-features. This correlation is expressed through the following technical mechanism: if a functional head H (a probe) has uninteqiretable f-fcatures, II probes its domain looking for a goal, an element with an identical (identical in die choice of the features, not in their values) set of f-fcatures. As soon as the goal is found, the operation Agree takes place, deleting the f-featurcs of II and the Case feature of the goal. If II also has an EPP feature, the goal is pied-piped and merged widi HP.

I maintain the correladon between agreement in f-fcatures and Case checking, but 1 retain die ‘old’ idea that structural Case is checkable only in a spec-head configuration. To do so, 1 assume (6Gb): a goal can move to the specifier of a probe and check its Case feature only if they both agree in f-features. Intuitively, the idea is that Case is a feature that allows an clement to move to a functional projection. What f-fcature agreement does is to identify the element before allowing the actual movement to take place.

To give a concrete example of how this works, consider die sentence in (67a). The relevant step in die derivation of (67a) is represented in (67b). John was merged widi VP, checking the dicta role of V. When T was inserted, T established agreement with John, and the f-featurcs of T were deleted. But die Case feature of John was not deleted because it requires John to enter TP. The movement per se has already been authorized since T and John agreed in f-fcatures. Note that the raising of John is also demanded by die EPP feature of T. When John moves to spec of 'IT, its Case feature and die EPP feature of T are checked, forming a convergent phrase marker (67c):

If this is on the right track, 3"1? null subjects in BP and Finnish arc derived as exemplified below. The derivation of (65a) (= (68)), starts widi the numeration in (69).

(69) NUm = {T2, o1, João1, disse1, que1, comprou1, um1, carro1 }

The verb comprou ‘bought’ merges with its internal argument, the DP um carro ‘a car’, wliich checks the Thcta-rolc of the verb. Next, die DP o João ‘die John’ is built and merged with VP. At this step of die derivation, the embedded VP (70a) is formed. After diat, T is selected and the verb adjoins to T-, carrying Agr diat has a defeedve set of <{>-features and a D-feature. Thus, Agr checks die EPP feature of T, but not the (^features, which prohe the complement domain of T, looking for a goal. The DP o João ill spec of VP is localized and die agree operation takes place, deleting the (^features of T (70b):

At this point of die derivation, die system can either move o João to spec of TP or continue the derivation by merging the next item in the numeration, namely the complementizer que. 'laking greed as enlightened self-interest, the movement of o João to Spec TP is legitimate; though T does not have any feature requiring the movement. the Case feature of DP demands die movement. But, if the DP is moved, it becomes inactive to further computations, and the derivation will crash at the matrix level because there will be no item available to saturate die external argument position of the matrix predicate and check die f-features of T. On die odier hand, if the system selects and merges the complementizer qut, delaying the deletion of the Case feature of o João, the derivation converges. At the stage represented in (71), the numeration contains only a T. Thus, the system must apply Move, moving o João to the matrix spec of VP.

When the matrix T is inserted, the matrix verb adjoins to T, Agr checks die F.PP feature of T, but fails to delete the (^ features of T. These agree with the copy of o João in the matrix Spec VP and the DP is moved to [spec, TPJ, where its Case feature is checked.

Consider now a case in which an overt pronoun is inserted as the subject of the embedded clause, as in (73):

The first relevant step of the derivation of (73) is die one in (74a). The embedded clause has been constructed and the pronoun ele was selected from the numeration and inserted as the external argument of the embedded verb. When T enters into the derivadon, the verb adjoins to T and Agr checks the EPP features of T, but not the <j>- features. Hence, T is free to establish agreement with ele. As a consequence, the pronoun can move to [spec, 'IT] and check its Case. Tf die system did not apply movement at this point, the derivation would not converge. The Case feature of ele would remain unchecked since at the matrix level the DP o João is inserted as the subject. Therefore, the only way to ensure convergence is by moving the pronoun ele to die embedded spec of TP, forming the CP represented in (74b). At the matrix level, o João is inserted in spec of VP, and then moving to spec of TP to check its own Case feature (74c).

In brief, the analysis proposed here accounts for the presence of null subjects in BP and Finnish by taking rhem to be formed via category movement The loss of referential null subjects in these grammars is due to die fact that Agr lost its status as an independent D item, being demoted to a verbal affix, conserving its D-feature, but having undergone degradation of f-featurcs. Movement of die external argument of an embedded VP to the subject position of an immediately higher clause is possible because, in die current stage of BP and Finnish, V moves to T, carrying Agr, which checks the EPP- fcaturc of T.20,21

 6. Conclusions

In this paper, I have explained the loss of referential null subjects in Brazilian Portuguese by correlating it with the loss of verbal agreement morphology. The subject gaps allowed in finite clauses are residues of movement rather than null pronouns. Matrix null subjects are formed via movement to a topic position plus topic deletion. Embedded 3rdSg null subjects are formed via subject-to-subject movement, which is possible because the verbal agreement morpheme retained the D-fcature, though underwent f-feature degradation.

The idea of movement defended here raises some theoretical challenges: If theta-roles are the result of configurations, and movement into theta-positions is not allowed (Hale and Keyser 1993, Chomsky 2000), it is difficult to see how one could analyze the phenomenon depicted here. Moreover, if the Chain Condition is part of UG (Chomsky and Lasnik 1993),22 that would present problems for the subject-to-subject movement used here. To that extent, the analysis given here is evidence against both the Chain Condition and the configurational view of theta-roles.

* This paper is a short version of Chapters 3 and 4 of my rhesis (Kodrigues, 2004a). Thus, I am thankful to Nocbert 1 lornstein and David Lighlfoot for their guidance during my PhD. I am also grateful to Jukka Enqvist, Tor Aschan, Tarja Hcinonen and Urpo Nikannc for their judgments and comments on Finnish data.

The term partial pro-drop has been ambiguously use to characterize two types of grammars: (a) grammars like German where expletive subjects can be omitted, even though referential null subjects cannot; and (b) grammars that drop referential subjects, bur only in a restricted set of configurations. In this paper, I shall employ the term partial pro-drop unambiguously in reference to grammars of type (b), which are also called semi null subject languages.

The term ‘trace’ should be understood as a short-hand form for ‘silent copies formed by copy-and-delction.’

In Finnish, there is no verbal suffix for the future tense.

Synchronically and diachronically Finnish follow's a general /ti/ —>/si/ rule of assibilntion.

Jaeggli and Saiir (1989:30) state morphological uniformity in the following wag.

(i) Morphological Uniformity

An inflectional paradigm P in a language L is morphologically uniform iff P has either only undciived inflectional forms or only derived inflectional forms.

61 will distinguish Colloquial from Standard Finnish whenever necessary, otherwise 1 will refer to both dialects as Finnish.

For a discussion on null expletive in BP, see Viotti (1999).

Finnish is a SVO language that avoids verb iniual declarative sentences (cf. Vilkuna 1989 and Vainikka 1989). llius, existential and generic constructions are acceptable as longas a locative adverb or an indefinite DP is fronted.

Guimarães and Rodrigues (2002) suggest that the locative adverb functions as the sentential subject of these generic sentences. This would explain the obligator)' presence of this adverh and also the fact that it. has to precede the verb.

SOSis the partitive form of rt,

In DP there arc particular eases in which this preference either docs not arise or is blocked:

(a) null object pronouns are allowed in a broad range of configurations (c£ Cyrioo, 1996);

(b) in accordance to Montalheni’s (1984) Overt Pronoun Constraint, a >nfigurations involving coindexing with an A-bar antecedent block overt pronouns. In (ia), for instance, the overt pronoun tie ‘he’ must be free in reference. For a discussion of Montalbctri’s constraint in BP, see Negriio (1999): It is worth mentioning, however, that a quantifier expression with die format a*&-NP ‘every NP* can bind an overt pronoun (ic).

(i) a. Quem,/ninguém, acha que ele.1/2 é um gênio

who} nr,Indy tbink-iSg that he is-lSga genius

‘Who/nobody thinks that he is a genius’

b. Toda criança, acha queela,/2é um gênio every cf»U fhink-3Sg that sbt a genius

‘livery child thinks that she is a genius'

Co-indexing with an overt pronoun is also blocked in coordinated structures as

(ii) , though these structures arguably involve an A-autecedent (die matrix subject). See Rodrigues (2004a), where it is proposed that these are cases of coordination of VPs.

(i) O João, cantou c clc.,rt dançou the João sang-3Sg and be danced-3Sg ^oão sang and he danced*

A topic deletion analysis can also explain the possibility of having BP referential 3rdPSg null suhjects in question-answer contexts like (i) (cf. Modesto, 2000).

(i) A: OJoão telefonou? B: t telefonou

the João caUcd-3Sg calltd-3Sg

‘Did João call’ *He called’

Holmbctg and Nikanne propose rhat in (21) wh-mnvement and topic-movement do not interact because they are driven by different features.

European Portugese has strong preference for the readings in which the null subject is cointerpreted with the DP with index 1 (Joao Costa, personal communication). See Calabrese (1986) for Iralian. Thus, die real difference between null subject grammars and BP and Finnish is dial in the former the co-indexation shown in (29) is possible diough not preferred, while in the latter it is impossible.

For arguments in favor of a raising analysis sec Kayne (1994) and Bianehi (1999).

This is spoken Finnish. To avoid an overflow of irrelevant information, 1 cut off hesitations and repetitions.

It is still unclear to me whether 3rdP null subjects in Finnish forces a dt se interpretation or nor Thus, with respect to diis topic, I will discuss only BP.

I am omitting here other arguments offered in Rodrigues (2004a) in favor of a movement analysis of BP 3,JP referential null subjects. These arc arguments arc: BP 3,4P null subjects arc not allowed within embedded clause headed by the connective com (cf. (i), which Torrego and Uriagcrcka (1995) analyze as an instance of para tactic dependency, that is structures in which die embedded clause remains structurally disconnected to die main spine of the tree throughout the derivation, (ii) In BP a past participle or a floating quantifier within the c-command domain of die embedded null subject must agree in gender with the antecedent of the null subject (cf. (ii). ’litis gender agreement is not observed in full pro-drop languages.

(i) OJo»o,vai ver comocle/^e, tem razão

the João will-3Sg see-inf bow bt bas-3Sg right ‘João will see how he is right.’

(if) a. A vitima, disse que r, foi atacada/*atacado

the vidim-Ftm said-3Sg that was-3Sg attatked-Fmfattacked-hiasc The victim said that she was attacked’ b. As vítimas, falaram que e, toda.sI/??todos1 vao depor the v'utms-Tem said-3P/ that all-fsmf aU-Musc W///-3P testify-!nf The victims said that they will all testify at die same day*

On the f-fcatures of T, see Chomsky (2000).

For alternative analyses, see Kato (1999),\h>desto (2000) and Ferreira (2000) on BP, and Vainikka and Levy (1999) on Finnish. Sec also Rodrigues (2004) for a critical review of these analyses.

3rdP null subjects arc also observed inside finite adjunct clauses. Rodrigues (2004a) accommodates this under a movement analysis by assuming sidewards movement (Nunes, 1995, 2001). Rodrigues also suggests that eases like (i) are an instance of null subjects inside adjuncts clauses. In (i), the in-situ objects fail to be the antecedents, but the fronted object is a perfect antecedent. (i) a *OJoSo convenceu aMaria2quc f2 tinha de sair lhe João conmcd-3Sg the Gloria that bad-3Sg of lum-lnf ‘João convinced Maria that she had to leave’ b. A Maria^ o João convenceu l2 que el tinha de sair the Maria, the João conrincrd-3Sg t)>at bad-3Sg of Itavt ltf ‘As for Maria, João convinced her that she had to leave* Modesto takes diis to be evidence that/ro in BP is a h< >und variable at Lh’ taking die closest A-bar phrase as its antecedent. 'lbereforc in (ia), though the object is the closest potential antecedent, it cannot be die antecedent because it is not in an A-bar position. I lowever, as Rodrigues (2004a) points out die embedded clause of (i) mighr be an adjunct, which is not c-commanded by the matrix object position. First, wh-cxtraction from inside this complement is not possible (joa); second, as shown by Ferreira (2000), the subjoct of the complement clause can be an epithet rcferriiigbackto die matrix object (lib). Ifephhets arc subject to Principle C, in (jib) the matrix object does not c-cotnmand die embedded subject: (ii) a. ?? Quem o João convenceu a Maria que /vemhojc? who tbf João cnnvinced-3Sg tbt A{aria that (ome-3Sg today * Who did João convince Maria diat he is coming today?’ b. O Ira convenceu o Diogt>2 que o bobão, não deveria comprar o carro tbe Ira convimed-3Sg the Dingo that the sillj not should-3Sg buy inf the car ‘Ira convinced Diogo that the silly shouldn’t buy the car’ The contrast in (i) is similar to parasitic gap (PG) constructions, in PG constructions a resumptive pronoun cannot be inserted as the PG licenser (iia) (Chomsky 1982). The same is true for the constructions in (i). As pointed out by Modesto, if a resumptive pronoun is inserted in matrix object position (iib), the DP a Mann cannot be the antecedent anymore (iiib): (iii) a. El rcloj dc que me Iwbfctfic, epic (*ki) lian conseguidu anrgar /sin mover 1XJ, ha q «tUki muy ban Mhc dock you spoke to me al>oui, which they got to fix (it) without moving now weeks vety wcJT b. * A Mariano João convenceu el^ que e.l linlia dc sair the Maria the João (on\inctd-3Sg Ur tfsat b<ul-3Sg of ieavt-lnf *As for Maria, João convinced her diat she had to leave' Besides, as in PGs (iva), the gap in (i) cannot occur within an island inside an adjunct clause (ivb):

(i) a. * Which book did you read t (before John asked Bill (whether you review PG|| b. *Qucmr o João convenceu i} [que a Maria perguntou o Paulo [se e2 pcxlia sairjj who the Joan a?nmced-3Sg that ike Maria a.fks,d-3Sg the Pauli f could-3 Sg kavtAnf ‘ Who did João convince him that Maria asked Paulo if he could leave’

If the constructions in (t) are eases of parasitic gaps, it is possible to analyze them as formed via movement, as proposed by Nunes (1995, 2001).

The Chaia Condiuon states that an A-chain has its foot in a dteta-position and its head on a Case position. Hence, it is evident that the movement analysis ! suggested here is incompatible with such a condition. The null subjects under question arc taken to be formed by movement from and into theta-positions.

 491 539 16 1998 Alexiadou A Anagnostopoulou E Natural Language and Linguistic Theory Parametrizing Agr: Word Order, V-Movement, and EPP CHecking MIT 1995 Barbosa P Null Subjects Berlim Mouton de Gruyter 1999 Bianchi V Consequences of Antisymmetry: Headed Reative Clauses 375 416 17 1986 Borer H Linguistic Inquiry I-Subjects Dordrecht Kluwer 1989 Borer H Jaeggli O Safir K The Null Subject Parameter Anaphoric Agr. 247 286 24 1994 Boskovic z Linguistic Analysis D-structure, Theta Criterion, and Movement into Theta-Positions 347 366 29 1998 Boskovic z Takahashi D Linguistic Inquiry Scrambling and Last Resort MIT 1987 Browning M Null Operator Consructions MIT Working papers in Linguistic 8 Cambridge, Mass MIT 1986 Calabrese A Pronomina 130 157 7 1966 Castañeda H. N Ratio 'He': A Study in the Logic of Self-Consciousness Dordrecht Foris 1990 Chierchia G Bartsch R van Benthem J Van Emde Boas P Semantics and Contextual Expression Anaphora and attitudes de se. Dordrecht Foris 1981 Chomsky N Lectures on Government and Binding Cambridge, Mass. MIT Press 1982 Chomsky N Some Concepts and Consequences of the Theory of Government and Binding Cambridge, Mass. MIT Press 1993 Chomsky N Hale K Keyser S. J The View from Building Twenty A Minimalist Program for Linguistic Theory Cambridge, Mass. MIT Press 1995 Chomsky N The Minimalist Program Cambridge, Mass. MIT Press 2000 Chomsky N Martin R Michaels D Uriagereka J Step by Step: Essays on Minialist Syntax on Honor of Howard Lasnik Minimalist Inquiries:the Framework 1 52 Cambridge, Mass. MIT Press 2001 Chomsky N Kenstowicz M Ken Hale: A Life in Language Derivation by Phase Berlin Walter de Gruyter 1993 Chomsky N Lasnik H Jacobs J von Watechow A Sternefeld W Vennemann T Syntax: An International Handbook of Contemporary Research The Theory of Principles and Parameters New York Academic Press 2001 Cole P Hermon G Huang C. T. J Cole P Hermon G Huang C. T. J Long Distance Reflexives. Introduction Universidade de Campinas 1995 Duarte M. E A Perda do Princípio Evite Pronome no Portugues Brasileiro Campinas Universidade de Campinas 1996 Duarte M. E Kato M Roberts I Portugues Brasileiro: Uma Viagem Diacronica Do Pronome Nulo ao Sujeito Pleno 57 77 26 1990 Fabb N Journal of Linguistics The Difference between English Restrictive and Nonrestrictive Relative Clauses Campinas Universidade de Campinas 2000 Ferreira M Argumentos Nulos em Portugues Brasileiro Campinas Editora da Unicamp 1996 Figueiredo Silva M. C A Posição do Sujeito no Português Brasileiro 1996 Galves C Roberts I Kato M O Enfraquecimento da Concordânia no Português Brasileiro Ms. University of Maryland College Park 2002 Guimarães M Rodrigues C Locative Deteriner and Argumental Pseudo-Adverbs University of Delaware 1999 Gutman R Null-Subjects: A Theory of Syntactic and Discourse Identification Cambridge, Mass. MIT Press 1993 Hale K Keyser J Hale K Keyser J The View from Building Twenty On Argument Structure and the Lexical Expression of Syntactic Relations Helsinki Otava 1979 Hakulinen A Suomen kielen rakenne ja kehitys [The Structure and Development of the Finnish Language]. Neljäs, korjattu ja lisätty painos. 1994 Helasvuo M. L Holmberg A Nikanne U Are relative clauses either restrictive or non-restrictive? A study of relative clauses in the Finnish Pear Stories Cambridge, Massachusetts Ms, MIT 1992a Higginbotham J Anaphoric Reference and Commom Reference Dordrecht Kluwer 1992b Higginbotham J Larson R Iatridou S Lihari U Higginbotham J Control and Grammar Reference and Control Ms Research Institute for the Languages of Finland 1995 Hoinenen T Null Subjects in Finnish: From Either-Or to More-Or-less. Ms University of Oslo 2003 Holmberg A Null Subjects and Uninterpretable Features: Evidence from Finnish Oxford Oxford University Press 2002 Holmberg A Nikane U Svenonius P Subjects, Expletives, and the EPP Expletives, Subjects, and Topics in Finnish 173 187 24 1994 Holmberg A Nikane U Proceedings of NELS Expletives and Subject Positions in Finnish Berlin Mouton de Gruyter 1993 Holmberg A Nikane U Holmberg A Nikane U Case and Other Functional Categories in Finnish Syntax Introduction 69 96 30 1999 Hornstein N Linguistic Inquiry Movement and Control Oxford Blckwell 2001 Hornstein N Move! A Minimalist Theory of Control 531 574 15 1984 Huang C. T. J Linguistic Inquiry On the Distribution and Reference of Empty Pronouns Dordrecht Kluwer 1989 Jaeggli O Safir K The Null Subject Parameter 1 37 2 1999 Kato M Probus Weak and Strong Pronominals Frankfurt Vervuert-Iberoamericana 2000 Kato M Kato M Negrão E Brazilian Portuguese and the Null Subject Parameter The Partial pro-drop Nature and the Restrictec Word Order MA MIT Press 1994 Kayne R The Antisymetry of Syntax 1 32 Tokyo Hitizi Syobo 1995 Lasnik H Haraguchi S Funaki M Minimalism and Linguistic Theory Last Resort University of South California 2000 Modesto M Null Subjects wothout 'Rich' Agreement MIT 1984 Montalbetti M After Binding Ms Universidade de São Paulo 1999 Negrão E O Português Brasileiro: Uma Língua Voltada para o Disccurso College Park University of Maryland 1995 Nunes J The Copy Theory of Movement and Linearization of Chains in the Minimalist Program Cambridge, Mass. MIT Press 2004 Nunes J Linearization of Chains and Sideward Movement Dordrecht Foris 1982 Rizzi L Issues in Italian Syntax Generals Paper College Park University of Maryland 2000 Rodrigues C Deriving Null Subjects from Movement 160 178 Oxford Oxford University Press 2002 Rodrigues C Lightfoot D Syntactic Effects of Morpgological Change Loss of verbal Morphology and Referential Subjects in Brazilian Portuguese College Park University of Maryland 2004a Rodrigues C Weak Morphology and A-movement out of Case Domains 123 147 20 2004b Rodrigues C Delta Thematic Chains Annual Meeting of The Linguistic Society of America San Diego, California 1982 Ross J. R Pronoun Deletion in German 663 689 17 1986 Safir K Linguistic Inquiry Relative Clauses in a Theory of Binding and Levels 401 429 27 1986 Salmon N Notre Dame Journal of Formal Logic Reflexivity London Routledge 1992 Sulkala H Karjalainen M Finnish Indiana Indiana University Linguistics Club 1980 Taraldsen K. T On the Nominative Island Condition, Vacuous Application and The That-Trace Filter Ms University of Maryland and Umass Boston University 1995 Torrego E Uriajereka J Parataxis Mass. GLSA, University of Massachusetts, Amherst 1989 Vainikka A Deriving Syntactic Representations in Finnish 613 671 17 1999 Vainikka A Levi Y Natural Language and Linguistic Theory Empty Subjects in Finnish and Hebrew MIT 1974 Vergnaud J. R. French Relative Clauses Helsink Soumalaisen Kirjallisuuden Seura 1989 Vilkuna M Free Word Order in Finnish: Its Syntax and Discourse Functions Universidade de São Paulo 1999 Viotti R A Sintexe das Sentenças Existenciais do Português do Brasil