1

On the notion of partial (non-) pro-drop in Romance and beyond1

Andrés Saab

Instituto de Filología y Literaturas Hispánicas “Dr. Amado Alonso” (Universidad de Buenos

Aires / CONICET)

al_saab75@yahoo.com.ar

Abstract: This paper centers on the nature of partial pro-drop in Northern Italian dialects and Brazilian Portuguese. I propose that the distribution of overt and null subjects in these languages provides evidence for an analysis of the null subject parameter in terms of ellipsis. In particular, I defend the idea that nominative subjects are deleted at PF in consistent null subject languages through what I call head ellipsis, a morphological operation subjected to the same locality conditions as other post-syntactic operations. Head ellipsis dictates when and how a particular syntactic object should be pronounced. Thus, Northern Italian dialects are seen as null subject languages with some subjects phonetically realized by the same reasons that apply in other cases of multiple copy realization. In turn, Brazilian Portuguese is a non-null subject language with some null subjects. Crucially, these subjects are not the result of morphological head ellipsis, but of another syntactic mechanism of licensing. A fundamental division between morphological oriented and syntactically oriented languages is then made. As I will show, three related strategies are available for syntactic oriented languages, especially, partial and radical pro-drop ones, namely: (i) NP-anaphora (Barbosa 2010, 2013), (ii) Copy deletion in the syntax (Ferreira 2000 and Rodrigues 2004, among others), and (iii) DP-ellipsis (Tomioka 2003, Saito 2007 and Takahashi 2010, among many others). The different strategies in (i), on the one hand, and (ii)/(iii), on the other, boil down to the general distinction between deep and surface anaphora. In turn, Copy deletion and DP-ellipsis are shown as forming a natural class of surface anaphora phenomena. The fact that these strategies are commonly attested in partial and radical pro-drop languages is not casual, but follows from the general properties of the agreement and case system of such languages. Therefore, that languages allowing for DP-ellipsis generally license null NP-anaphora as well is connected to the fact that both elliptical DPs and NP-anaphora are caseless and, as such, fail to be pronounced. This option is legitimate only in those languages where there is no case-agreement connection. I show how Oku’s observation and similar phenomena follow directly under this conception of null subjects across languages.

1 This paper is partially based on Chapter 6 of Saab (2009), except for the discussion on Oku’s observation which was not included in that work. Part of this material was also presented at the IV Encuentro de Gramática Generativa (Mendoza, 2007), Romania Nova III (Montevideo, 2008), Romania Nova IV (Campos de Jordão, 2010), the Linearization Workshop (Berlin, 2010), Romania Nova V (Alcalá de Henares, 2011), the II Jornadas de Jóvenes Lingüistas (Buenos Aires, 2013) and the II Encuentro Iberoamericano de Historia y Filosofía de la Lingüística Generativa (Santa Fe, 2013). I would like to thank the audiences of these conferences for stimulating comments and discussion. For specific comments there and elsewhere I am grateful to Pilar Barbosa, José Camacho, Sonia Cyrino, David Embick, Mary Kato, Mercedes Pujalte, Jairo Nunes and Pablo Zdrojewski. For Brazilian Portuguese judgments, thanks to Sonia Cyrino, Mary Kato, Rafael Minussi, and Jairo Nunes. For data and discussion on Hungarian, I am in debt with Anikó Lipták. I would like to extend my most sincere gratitude to the students of my seminar Sintaxis de los silencios (UBA, 2013) for many hours of discussion and feedback. The comments provided by two anonymous reviewers helped me to clarify and expand some unclear aspects of a previous version of this paper. Space reasons prevented me to do justice of all their comments, which I hereby acknowledge and to which I hope to come back in a near future. Special thanks goes to Mary Kato and Paco Ordóñez for organizing the Romania Nova workshop during the last years and for providing the best environment for a fruitful scientific dialogue.

2

1. Introduction

In this paper, I explore the null subject parameter (NSP henceforth). In particular, I will focus

on the notion of partial pro-drop in the empirical domain of Northern Italian dialects (NIDs)

and Brazilian Portuguese (BP). These languages can be descriptively seen as partial pro-drop

in the sense that the distribution of their null subjects is considerably more restricted than in

consistent null subject languages (NSLs) like Spanish, standard Italian or European

Portuguese. However, while NIDs share the relevant syntactic properties of consistent NSLs,

such as free inversion or absence of that-trace effects (Rizzi 1986b, Brandi & Cordin 1989,

among others), BP patterns like consistent non-NSLs (see in particular Barbosa, Duarte &

Kato 2005) in some respects (absence of free inversion of referential subjects) but like radical

pro-drop languages as far as the distribution of some of their null subjects is concerned

(Barbosa 2010, 2013)2.

Here, I will try to show that the empirical scenario these languages display provides

evidence for a particular approach to the NSP, one that integrates the null subject problem

into a general theory of ellipsis. Concretely, I will adopt and defend Saab’s (2009) theory,

according to which ellipsis is an all-the-way-down operation that can apply at syntax or

morphology (in the sense of Distributed Morphology) under a unique identity condition but

under different licensing/locality conditions depending on the component of the grammar

involved (i.e., syntax or morphology). In this respect, the qualitative differences between

partial and consistent pro-drop languages follow from the fact that referential null subjects in

NSLs like Spanish or NIDs are cases of ellipsis entirely resolved at PF, whereas null subjects

in partial NSLs like BP are exclusively resolved in the syntax. It is an open issue whether NSs

in partial pro-drop languages of the BP type are also cases of ellipsis in the sense to be

defined below. I will try to show that a fundamental distinction between deep and surface

anaphora is needed for syntactically licensed null subjects. Concretely, both an empty

anaphora analysis and a DP-ellipsis approach seem to be required within and across

syntactically oriented languages. Oku’s effects in Japanese and related languages and the

distribution of null generics in partial pro-drop languages unequivocally lead us to this

conclusion.

The paper is organized in the following way. In section 2, I briefly summarize the

debate on the NSP and present an empirical argument against the hypothesis that agreement is

interpretable in NSLs. In section 3, I introduce the basic definitions of the theory of ellipsis

2 So, it is tempting to say that Northern Italian dialects are partial NSLs, whereas BP is a partial non-NSL. However, I will restrict the term partial pro-drop or partial NSL just for BP, as usual in the current literature.

3

defended in Saab (2009). In section 4, this theory is illustrated with reference to NSLs. I show

there how this system predicts NSLs with some obligatory subjects. This case is particularly

instantiated by NIDs. In section 5, I explore the reverse case, namely, a non-NS language with

some null subjects in finite contexts. This case is illustrated with reference to BP, a language

that is losing the set of properties that characterizes NSLs, although still allowing for some

null subjects in certain restricted contexts. It is shown that these null subjects are not cases of

morphological ellipsis, but cases of null subjects syntactically licensed. In section 6, I advance

some conjectures on the nature of such syntactic licensing mechanisms by discussing the

nature of the so-called Oku’s observation. Section 7 contains the conclusion.

2. Three approaches to the NSP in the generative framework

Three general approaches to the NSP are in the focus of debate in the generative framework.

Putting aside differences in implementation, they can be summarized as follows:

(A) The GB style approach (Chomsky 1982, Rizzi 1982, 1986a, and much subsequent work),

according to which consistent NSLs license a special type of empty category, namely, pro.

Different versions of the minimalist program, including Chomsky’s own work, have adapted

this theory to minimalist considerations (see Camacho 2013 for a recent overview and a

proposal).

(B) The pronominal agreement approach (Jelinek 1984, Barbosa 1995, 2010, Ordóñez 1997,

Alexiadou & Anagnostopoulou 1998, Kato 1999, Barbosa, Duarte & Kato 2005, among

others), according to which agreement on tense is pronominal / interpretable in consistent

NSLs.

(C) The deletion approach (Perlmutter 1971, Holmberg 2005, 2010a, Saab 2009, Roberts

2010, among others), according to which at least some null subjects are a (sub-)case of

ellipsis.

Different arguments have been provided in favor or against the approaches

summarized above. The theory of pro has been criticized for those that believe that pro does

not follow from minimalist considerations (Manzini & Savoia 1997 and Alexiadou &

Anagnostopoulou 1998, among many others), or is conceptually incompatible with some

versions of the theory of Agree (Holmberg 2005). In Saab (2009), I argue that arguments

4

should not be constructed on the basis of this kind of conceptual considerations, but on the

basis of the predictive power of a given theory to capture broad generalizations about the

cross-linguistic distribution of empty categories in general. In this respect, the weakness of

the theory of pro is that it loses the generalization that null subjects are not inherently null. In

other words, we find instances of phonetic realization of pro. Indeed, as argued in Saab

(2009), the conditions that regulate when and how a particular syntactic object is (not)

pronounced are quite general and extend to apparently not related phenomena (ellipsis, copy

realization, null subjects, etc.). Put differently, there are empirical reasons that force us to

capture the nullness of null subjects in derivational terms. This argument against the theory of

pro can be similarly extended to the approach (B) mentioned above. Indeed, I think the

problem with this approach is even more severe than with pro theory (see also Camacho

2013). Let us see this problem in some detail.

The assumption that agreement is interpretable on T in consistent NSLs makes some

predictions in connection with the identity condition on ellipsis. As is well known,

interpretable features or features controlling agreement do not tolerate difference between a

given antecedent and some potential elliptical constituent in the linguistic environment. This

observation can be stated as follows (Saab 2009 for extensive discussion and references and

Merchant forthcoming):

(1) Interpretable features (or features controlling agreement) never trigger partial identity

effects under ellipsis whenever they are part of an elliptical gap.

This can be easily tested in the case of TP-ellipsis in Spanish where the interpretable

tense features on T cannot differ in the antecedent and the elided phrase. Thus, (2), where the

antecedent TP is in the past and the elided TP is in the future, gives an ungrammatical result

even when the adjunct en el futuro should be enough to recover the missing information:

(2) *María ha leído mucho y Elena en el futuro

María has read a-lot and Elena in DET future

habrá leído mucho también.

will-have read a-lot also

‘María has read a lot and Elena in the future will have too.’

(Murguia 2004: 86)

5

The behavior of gender is even more revealing in this respect: gender differences

under ellipsis are only attested when gender is uninterpretable on the elided phrase (see

Depiante & Masullo 2001, Saab 2004, 2009, Nunes & Zocca 2009, and Merchant

forthcoming, among many others).

NP-ellipsis: Gender strict identity effects

(3) a. *Juan visitó a su tío y Pedro prometió visitar a la tía de él.

Juan visited his.sg uncle and Pedro promised visit the.f.sg aunt of his

b. *Juan visitó a su tía y Pedro prometió visitar al tío de él.

Juan visited his.sg aunt and Pedro promised visit acc.the.m.sg uncle of his

Predicate ellipsis: Partial identity effects for agreeing adjectives

(4) a. Juan es alto y María también es alta.

J. is tall.masc and M. also is tall.fem

b. María es alta y Juan también es alto.

M. is tall.fem and J. also is tall.masc

If T had interpretable φ-features or agreement was interpretable by itself as claimed by

the proponents of the approach (B), NSLs would have strict identity effects in contexts of

ellipsis. However, this prediction is not borne out.

Partial identity of subject agreement:

(5) a. Juan fue al cine y nosotros también fuimos al cine.

J. went.3SG to-the cinema and we also went.1PL to-the cinema

b. Nosotros fuimos al cine y Juan también fue al cine.

we went.1PL to-the cinema and J. also went.3SG to-the cinema

c. Juan fue al cine y yo también fui al cine.

J. went.3SG to-the cinema and I also went.1SG to-the cinema

d. Yo fui al cine y Juan también fue al cine.

I went.1SG to-the cinema and J. also went.3SG to-the cinema

6

As far as I know, this is the most relevant empirical argument against theories of

interpretable agreement3. Notice that the problem cannot be resolved under the approach (B)

just claiming that agreement is a relational / redundant feature, whereas tense is not (see for

instance Brucart 1987 and Murguia 2004). As the contrast in (3) and (4) shows, the problem is

not in the intrinsic nature of features but in their distributional properties. That is, gender can

be obviated whenever is not the controller of agreement.

On the basis of the discussion of this section, I think that the deletion approach to the

NSP is the most promising one and, in what follows, I present a specific implementation of it

(see Holmberg 2005, 2010a and Roberts 2010 for other implementations).

3. Syntactic ellipsis vs. morphological ellipsis

A particular deletion approach to the NSP is proposed in Saab (2009) in the general

framework of Distributed Morphology. According to this theory, ellipsis can apply all-the-

way-down from syntax to PF. As shown in (6), syntactic ellipsis only affects phrases under c-

command or selection (E-selection in Merchant’s 2001 terms), among other conditions to be

discussed below (see especially section 6), whereas morphological ellipsis only affects heads

under the same morphological conditions that apply to post-syntactic displacements, namely,

immediate locality and adjacency.

(6)

Syntax ←Phrasal ellipsis (under different syntactic conditions)

←Head ellipsis (under immediate locality or adjacency)

PF

Here, ellipsis is understood as an instruction for non-pronunciation. A [+I] feature is

added to elliptical heads/phrases under formal identity. The adding of the [+I] feature blocks

the lexical insertion rules that, otherwise, would automatically apply on terminal nodes. The

relevant definition is as follows:

3 In Saab (2009) I have shown that Holmberg’s (2005) argument against approach (B) based on the distribution of null subjects and overt expletives in Finnish is not conclusive.

7

Non- Insertion:4

(7) No Lexical Insertion Rule, IR, applies in the domain of X0, X0 a MWd, if X0, or some

projection of X0, is specified with a [+I] feature.

From this definition, the following corollary is obtained:

Sub-word Deletion Corollary:.

(8) No SWd can be subject to Non-insertion if the MWd that contains it is not I-assigned.

So, in a situation like the one in (9c), Y0 has to be pronounced:

(9) a. XP[+I] b. XP c. XP ! ! ! X0 X0 [+I] X0 2 2 2 Y X Y X Y[+I] X

In other words, the theory explicitly states when a given syntactic object has to be

pronounced even when the identity condition on ellipsis is satisfied. As we will see in the next

section, (9c) is instantiated in the case of Northern Italian dialects.

Crucial for my present purposes is the definition of morphological I-Assignment or

Head Ellipsis given in (10):

Morphological I-Assignment (Head Ellipsis):

(10) Given a morphosyntactic word (MWd) Y0, assign a [+I] feature to Y0 if and only if

there is a node X0 identical to Y0 contained in an MWd adjacent or immediately local

to Y0. (where the notion of contained is reflexive)

Immediate locality is the relation between a head and the head of its complement. It is

the structural condition that applies for affix hopping in English, where adjacency is

4 The associated definitions are formulated in (i)-(iii) (ii and iii from Embick & Noyer 2001: 574):

(i) The domain of X0, X0 a MWd, is the set of terminal nodes reflexively contained in X0. (ii) At the input to Morphology, a node X0 is (by definition) a morphosyntactic word (MWd) iff X 0 is the highest segment of an X0 not contained in another X0. (iii) A node X0 is a subword (SWd) if X0 is a terminal node and not an MWd.

8

irrelevant, as the intervention of adverbs between the inflectional affix and the verbal base

shows (the traces t in 11 and 12b have only an expository status without any theoretical

content):

(11) John [TP t [vP completely destroy-ed the opposition...

(adapted from Embick & Noyer 2001: 585)

Other cases of affixation at PF require adjacency between the targets of the movement.

This kind of post-syntactic movement is called Local Dislocation (LD) in Embick & Noyer’s

(2001) framework. A well-known case of LD is the synthetic comparative/superlative

formation in English:

(12) a. Mary is the mo-st amazingly smart person . . .

b. *Mary is the t amazingly smart-est person . . . (Embick & Noyer 2001: 565)

The difference between Lowering and LD can be derived if post-syntactic operations

can take place before or after the introduction of linearization statements. Once a linearization

statement is introduced into the structure, adjacency becomes the relevant condition for

displacement. Therefore, the or part in the definition in (10) is not stipulated but follows from

the derivational property of the computational system.

The system briefly sketched above predicts a set of interactions between syntax, head

ellipsis and other post-syntactic operations -such as Agreement and Local Dislocation.

Abstractly, if Y0 and X0 are MWds and X0 = Z0, head ellipsis can apply to X0 in a

configuration like (13). However, if X0 is adjoined to Y0 like in (14), I-Assignment to X0 is

left without effect and Lexical Insertion has to apply to every terminal node contained in Y.

(13) [X] ^ [Y Y ⊕ Z] ( ^ = concatenation of MWds, ⊕ = concatenation of SubWds)

(14) [X] ^ [Y Y ⊕ Z] → [Y [X] ⊕ [Y ⊕ Z] ]

As I will show in the next section, (13) and (14) are concretely instantiated in the

domain of NSLs. Specifically, (13) is the typical situation in consistent NSLs, whereas (14) is

the particular situation attested in Northern Italian dialects, where incorporation of the subject

on T, an MWd, via LD produces a case of copy realization.

9

4. Null subjects as morphological ellipsis

As already mentioned, I would like to propose that null subjects in consistent NS languages

instance a case of I-Assignment at PF, which is fed by the introduction of a dissociated

morpheme of agreement (in the sense of Embick & Noyer 2001) at that level and under an

adjacency condition. Le us assume that for a sentence like compramos ‘bought.1pl’ we have

the following syntactic tree:

(15) TP 3 DP T’ # V D0 T0 vP # V ... [1PL] v0 T0 V # √0COMPR v0 [past]

Unde this analysis, a null subject is just a maximal D projection in Spec,TP, although

the analysis will remain essentially unaltered if more structure internal to the DP is posited.

This is all the (narrow) syntax we need. The crucial steps of the derivation take place in the

PF branch. Following Halle & Marantz (1993), Embick & Noyer (2001) and Bobaljik (2008),

I assume that agreement is implemented exclusively at PF through the introduction of a

dissociated morpheme. This morpheme is simply a copy of the formal features of the subject

(16a). After the linearization statement ^ is introduced in (16b) (see Embick 2007 for a precise

definition), we get a situation identical to the abstract representation in (13). In other words,

(16b) satisfies head ellipsis as defined in (10), given that T0 and D0 are both MWds and T0

contains an identical antecedent for the node D0, namely, the agreement morpheme. In (16c),

then, a [+I] feature is added on the D0 node indicating that the lexical insertion rules for this

node are blocked in consonance with (7)5:

5 Notice that the D0 node in subject position is a MWd, but the dissociated morpheme is not. Therefore, the Sub-word Deletion Corollary in (8) predicts that the subject can be elided (i.e., not pronounced), but the dissociated morpheme cannot as shown in (i).

(i) a. D(P)[1pl] compramos un libro.

bought.1PL a book b. *Nosotros compramos un libro. we bought a book

‘We bought a book.’

10

Agreement LIN- Introduction (16) a. TP →→→→ b. ( D0 ^ T0 ) →→ 3 # V DP T’ [1PL] T0 Agr0 # V V # D0 T0 vP √ 0 +v0 T0 [1PL] # V ... [1PL] T Agr 0 V [1PL] v T0 V

√0 v0

I-Assignment →→→→ c. ( D[+I] ^ T0 ) # V [1PL] T0 Agr0 V # √ 0 +v0 T0 [1PL]

Assume now that a language is identical to Spanish in the relevant respects (i.e., they

have referential post-verbal subjects, “rich” agreement, and a nominative pronoun in

Spec,TP). They have, however, a crucial difference: the head of the pronominal DP is a clitic

incorporated on T0 through Local Dislocation. This situation is illustrated in (17)-(18):

(17) TP 3 DP T’ # V D0 T0 vP

# V [ϕ] v0 T0 V # √0 v0 [tense] Morphology: Agreement LIN Introduction (18) a. TP →→→→ b. ( D0 ^ T0 ) →→ 3 # V DP T’ [ϕ] T0 Agr0 # V V # D0 T0 vP √ 0 +v0 T0 [ϕ] # V ... [ϕ] T0 Agr 0 V [ϕ] (D = Agr)

11

v T Local Dislocation: D adjoins to T and D has to be pronounced →→→→ c. T0 3 D([+I]) T0 # V [ϕ] T0 Agr0 V # √0 +v0 T0 [ϕ]

NIDs seem to have the relevant properties of this kind of languages (see Safir 1986,

Rizzi 1986b, Brandi & Cordin 1989, and Poletto 2000, among many others). As specifically

argued by Rizzi (1986b) and Brandi & Cordin (1989), these languages have the classical

properties associated to NSLs (free inversion as in 19d) although they have some obligatory

clitic subjects (19b vs. 19c), a fact predicted by the system proposed here, if these clitics are

analyzed as SWds:

(19) a. el Mario el magna. Trentino

the Mario he-eats

b. el magna.

he-eats

c. *magna.

eats

d. magna el Mario

eats the Mario (Safir 1986: 336)

That clitic subjects in Trentino are SWds (clitics, in Cardinaletti & Starke’s 1999

sense) can be corroborated independently by well-known tests that distinguish weak pronouns

and clitics:

Italian vs. Trentino:

(20) a. Lui mangia della zuppa e - beve del vino. It.

Egli mangia della zuppa e - beve del vino.

he eats of-the sopu and drinks of-the wine

*La canta e – bala Trent.

she sings and dances

12

b. Lui e la ragazza del bar sono gli unici ad apprezzare tutto questo. It.

he and the girl of-the bar are the only to appreciate all this

*La e la Maria è vegnude algeri. Trent.

she and the Mary are come yesterday

(Cardinaletti & Starke 1999: 166-7)

Thus, the theory proposed captures a very intriguing fact: Why cannot clitics -besides

being associated with potential antecedents at PF- be elided? This is a very important question

and I do not know of any explicit answer to it. Theories assuming that subject clitics in NIDs

are just agreement markers (Rizzi 1986b and Brandi & Cordin 1989, among others) have to

explain why there is reduplication of agreement markers (agreement clitic plus agreement

morpheme). In my theory, this follows straightforwardly from the Sub-word Deletion

Corollary. Even more important, subject clitics in these languages form a natural class with

other well-known cases of multiple copy realization discussed in detail by Nunes (2004) and

Saab (2009)6.

6 The claim that null subjects are deleted by the same mechanism that underlies copy deletion is independently made by Roberts (2010), although the implementations and the empirical domains in each case are different. Probably, both theories also differ in the predictions they make, but this is hard to evaluate given that Robert’s theory does not explicitly address the issue of copy deletion/pronunciation beyond null subjects and related phenomena. As I have shown in Saab (2009), instead, the I-assignment directly correlates subject clitics in NIDs with well-known cases of multiple copy realization. Consider for instances multiple-wh realization in examples like the following:

German: (i) Wem glaubt Hans wem Jakob gesehen hat? whom thinks Hans whom Jakob seen has ‘Who does Hans think Jakob saw?’

Romani: (ii) Kas misline kas o Demiri dikhlâ? whom you-think whom Demir saw ‘Who do you think Demir saw?’ (cf. Nunes 2004: 38 and the references therein)

Nunes’s (1999, 2004) observation is that multiple realization of wh-copies and beyond is allowed only if at least one copy in a given nontrivial chain has suffered morphological reanalysis (i.e., Fusion as defined in DM). So, for the examples (i) and (ii), Nunes proposes to derive the basic facts in two steps. First, he claims that “successive-cyclic wh-movement in these languages may proceed by adjunction to an intermediate C0” and second, “Morphology in these languages may convert the adjunction structure [C

0 WH [C0 C0]] [...] into a single

terminal element.” (Nunes 2004: 40) Once a copy has been fused, it does not “count” as a copy but as a distinct element and, consequently, it cannot be affected by Chain Reduction as defined by Nunes. Nunes’s analysis is illustrated in (iii). Morphological Reanalysis (Nunes 2004:41): (iii) [ CP WHi [C’ Q [TP T [VP V [CP [C

0 #WH i C0# ] [TP [VP…WHi…

A prediction of this analysis is that there should not be multiple realizations of complex wh-phrases. This is correct: German:

13

5. From null subjects to partial pro-drop

As I already mentioned, the impossibility of null subjects in NIDs follows in part from a

licensing condition on Head Ellipsis. However, Head Ellipsis could also fail because of the

identity condition on ellipsis, as formulated in Saab (2009):

Identity :

(21) An abstract morpheme α is identical to an abstract morpheme β if and only if α and β

match all their semantic and syntactic features.

This is the general case in consistent non-NS languages like English or French, where

the head of a subject DP cannot be elided at PF just because agreement cannot feed the

identity condition on morphological ellipsis. I would like to claim that this is also the case in

partial pro-drop languages like Brazilian Portuguese. Almost without exceptions, the

literature on BP claims that this language is indeed a non-NSL (in the relevant sense) with

some instances of null subjects7.

Through a detailed diachronic study, Duarte (1993, 1995 and 2000) shows how the

loss of verbal morphology (probably triggered by an impoverishment of the pronominal

system; see Kato 1999 and below) had as a consequence the decreasing in the use of null

subjects. In effect, BP suffered at least two reductions of the verbal paradigm: from six

distinctions to four in a first stage, and finally to only three distinctions in the last generations

(Duarte 2000:18). According to Duarte, the first replacement affected the second person

singular and plural, tu and vós, respectively. The forms that replaced them are você and vocês.

These forms are combined with third person verbs just because they historically derive from a

(iv) *Wessen Buch glaubst du wessen Buch Hans liest? Whose book think you whose book Hans reads ‘Whose book do you think Hans is reading?’

Romani: (v) *Save chave mislinea save chave o Demiri dikhlâ? which boy you-think which boy Demir saw ‘Who do you think Demir saw?’ (McDaniel 1986 apud Nunes 2004: 18-9) Notice now that under the I-assignment system the basic facts follow from the Sub-word Deletion Corollary without the need of assuming Fusion. Put differently, we eliminate one step in Nunes’s analysis, because wh-adjunction to an intermediate C0 is enough to convert the wh-phrase into a SubWd. Therefore, not only does my system simplify Nunes’ analysis but it also provides an explicit correlation between multiple realizations of wh-copies and multiple realizations of subjects in NIDs and beyond. The difference between both types of phenomena is reduced to be subjected to syntactic or morphological copying (wh-copy vs. agreement, respectively). The rest follows from the Sub-word Deletion Corollary, a very welcome result for the ellipsis approach to the NSP we are suggesting here. 7 See, among many others, Duarte (1993), (1995), Kato (1999, 2000, 2011), Modesto (2000), Ferreira (2000), Rodrigues (2004), Barbosa, Duarte & Kato (2005), Barbosa (2010, 2013), and Camacho (this volume).

14

third person expression8. In many current dialects, the first person plural pronoun nós was

also replaced by a formally third person singular expression, a gente ‘the folk’. See the

following Table from Duarte (2000):

Table 1 Person Pronouns Paradigm 1 Paradigm 2 Paradigm 3 1sg Eu am o am o am o 2sg Tu

Você am a s am a

--- am a

--- am a

3sg Ele/Ela am a am a am a 1pl Nós

A gente am a mos ---

am a mos am a

--- am a

2pl Vós Vocês

am a is am a m

--- am a m

--- am a m

3pl Eles/Elas am a m am a m am a m

A particular interpretation of the paradigm 3 in BP is given by Nunes (2008), for

whom BP is a language with no syntactic specification for person features on T; only number

is encoded on the first person and on all the plural forms. For the rest of the singular forms,

number is just a default value. According to Nunes, the realization of the first person singular

is obtained by a redundancy rule at PF. Such a rule specifies that a first person value is added

whenever Number is SG; otherwise, a default value is obtained. See Table 2 from Nunes

(2008):

Table 2. Verbal agreement paradigm in (colloquial) Brazilian Portuguese cantar ‘to sing’: indicative present Eu (I) canto P:1.N:SG você (you.SG) ele (he) ela (she) a gente (we)

canta P: default, N: default (=3SG)

vocês (you.PL) eles (they.MASC) elas (they.FEM)

cantam P:default, N:PL(=3SG)

8 Vossa Mercê. The evolution is as follows: vossa mercê > vossemecê > vosmecê > você. Interestingly, the same change took place in the second person plural in the dialects of American Spanish although any relevant change in the NS property is attested. For different reasons, some Caribbean dialects did suffer some changes in the NS property (see Camacho, this volume, for a comparison between these dialects and BP).

15

Therefore, BP cannot license null subjects of the type attested in Spanish or European

Portuguese (i.e., via Head Ellipsis). However, this language does have another kind of null

subjects. In what follows, I will claim that these null subjects are licensed in the syntax.

5.1. Basic properties in BP

As other partial pro-drop languages (see, among others, Rodrigues 2004, Holmberg 2005,

2010a, Holmberg, Nayudu & Sheehan 2009 and Barbosa 2010, 2013), BP has some of the

following properties. As for the distribution of null subjects, BP does not have referential null

subjects in out-of-the-blue contexts9:

(22) a. *(Eu) como pizza.

I eat+1st pizza

‘I eat pizza.’

b. *(você) come pizza.

you eat+S3rd pizza.

‘You eat pizza.’

c. *(Ele) come pizza.

he eat+S3rd pizza.

‘He eats pizza.’

d. *(A gente) come pizza.

the folks(=we folks) eat pizza. (Kato 1999: 5)

However, other instances of null subjects are attested in BP such as quasi-arguments

or expletives (23) and generic/impersonal subjects (24):

(23) a. Tá chovendo.

be+S3rdp raining

‘It is raining.’

b. Tem novidade.

9 Some exceptions to this pattern (see i) are argued to be cases of Topic Deletion (see Ferreira 2000 and Rodrigues 2004): (i) e falei com o João ontem a noite (BP) spoke-1Sg with the João yesterday at night ‘I spoke with João yesterday night’ (Rodrigues 2004: 81)

16

have+S3rd news

‘There is news.’

c. Parece que vai chover.

seem+S3rd that go+3rd rain

‘It seems that it is going to rain.’ (Kato 1999: 5)

(24) a. Aqui pode fumar.

here can+S3rd smoke

‘You/can smoke.’

b. Aqui conserta sapatos.

here repair+S3rd shoes

‘One repairs shoes.’ (Kato 1999: 5)

Null third person subjects are also allowed in embedded contexts if some locality

constraints are obeyed. In effect, these null subjects require some condition of closeness and

c-command (although things are more complex, see Holmberg 2005). Compare (25) and (26):

(25) a. Ninguém acha que [e] é estúpido.

nobody thinks that is stupid

‘Nobodyi thinks that hei is stupid.’

b. O João disse que [e] comprou um carro.

the John said that bought+S3rd a car.

‘Johni said that hei has bought a car.’ (Kato 1999: 5)

(26) a. *O João disse [que a Maria acha [que e é bonito]]

the J. says [that the M. believe [that e is pretty]]

b. *A mãe do João acha [que e é bonito]

the mother of J. believes [that e is pretty]

(Ferreira 2000: 20)

Crucially, null subjects in partial pro-drop languages are not cases of morphological

ellipsis as in NSLs. Whatever their licensing mechanism is, it should be related to the

semantic/syntactic component. In other words, null subjects in BP are licensed in the syntax.

There are several proposals in the literature on BP and other partial pro-drop languages: third

17

null subjects are PRO (Kato 1999), Trace/Copy (Ferreira 2000, Rodrigues 2004), defective ϕP

in Spec,TP (Holmberg 2005 for Finnish), or null NP anaphora (Barbosa 2010, 2013)10.

If the movement analysis for the finite control cases in (25) is on the right track

(Ferreira 2000 or Rodrigues 2004), syntactic null subjects are just cases of syntactic ellipsis

under c-command (cf. 6 and section 6.3 for more discussion):

(27) [O João] disse que [o João][+I] comprou um carro.

Of course, this analysis does not extend to null quasi-arguments/expletives (23) or to

generic/impersonal null subjects (24). An alternative analysis could be to adopt Barbosa’s

(2010) approach according to which both controlled null subjects in finite contexts and

generic/impersonal ones are cases of null NP-anaphora11 (more on this below). In any case, it

is clear that the licensing mechanism should be syntactic and not morphological. I do not

have a particular commitment to any of these approaches at this point. As I will discuss in

section 6, it seems that both mechanisms are in principle available in partial pro-drop

languages of the BP type and beyond. I will then postpone the discussion to that section and

assume for the time being that a null NP approach and a movement analysis for empty

subjects are both compatible with the system I am proposing here.

As for overt subjects in BP, they share all the properties of weak pronouns in the sense

of Cardinaletti & Starke (1999) (see Kato 1999 and Barbosa, Duarte & Kato 2005). In

particular, they can be used to refer both to animate or inanimate referents (28) and have a

bound variable behavior (29):

(28) a. E elei precisou ir ao banheiro. Quando elei viu o que

And he needed go to-the bathroom. When he saw that

que era o banheiro, elei ficou apavorado.

that was the bathroom he was terrified

‘And he had to go to the bathroom. When he saw that the bathroom looked like he was

terrified.’

b. [A casa]i virou um filme quando elai teve de ir abaixo.

10 These two last options seem to differ only in the level of projection of a given empty NP. 11 See also Tomioka (2003). For an empty NP (or equivalently nP), Barbosa understands a type of semi-functional nominal category. In this respect, her approach to null NP anaphora resembles Panagiotidis’ (2002) theory of empty nouns. To derive the existential/generic reading of empty NPs, on the one hand, and the anaphoric reading of embedded subjects, on the other hand, she also postulates two different semantic rules, namely, existential closure and type-shifting, respectively.

18

the house turned-into a movie whet it had to go down

‘The house became a movie when it was demolished.’ (Barbosa et al 2005: 15-6)

(29) a. [Ninguém no Brasil]i acha que elei é prejudicado pelo Governo.

no-one in Brazil thinks that he is harmed by the government

‘No-one in Brazil thinks that he is harmed by the government.’

b. [Nenhuma criança]i acha que elai é burra.

no child thinks that she is stupid.

‘No child thinks that s/he is stupid.’

c. [Algum professor]i vai achar que elei é o responsável.

some teacher will-think that he is the responsible

‘Some teacher will think that he is the responsible one.’ (Barbosa et al 2005: 44-5)

As is well known, parallel sentences in European Portuguese or Spanish must replace

these overt pronouns by a null subject. Weak pronouns in BP pattern, then, as null subjects in

NSLs with respect to all these interpretative/ distributional properties.

From this set of facts, an important conclusion emerges: there is a correlation between

the arising of weak subjects and the loss of the pro-drop properties in BP. I will call this

correlation Kato’s observation (see in particular Kato 1999):

Kato’s observation:

(30) The arising of weak forms correlates with the losing of pro-drop properties.

Kato’s observation raises important questions regarding the nature of (partial) pro-

drop languages, the direction of linguistic change and related issues (see the next section). At

the same time, though, this observation reinforces the idea defended in this paper that the

nullness of null subjects cannot be a lexical, inherent property. As we have seen in the case of

NIDs, some potential null subjects have to be phonetically realized because of the Sub-word

Deletion Corollary (see 8), which prevents deletion of parts of words. In the case of BP, it is

the identity condition on ellipsis what prevents subject ellipsis. The arising of weak forms is

then a natural consequence of the deletion approach. Put differently, Kato’s observation

follows if weak forms and null subjects are the same syntactic object, as extensively argued

by Roberts (2010). Under my view, BP lost the rule introducing dissociated morphemes at PF

or the features of the dissociated morphemes are impoverished with respect to the features of

19

the pronominal subject. The natural conclusion is that weak pronouns in BP are the “visible”

reflex of the old null subjects.

Summarizing what has been discussed so far, the ellipsis approach to the NSP I am

exploring derives the occurrence of clitic subjects from the Sub-word Deletion Corollary and

the arising of weak pronouns in partial pro-drop languages as BP from the identity condition

on ellipsis. Both phenomena point to the same conclusion, namely, that the nullness of null

subjects cannot be a lexical matter, in consonance with Holmberg (2005) and subsequent

works. I do not see how the important generalizations and correlations discussed in this paper

can be captured under the classic theory of pro and its minimalist variants.

5.2. On the direction of linguistic change

As the reader may have noted, a purely morphological approach to the NSP as the one

defended here leads us to the question about the syntactic effects of the NSP. The correlation

between head ellipsis and free inversion, for instance, seems to lack any deep motivation. It

should be observed, however, that the general correlation between morphological richness

and syntactic freedom is a long-standing problem in the generative program. Other

approaches to the NSP in the framework of DM assume that some morphological operations,

like impoverishment, take place pre syntactically (Müller 2005 and Roberts 2010 for detailed

discussion). I will not take such a radical move here. I think that the basic mechanics of post-

syntactic operations are well-motivated in DM and I do not see any reason to modify such

architectural assumptions. Indeed, I believe that the system proposed here has the potential to

account for the syntactic effects of the NSP. Let us see this in some detail.

A way to address this issue is to explore the nature of EPP across languages. A

plausible conjecture could be that the introduction of dissociated morphemes and the EPP

(i.e., the need of a specifier for some functional head) are in complementary distribution.

Conjecture:

(31) The introduction of dissociated morphemes (i.e., morphological agreement) and the

EPP are in complementary distribution.

Let us assume that the EPP is just the requirement that T is associated with a nominal

feature (maybe, a D or N feature) as in Chomsky’s (1995) approach.

20

EPP:

(32) T must be associated with a nominal (D/N) feature at syntax or PF.

Non-NSLs satisfy the EPP in the usual way, merging the DP subject with T (33),

whereas NSLs satisfy the EPP in the morphology via the introduction of an agreement node

within the T head (34).

Non pro-drop Languages: syntactic EPP (33) TP V DP[D] T’ V T vP [D] V t v’ V ... Pro-drop Languages: morphological EPP (34) TP V T’ V T vP V V

T [AGR]

Notice that this simple idea straightforwardly accounts for the basic correlation

between morphological richness and word order (free inversion, specifically): Morphological

agreement yields the null subject property via head ellipsis as explained in the previous

sections. Free inversion is now a natural consequence of the fact that Spec,TP does not need

to be filled for EPP reasons. This hypothesis seems to be connected to Alexiadou &

Anagnostopoulou’s (1998) analysis of the NSP at least in a broad sense (i.e., the NSP follows

in part from the EPP). Nevertheless, it also differs in non-trivial aspects. In Alexiadou &

Anagnostopoulou’s system NSLs just do not make use of Spec,TP, but according to the

conjecture made here Spec,TP is perfectly available in NSLs. In this sense, movement to

Spec,TP and the EPP should also be dissociated.

The NSP is derived then as a problem of the syntax-morphology connection. I think

that the conjecture I am sketching here provides a clear way of stating the problem of

linguistic variation. In effect, we can think of the change suffered by BP as a change that goes

21

from morphology to syntax: because of the loss of morphological agreement, the EPP in BP

has to be satisfied in the syntax. The direct consequence of this change is the loss of free

inversion among other typical properties of NSLs (see Barbosa, Duarte & Kato 2005).

(35)

Syntax ← EPP (via Spec-head relation) Non-NSLs

Path of change

←EPP (morphological agreement) NSLs

PF

An additional advantage of this way to address the problem is connected to the

problem of generic / impersonal null subjects in partial NSLs, an issue that has received

considerable attention in the last years (see, in particular, Holmberg 2005, 2010b, Holmberg,

Nayudu & Sheehan 2009 and Barbosa 2010, 2013, among others). As claimed by Holmberg

(2010b), the occurrence of a type of null generic subject in finite contexts is a property of

partial NSLs, which quite surprisingly is not attested in consistent NSLs. Compare in this

respect the impersonal sentences in (24) (repeated as 36 below) with parallel sentences in

Spanish:

(36) a. Aqui pode fumar.

here can+S3rd smoke

‘You/one can smoke.’

b. Aqui conserta sapatos.

here repair+S3rd shoes

‘One repairs shoes.’ (Kato 1999: 5)

(37) a. *(Se) castiga a los culpables.

SE punishes to.acc the culprits

‘Someone punishes the culprits.’

b. *(Se) trabaja duro.

SE works hard

‘One works hard.’

22

As mentioned above, the question is why partial NSLs license a type of null subject

that is not licensed in NSLs. This is not a trivial question as the recent literature on this topic

shows.

Yet, it should be noted that it is not the case that null generic subjects are not attested

in general in NSLs and in consistent non-NSLs. What is particular to BP and other partial

pro-drop languages is just the licensing of null generics in finite contexts. Crucially,

impersonal SE in Spanish is disallowed in non-obligatory control contexts in consistent NSLs

(see Saab 2002 and the references therein), where some type of null generic is used instead.

(38) Juan dice que castigar(*se) a los culpables es necesario.

J. says that to.punish(*SE) to.acc the culprits is necessary

‘John says that to punish the culprits is necessary.’

Therefore, it is tempting to assume that the underlying null subjects in (36) and (38)

are the same category. Under this account, partial NSLs would not have a special type of

pronoun (say, a null NP in Barbosa’s 2010 terms or ϕP as in Holmberg 2010b). In this sense,

the answer to the problem posed by cases like (36) consists in answering what licenses (non-)

obligatory control in finite configurations. However, the actual scenario is considerably more

complex, first, because it is not the case that null generics in finite contexts in BP behave like

subjects of non-obligatory control clauses and, second, because their behavior is not uniform

across partial pro-drop languages.

As for the first problem, null generics in BP have the same distribution as impersonal

SE with respect to certain binding tests. Notice first that pronominal binding in contexts of

impersonal se constructions is impossible (thanks to Francisco Ordóñez for pointing out this

fact, see Saab 2012 and Pujalte 2013 for more discussion). Thus, (39) is ungrammatical under

the reading that one can leave one´s coat here.

(39) Aquí se puede dejar su saco.

here SE can leave his coat

Intended reading *: ‘Here, onei can leave hisi coat.’

Crucially, the relevant reading is perfectly available in non-obligatory control

contexts:

23

(40) Está permitido traer su mascota a la fiesta.

is allowed to-bring his pet to the party

Intended reading OK: ‘It is allowed to bring one’s pet at the party.’

My informants confirm that the BP counterpart of the Spanish sentence in (39) is also

ungrammatical under the pronominal binding reading12:

(41) a. Aqui pode deixar seu casaco

here can leave his coat

Intended reading*: ‘Here, onei can leave hisi coat.’

As for reflexive binding in Spanish, the result is the same: whereas non-obligatory

control allows for it, impersonal SE constructions do not (see Pujalte 2013 for recent

discussion):

(42) a. Aquí se puede lavar las manos.

here SE can wash the hands

‘One/you can wash his/your hands here.’

(*impersonal reading, OK referential/generic second person reading)

b. Juan dijo que mirarse a sí mismo en el espejo

J. said that to.look-SE to himself in the mirror

le causa terror.

CL.DAT cause horror

‘Juan said that to looking at himself in the mirror caused him horror.’

Again, my BP informants do not accept the generic reading whenever a reflexive

pronoun like SE is present (avoid the generic você reading)13:

(43) *Aqui pode-se lavar as mãos.

here can-SE wash the hands

12 For some of my informants, the sentence is grammatical only if interpreted with an underlying “você” reading. The same effect is attested in Spanish. 13 The fact that BP speakers do not allow for reflexivization of sentences like (36) also sheds light on the nature of the phenomenon in Spanish, because now the impossibility of reflexivizing impersonal SE constructions in (39) cannot be directly attributed to some ban on the co-occurrence of two types of SE (impersonal and reflexive). This supports Pujalte’s (2013) approach to the problem.

24

Therefore, we can conclude that null generics in BP behave exactly like Spanish

impersonal SE in the relevant binding tests14. Following the spirit, although not the letter, of

Pujalte & Saab’s (2012) approach to SE-insertion in Spanish with the qualifications made in

Saab (2012) and Pujalte (2013) I will take the impossibility of pronominal/reflexive binding

as an indisputable indication of absence of any syntactic import for the external argument. In

other words, there is no little proarb in this type of configurations and the arbitrary reading

arises as a repair strategy at the semantic-pragmatic interface (see Saab 2012 for detailed

discussion). Now, absence of an external argument constitutes a flagrant violation of the EPP,

at least in its more radical formulation (see below). As shown in detail by Pujalte & Saab

(2012), SE insertion is precisely the PF mechanism that resolves this EPP conflict. Put

differently, SE-insertion, which applies in a broad set of environments beyond impersonal

constructions, is nothing else but a type of dissociated morpheme. In this respect, then

Spanish makes use of the typical mechanism available in the language for EPP checking,

namely, PF insertion of dissociated material, in consonance with the conjecture in (31):

14 Spanish impersonal SE can control the subject of a non-finite clause and, under a generic, inclusive reading (but not under an existential one), be the subject of a secondary predicate. (i) Se castigó a los culpables para demostrar autoridad. SE punished to the culprits to show authority ‘The culprits were punished to show authority.’ (ii) Se puede entrar borracho aquí. SE can enter drunk here ‘One can enter drunk here.’ (iii) *Se castigó a María borracho. SE punished to Mary drunk Intended reading *: ‘Someonei punished Mary drunki.’

As for (i), I follow, among others, Landau (2010) and, especially, Saab (2012) and Pujalte (2013), for whom control cannot be taken as evidence for a syntactically projected argument. Pace Landau, however, I think that secondary predication is not a signal for syntactic activity, either. The contrast between (ii) and (iii) is, indeed, hard to capture by positing some syntactic entity. As shown by the English glosses, both inclusive and exclusive pronouns allows for secondary predication when overt. This apparent puzzle vanishes if, as suggested in Saab (2012) and others, absence of an external argument implies some type of existential closure at LF or beyond LF (see Heim 1982). This automatically produces the ungrammaticality of (iii), given that to existentially close both subject positions yields to a disjoint reference effect which is incompatible with the need that both predicates are predicates of the same subject. Under the generic, inclusive reading of (ii), instead, a generic operator can unselectively bind the subjects of both predicates and the desired reading is obtained. See Pujalte & Saab (2012), Saab (2012) and Pujalte (2013) for extensive discussion and arguments in favor of not projecting an external argument in impersonal SE constructions in Spanish.

25

(44) TP V T’ V T vP V V

SE T

Now, this analysis seems to provide an explanation of the lack of explicit morphology

for null generics in finite contexts in BP and partial pro-drop languages in general. Given that

BP has lost the general mechanism to introduce dissociated morphemes of the D type, a

syntactic way to check the EPP has arisen in the language. For the case of generic sentences,

BP behaves as Finnish where the locative/adverbial phrase checks the EPP when present,

otherwise, the EPP itself is suspended (see Holmberg 2005, 2010a,b for extensive discussion

and references). Compare in this respect (45a) in BP, where the locative is forced to move to

Spec,TP (cf. 36b), with (45b) in Spanish where the locative can occur in post-verbal position.

(45) a. *Conserta sapatos aqui.

repair shoes here

b. Se arreglan zapatos aquí.

SE repair shoes here

In other words, as a consequence of losing general PF mechanisms for EPP checking,

BP makes use of a syntactic way of satisfying it.

(46) TP V aqui T’ V T vP V v’ V ...

However, unlike English, and other consistent non-NS languages, partial pro-drop

languages seem to be more liberal with respect to the EPP (see e.g., the sentences in 23). It is

not my purpose to present a full articulated theory of EPP-checking in partial pro-drop

languages given that my goal here is just to show that the underlying mechanism is not

26

morphological, but syntactic. We can then think of consistent NS languages as being

morphologically oriented and (partial) non-NS languages as being syntactically oriented. As

noticed above, qualifications in the way the EPP is checked in the syntax are required in order

to capture the difference between partial and consistent non-NS languages (see Holmberg

2010b and the references therein). Yet, qualifications seem to be also necessary within partial

pro-drop languages. As shown by Holmberg (2010b) (see also Johns 2005 for discussion),

Finnish, unlike BP, allows for both pronominal and reflexive binding, in a similar way to

what was observed for non-obligatory clauses above:

(47) a. Shelliasemalla voi pestä auto-nsa.

Shell-station-ADE can-3SG wash car-PX

‘You can wash your car at the Shell station.’

b. Sitä ei kuulu ottaa itseään liian vakavasti.

EX not-3SG should take SELF-PX too seriously

‘One shouldn’t take oneself too seriously.’

(Holmberg 2010b: 205)

It seems then that at least some syntactic representation for the external argument is

necessary in Finnish (although see Johns 2005). According to Holmberg this category is a ϕP,

which is deleted under Agree with T. This would explain the binding facts in (47) and the fact

that generic null subjects, unlike other null subjects in Finnish, are incapable of checking the

EPP, assuming that the EPP minimally requires categories of the D type (see Landau 2010 for

discussion and a proposal). A null NP anaphora account (Barbosa 2010) could be also

compatible with the basic facts depending also on some assumptions on binding and the EPP.

At any rate, both approaches have to capture the contrast between BP and Finnish in this

respect15. The problem is a nontrivial one because to postulate the presence or absence of a

syntactically projected subject in one language but not in the other leads to the important

15 An alternative approach to account for the contrast between BP and Finnish could be to postulate that both languages project indeed an external argument, but they differ in the level of projection. Thus, whereas BP only projects an empty NP, Finnish projects a ϕP, as already suggested by Holmberg. The next assumption is that only DPs or ϕP are legitimate candidates for binding (although see Landau 2010 for a different conclusion), NPs being defective in this respect. If this is correct, we can extend this alternative to Spanish and postulate also a type of empty NP anaphora underlying impersonal SE constructions (see indeed Rivero 2001 for a similar approach). I think, however, that this type of approach loses the basic generalization that there is a pattern of systematic syncretism that connects impersonal SE constructions in Spanish to other SE constructions. See Pujalte & Saab (2012), Saab (2012) and Pujalte (2013) for extensive discussion.

27

question of what would be the underlying property that motivates such a difference across

languages. I leave the problem unresolved here.

In sum, the purely morphological account of the NSP I have proposed here can

account for the syntactic effects of NSLs provided we accept the conjecture in (31). BP and

partial NSLs in general can be seen as cases where the availability of morphological EPP

(plainly, morphological agreement) vanishes and the resort to syntactic EPP arises as the only

legitimate option. I have also shown that this view can elegantly account for the occurrence of

a morphological expletive in impersonal constructions in consistent NSLs, namely, the

expletive SE.

6. On the typology of elliptical subjects

So far, I have argued that whereas consistent NS languages of the Spanish type license head

ellipsis of subjects triggered by agreement at PF, partial pro-drop languages such as BP have

syntactically licensed null subjects. As mentioned in section 5, two main proposals are found

in the literature for anaphoric third person subjects: (i) the A-movement analysis (Ferreira

2000 and Rodrigues 2004, among others), and (ii) the null category analysis (where the null

category is PRO as in Kato 1999, a variable as in Modesto 2000, a null NP as in Barbosa

2010, 2013 or even a ϕP as in Holmberg 2005). For the case of null generics or expletives, of

course, both analyses also clearly differ. For instance, under the empty NP anaphora analysis

a rule of existential closure or the introduction of a generic operator produces the relevant

readings in generic/existential contexts (see footnote 11). Things are more complex in the

case of pure expletive constructions, but see Barbosa (2013) for some suggestions. For the

proponents of the movement analysis, instead, a null category should be inserted in those

contexts where movement is not allowed, in consonance with the general approach to non-

obligatory control under the Movement Theory of Control (Hornstein 1999 and subsequent

works). Evidence adduced in favor of the movement analysis is the fact that locality effects

and other movement diagnostics seem to be attested in BP and related languages (see the

locality effects in sentences like 26). Evidence in favor of a null category approach is

provided exactly for the opposite reasons: there seem to be contexts where movement is

prevented but null subjects are instead attested (see Duarte, Barbosa & Kato 2005, where

some examples of referential null subjects taking their reference across the discourse are

presented). Moreover, uniformity reasons are claimed as being in favor of such type of

approach. The argument has the following form: given that a null category approach is

unavoidable for generic sentences, then an analysis that accounts for both null generics and

28

anaphoric null subjects is preferable (Holmberg 2005 and Barbosa 2010, 2013 are good

illustrations of this type of arguments). If anaphoric third person NSs in embedded clauses

(see 25 in section 5.1) are also cases of empty nPs, or ϕP, then we can safely conclude that BP

only licenses a type of inherently null category, maybe universally available under some

conditions. In this respect, partial pro-drop phenomena (embedding of anaphoric null subjects

and null generics/expletives) are not part of the set of facts to be accounted for the theory of

ellipsis as defined in this paper. Of course, if the movement theory of null embedded subjects

is on track (see 27 in section 5.1) then these subjects are accounted under the theory of

ellipsis. Notice, however, that even thus there is still a residue of empty subjects in finite

clauses that are not accounted for under ellipsis. As I will try to show, in principle, it seems to

me that both phenomena are allowed for UG; differences between them boil down to the well-

known distinction between deep and surface anaphora (Hankamer & Sag 1976).

As the reader might have already noticed, this situation opens the question whether or

not actual cases of elliptical DPs beyond copy deletion are attested in natural language. The

answer seems to be positive and is connected to the well-known phenomenon of argument

ellipsis in radical pro-drop languages of the Japanese type. In what follows, then, I will

complete the picture by providing an analysis of the so-called Oku’s generalization. By doing

this, I will address Duguine’s (2013) recent objection to my I-assignment system, on the one

hand, and, on the other, I will discuss Barbosa’s (2010) claim that partial pro-drop should be

assimilated to radical pro-drop, both phenomena falling under her NP-anaphora analysis. As

we will see NP-anaphora and subject ellipses, regardless of their syntactic or morphological

nature, are independent phenomena.

6.1. A challenge: On Oku’s observation and the nature of radical pro-drop

Oku (1998) first observed that Spanish and Japanese differ in non-trivial ways as far as the

interpretative properties of null subject arguments are concerned. Thus, while the null subject

in the Japanese example in (48) is ambiguous between a strict and a sloppy reading, according

to which either John thinks that Mary’s proposal will be accepted or his(=John) own proposal

will, the null subject in (49) only admits the strict reading, according to which the empty

subject can only refer to María’s proposal and not to John’s:

Japanese: strict reading OK, sloppy reading OK

(48) a. Mary-wa [zibun-no teian-ga saiyo-sare-ru-to] omotteiru.

Mary-TOP [self-GEN proposal-NOM accept-pass-pres-comp] think

29

‘Mary1 thinks that her1 proposal will be accepted.’

b. John-mo [e saiyo-sare-ru-to] omotteiru.

John-also [ e accept-pass-pres-comp] think

Lit. ‘John also think e will be accepted.’

Spanish: strict reading OK, sloppy reading *

(49) a. María cree que su propuesta será aceptada.

Maria believes that her proposal will-be accepted

‘Maria believes that her proposal will be accepted.’

b. Juan también cree que e será aceptada.

Juan also believes that it will-be accepted

‘Juan also believes that it will be accepted.’

[Oku 1998: 165]

Two main approaches have been proposed to account for this intriguing contrast. The

first approach is given by Oku, who conjectures that such a contrast correlates with the

presence of long scrambling in Japanese and its absence in Spanish. Oku proposes to connect

these facts in the following way. Assuming that theta roles are features that must be checked,

a division is made between languages in which theta roles can (and must) be checked at LF

and languages in which theta roles have to be checked at syntax. According to Bosković &

Takahashi (1998), in the first type of languages theta features are weak (in Chomsky’s 1995

sense), but strong in the latter. Japanese is an instance of the first type of languages, whereas

Spanish or English are languages with strong theta features. Long distance scrambling can be

analyzed with the scrambled constituent in its surface position. At LF, this constituent moves

to its associated theta position to check the theta feature of some lexical or functional head

(see 50). This is possible only in languages in which theta features are weak as Japanese, but

not in English or Spanish. The scenario in (50), with a strong theta feature, produces a crash at

PF inasmuch the strong feature remains unchecked at syntax.

(50) Scrambled XP [ Yϴ ] (LF checking)

Adopting this approach to scrambling, Oku proposes that (48b) is generated without

an external argument in the syntax. This does not produce any syntactic conflict to the extent

that theta roles need not to be discharged at syntax. At LF, the embedded subject in the

30

antecedent clause can be copied onto the embedded subject position in (48b). Thus, we obtain

the sloppy reading straightforwardly:

(51) a. Mary-wa [zibun-no teian-ga saiyo-sare-ru-to] omotteiru.

b. John-mo [ <zibun-no teian-ga> saiyo-sare-ru-to] omotteiru.

The same procedure is not available in Spanish; a structure without an external

argument will produce a crash at PF given that thematic features are strong in this language.

Therefore, Spanish makes use of an empty pro, a syntactic entity that can satisfy the theta

feature of v. As the English glosses for (49b) indicate, the introduction of a referential

pronoun blocks the sloppy reading automatically.

The second approach connects argument ellipsis to lack of agreement. Or put

conversely, agreement blocks argument ellipsis. This is called the anti-agreement analysis for

argument ellipsis. With some differences in implementation this is the view pursued by Saito

(2007), Şener & Takahashi (2010), Takahashi (2008a,b, 2010, 2013), among others.

Simplifying somewhat, assume that the elliptical gaps in Japanese are filled at LF by copying

the argument antecedent.

(52) a. Mary-wa [zibun-no teian-ga saiyo-sare-ru-to] omotteiru.

Mary-TOP [self-GEN proposal-NOM accept-PASS-PRES-COMP] think

‘Mary1 thinks that her1 proposal will be accepted.’

b. John-mo [ zibun-no teian-ga saiyo-sare-ru-to] omotteiru.

John-also [self-GEN proposal-NOM accept-PASS-PRES-COMP] think

Lit. ‘John also think e will be accepted.’

According to Saito (2007), this copy operation is only available in languages in which

agreement is absent and case assignment is determined by other means. In languages in which

there is a correlation between agreement and case, LF copy is prevented by the activation

condition on Agree (Chomsky 2000, 2001). In other words, given that the Case feature on the

argument antecedent is deleted by Agree before copying it into the elliptical gap, the copied

argument cannot check and delete the unvalued agreement features on the Probe with which is

related (i.e., T in the case at hand).

(53) a. Antecedent: [ Probeφ DP{φ, K} ]

31

b. Elliptical gap: *[ Probeφ DP{φ} ]

With the exception of the recent work by Duguine (see Duguine 2013 and below for

more discussion), most researchers draw a fundamental division between Japanese and

Spanish. Concretely, it is claimed that given that argument ellipsis is not attested in Spanish, it

should be the case that this language, and consistent pro-drop languages in general, are not

amenable to an ellipsis analysis but only to a pro one. Put in a more general way, it seems that

the division between Japanese and Spanish can be derived under the well-known distinction

between surface and deep anaphora (Hankamer & Sag 1976): whereas Japanese null subjects

are instances of surface anaphora (i.e, ellipsis), Spanish makes uses of deep anaphora in

subject position (i.e., pro):

(54) a. Japanese: [IP ... DPSubject ... ] b. Spanish: [IP ... proSubject ... ]

At first sight, Oku’s observation seems to present a challenge to the view I am

defending here and, virtually, to any existent ellipsis analysis of consistent null subject

languages in general. However, under closer inspection, it will turn out that the basic facts are

directly accounted for under the I-assignment system in a somewhat unexpected way.

Concretely, the morphological ellipsis analysis I have proposed for consistent null subject

languages automatically accounts for the fact that Spanish does not allow for phrasal ellipsis

of their subjects, so, the DP su propuesta in (49) cannot be a target for morphological ellipsis.

The only syntactic object that morphological agreement allows to elide is the ϕ-set of which

agreement itself is a mere copy. But notice now that a ϕ-set can only be a pronoun and

nothing else. Thus, absence of sloppy readings in sentences like (49) follows without any

further ado. Of course, we are now left with the crucial question of what prevents, in the

general case, phrasal DP ellipsis in subject position in Spanish, but allows it in Japanese. In

what follows, I will provide an answer in the spirit of the anti-agreement hypothesis which is

fully consistent with the general theory of ellipsis I am proposing here16.

16 The anti-agreement hypothesis has been proven as correct in several languages, although there are remaining problems. It seems for instance that the correlation between presence/lack of agreement and absence/presence of argument ellipsis is not attested in some languages. One of such languages is Hindi where overt subject agreement does not block argument ellipsis (see Simpson Choudhury & Menon 2012). This, however, does not invalidate the hypothesis. Suppose for instance that the activation condition is subject to parametric variation in such a way that for some languages at least there is not need for Agree to take place between a probe and active goal, but just between a probe and a goal with the relevant set of interpretable features capable of valuating the unvalued features of the probe. If the language has overt agreement, then argument ellipsis with overt agreement is a possible result. Crucially, the elimination of the activation condition has been proposed in the literature

32

However, before entering into the details of my analysis, let me briefly comment on

Duguine’s (2013) recent critique to my approach. According to her, the fact that we find cases

of phrasal subject ellipsis like (48b) goes directly against my head ellipsis approach to null

subject phenomena. Instead, she proposes a general and unified phrasal ellipsis account under

which all cases of null subjects across languages are just cases of elliptical DPs. First, it is

important to note that the I-assignment system is fully compatible with both the existence of

syntactic (i.e., phrasal) and morphological (i.e., head) ellipses; indeed, as I have tried to show

in Saab (2009) several paradoxical facts are directly explained if we accept this division in the

first place. So, I-assignment to heads by morphological agreement is just one of the options

my system legitimately allows. As we will see in the next section, Oku’s observation can be

derived under such an approach in an elegant fashion.

Second, crucial to Duguine’s analysis, of course, is the very nature of Oku’s

observation, because, if correct, her unified account would not be able to derive the attested

patterns across languages. In other words, if all null subjects are elliptical DPs, then the

absence of sloppy readings in Spanish for cases like (49) are not correctly ruled out in her

system. This is the reason, I think, that leads Duguine to directly attack Oku’s generalization.

In effect, according to her this is a spurious observation. The point, she argues, is that adding

an objective pronoun in the embedded clause in (49b) co-referential with the main subject

makes the sloppy reading available (Duguine 2013: 442).

(55) A: María cree que [su propuesta le será aceptada (a ella)].

Maria believes that POSS proposal cl.3sg(DAT) will.be accepted to her

Lit. 'Maria believes that her proposal will be accepted to her.'

B: Juan también cree que [[e] le será aceptada (a él)].

Juan also believes that cl.3sg(DAT) will.be accepted to him

Lit. 'Juan also believes that [e] will be accepted to him.'

Sloppy reading OK

In view of this fact, she proposes a new generalization on sloppy readings for null

subjects in Spanish:

precisely on the basis of long-distance agreement facts in Hindi (see Bhatt 2005). Other alternatives, of course, are available. At any rate, I think that the anti-agreement hypothesis deserves careful exploration across languages.

33

(56) Generalization on the sloppy reading in Spanish:

Possessive pronouns embedded within elided DPs fail to give rise to a sloppy reading

when they do not have a local antecedent. (Duguine 2013: 441)

This is a curious observation that does not seem to follow from any obvious constraint

on sloppy readings in, for instance, well-known ellipsis contexts. Indeed, as Duguine

acknowledges, the sloppy reading in (49b) automatically reappears whenever the embedded

clause is part of an elliptical TP (see Duguine 2013: 444, footnote 33). But this fact is

obviously derived under the (rough) analysis in (57b) below, where su propuesta, which can

be co-referential with the matrix subject, is not a null subject, but a full DP contained in the

elliptical TP.

(57) a. María cree que su propuesta será aceptada.

Maria believes that her proposal will-be accepted

‘Maria believes that her proposal will be accepted.’

b. Juani también [cree que sui propuesta será aceptada]

Juan also believes that it will-be accepted

‘Juan also believes that it will be accepted.’

Therefore, (57) has nothing intriguing; it is just a typical case of sloppy reading under

ellipsis. It would be puzzling only if we accepted that Spanish null subjects are elliptical DPs,

as Duguine proposes. Under my proposal for instance (but the same under a pro analysis)

there is nothing special here. At any rate, what seems to be spurious is not Oku’s observation

but the generalization in (56). Let us see why.

First, for my informants, but apparently not for Duguine’s, it is important to have

some contrast between the object DPs in parentheses. Without this contrast, the sloppy

reading is clearly disfavored and, even thus, speakers’ reactions are quite unstable. Duguine’s

informants, instead, prefer that the DP object to be null, at least in very similar examples (see

Duguine 2013: 439, footnote 23). In any case, the judgments are not consistent among

speakers.

Second, and even more important, speakers’ judgments are entirely consistent in cases

like the followings, which do not allow for sloppy readings even when they observe the

condition in (56):

34

(58) A: Juanj cree que suj noviai loj ama (a élj).

J. believes that his girlfriend him loves to him

‘Juanj believes that hisj girlfriendi loves him.’

B: Pedrok también cree que [e]i lok/j ama (a élk/j)

P. also believes that [e] him loves (to him)

‘Pedro also believes that shei loves him.’

(59) A: Juanj dice que suj madre loj criticó (a élj).

J. says that his mother Cl.acc criticized (to him)

‘Juani says that hisj motheri criticized himj.’

B: Maríai también dice que [e] lai criticó (a ellai).

M. also says that Cl.acc.fem criticized (to her)

‘María also says that shei criticized her.’

(60) a. A Juani lei pegó sui madrej.

to J. CL.DAT hit his mother

‘Hisi mother hit Johni.’

b. Pedrok espera que [e] no lek pegue a élk.

P. hopes that not CL.DAT hits to him

‘Pedro hopes shei does not hit him. ’

(61) A: Juan cree que su madrei le regaló un libro.

J. believes that his mother CL.DAT gave a book

‘Juan belives that his mother gave him a book. ’

B: Pedro también cree que [e] le regaló un libro.

P. also believes that CL.DAT gave a book

‘Pedro also believes that shei gave him a book.’

(58-61: [e]= strict reading)

One can construct myriads of examples like these and the result would be that the

generalization in (56) is falsified. Of course, in each case, the sloppy reading arises whenever

the full possessive DP occurs in the relevant subject position and whenever TP ellipsis is

allowed.

35

So far, it seems that Duguine’s observation does not hold. There is still a set of data,

those patterns like the example in (55B), that produces particular reactions in the speakers.

But this, of course, does not lead us to generalize the worst case scenario, as is well known

that sloppy readings are also attested for deep anaphora (i.e., pronouns) under some particular

conditions (see Saab 2009 and the references therein). In effect, my own impression is that

data like (55B) and similar ones suppose some type of pragmatic accommodation. The fact

that some speakers react allowing a sloppy reading is due to the fact that the strict reading for

those particular examples is at odds with our common sense that someone will accept John’s

proposal to Peter, although the relevant context can be constructed. Recall that my informants

prefer the embedded objects to be contrasted, showing that we are talking about different

alternatives for the variable in “x’s proposals”. For those speakers that accept the sloppy

reading when the embedded indirect objects are null, it seems that they have constructed a

previous background according to which we were talking about different proposals (John’s,

Peter’s and so on) to be accepted.

At any rate, Duguine’s uniform analysis also fails to account for Takahashi’ (2008a,b

and 2010) additional observation that, whereas null subjects in Japanese can be ambiguous

between a quantificational and an E-type reading (see 62), Spanish does not, as shown in (63):

(62) a. Sannin-no mahootukai-ga Taroo-ni ai-ni kita.

three-GEN wizard-NOM Taroo-DAT see-to came

‘Three wizards came to see Taroo.’

b. [e] Hanako-ni-mo ai-ni kita.

Hanako-DAT-also see-to came

‘lit. e came to see Hanako, too.’

[e] = the set of wizards are coincident (E-type reading).

[e] = the set of wizards can be divergent (quantificational reading)

(63) a. Tres magos vinieron a ver a Juan.

three wizards came to see J.

b. [e] Vinieron a ver a Pedro también.

came to see to P. also

(only E-type reading)

Thus, null subjects in Spanish behave (again) as English weak pronouns:

36

(64) a. Three wizards came to see Taroo.

b. They came to see Hanako, too. (only E-type reading)

The quantificational readings, as argued by Takahashi at length, are directly derived

under the DP ellipsis analysis:

(65) Sannin-no mahootukai-ga Hanako-ni-mo ai-ni kita.

three-GEN wizard-NOM Hanako-DAT-also see-to came

It is easy to see now that Duguine’s uniform analysis cannot account for the absence

of quantificational readings in examples like (63b) or, put differently, her analysis

overgenerates quantificational readings in contexts where they are clearly impossible17.

17 Additional evidence in favor of the distinction between phrasal DP ellipsis and head ellipsis (or pronoun ellipsis) comes from another consistent pro-drop language like Hungarian. Notice first that in examples like (i) only the strict reading is possible: (i) A: Mari azt hiszi, hogy eltört a lába.

Mari that.acc believes that broke the foot.poss3sg ‘Mari believes her foot is broken.’

B: Péter is azt hiszi, hogy eltört. Peter also that.acc believes that broke

‘Péter also believes her foot/*his foot is broken.’ (only strict reading)

As for Takahashi’s observation, notice now that only the E-reading is grammatical: (ii) A: Három varázsló meglátogatta Jánost. three wizard visited.3sg János.acc ‘Three wizards visited János.’ B: Meglátogatták Pétert is. visited.pl Péter.acc too ‘They visited Péter, too.’

As Anikó Lipták (p.c.) points out to me the conjugation on the verb in (iiB) has to be plural. In (iia) it is singular, because the noun ‘wizard’ is singular (after numerals, Hungarian requires singular nouns). In (iiB), however, you cannot have singular agreement as a new utterance, because the reference is plural: (iii) * Meglátogatta Pétert is. visited.sg Péter.acc too ‘He visited Péter, too.’

This pattern is compatible with the pro or the head ellipsis analysis for consistent NS languages, but not with Duguine’s uniform analysis in terms of DP-ellipsis. Concretely, under a DP-ellipsis analysis, (iii) should be grammatical with a singular verb because the elliptical subject is singular: (iv) * Három varázsló meglátogatta Pétert is. three wizard visited.sg Péter.acc too

37

Having shown that Oku’s observation indeed holds and, more importantly, that my

analysis correctly rules out the basic patterns, I will provide now a more explicit answer as for

why DP-ellipsis is allowed in Japanese but not in Spanish subject position, where in general

only D heads can be subject to ellipsis, as defined in this paper (see section 3).

6.2. Deriving Oku’s observation

The crucial fact to be accounted under an ellipsis approach to null subjects is why argument

ellipsis (i.e., ellipsis of full DPs) is not allowed in Spanish. I already have shown that the

difference between radical and consistent pro-drop does not reduce to an ellipsis vs. a pro

analysis; the minimal assumption we need is that, whereas Japanese and related languages

allow for DP-ellipsis in the syntax, Spanish only licenses ellipsis of ϕ-sets at PF (modulo

other cases of topic-drop such as indefinite object drop and similar phenomena).

My implementation of Oku’s observation follows the spirit of previous works framed

under the so-called anti-agreement hypothesis that correlates absence or presence of

agreement as a crucial ingredient of the theory of argument ellipsis (see, among many others,

Saito 2007 and Takahashi 2010). The minimal assumptions we need to account for the basic

patterns are listed below, some of which are rather uncontroversial:

(A) There is a principle of recoverability.

(B) Nominal arguments with phonetic content must have case (K) at PF (i.e., Case Filter)

(C) Agreement is a PF phenomenon, as assumed in this paper (see also Bobaljik 2008,

among others)

(D) In Spanish, but not in Japanese, agreement is parasitic on K. This is the case either

because Japanese lacks agreement (see Saito 2007 and references therein) or because it

does have abstract agreement but it is not parasitic on K.

Assumptions (A) and (B) do not require more elaboration, as I assume they are fairly

uncontroversial. The assumption in (C) has already been made in this paper, so it is mainly

required by internal considerations. Other implementations in the spirit of my system could be

done in a syntactic approach to agreement, but I will not follow this route of analysis here.

Finally, the claim in (D) is at the heart of the contrast between radical and consistent pro-

drop. It is by no way a novel assumption, as the reader can confirm by comparing previous

work under the anti-agreement hypothesis (see, among others, Saito 2007, Şener & Takahashi

2010, Takahashi 2008a,b, 2010, 2013, and Miyagawa 2013). Put in a general way, my own

38

view about this assumption is that there is no syntactic difference as far as the mechanism that

assigns [nominative] in the syntax (or PF, see Bobaljik 2008) is concerned; the difference is

that Spanish and related languages have a PF mechanism of agreement that adds a dissociated

agreement morpheme on the basis of a K-marked DP. Simplifying the analysis, we can

illustrate the difference between both languages as follows:

Japanese:

(66) [TP DPK[?] T ] � DPK[nominative] Syntax

Spanish:

(67) a. [TP DPK[?] T ] � DPK[nominative] Syntax

b. [TP DPK[nominative] T+Agr] Agreement at PF

Notice now that even when assumption (B) is quite uncontroversial, it implicitly

contains a corollary that has not been stressed in the literature on Case, namely, given

minimalist assumptions, K is freely assigned to DPs, i.e., DP(K). Put differently, nothing goes

wrong with a configuration like this as far as syntax is concerned:

(68) [TP DP T ] Syntax

Of course, given the assumption in (B) such a configuration will produce a PF crash.

We obtain then the following corollary:

(69) Corollary : Don’t spell out (i.e., don’t pronounce) a K-less DP.

This situation gives a legitimate result in the syntax/LF, provided that the principle of

recoverability is satisfied (see assumption A). At PF, however, the result is divergent

depending on the language: whereas the object (70b) is legitimate in Japanese, a language

without morphological agreement, it is illegitimate in Spanish (71b), because by assumption

D the morphophonological properties of T cannot be satisfied at this level.

Syntax/LF PF

(70) a. TP b. TP 2 # Japanese DP T T

39

Syntax/ LF PF (71) a. TP b. * TP 2 # Spanish DP T Tφ

Put in the terms of this paper, we can assume that a K-less DP is I-assigned in the

syntax. This is locally determined by the computational system: a K-less argument DP is

automatically I-assigned in the syntax by local inspection internal to the DP structure. Of

course, other alternatives are also conceivable. At any rate, what is worth noting now is that

the direct prediction of this analysis is that argument ellipsis is ellipsis of a K-less argument.

This prediction was already confirmed in the literature by Saito (2007) who also claims that

null arguments in Japanese are caseless18. Thus, argument ellipsis is derived as case of phrasal

I-assignment in the syntax.

18 Saito convincingly shows that this is indeed the case in Japanese on the basis of the well-known alternation between genitive and nominative subjects in this language (all data from Saito 2007): (i) a. [ Taroo-ga /-no itta ] tokoro -NOM/-GEN went place

‘the place where Taroo went’ b. Taroo-ga /*-no soko –e itta

-NOM/ -GEN there-to went ‘Taro went there.’

As Saito shows, the occurrence of an accusative argument in cases like (ia) prevents the occurrence of a

genitive subject: (ii) *[ Taroo-no hon –o katta ] mise

-GEN book-ACC bought shop ‘the shop where Taroo bought a book’

Compare with the cases in (iii), where no accusative argument is present: (iii) a. [ Taroo-no kino itta ] tokoro

-GEN yesterday went place ‘the place where Taroo went yesterday’

b. [ Taroo-noi ti taihosareta ] tokoro -GEN arrested-was place

‘the place where Taroo was arrested’ c. *[ hon –oi Taroo-no ti katta ] mise

book-ACC -GEN bought shop ‘the shop where Taroo bought a book’

Now, the contrast in (v) clearly demonstrates that null objects in Japanese have not accusative case,

confirming the idea that elliptical argument in Japanese are caseless: Context: (iv) Ziroo-ga hazimete Nagoya-ni kuru -node, minna-ga -NOM for the first time -to come-since all -NOM

iroirona basyo-ni kare-o turete iku yotei-desu various place –to he -ACC take plan -is

40

An immediate advantage of this approach is that it explains why the phenomenon is

not attested in adjunct position. In effect, in a sentence like (72b) (taken from Takahashi

2010), you cannot interpret that John didn’t wash a car carefully, but only that he washes a

car:

(72) a. Bill-wa kuruma-o teineini aratta.

Bill-TOP car-ACC carefully washed

‘Bill washed a car carefully.’

b. John-wa e arawanakatta

John-TOP washed.not

‘lit. John didn’t wash e.’

As far as I know, this is a very general pattern across languages. If the proposal I am

making here is correct, then the ban of adjunct ellipsis follows because adjuncts simply do not

have a structural K feature19.

Notice now that the reasons that prevent DP-ellipsis in Spanish license the so-called

pro-drop property: a pronoun induces agreement at PF and, precisely, in virtue of such an

‘Since Ziroo is coming to Nagoya for the first time, the plan is for everyone to take him to various places’

(v) a. *[ Hanako-no kare-o turete iku ] tokoro-wa Nagoya-zyoo -desu -GEN he -ACC take place -TOP Nagoya Castle-is

‘The place that Hanako is taking him is the Nagoya Castle.’ b. [ Hanako-no kare turete iku ] tokoro-wa Nagoya-zyoo -desu

-GEN take place -TOP Nagoya Castle-is 19 Takahashi (2010) conjectures that argument ellipsis must be licensed by some selecting head, which is the case with arguments but not with adjuncts. So, the problem with adjuncts is not identification but licensing. However, this conjecture cannot be on the right track. Consider the English example in (i): (i) *John loves his son and Peters also loves. Under the antiagreement hypothesis, the ungrammaticality of (i) has to be connected to the fact that direct objects in English participate in agreement with v and consequently cannot be elided. Now consider (ii): (ii) A: The solution to Johni’s problems depends on hisi son. B: *The solution to Peterk’s problems also depends on hisi/k son.

What accounts for the ungrammaticality of (iiB) now? I see no obvious solution for Takahashi’s account of (72) because: (a) the prepositional complement in (ii) is obviously selected by the main verb and, (b) arguably, no agreement relation is at play here between the PP complement and little v. Of course, the fact that there is no agreement here can be demonstrated as false, but this would require strong empirical evidence. Under the analysis proposed here, instead, this follows just because prepositional complements of this type are not endowed with a K feature.

41

operation this pronoun can be I-assigned under formal identity. The trees in (73) illustrate the

derivation of null subjects in Spanish in a simplified way (see section 4 for details; 73b does

not make reference to linearization only for expository convenience):

Syntax/LF PF (73) a. TP b. TP 2 2 Spanish DP T D T 2 Head ellipsis � T Agr (D = Agr)

As observed above, agreement only licenses I-assignment of pronominal entities (i.e.,

ϕ-sets) under identity. Crucially, phrases cannot be deleted under agreement. Hence, we

derive Oku’s observation in a theory that conceive both types of null subjects in Spanish and

Japanese as derived via ellipsis and that, at the same time, avoid all the problems mentioned

in connection with Duguine’s theory.

(74) a. María1 cree que su1 propuesta será aceptada.

b. Juan también cree [que pro será aceptada].

(75) a. Mary-wa [zibun-no teian-ga saiyo-sare-ru-to] omotteiru.

Mary-TOP [self-GEN proposal-NOM accept-PASS-PRES-COMP] think

‘Mary1 thinks that her1 proposal will be accepted.’

b. John-mo [ zibun-no teian-ga saiyo-sare-ru-to] omotteiru.

John-also [ self-GEN proposal-NOM accept-PASS-PRES-COMP] think

Lit. ‘John also think e will be accepted.’

In sum, we obtain the following scenario, fully compatible, and indeed predicted, by

the theory defended in Saab (2009) and in this paper:

(76) Syntax/LF: Argument ellipsis XPs (Japanese)

Agreement

PF: Null subjects Xs (Spanish)

42

6.3. Partial pro-drop vs. radical pro-drop: Some speculations

Partial pro-drop languages of the BP type have two main properties, namely: (i) null generics

and (ii) anaphoric third person null subjects in embedded position. It seems then that Oku’s

observation cannot be tested in partial pro-drop languages (see also Barbosa 2013). I have

obtained however two types of reactions depending on the speaker, those that apparently

behave as Spanish speakers (although see below) and those that react as Japanese ones.

(77) A: João disse que sua proposta será aceita.

J. says that his proposal will-be accepted

‘J. says that his proposal will be accepted.’

B: Pedro também disse que será aceita.

P. also says that [e] will-be accepted

‘P. also says that it will be accepted.’

[Strict reading Ok, sloppy reading: %]

Crucially, all the consulted speakers are partial pro-drop in the sense that they allow

null generics in the relevant contexts (see the examples in 36). There are various speculations

to be done in this respect. First, it is important to stress at this point that, as observed by Kato

(2011), null generics seem to be part of the core grammar acquired by BP children, but

anaphoric null subjects are acquired late not as part of the acquisition process but because of

schooling. But interestingly schooling does not convert speakers in consistent pro-drop ones,

because there is not available mechanism such as head ellipsis in the core grammar to produce

the correct output. At least three strategies seem to be available in general for non-NS

speakers: (i) empty NP-anaphora, (ii) Copy Deletion, and (iii) phrasal DP ellipsis of the

Japanese type. Depending on the language and other overlapping factors, the three strategies

are indeed attested and have already proposed in the literature, as discussed previously in this

paper. As noticed by Barbosa, the DP ellipsis analysis seems to be unavoidable in Japanese if

we want to derive Oku’s observation (and also Takahashi’s one), but Chinese, instead, where

sloppy readings are not attested in subject position (see 78 from Takahashi 2008a, 2010 and,

see also, Miyagawa 2013, among others, for discussion), seems to be more amenable to an

empty NP anaphora analysis.

(78) a. Zhangsan shuo ziji de haizi xihuan Xiaohong.

43

Zhangsan say self of child like Xiaohong

‘Zhangsan said his child liked Xiaohong.’

b. Lisi shuo e xihuan Xiaoli.

Lisi say like Xiaoli

‘lit. Lisi said e liked Xiaoli.’ (only strict reading)

I agree with Barbosa’s conclusion: even when the reasons licensing NP-anaphora or

DP-ellipsis are obscure under our current understanding (see, however, Takahashi 2010 and

Miyagawa 2013). As for BP, given the reactions provided by my BP informants, it seems that

there is a split between those who react as Japanese speakers and those who react as Chinese

ones. Although of course further research is needed in this respect, it is interesting to note that

BP speakers do not seem to react as pro-drop speakers20.

Now, I see no empirical or conceptual reason to reject cases of phrasal movement in

hyper-raising contexts to account for some types of embedded null subjects, in consonance

with proposals like Ferreira (2000) or Rodrigues (2004). Indeed, I think that we can get a

better understanding of null subject phenomena across languages if we accept that the three

strategies are in principle available for non-NS languages in general and, specifically, for

partial and radical pro-drop ones. Under closer inspection, there are good reasons to believe

that the three types of null subjects form a natural class connected with general properties of

the agreement-case system of a given language. On the one hand, it should be the case that

bare NPs and even ϕPs have to fail to be pronounced (in Holmberg’s sense) because of the

Case Filter (see Assumption B): Case is a property of DPs. Of course, K-less DPs fail to be

pronounced for the same reason. So, empty NP anaphora and DP-ellipsis form a class of

natural phenomena as far as their PF nature is concerned; the difference between them boils

down to the difference between deep and surface anaphora, respectively. In this way, we also

capture the observation by Tomioka (2003), stressed and further elaborated by Barbosa (2010,

2013), that there is a correlation between the productive distribution of bare NPs in a given

language and radical pro-drop. Given our assumptions between agreement and case, the

availability of the phenomenon should be allowed only in those languages in which there is

no agreement at all or in which agreement and case are not connected in the way stated by

assumption (D). In sum, NP anaphora and DP-ellipsis are allowed to different extents in

languages in which K is not a prerequisite for agreement to take place. Then, the extent to

20 For those speakers that only accept the strict reading, it can be also conjectured that they are bilingual (i.e., full pro-drop and partial pro-drop), as Kato (2011) suggests, because of independent reasons.

44

which every language allows for DP-ellipsis or empty NP-anaphora (and different projections

of empty Ns) or both should be explored in a case by case fashion.

As mentioned, some part of the literature is also committed with the idea that certain

anaphoric subjects both in finite and non-finite contexts are derived by A-movement.

Roughly, under such an approach, a case of hyperraising is analyzed in the following way:

(79) Os meninos parecem que os meninos gritam.

the children seem.PL that shout.PL

‘The children seem to shout.’

The question is what the mechanism that deletes copies is. If the system we presented

in section 3 is on track, then traces of arguments are just elliptical DPs: Movement is Copy

plus I-assignment to lower copies. Suppose that a DP with a K feature moves in order to value

this feature. The minimal assumption is that the K feature is valued only for the copy in the

landing position, but not for the lower one, which is, I think, the default hypothesis (see

Nunes 2004 for extensive discussion). Now, a copy with an unvalued K feature fails to be

pronounced by the Case filter. Or in the terms of this paper, a DP with an unvalued K feature

is I-assigned. This is the simpler way in which a copy of a nominal argument is elided,

because it only requires local inspection within the structure of the DP (see above). Once the

system recognizes a DP copy with an unvalued K feature automatically assigns an I-feature.

This way of copy deletion can now be extended to all copies with a K feature, regardless of

valuation. In other words, copy deletion for a K-specified argument is deletion of its K

feature. I think this is a natural conclusion if economy plays some role in the UG design,

where local operation wins over non-local or less local ones. Again, this forces us to draw a

fundamental division between arguments and adjuncts, because for adjunct copies to be

deleted we need a different mechanism, one that is not so local and, as a minimum, requires

searching an antecedent to delete the relevant copies. Therefore, we expect argument-adjunct

asymmetries in the realm of copy deletion, similar to what we find in the realm of DP-ellipsis

in Japanese where adjuncts cannot be subject to ellipsis (see 72). Such asymmetries are

indeed well known and involve for instance relativized minimality effects:

(80) a. Qué te preguntás quién compró que+I?

what CL.2p wonder.2p who bought

‘?What do you wonder who bought.’

45

b. *Cuándo te preguntás quién compró un auto cuándo+I?

when CL.2p wonder.2p who bought a car

‘When do you wonder who bought a car.’ (ok under the short construal reading)

For the object copy to be deleted in (80a) and antecedent is not required because the K

feature of the DO copy is enough to induce I-assignment. Such a possibility of course is not

available with an adjunct copy which must be deleted under I-assignment through the

localization of a local, c-commanding antecedent (i.e., the higher copy). Closeness plays a

role here as in other cases of ellipsis. In the case at hand, a wh-element like quién blocks I-

assignment for the adjunct. It seems then that we have empirical reasons to think of argument

ellipsis and copy deletion as forming a natural class of phenomena21.

In sum, null arguments in the syntax can be the result of three independent available

mechanisms: (i) NP-anaphora, (ii) DP-ellipsis or, (iii) Copy Deletion. Strictly speaking, the

first strategy goes from having a non-projected argument in the syntax to different sorts of NP

projections. Different tests, like binding, should be constructed in order to show what level of

projection a given NP anaphora has in a particular language (see section 5.2. above). At any

rate, this strategy reduces to a type of deep anaphora strategy. Instead, DP-ellipsis and Copy

Deletion should be thought of a type of surface anaphora phenomenon. The distinction

between both can be detected under usual tests of movement and ellipsis (island effects, for

Copy Deletion, or occurrence across the discourse, for DP-ellipsis). Both phenomena,

however, display a similar behavior in other relevant domains. I have shown, for instance, that

argument-adjunct asymmetries are attested for DP-ellipsis and Copy Deletion. In a broader

perspective, then, different types of syntactically licensed null subjects boil down to the

indubitable existence of deep and surface anaphora across languages. In this respect, again, a

general theory of ellipsis seems to be superior to its competitors. A crucial novelty of this

paper, however, is that null subjects of consistent pro-drop languages should not be confused

with any of the syntactic strategies discussed in this section. In terms of the deep vs. surface

anaphora distinction, these null subjects are surface anaphora of an underlying pronoun (i.e.,

deletion of a deep anaphora!) and this phenomenon must be kept apart from the NP-anaphora

phenomena. The final picture we get is illustrated as follows, where the ordering between

phrasal ellipsis and NP-anaphora is irrelevant:

21 The connection between relativized minimality effects and case is not new: Kitahara (1999) also relates the contrast observed in (80) to Case theory, although his implementation is clearly different to the suggestion made here that the underlying reason that explains such a contrast is locality for phrasal ellipsis.

46

(81) Syntax/LF Phrasal ellipsis DP ellipsis Copy deletion

NP-anaphora

Head ellipsis

PF 7. Conclusion

In this paper, I have defended the idea that the null subjects of NSLs are instances of

morphological ellipsis. This type of ellipsis is subjected to the same conditions that apply to

other kinds of post-syntactic operations. In the case of null subjects, adjacency is the crucial

locality condition. The theory predicts several situations where a potentially elliptical subject

has to be pronounced. We have seen this case instantiated in NIDs, which are NSLs with

some obligatory overt subjects. In particular, this type of clitic subjects arises as a

consequence of the Sub-word Deletion Corollary, which prevents sub-words of being affected

by Non-insertion. Clitic subjects in NIDs form then a natural class with other cases of copy

realization across languages (see footnote 6 for an illustration).

Partial pro-drop languages of the BP type are, instead, cases of non-NSLs with some

null subjects licensed in the syntax. Crucially, the impoverishment of the verbal paradigm in

BP caused that the EPP in this language cannot be resolved morphologically. A syntactic way

of EPP checking, like in non-NSLs, then arose and, as a consequence, other pro-drop

properties also vanished. I have shown how this conjecture has some positive results when we

consider the complementary distribution between impersonal null subjects and the impersonal

clitic SE in Romance.

As for the syntactic mechanism that licenses null subjects, I have suggested that a

fundamental division between deep and surface anaphora is required to capture the full

distribution across and within partial (and also radical) pro-drop languages. Different

projections of empty Ns (from zero to ϕ and maybe to D, a case which we did not discuss

here) accounts for the pronominal behavior of some null subjects in partial and radical NS

languages. The emptiness of such entities follows from case theory, as discussed. As for

elliptical subjects (i.e., those derived as surface anaphora), they can come in two guises: (i) as

mere copies of their antecedents or (ii) as K-less DPs. In both cases, the PF result is the same

47

that we obtain with respect to deep anaphora: A K-less DP fails to be pronounced. Put

differently, if deletion of argument copies is deletion of their K feature, then both phenomena

form a natural class as their PF status is concerned. Well-known argument-adjunct

asymmetries in both domains show that this conjecture could be on the right track. However,

more research is needed in order to know when DP-ellipsis or Copy Deletion across finite

domains (i.e., hyperraising) is allowed in a given language. At any rate, we predict that radical

and partial pro-drop languages should generally allow for empty NP anaphora and DP-ellipsis

(in one of the two forms discussed here).

Finally, an important theoretical result of this work is that it integrates the long-

standing problem of the NSP into a general theory of ellipsis, whose most striking property is

that grammatical silences are derived in the course of the syntactic derivation under the same

conditions that are needed independently for other syntactic mechanisms (displacement, for

instance). As I have shown elsewhere (Saab 2009; see also footnote 6), this theory

successfully extends to domains that go beyond the particular domain of null subjects.

References

Alexiadou, Artemis, & Elena Anagnostopoulou (1998) Parametrizing AGR: Word order, V-

movement, and EPP-checking. Natural Language & Linguistic Theory 16:491–540.

Barbosa, Pilar (1995) Null subjects. Phd dissertation, MIT.

Barbosa, Pilar (2010) Partial pro-drop as null NP anaphora. Talk given at NELS 41, UPenn.

Barbosa, Pilar (2013) 'pro' as a minimal NP: towards a unified theory of 'pro'-drop. Ms.,

available at http://ling.auf.net/lingbuzz/001949

Barbosa, Pilar, Maria Eugênia Duarte & Mary Kato (2005) Null Subjects in European and

Brazilian Portuguese. Journal of Portuguese Linguistic 4: 11-52.

Bhatt, Rajesh (2005) Long Distance Agreement in Hindu-Hurdu. Natural Language &

Linguistic Theory 23: 757-807.

Bobaljik, Jonathan (2008) Where's Phi? Agreement as a Post-Syntactic Operation. In Daniel

Harbour, David Adger, and Susana Béjar (eds.) Phi-Theory: Phi features across interfaces

and modules, 295-328. Oxford: Oxford University Press.

Bošković, Željko & Daiko Takahashi (1998). Scrambling and Last Resort, Linguistic Inquiry

29: 347-266.

Brandi, Luciana & Patrizia Cordin (1989) Two Italian Dialects and the Null Subject

Parameter. In Osvaldo Jaeggli & Ken Safir (eds.) The null subject parameter, 111-142.

Dordrecht: Kluwer.

48

Brucart, José (1987) La elisión sintáctica en español. Barcelona: Bellaterra.

Camacho, José (2013) Null Subjects. Cambridge: Cambridge University Press.

Camacho, José (this volume) The Null Subject Parameter revisited. The evolution from null

subject Spanish and Portuguese to Dominican Spanish and Brazilian Portuguese.

Cardinaletti, Anna & Michal Starke (1999) The Typology of Structural Deficiency. In Henk

van Riemsdijk (ed.) Clitics and other functional categories in European languages, 145-

233. Berlin: Mouton de Gruyter].

Chomsky, Noam (1982) Some concepts and consequences of the theory of government and

binding. Cambridge, Mass.: MIT Press.

Chomsky, Noam (1995) The Minimalist Program. Cambridge, Mass.: MIT Press.

Chomsky, Noam (2000) Minimalist inquiries: The framework. In R. Martin, D. Michaels & J.

Uriagereka (eds.) Step by step: Essays on minimalist syntax in honor of Howard Lasnik,

89-155. Cambridge, Mass.: MIT Press.

Chomsky, Noam (2001) Derivation by phase. In M. Kenstowicz (ed.) Ken Hale. A life in

language, 1-52. Cambridge, Mass.: MIT Press.

Depiante, Marcela & Pascual Masullo (2001) Género y número en la elipsis nominal:

consecuencias para la hipótesis lexicalista. Talk given at the I Encuentro de Gramática

Generativa, Gral. Roca, November 22-24.

Duarte, Maria Eugênia L. (1993) Do pronome nulo ao pronome pleno: a trajetória do sujeito

no português do Brasil. In Ian Roberts & Mary A. Kato (eds.) Português Brasileiro: Uma

viagem diacrônica (Homenagem a Fernando Tarallo), 107-128. Campinas: Editora da

UNICAMP.

Duarte, Maria Eugênia L (1995) A perda do princípio “Evite pronome” no Português

Brasileiro. Phd dissertation, UNICAMP.

Duarte, Maria Eugênia L (2000) The loss of the ‘Avoid Pronoun’ principle in Brazilian

Portuguese. In Mary Kato & Esmeralda Negrão (eds.) The Null Subject Parameter in

Brazilian Portuguese, 17-36. Frankfurt-Madrid: Vervuert-Iberoamericana.

Duguine, Maia (2013) Null Arguments and Linguistic Variation: A Minimalist Analysis of

pro-drop. Phd dissertation, University of the Basque Country/University of Nantes.

Embick, David (2007) Linearization and Local Dislocation: Derivational Mechanics and

Interactions. Linguistic Analysis 33: 2-35.

Embick, David & Rolf Noyer (2001). Movement Operations after Syntax. Linguistic Inquiry

32: 555-595.

49

Ferreira, Marcelo (2000) Argumentos Nulos em Português Brasileiro. Master Thesis,

UNICAMP.

Halle, Morris & Alec Marantz (1993) Distributed Morphology and the Pieces of Inflection. In

Kenneth Hale & Samuel Keyser (eds.) The view from Building 20, 11-176. Cambridge,

Mass.: MIT Press.

Hankamer, Jorge & Ivan Sag (1976) Deep and Surface Anaphora. Linguistic Inquiry 7: 391-

426.

Heim, Irene (1982) The Semantics of Definite and Indefinite NP’s. Ph.D. dissertation,

University of Massachusetts at Amherst.

Holmberg, Anders (2005) Is There Little Pro? Evidence from Finnish. Linguistic Inquiry 36:

533-564.

Holmberg, Anders (2010a) Null Subject Parameters. In Theresa Biberauer, Anders Holmberg,

Ian Roberts & Michelle Sheehan (eds.) Parametric Variation: Null subjects in minimalist

theory, 88-124. Cambridge, UK: Cambridge University Press.

Holmberg, Anders (2010b) The null generic subject pronoun in Finnish: a case of

incorporation in T. In Theresa Biberauer, Anders Holmberg, Ian Roberts & Michelle

Sheehan (eds.) Parametric Variation: Null subjects in minimalist theory, 200-230.

Cambridge, UK: Cambridge University Press.

Holmberg, Anders, Aarti Nayudu & Michelle Sheehan (2009) Three partial null-subject

languages: a comparison of Brazilian Portuguese, Finnish and Marathi. Studia Linguistica

63: 59-97.

Hornstein, Norbert (1999) Movement and control. Linguistic Inquiry 30: 69-96.

Jelinek, Eloise (1984) Empty categories, Case, and configurationality. Natural Language &

Linguistic Theory 2:39–76.

Johns, Christopher (2005) Unassigned theta-roles in Finnish: the null generic subject. Talk

given at the Linguistics Association of Great Britain (LAGB) Annual Meeting, University

of Cambridge.

Kato, Mary (1999) Strong and weak pronominals in the Null Subject Parameter. Probus 11:

1-37.

Kato, Mary (2000) The Partial Pro-Drop Nature and the restricted VS Order in Brazilian

Portuguese. In Mary Kato & Esmeralda Negrao (eds.) Brazilian Portuguese and the Null

Subject Parameter, 223-258. Frankfurt-Madrid: Vervuert-Iberoamericana.

50

Kato, Mary (2011) Acquisition in the context of language change: the case of Brazilian

Portuguese. In E. Rinke & T. Kupisch (eds.) The development of Grammar: Language

Acquisition and Diachronic Change, 309-330. Amsterdam: John Benjamins.

Kitahara, Hisatsugu (1999) Eliminating * as a Feature (of Traces). In Samuel D. Epstein &

Norbert Hornstein (eds.) Working Minimalism, 77-93. Cambridge, Mass.: MIT Press.

Landau, Idan (2010) The explicit syntax of implicit arguments. Linguistic Inquiry 41(3): 357-

388.

McDaniel, Dana (1986) Conditions on wh-Chains. PhD dissertation, the City University of

New York.

Manzini, M. Rita & Leonardo Savoia (1997) Null Subjects without pro. UCL Working Papers

in Linguistics 9: 301-313.

Merchant, Jason (2001) The syntax of silence: Sluicing, islands, and the theory of ellipsis.

Oxford: Oxford University Press.

Merchant, Jason (forthcoming) Not all genders are created equal: Evidence from nominal

ellipsis in Greek. Lingua.

Modesto, Marcello (2000) Null Subjects without ‘Rich’ Agreement. PhD dissertation,

University of South California.

Miyagawa, Shigeru (2013) Surprising agreements at C and T. Talk given at Romania Nova

VI, Natal, Brazil.

Müller, Guereon (2005) Pro-drop and impoverishment. In P. Brandt and E. Fuss (eds.) Form,

structure and grammar. A Festschrift presented to Günther Grewendorf on the occasion of

his 60th birthday, 93-115. Tübingen: Narr.

Murguia, Elixabete. 2004. Syntactic identity and locality restrictions on verbal ellipsis. PhD

dissertation, University of Maryland.

Nunes, Jairo (2004) Linearization of chains and sideward movement. Cambridge, Mass.: MIT

Press.

Nunes, Jairo (2008) Inherent case as a licensing condition for A-movement: The case of

hyperraising constructions in Brazilian Portuguese. Journal of Portuguese Linguistics 7:

83–108.

Nunes, Jairo & Cynthia Zocca (2009) Lack of morphological identity and ellipsis resolution

in Brazilian Portuguese. In Jairo Nunes (ed.) Minimalist essays on Brazilian Portuguese

syntax, 215–236. Amsterdam: John Benjamins.

Oku, Satoshi (1998) A Theory of Selection and Reconstruction in the Minimalist Perspective.

PhD dissertation, UConn.

51

Ordóñez, Francisco (1997) Word Order and Clause Structure in spanish and Other Romance

Languages. PhD dissertation, the City University of New York.

Panagiotidis, Phoevos (2002) Pronouns, clitics and empty nouns. Amsterdam: John

Benjamins.

Perlmutter, David (1971) Deep and surface constraints in generative grammar. New York:

Holt, Rinehart and Winston.

Poletto, Cecilia (2000) The Higher Functional Field in the Northern Italian Dialects. Oxford:

Oxford University Press.

Pujalte, Mercedes (2013) Argumentos (no) agregados. Indagaciones sobre la morfosintaxis

de la introducción de argumentos en español. PhD dissertation, Universidad de Buenos

Aires.

Pujalte, Mercedes & Andrés Saab (2012) Syncretism as PF-repair: the case of SE-insertion in

Spanish. In Cristina Cuervo & Yves Roberge (eds.) The end of argument structure? Syntax

& Semantics, 229-260. Bingley: Emerald.

Rivero, María Luisa (2001) On Impersonal reflexives in Romance and Slavic and semantic

variation. In Joaquim Camps & Caroline R. Wiltshire (eds.) Romance Syntax, Semantics

and L2 Acquisition, 169-195. Amsterdam: John Benjamins.

Rizzi, Luigi (1982) Issues in Italian syntax. Dordrecht: Foris.

Rizzi, Luigi (1986a) Null objects in Italian and the theory of pro. Linguistic Inquiry 17:501–

557.

Rizzi, Luigi (1986b) On the status of sublect clitics in Romance. In Jaeggli Osvaldo &

Carmen Silva-Corvalán (eds.) Studies in Romance Linguistics, 391-419. Dordrecht: Foris.

Roberts, Ian (2010) A deletion analysis of null subjects. In Theresa Biberauer, Anders

Holmberg, Ian Roberts & Michelle Sheehan (eds.) Parametric Variation: Null subjects in

minimalist theory, 58-87. Cambridge, UK: Cambridge University Press.

Rodrigues, Cilene (2004) Impoverished Morphology and A-Movement out of Case Domains.

PhD dissertation, University of Maryland.

Saab, Andrés (2002) Asignación de caso nominativo en construcciones con se impersonal.

Proceedings of the IX Congreso de la Sociedad Argentina de Lingüística (CD-ROM).

Córdoba: Universidad Nacional de Córdoba.

Saab, Andrés (2004) El dominio de la elipsis nominal en español: Identidad estricta e

inserción tardía. Master Thesis, Unversidad Nacional del Comahue.

Saab, Andrés (2009) Hacia una teoría de la identidad parcial en la elipsis. PhD dissertation,

Universidad de Buenos Aires.

52

Saab, Andrés (2012) Syntax or Nothing: Implicit arguments as absence of Merge. Ms.,

University of Buenos Aires.

Safir, Ken (1986) Subjects Clitics and the NOM-DROP Parameter. In Borer H. & Y.

Grodzinsky (eds.) Syntax and Semantics 19: The Grammar of Pronominal Clitics, 333-

356. New York: Academic Press.

Saito, Mamoru (2007) Notes on East Asian argument ellipsis. Language Research 43: 203-

227.

Simpson, Andrew, Arunima Choudhury & Mythili Menon (2012) South Asian Perspectives

on Argument Ellipsis. In Bum-Sik Park (ed.) Three factors and Syntactic Theory.

Proceedings of The 14th Seoul International Conference on Generative Grammar, 370-

389. Hankuk Publishing Co.

Şener, Serkan & Daiko Takahashi (2010) Ellipsis of arguments in Japanese and Turkish.

Nazam Linguistics 6: 79-99.

Takahashi, Daiko (2008a) Noun phrase ellipsis. In Shigeru Miyagawa & Mamoru Saito eds.

The Oxford Handbook of Japanese Linguistics, 394-422. Oxford: Oxford University Press.

Takahashi, Daiko (2008b) Quantificational null objects and argument ellipsis. Linguistic

Inquiry 39(2): 307-327.

Takahashi, Daiko (2010) Argument Ellipsis, Anti-Agreement, and Scrambling. Ms. Tohoku

University. [To appear in Mamoru Saito ed. Japanese Syntax in Comparative Perspective.

Oxford: Oxford University Press.]

Takahashi, Daiko (2013) Argument Ellipsis in Japanese and Malayalam. Nazam Linguistics 9:

173-192.

Tomioka, Satoshi (2003) The semantics of Japanese null pronouns and its cross-linguistic

implications. In K. Schwabe & S. Winkler (eds.) The interfaces: Deriving and interpreting

omitted structures, 321-40. Amsterdam: John Benjamins.