Unifying Relativization and Control in Bangla

[Published in:

Manabendu Banerjee et al. (eds.) Nyaya-Vasistha: Felicitation Volume of Professor V.N. Jha. Kolkata: Sanskrit Pustak Bhandar. 138-170. 2006.]

In this study of relative clauses and pronominal control phenomena, the Indo-Aryan language Bangla (a.k.a. Bengali) is used for exemplification, but many points carry over to other Indo-Aryan languages present and past. The work done here is expected to converge with advances in the study of Dravidian relative participles and Eastern Indo-Aryan classifier systems, but it is too early to place those considerations directly on the agenda. The present study takes up only the core issues, and even these are difficult.

Several pieces of evidence converge on the hypothesis that the relative pronoun shares crucial grammatical properties with a zero pronoun it alternates with. Consider this case of free variation in Bangla between the relative /jaa/ ‘which’ and a zero pronoun (our notation employs /tx dx rx/ for retroflex sounds, /Vx/ to nasalize the V[owel], /ea/ to represent a low front monophthong, /a/ for a low back rounded vowel, /aa/ for a low back unrounded vowel, /ng/ for the velar nasal, /s/ for a phoneme realized as a palato-alveolar except in certain clusters, and Emph to gloss an emphatic particle):

(1) uni [jaa] nije deakhen ni eamon boi gronthoponji theke baad dilen

(2) uni [ZERO] nije deakhen ni eamon boi gronthoponji theke baad dilen

‘He excluded from the bibliography books [(1) which / (2) zero] he had not seen himself’

A second, more familiar alternation associates /jegulo/ ‘which’ with zero:

(3) tumi jegulo sikhecho sei surgulo baajiyo

‘Play the tunes which you have learnt’

(4) tomaar ZERO sekhaa surgulo baajiyo

‘Play the tunes ZERO learnt by you’

Here the overt relative pronoun in the clause /tumi jegulo sikhecho/ ‘which you have learnt’ bearing finite inflection is in complementary distribution with the zero pronoun in the participial construction /tomar ZERO sekhaa/ ‘ZERO learnt by you’. Following contemporary conventions, we shall write this zero as Pro, standardly called control Pro and described as “controlled” by its antecedent, here /surgulo/ ‘the tunes’.

We propose that Pro is a pronominal anaphor whose control relation is mediated by the Comp (complementizer) node of its clause. On this basis we are able to link (3)-(4) to (1)-(2), if (2)’s zero also instantiates Pro. All the relevant constructions must be investigated to clarify how Pro and the relative pronoun are related in each case. The free variation at (1)/(2) is harder to study than the complementary distribution at (3)/(4). Bangla relatives of the type observed at (3) are taken up for consideration here, without reference to the element /eamon/ ‘such’ on which the (1)/(2) pattern depends.

Our analysis of (3)/(4) develops further the proposal (Dasgupta 1980, 1989b) that the relative state is an inflectional state of the clause. The more specific proposal that the relative pronoun is a pronominal anaphor (Dasgupta 1996a) fleshes out the idea of clausal inflection. The present sequel to those texts unifies relativization with control, thus elaborating the idea that clauses inflect to establish interclausal relations, whereas verb inflection manages clausal argument structure. Current research takes for granted the Indian tradition’s account of the inflection-driven relations between a verb and its arguments, its kaarakas. The inflection concept proves useful also in the context of expanding the research agenda to cover interclausal relations. Research done today is in a position to combine insights from our Indian inheritance with the suggestion, implicit in the Latin translation relativus of the Greek word anaphorikos, that a relative pronoun is an anaphoric pronominal.

The innovations in the present account include the proposal that Pro moves from subject to Comp. Pro needs to be in Comp to be interpreted; the contradictory demands on a pronominal anaphor can be met nowhere else. This proposal competes with such ideas as that Pro exhibits a special null Case. One advantage of the present theory lies in its associability with dual interface semantics and the biplanar syntax account of Comp-managed embedding, matters not explored here.

Our empirical project is to analyze Bangla relatives within a unification of relatives and control. We show that this account of relatives is rooted in early parametric work. We then survey facts any theory must handle, dividing Bangla relatives into two descriptive classes. This survey is followed by a formal description of these classes, Headed Relatives and Sequential Relatives. Finally, an explanatory account underwriting the description is offered, elaborating some entailments of our assumptions with reference to complementation and control.

Our analysis is an alternative to the minimalist postulate of a null Case associated with control Pro. That claim appears to be a coding manoeuvre that serves to express dependencies earlier captured by binding theoretic methods now abandoned. The manoeuvre contradicts minimalism’s inclusiveness goal, the goal of ensuring that the syntax operates only with features demonstrably present at the lexical item level. No lexical grounds warrant the “null Case vs nominative vs absence of Case” trifurcation. But the lexically distinct binding properties of reflexives, reciprocals, personal pronouns, and the zero pronoun Pro have long supported descriptive and theoretical interventions. They are the appropriate site for reopening the issue of how to treat relative pronouns in formal syntax, an issue long neglected in syntactic theory’s parametric past and minimalist present.

Current minimalist concern about relatives reflects a need to reconcile classical trees with antisymmetry and has favoured the strategy of raising a unique intraclausal pivot nominal to a left peripheral position. Such a strategy misses the point that South Asian languages feature both sequential relatives with multiple pivots and headed relatives with a unique pivot.

The present account addresses issues surrounding the feature content and distribution of relative pronouns. These issues are independent of characteristic minimalist themes such as antisymmetry, null Case, or the highlighting of proper government at the cost of a binding theory relegated entirely to extra-syntactic interpretation. In order to point up the consequences of highlighting the relative pronoun itself, the present account, tactically eschewing the minimalist austerity drive, employs the Barriers variant (Chomsky 1986) of the proto-parametric architecture. These moves reflect more than matters of expository strategy (cf Dasgupta 1997, 1998, 2000, Dasgupta, Ford & Singh 2000).

Why seek to unify relativization and control? We begin with empirical reasons pertaining to the distribution of relative pronouns and control Pro, choosing examples from Bangla. This exposition leads into the theoretical logic of the unification.

The present approach responds to the observable parallel between the asymmetric behaviour of relatives in Bangla and the general asymmetry of control clauses. The parallel is apparent in the core facts about control. Consider the null subject Pro in the gerundial clause in (5) and (6) as well as in the infinitival clause in (7).

(5) Prosaad [Pro raastxropotike citxhi lekhaa] suru korlo

‘Prasad started [writing letters to the president]’

(6) [Pro raastxropotike citxhi lekhaa] Diliper cokhe bokaami

‘[Writing to the president] looks stupid to Dilip’

(7) Prosaad [Pro raastxropotike citxhi likhte] suru korlo

‘Prasad began [to write letters to the president]’

The general pattern is that Pro is controlled in cases like (5) and (7) by an antecedent, here /prosaad/, but uncontrolled in cases like (6), where it “freely (co)refers”, i.e. where it either has an arbitrary index or is freely coreferential with a non-c-commanding category. Classical accounts of control like Manzini 1983 associate the contrast between the controlled Pro in (5) and (7) and the arbitrary or free Pro in (6) with the fact that its clause is governed in (5) and (7) but ungoverned in (6). Translating into systems without government, Pro is controlled in a clause that serves as a complement to a lexical head selecting it and uncontrolled where its clause is headless.

This asymmetry concerning control clauses resembles the headedness-linked asymmetry in the behaviour of Bangla relatives:

(8) Biren jebhaabe taakaae, keu or sangge sacchonde kathaa bolte paare naa

‘The way Biren stares, nobody can talk to him comfortably’

(9) Bijae jerakom ceaxcaae, keu oke jene-sune catxaae naa

‘The way Bijoy shouts, nobody knowingly annoys him’

(10) Rinaar jaa txaakaa, oke to anekei biye korte caaibe

‘Given the money Rina has, there will of course be many who want to marry her’

(11) tumi je-sab songgi jotxaaccho, aamraa keu aar tomaar sangge miste paarbo naa

‘With the company you are keeping, we won't be able to associate with you any more’

Examples (8)-(11) feature clauses whose relative element corefers with no other element in the sentence. The terms Free Relative and Headless Relative are used for such clauses. Elements like /je-sab/ ‘which-all’ in (11) are free: they have no formal head or antecedent. It seems natural to draw a parallel here with the arbitrary reading available for Pro in some clauses unselected by any head.

But (12)-(14) show classical relativization. Antecedents and their relative bindees carry shared indices shown in double square brackets.

(12) Rupaar [se boi][[i]] bhaalo laagbe naa [jetxaa][[i]] Niraar bhaalo laage

‘Rupa won’t like the book Nira likes’

(13) Subhraa maancitre [sei sahortxaa][[i]] khuxjche [jetxaar][[i]] tin dike tintxe nodi

‘Subhra is trying to find on the map that city which has three rivers surrounding it’

(14) loktxaa [kono baarxi][[i]] khuxje pelo naa [jaate][[i]] keu kakhono maaraa jaae ni

‘The man couldn’t find a house in which nobody had ever died’

In such classical relativizations, the relative element always has a formal antecedent. Notice that in these examples its antecedent precedes and (in Safir’s 1986 sense) R-binds it. But the construction in which the relative element is free, as in (8)-(11), or corefers freely, as in the cases we shall examine next, places the relative clause at the left periphery of the matrix structure. Consider some cases of free coreference:

(15) tumi jaake[[i]] jetxaa[[j]] dite bolle aami taake[[i]] setxaa[[j]] dilaam

you whom what to-give told I him/her that gave

‘I gave what you told me to, to the person you told me to give it to’

(16) rumaa je[[i]] somaye je[[j]] klaase jaa[[k]] parxaato eakhon se aar taa[[ijk]] kare naa

Ruma which time-Loc which class-Loc what taught now she any-more it does Neg

‘What Ruma used to teach, when she used to teach it, to which class – she does not do it any more’

(17) je[[i]] baas je[[j]] raastaa diye colto eakhon se-sab[[ij]] paaltxe geache

which bus which route along ran now that-all changed has

‘The buses that used to run, the routes they used to follow, all that has changed now’

An example like (15) might appear to be simply a classical relativization structure with two special features – unusual order, with relative preceding antecedent, and multiple (here double) pairing of relatives with antecedents. But (16) and (17) instantiate a point from Dasgupta 1980 not picked up in the later literature, a point that has made the present intervention necessary. Namely, this construction, with the relative clause on the left periphery of the matrix, gives so much latitude to the relation between relative elements and their apparent antecedents that, even in cases like (15), /taake/ ‘him’ and /setxaa/ ‘that’ cannot count as antecedents. Such a matrix element that corefers, freely, with a relative element can be called a Sequel. The Sequent type of determiner used in sequels is morphologically distinct from both the this-type Proximals and the that-type Distals that work with deixis.

To see that such coreference is free in the sense of not obeying standard constraints, reexamine (16) and (17). In (16), the sequent pronominal /taa/ ‘it’, on its own, responds to and imprecisely satisfies the referential expectations initiated by /je somaye/ ‘at which time’, /je klaase/ ‘in which class’, and /jaa/ ‘what’. In (17) the sequel phrase /se-sab/ ‘all that’ satisfies, again imprecisely, the expectations initiated by /je baas/ ‘which bus’ and /je raastaa/ ‘which route’. The imprecise satisfaction of such expectations is evidently a looser relation than the split antecedent phenomenon familiar from Mary talked to Linda about their joint work. Examples (8)-(11) also indicate that such relatives need not initiate any referential expectations for a sequel to satisfy, imprecisely or otherwise.

Summarizing, a relative clause in Bangla to the right of its head behaves classically. It contains just one relative element, coindexed with a head. In contrast, a matrix-initial relative clause may contain one or more relative elements. These exhibit either arbitrary reference or free coreference with some sequel element/s. This behavioural asymmetry between headed and matrix-initial relatives resembles the asymmetry, familiar from work on control, between antecedent-controlled Pro and arbitrary Pro. Hence the conjecture that relative pronouns and Pro carry identical features (modulo the factors that make Pro come out as zero and the relative pronouns get pronounced) whose study is likely to throw light on how relative elements are related to their real or apparent antecedents.

We turn now to reasons for the relative-Pro unification proposal that pertain to the nature of these elements.

Control Pro occurs only as the subject of a nonfinite clause of the infinitive, gerund, or participle type. Pro takes scope over its clause, and in the core cases is bound by an extraclausal antecedent. Notice that Pro shares this dual nature with that Agr bundle of phi features at the Infl head of IP with which proto-parametric work identified Pro.

This dual nature of Pro, especially clear in participial clauses with a Pro subject, has to do with the way Pro sets up a predication. The prominent functional element X that enables a predication XP to be predicated of an external argument Arg must take scope over XP and be bound by Arg. Parallel to this is the dual nature of the relative operator, explicitly noticed in proto-parametric research. The relative clause too is an XP in the core cases, predicated of an Arg, and the relative pronoun is the enabling X.

Proto-parametric conceptions established both the relative pronoun’s bindee status as an X (bound by whatever XP is predicated of) and its binder status as an operator (a relative pronoun binds a variable). On the one hand, a relative pronoun is an X – Chomsky (1982: 13, 92-5) explicitly says a relative clause is predicated of its antecedent and logically counts as an open sentence – and in this role is bound from outside its clause. Wearing another hat, however, relative pronouns are “definite or indefinite operators” (1981: 102), bind variables and “perhaps”, a note adds, “are also quantifiers” (146 n. 96).

To stress the predication-forming and antecedent-bound character of the relative pronoun involves taking as a point of departure Safir’s 1986 unpacking of Chomsky 1982 on relatives as predications. Developing earlier and unsuperseded work by Cinque 1982, Safir proposes that the antecedent “R-binds” the relative pronoun. He contributes the suggestion that, in cases of pied piping where the complex wh-phrase that A-bar-binds the intraclausal variable is distinct from the relative pronoun proper, this pronoun overtly moves or LF-moves to a position where it becomes accessible to the R-binder.

If we want relative operators mainly to bind variables, we focus on Koopman and Sportiche’s 1982 Bijection Principle, requiring variables to be singly paired with operators locally A-bar-binding them. Chomsky’s 1982 description of parasitic gaps suggested that the Bijection Principle might be too strong. But the Chomsky 1986 revision proposes that, wherever alpha seems to locally A-bar-bind both a proximate variable and a remote variable, actually some closer A-bar-binder beta protects the latter from A-bar-binding by alpha and violation of bijection. Chomsky proposes a chain composition device unifying the A-bar-chains at LF. Pursuing this insight, one might call the minor operator beta a “bound operator”.

These lines of proto-parametric thought meet. The minor operator in Chomsky 1986 is O, equatable (Chomsky 1982) with Pro in Comp and thus (Chomsky 1981) with the relative operator. To unify the relative pronoun as an R-bindee with the relative pronoun-operator as an A-bar-binder, we develop the idea that this pronoun is an anaphoric pronominal of the type familiar from control clauses and, as an X in an open sentence or predicate, both predicates this open sentence of an antecedent (serving as a bindee) and takes scope over this clause (A-bar-binding a variable within it).

Ideas about predication (Rothstein 1983) thus meet the proposal that the relative pronoun is semantically a variable (Dasgupta 1980) and syntactically a pronominal anaphor. It follows that the relative pronoun is a bound operator. This account can be easily extended from overt pronouns to the null operator O. The present paper proposes further that control Pro in a man Pro to do the job (where the clause is predicated of a man) as well as Mary is eager Pro to do the job involve Wh-Movement of Pro from subject to Comp. This move is also rooted in earlier work and involves equating control Pro in subject position with the operator O preposed to Comp from the object position of a Complement Object Deletion clause.

The present article argues, then, for an analysis in which the relative pronoun is (i) bound, (ii) an operator-binder, and (iii) the crucial X that builds a certain XP and predicates it of an antecedent. Since (ii) and (iii) are uncontroversial, what is argued for is point (i), that the relative pronoun is formally a bound element in core cases. Specifically, it is argued to be a pronominal anaphor, thus assimilating R-binding in the sense of Safir 1986 to control. We assume as a point of departure Manzini’s 1983 reduction of control to A-binding of the controlled nominal, under Borer’s 1989 restatement in terms of A-binding of the Infl coindexed with the controlled nominal. The proposed reduction of R-binding to control thus amounts to subsuming R-binding within the binding module, wherever this may be located, in the syntactic computation or the pragmatics, once the final boundaries are drawn.

We begin the empirical work by surveying Bangla relative phenomena. Headed Relatives like (18) below, where the head nominal /anekke/ ‘many people’ R-binds /jaader/ ‘whom’, contrast with such non-head-selected Sequential Relatives as (19) and (20). Within sequentials we distinguish the Free Relative subtype as in (20) from the Correlative subtype as in (19). Useful accounts focused on Hindi-Urdu include Kachru 1980, Subbarao 1984, Davison 1986, 1987, 1988, Srivastava 1990. A Free Relative has arbitrary reference. A correlative corefers freely with some sequel. Both are Sequential, for both constructions put the relative clause to the left of the rest of the matrix, whether or not it contains a coreferential sequel. Even in cases (not discussed here) where a sequential relative clause is overtly both preceded and followed by matrix material, coreferential nominals follow the relative.

(18) Headed Relative

Hiren ekhaane anekke cene jaader Torun cene naa

‘Hiren knows many people here whom Torun doesn’t know’

(19) Sequential Relative: Correlative Subtype

Belaa je sahore jaae Hiren se sahore jaae naa

‘Hiren doesn’t go to the town/s that Bela goes to’

(20) Sequential Relative: Free Subtype

Biren je sahore-i jaak, Rupen or khoxj raakhe

‘Whichever town Biren may visit, Rupen keeps track of him’

We note some differences now between the Headed Relative of (18) and the Sequential Relative, for which we choose examples from subtype (19). A headed relative clause normally occurs at the end of the matrix and never precedes its antecedent; hence the (21)/(22) contrast; and a sequential normally occupies matrix-initial position, as in (23)/(24).

(21) Hiren kichu chobi dekheche jaa Ronjit deakhe ni

‘Hiren has seen some films Ranjit hasn’t’

(22) *jaa Ronjit deakhe ni Hiren kichu chobi dekheche

lit.: which Ronjit seen hasn’t Hiren some films has-seen

(23) Belaa je chobi deakhe Hiren se chobi deakhe naa

‘Hiren doesn’t see the film/s which Bela sees’

(24) ??Hiren se chobi deakhe naa Belaa je chobi deakhe

lit.: Hiren that film sees Neg Bela which film sees

A second difference concerns the composition of the clause. A headed relative clause never contains two or more referentially distinct relative phrases. Hence the ill-formedness of (25). In (26), the phrases are coreferential and thus legitimate. In contrast, a sequential may have two or more distinct relative phrases, as in (27).

(25) *se[[i]] taake[[j]] samossaatxaa bhaalo kore bujhiyeche jaar[[i]]

s/he him/her the problem adequate -ly has-explained whose

chele-meyer beabohaar niye jaar[[j]] protibesiraa birokto hae

children’s behaviour about whose neighbours annoyed get

Intended reading as in (27)

(26) e paarxaae eakjon[[i]] aache jaar[[i]] chele-meyer beabohaar

this area-Loc someone there-is whose children’s behaviour

niye jaar[[i]] protibesiraa birokto hae

about whose neighbours annoyed get

‘There is someone in this area whose neighbours get annoyed with

his/her (“whose”) children’s behaviour’

(27) jaar[[i]] chele-meyer beabohaar niye jaar[[j]] protibesiraa birokto

whose children’s behaviour about whose neighbours annoyed

hae se[[i]] taake[[j]] samossaatxaa bhaalo kore bujhiyeche

gets s/he him/her the-problem adequate -ly has-explained

‘Given x, y, such that x’s children’s behaviour annoys y’s neighbours,

x has adequately explained the problem to y’

A third difference concerns the niche that the relative phrase occupies within the clause. It prefers the clause-initial position in the headed relative, as in (28) vs (29), but has no preference for initial position in the sequential type, cf. (23), (30). This account predicts that (29) should be ill-formed. Why does it sound relatively acceptable? Perhaps this is because it is open to construal as an afterthought-like sequential, or as an appositive with the clause boundary pause weakened or omitted.

(28) Gitaa eaktxaa pad raaxdhte caae jaa Rebaa kakhono raaxdhe ni

‘Gita wants to cook a dish that Reba has never cooked’

(29) ?Gitaa eaktxaa pad raaxdhte caae Rebaa jaa kakhono raaxdhe ni

lit.: Gita a dish to-cook wants Rebaa which ever cooked hasn’t

(30) Ronjit jekhaane jaae durghatxonaa sekhaane-i ghatxe

‘Where Ronjit goes, accident follows’

The relative preposing at (28) poses analytical issues, for neither the sequential construction of (30) nor constituent questions exhibit overt wh-movement in Bangla. We return to these issues.

A fourth point is that relative phrases in the sequential type can co-occur with /aar/ ‘else’, a property they do not share with headed relatives:

(31) biyete aar-jaaraa esechilo taader sabaaike aami cini

‘The others who had come to the wedding, I know them all’

(32) *sabaaike /taader sabaaike aami cini aar-jaaraa biyete esechilo

lit.: all / them all I know else-who wedding-Loc had-come

Our fifth observation considers the composition of the relative phrase. This phrase may be a full lexical nominal in the sequential type, see (23). In the headed type, it usually consists of just a functor like /je/ ‘who’ or /jaa/ ‘what, which’, as in (21) vs the less well-formed (33).

(33)??Hiren kichu chobi dekheche je-sab chobi Ronjit deakhe ni

‘Hiren has seen some films which films Ranjit hasn’t seen’

Our sixth and final observation brings out a property that relatives of the two types share, but which in the case of sequentials becomes surprising in a way not discussed in the literature. The common feature is that relative phrases in Bangla may trigger first person agreement, as in sequential (34) or headed (36), or second person agreement, as in sequential (35) or headed (37). This is surprising because in (34) and (35) we find first and second person agreement without a triggering first or second person pronoun anywhere; such a pronoun is in fact excluded there.

(34) jaaraa kambol aani ni taaraa sabaai kurxi txaakaa debo

who blankets brought haven’t they all twenty rupees will-pay

‘Those of us who have not brought blankets will all pay Rs 20’

(35) jaaraa kambol aano ni taaraa sabaai kurxi txaakaa debe

who blankets brought haven’t they all twenty rupees will-pay

‘Those of you who have not brought blankets will all pay Rs 20’

(36) aamraa sabaai kurxi txaakaa debo jaaraa kambol aani ni

we all twenty rupees will-pay who blankets brought haven’t

‘All of us will pay Rs 20 who haven’t brought blankets’

(37) tomraa sabaai kurxi txaakaa debe jaaraa kambol aano ni

you all twenty rupees will-pay who blankets brought haven’t

‘All of you will pay Rs 20 who haven’t brought blankets’

The points just made are summarized below, using standard terminology:

(38) Headed Relative Clause

A. RC (relative clause) is extraposed.

B. RC cannot contain two or more referentially distinct RPs (relative phrases).

C. RP must be RC-initial, in “a Comp position”.

D. RP is not a quantifier, cannot cooccur with ‘else’.

E. RP is only a functor.

F. RP may bear first or second person features, is a pronominal.

(39) Sequential Relative Clause

A. RC is topicalized or left-dislocated.

B. RC may contain two or more referentially distinct RPs.

C. RP need not be RC-initial, can stay in situ.

D. RP may (a quantifier property) cooccur with ‘else’.

E. RP may include contentives.

F. RP may bear first or second person features, is a pronominal.

Informally, these points add up to an obvious story, whose elements meet at point F: the pronominal character of all relative elements. This leads us to ask how the relative system, despite this featural unity, ends up exhibiting two contrasting clusters of properties. In investigating this, we move towards descriptive adequacy.

Recall that standard proto-parametric accounts treat RCs as antecedent-linked open sentences at LF. An RC is predicated of, and thus restricts the referential range of, the antecedent nominal at LF (Chomsky 1982: 92-5). Chomsky and others do not commit themselves to specific proposals about the composition of the RP. However, in empty RP cases, the bond connecting it to the RC’s antecedent is seen as an anaphoric bond in the empty operator analysis of a book O Pro to read t due to Chomsky (1981; 1982). The element O, also characterized as a Pro in Comp, A-bar-binds the variable and is anaphoric to the antecedent a book. Chomsky presents another analysis along the same lines – with an empty RP classified as a pronominal anaphor (1981: 167).

Cinque’s 1982 work also suggests an anaphoric pronominal analysis. He did formally postulate a pure anaphor, but only because his text, despite its publication date, was written in the “free deletion in Comp” framework. Moving into the account postulating Pro in Comp for a book to read, one needs to make adjustments for overt relatives. Such revision leads to the anaphoric pronominal analysis of the relative pronoun. One publication that directly proposed that overt relative pronouns had the syntactic features of control Pro was Hendrick 1982. But the details of his implementation, wedded to formal government, are dated in ways that preempt discussion in current terms; to work in continuity with that article is not an option. To summarize, the early parametric literature is generally supportive of the pronominal anaphor analysis of the RP at the thematic level. Let us now consider relevant data and develop the operative notions rigorously, beginning with a pied piping example to make it possible to take up issues that became important in Safir’s work. Consider (40), which contains the RC (41):

(40) Dipen eaktxaa prostaab koreche jaa niye Raajes haaxsche

‘Dipen has a made a proposal about which Rajesh is laughing’

(41) jaa niye Raajes haaxsche

‘about which Rajesh is laughing’

The adposition /niye/ ‘about’, in the proto-parametric architecture, governs and Case-marks the nominal /jaa/ ‘which’. In the formulation presented here, worries about government disappear, in keeping with the independent general dismantling of that apparatus since the early nineties. In our account, the Pro theorem is an old encoding of contradictory demands made by the Binding Theory on pronominal anaphors. The Binding Theory can be redone as follows for our purposes. Principle B says that a pronominal, capable of personal deixis, must maintain minimal referential independence from eligible binders near by. Thus, if a nominal c-commands a pronominal within a local domain, then the two must be contraindexed. Slight variants of this story need to be, but in this article will not be, constructed in order to reflect parametric variation.

Our retelling of Principle A directly counterpoints this. An anaphoric feature matrix is deixis-incapable and includes no intrinsic phi features, though anaphor subtypes do differ with respect to other intrinsic features. So an anaphoric feature matrix seeks a binder endowed with either phi features or some other referential base. Anaphors vary as to their degree of lexicality vs functionality. The more functional a given anaphor is, the more it is a creature of the grammar and must look up the tree’s Infl line to seek a suitable Agr site whose phi features it can pick up referential support from. Call such an antecedent an AT-binder, for “Accessible Topicoid”, where a Topicoid is either a “big Subject” in the proto-parametric sense or an Agr bundle of features located at C.

Combining these reformulations for a matrix with an anaphoric and a pronominal submatrix, we obtain the following composite picture that lies at the heart of this paper’s central empirical move – the identification of RP and control Pro. A pronominal anaphor, as a pronominal, is deixis-capable and ineligible for referential support from the tree’s Infl line. But, as a nonlexical anaphor, it must seek a source of referential features, and cleave to some Agr line in doing so. Since its pronominal character makes it recoil from the Agr sites on the line of Infl heads of IPs, the pronominal anaphor looks up not the Infl but the Comp line, seeking an Agr site at the nearest C-Agr available, to seal a referential pact with it, ready to move up for this purpose. We are using phrasing such as “referential support” in order to allow for the necessary latitude, for C-Agr in some cases offers a type of support that stops short of referential coindexing. This solution paves the way for a future rapprochement with pragmatics-based views of pronominals and anaphors (discourse associates topicoids with pragmatic correlates of biplanarity). So reconstructed, the Binding Theory looks like this:

(42) At S-structure,

A. An anaphor must be AT-bound (or, parametrically, if nonpronominal /lexical then A-bound) in its Local Domain (LD)

B. A pronominal must be A-free in its LD

(43)i. LD(E) for element E is the smallest maximal projection that contains E, a big Subject, and, if E is anaphoric, also an AT(E)

ii. A big Subject of a maximal projection of nominal character is an Agr (a set of phi features) in the head or the specifier

iii. An Accessible Topicoid accessible to E, AT(E), is a category c-commanding E

(p) that is the big Subject of (if E is nonpronominal) the lowest big-Subject-containing DP or IP dominating E, or (if E is pronominal) of the lowest CP dominating E

(q) and such that assigning to E the index of AT(E) would not trigger a violation of the i-within-i filter, which says a nominal can't be coindexed with any nominal it immediately contains

The approach adopted here ensures that a pronominal anaphor’s LD(anaphor) is broader than its LD(pronominal). Such an element seeks a controller located outside LD(pronominal) but still within LD(anaphor). Our empirical task now is to find it such a controller over the range of known cases.

We cannot unpack the ramifications of this account all at once. For the moment, we will bring this to bear on Headed RCs, focusing on traits that distinguish this construction from the Sequential RC. After we have done the relative examples as comprehensively as the state of our understanding permits, we will turn to complement clauses and control Pro. Throughout, we take for granted the standard understanding of how various pure (nonpronominal) anaphors in A-positions are bound, and consider only the new cases at stake here.

Our approach extends the by now widespread assumption that the behaviour of the simple (nonlexical) reflexive pertains to the IP system’s inflectional grammar. By analogy, we seek to construe the core cases of RPs in terms of what we see as the CP system’s inflectional grammar, thus unpacking the proposal that the relative state is an inflectional state of the clause.

The inflectional grammar of IP revolves around Cases, the verbal inflection responsible for them, and the argument structure they respond to. A reflexive clause links the subject and another argument slot to a distinctive verbal inflection that registers this linkage as coreference. The grammar of IP gives prominence to the subject, which is IP’s Topicoid site.

In contrast, the inflectional grammar of CP pertains to the staging of the event that the clause denotes, deciding how to embed the IP either in a higher clause or in the illocution. The Topic is the Topicoid site made prominent by CP’s inflectional system. Not every inflectional state of a CP harbours a Topic, though.

There are some important differences between the two inflectional grammars. The C head does not manage the grammar of CP primarily in Case terms. Nor does C provide referential anchorage to constituents of the CP. However, CP shares with IP the work of transmitting Case and referential dependencies up the tree. This sharing takes a particular form, examined in this paper, at the values for Anaphoricity and Pronominality in the C head’s Agr bundle. We argue here that this C-Agr feature bundle in the case of the Headed RC that we are right now looking at shares the nominal referential index of the antecedent nominal (of which the Headed RC is predicated) and, through binding, transmits it to the RP /jaa/ ‘which’. But, in the case of the Sequential RC, it becomes clear that C-Agr, when it receives no nominal referential index from an antecedent nominal, carries an arbitrary index that lacks referential identification powers. When C-Agr imposes that index on its bindees within the relative IP, the result is scope marking, not coreference.

What is crucial for present purposes is that the proposals just made allow the RP /jaa/ ‘which’ to be AT-bound by the nominal feature-bundle Agr in C, provided that the CP shown in (44), a syntactic representation for (41) reflecting a few concrete assumptions not motivated here, is the LD(anaphor) for /jaa/ (we use a bracket labelling convention that involves writing “IP[ ]” for a bracketed stretch belonging to the IP category):

(44) CP[ C[Agr[[I]]] IP[ PP[ DP[jaa][[i]] P[niye][[j]] IP[Raajes t[[j]] haaxsche]]]

which about Rajesh t is-laughing

Why does the relative phrase in a Headed RC prepose (a fact noted at (28) and (38C)), given that Bangla wh-phrases in questions and even in certain relatives (as in (30)) remain in situ instead of moving? The hypothesis we are now considering suggests that we seek an answer in terms of adjacency considerations. If /jaa niye/ were to stay in situ, it would be where the PP trace is in (44). Then Agr in C would have no access to the DP /jaa/ that it binds. But /jaa/ needs, in a Headed RC like (44), to pick up reference from C-Agr. Hence the necessity of adjoining the PP /jaa niye/ to IP; we take this to be scrambling, without commitment to any particular formal account. Once this adjunction takes place, yielding (44), Agr in C can AT-bind /jaa/.

The account developed so far makes sense of the occurrence of RP preposing in headed RCs. We will see later just why sequential RPs are exempted.

But there is more to say about S-structure (44), and to say it we will employ proto-parametric LF-movement (allowing heads to move instead of translating into the remnant movement format of analysis and raising orthogonal issues). For several reasons, some of them discussed by Safir 1986, the relative pronoun proper must, in pied piping cases, LF-move out of the relative phrase containing it. Assume, for concreteness, that it moves to C (choosing Spec of CP would leave open the issue of why RP in a Headed RC is consistently a head, in fact a function word, and not a fully expanded nominal phrase of the type available in a Sequential RC). The resulting LF is:

(45) CP[C[D[jaa][[i]] C[Agr][[i]] IP[PP[t[[i]] P[niye]][[j]] IP[Raajes t[[j]] haaxsche]]]

which about Rajesh is-laughing

This LF-movement serves two purposes. From the RP’s own viewpoint, it moves to take scope over IP. In the larger structural context, this movement occurs to allow the CP as a whole, through its specifier R-bound by an antecedent, to be predicated of that antecedent, as one would expect of an RC in the core case.

The extraction of RP at LF yielding the LF-representation (45) permits a straightforward derivation of property (39B) of headed RCs from our analysis. A headed RC must either have only one RP or ensure that any RPs that remain in situ (and escape LF-extraction) are connected, in the technical sense, to the network of paths initiated by the LF-moved RP, its trace in the relative phrase, and the trace of the relative phrase. Such a connected non-prominent RP must be coreferential to the prominent one, for it must share the (unique) index of the Spec of CP and participate in the predication if there is to be even vicarious exercise of its scope-taking rights. There are indeed examples of this sort:

(46) aami eakjonke cini jaar galaabaajir cotxe jaar porxsiraa atistxho hoye utxheche

‘I know someone whose neighbours have gone crazy because of his (“whose”) shouting’

We are deliberately setting aside property (38A), obligatory extraposition, since it varies parametrically: Hindi-Urdu does not share it. Property (38C) of the headed RC, the fact that it displays RP preposing, is already a central feature of the present analysis. Turning to (38D), then, we need to explain why the RP in the headed RC is not a quantifier (cannot coccur with /aar/ ‘else’) whereas in the sequential RC it does exhibit quantifierlike behaviour, point (39D).

Our explanation is based on the claim that the relative operator in the core case (under consideration here) is a minor operator and thus a bound element, which cannot quantify. We return to the fact the RP in a Sequential RC can nevertheless quantify.

We turn next to property (38E): the RP may have lexical content in the sequential type, but must be just a functor in the headed RC. This fact follows from our hypothesis that the RP in the core cases is an anaphoric pronominal: a nominal that includes a contentive noun would be an R-expression. What we have to explain, and will address below, is why sequential RCs untypically permit a relative phrase to have lexical content.

Our last point, (38F), was an observation that follows from the pronominality of the RP, and can be taken as an empirical argument for the otherwise theory-dependent claim that the RP is pronominal.

All the major observations are consistent with or follow from the analysis.

The hypothesis that an RP like /jaa/ ‘who’ is an anaphoric pronominal needs to be formulated more rigorously before the principles underpinning it can be extended to Sequential RC cases where the relative determiner /je/ ‘which’ is part of an R-expression such as /je jaatri/ ‘which passenger’:

(47) je jaatri je jaanlaar dhaare boste pelo se sekhaane thaaklo

which passenger which window near to-sit got s/he there stayed-on

‘For every x a passenger and every y a window such that x got a seat next to y, x stayed on next to y’

The sequential RC construction often contains such nominals; we begin by asking how the relative element /je/ interacts with the N. We shall tentatively call it a Det without implying that this is the head D of DP; work on the internal structure of the Eastern Indo-Aryan nominal is in too unsettled a state to be invoked directly in this text; it is already clear that classifiers play a crucial role, not yet rigorously understood.

For reasons given in Dasgupta 1992, we propose that the non-anaphoric demonstratives /e, o, se/ ‘this (proximal), that (distal), that (sequent)’ and the anaphoric /je/ are Determiners with a variable pronominality feature. We assume that each occurrence enters the tree with a plus or a minus for this feature attached and that the result meets criteria imposed by the rest of the tree, or fails to and leads to crash. For well-formed trees, it is as if the Det copies the minus pronominality feature from the N.

On this analysis a sequential RC differs very little, at S-structure, from a non-relative clause like ‘this passenger got a seat next to that window’. None of the nominals in the IP are A-bound or AT-bound. The internal grammar of this particular type of sequential RC is not what makes it relative. What distinguishes it from a normal clause is the external grammar, and in fact some specific aspects of the external grammar, having to do with the anaphoric character of the relative Det element.

We have sugggested that the relative Det /je/ retains its anaphoricity even in a nominal like /je jaatri/ 'which passenger' in (47) where the noun, being referential, makes the nominal an R-expression, anaphorically inert. How, then, can the anaphoric Det of a referential nominal be anaphoric to any antecedent? Our answer to this rests on the work of Pesetsky 1987, who notes that a wh-nominal containing a lexical noun strongly tends to be D-linked or discoursally anchored in the universe of reference. We indicate this status by having such a nominal coindexed with a C-Agr; this is not a relationship of coreference, but more like the one between French negative quantifiers like rien and personne and their scope marker ne. In contrast, Pesetsky observes, bare wh-words strongly tend to be quantificational, a point our analysis will also have to take on board.

Here is the resulting description of the relative CP of (47):

(48) CP[C-Agr[[i]] IP[[je jaatri][[i]] [je jaanlaar][[i]] dhaare boste pelo]]

which passenger which window near to-sit got

The coindexing shown in (48) is not referential, but an indication of the fact that we permit this C-located Agr element C-Agr to share scope indices freely with bindees, a feature of the unselective binding that seems to be as old as Baker 1970 on multiple interrogatives.

Once C-Agr has established scope marking, the relative phrases set up their referential points as R-expressions, for in each relative nominal what is anaphoric is only the relative Det /je/, not the nominal as a whole. The setting up of these referential points makes it possible and pragmatically expected, though not syntactically necessary, for one or more sequel phrases in the sequel clause to have more to say about those entities, either, as in (47), point by point, or split-coreferentially, as in cases mentioned earlier like (16) and (17) that force us to give up the otherwise plausible restriction to point by point cases. Even a total absence of sequel phrases is attested, as in (11) and (20), where the sequel clause merely presupposes, without directly drawing on it, the background set up by the RC. Such cases exemplify what we call the free subtype (as opposed to the correlative subtype) of the Sequential RC construction.

There are even cases where the sequel clause completely disappears:

(49) je biyer je montro

which wedding-Gen which liturgy

‘One has to do what the occasion demands’

(50) je jebhaabe bojhe

who which-way understands

‘People see things differently’

These forms are complete sentences. (49) is a proverb corresponding to ‘When in Rome, do as the Romans do’, and (50) is a near-proverbial full statement. We shall suppose that a Sequential RC normally invites one, as a matter of pragmatics, but that this invitation may be declined, as in (49) and (50). This hypothesis, which covers the facts, is most conveniently formalized as follows. We propose that, right above the CP structure of the sequential relative clause, there is a DP “shield” with an empty D head. It is the movement of C-Agr to D that has the phenomenal effect of making the sequential RC an independent entity inviting (but not compelling) follow-up. When there is a sequel clause, the relation between this DP-shielded CP and the following clause is a loose one. We take it to be Left Dislocation. Later in the discussion we contrast this analysis with the purely CP character of interrogative clauses. Throughout the discussion, we assume this additional DP shielding the sequential relative CP, but explicitly show it only where relevant.

To complete the analysis of sequential RCs, we examine bare wh-words; as noted by Pesetsky in the interrogative case, they behave like quantifiers. The main empirical contrasts appear in the data sets (51)-(54) and (55)-(56):

(51) je jaa caae se taa paae naa

who what wants s/he it gets Neg

‘For x, y such that x wants y, x does not get y’

(52) *jaa je caae taa se (/se taa) paae naa

what who wants it s/he (/s/he it) gets Neg

intended reading: same as (51)

(53) je silpi je chobi exkechen tini setxaa niye bolben

which artist which painting has-painted s/he it about will-speak

‘For x an artist, y a painting, such that x has painted y, x will speak about y’

(54) je chobi je silpi exkechen setxaa niye tini (/tini setxaa niye) bolben

which painting which artist has-painted it about s/he (/s/he it about) will-speak

same reading as (53)

(55) tumi aar-jaake dxekecho se aasbe naa

you else-whom have-called s/he will-come Neg

‘The other person you have called will not come’

(56) *tumi aar-je silpike dxekecho tini aasben naa

you else-which artist have-called s/he will-come Neg

‘The other artist you have called will not come’ (intended reading)

Consider (51)-(54) first. The contrast between (51) and (52) shows that, if the relative hierarchical positions of /je/ ‘who’ and /jaa/ ‘what’ are reversed, as in (52), they violate PCC (Pesetsky’s 1982 Path Containment Condition); this indicates that such elements undergo LF movement; the ill-formedness of (52) instantiates Crossing, a familiar type of PCC violation. The acceptability of both (53) and (54) shows that /je silpi/ ‘which artist’ and /je chobi/ ‘which painting’ do not LF-move. As a first guess, we may propose that the bare relative words are quantifiers while the full relative phrases are discourse-linked (“D-linked”).

We reach the same tentative conclusion when we note, at (55), that /jaake/ ‘whom’ cooccurs with /aar/ ‘else’, and at (56), that /je silpike/ ‘which artist’ does not so cooccur.

These are roughly the main facts to be handled. Less roughly, we can zero in on a more careful second approximation, making it easier to provide a descriptively adequate account. Let us distinguish two questions. One question is whether quantifiers are LF-moved and D-linked phrases are not; to this we can afford to give an affirmative answer. The second question is whether it is consistently bare relative words that are quantifiers and full relative phrases that are D-linked; here one finds only a strong tendency. Consider the following data:

(57) je jaake saahaajjo kare se taake cotxiye deae

who whom help does s/he him/her irritiate Aux

‘For persons x and y such that x helps y, x irritates y’

(58)??jaake je saahaajjo kare taake se cotxiye deae

whom who help does him/her s/he irritate Aux

(59) je jaake saahaajjo korechilo se taake cotxiye dilo

who whom help had-done s/he him/her irritiate Aux

‘For persons x and y such that x had helped y, x irritated y’

(60) jaake je saahaajjo korechilo taake se cotxiye dilo

whom who help had-done him/her s/he irritate Aux

same reading as (59)

The present tense in (57) and (58) induces a generic reading with a multiple helper-helpee pairing. On that clearly quantificational reading, (58) is excluded. That it is marginal and not completely excluded is due to the tenuously available particular person reading, with /je/ ‘who’ and /jaake/ ‘whom’ D-linked. (59) and (60) use narrative past tenses and encourage a particular person reading, with D-linking of /je/ and /jaake/. This is why turning (59) into (60) does not affect grammaticality.

There are also examples that go the other way – full nominals for which a quantificational reading is available, like /je chele/ ‘which boy’ as opposed to the classifier-augmented form /je cheletxaa/ ‘which particular boy’, with corresponding sequels /se chele/ ‘that (potentially generic) boy’ and /sei chele(txaa)/ ‘that (definite) boy’. There is no space in a paper like this to look closely at the role of classifiers and the augmented form /sei/ in Bangla nominal reference.

To summarize, the correlation between noun absence vs presence in a relative phrase and quantification vs D-linking is only a strong tendency. Our analysis says, following Pesetsky 1987, that if a given phrase is D-linked, it stays in situ (it is intolerant of ‘else’ and inert vis-a-vis the PCC diagnostic), while a phrase that counts as quantified undergoes LF-movement and behaves accordingly (it co-occurs with ‘else’ and responds to the PCC diagnostic). These are the facts to be handled.

We will assume that the relative or J-word is anaphoric in both types of relative phrase. If this anaphor is bound by an element that gives it reference, then we derive the incorrect prediction that two relative words in a clause end up coreferential with each other. We propose, then, that an anaphoric J-word is bound by an Agr in C whose referential properties (thanks to its association with D) say “Arbitrary”, and whose index shared with its bindee is thus only a scope marker offering what we have been loosely calling referential support. If this Agr in C binds i.e. scope-marks two or more J-words in a clause, the prediction is that there will be no scope interaction effects. This is what we find for what we take to be the unselective binding by C-Agr of the anaphoric J-words in (53) and (54). In contrast, the quantificational forms /je/ ‘who’ and /jaa/ ‘what’ in (57)-(60) show scope interaction, from which we conclude that they LF-move.

There are wrinkles here that one needs to get right. For concreteness, we are offering a set of options that probably need revision. If only single word relative constituents were LF-moving entities in this paradigm, one would move them to C. But there is some evidence that generic /je chele/ ‘which (but less particularizing than English “which”) boy, roughly: a boy who’ is quantificational in the relevant sense and must LF-move. We presume that this calls for LF-movement to Spec of CP.

To formalize this solution, we claim that D-linked J-words as in (53) and (54) are non-pronominal anaphoric determiners that play a referential role in association with lexical nominal material, and thus perform some sort of deixis that takes the scope delineated by the referentially arbitrary C-Agr scopally coindexed with them.

In contrast, we propose that quantificational J-phrases as in (57)-(60) are pronominal anaphors with variable specificity in the manner of existential quantifiers. When they choose the less usual specific reading, they produce the tenuously D-linked effect.

To summarize the package, we are tentatively postulating a three-way distinction between D-linking, implemented as C-based unselective binding of an unmoved relative phrase; single word quantification, implemented as LF-movement to C; and generic phrase quantification, implemented as LF-movement to Spec of CP.

Why should a pronominal anaphor be quantificational in a Sequential RC? Tentatively, we shall work with the answer that, as a pronominal, it has a stake in an intraclausal argument position, and as an anaphor that must seek anchorage at the clause periphery, it takes scope, the two core properties of a quantified nominal. It was argued earlier that an RP in Headed RCs cannot quantify because it is a minor or bound operator there, in keeping with its role of a unique relative constituent arranging for its clause to be predicated of a unique antecedent. This implies that in a Sequential RC, where it is not bound by any element with independent referential content, there is nothing to stop it from quantifying.

How are the sequel clause’s sequents related to the relative phrases referentially? Our conjecture is that the sequent phrases merely follow up, loosely in the sense of working in the discourse and not the syntax, whence the non-unique matching in examples (16) and (17) above between multiple relatives and a single sequent. The syntax, as far as we can judge, treats the Sequential RC as either a Left Dislocation loosely preceding its sequel clause or a coordinate conjunct conjoined to the sequel clause, leaving all coreference-like relations to the discourse. Our evidence for this comes from the following minimal pair, where the paired Emphasizers marking coordination are acceptable in the Sequential RC example (61) but not in the Headed RC example (62):

(61) tumi-o jetxaa kheyecho aami-o setxaa kheyechi

you-also which have-eaten I-also that-one have-eaten

‘The one you have eaten I too have eaten’

(62) *aami-o setxaa kheyechi jetxaa tumi-o kheyecho

I-also that-one have-eaten which you-also have-eaten

This contrast makes it clear that the coupled Emphasizers /o... o/ ‘also’ can function across a sequential RC boundary but not across a headed RC boundary. Nobody has a full theory of these coupled Emphasizers yet, but they do indicate here that the Sequential RC is not strictly embedded within the sequel clause, whereas the Headed RC is.

Does this analysis cover the facts of (39)?

Point A in (39) says a Sequential RC is initial in the matrix sentence. This fundamental property of the construction has now been built into the analysis.

Point B, the availability of the referentially distinct multiple RP option in a Sequential RC, follows in our account from the factor that, unlike the Headed RC where the necessity of predicating the RC of a unique antecedent prevents multiplicity, in the Sequential RC there is nothing to prevent it. The details of the availability of index multiplicity at C may be formalized in a semantically rigorous account in terms of absorption or some cognate device.

Point C, the lack of RP preposing in the Sequential RC, follows from the availability of alternative guarantees of well-formedness. The C-Agr AT-binds or scope-marks each relative constituent. Thus the RP need not move to the edge of the IP to give C access.

Point D, the compatibility of a Sequential RP with /aar/ ‘else’, holds only of quantificational cases, where this is expected.

Point E, the usability of contentives in a Sequential RP, has also been discussed; it is on the whole restricted to D-linked cases.

Finally, point F, the pronominality of RPs (as seen in person agreement facts), is restricted to pronominal examples (and thus associated with quantification). This point was built into the analysis by stipulating that single word relative phrases in a Sequential RC are pronominal.

We have developed above an account of RCs that describes the differences between the Headed RC and the Sequential RC while maintaining throughout that a relative functor is consistently anaphoric and variably pronominal.

We now present systematically the principles underwriting this description and extend the description to other constructions.

The analysis so far rests on the core assumption that the relative functor, whose overt form in Bangla is typically /je/, is intrinsically specified as anaphoric, is normally pronominal (this is overridden if associated nominal material imposes non-generic referentiality and therefore non-pronominality), and is coindexed with C-Agr. In the core case, that of the Headed Relative Clause, this C-Agr is in turn coindexed with an antecedent nominal in the matrix clause. C-Agr thus passes on to the RP in the RC the referential index of the antecedent.

The Sequential RC exhibits different properties. Here, C-Agr is c-commanded by and coindexed with no external antecedent nominal. C-Agr is itself a pronominal anaphor. Unbound, it moves to the shield DP’s head D, i.e. it defaults into an arbitrary referential index. When it imposes this index on anaphors it c-commands, it scope-marks them, making it necessary for such bindees to count as scope-takers in order to avert uninterpretability and crash.

Recall that the project of this paper is to unify relativization with control, and that to this end we have chosen to move every classical Pro, whether controlled by an antecedent or bearing arbitrary reference, to the nearest superordinate C. Where does this treatment leave control? Do we have an account now of the familiar behaviour of Pro in nonfinites?

In order to have space to move Pro to C, we need to postulate a C node (bearing whatever the right feature bundles are, possibly not meriting the designation “Agr” every time, a matter left open) in some places where standard accounts don’t give us one. Gerunds and Participles are still usually taken to be bare IPs, though it has been known since at least Barriers that a Gerund needs a Comp to handle a parasitic gap, an insight not incorporated into standard operative descriptions. Our theory forces us to make every Gerund a DP. Its head D, like the shield DP’s head D in a sequential RC, takes a complement CP to whose head the Pro subject of a Gerund must move. Likewise, assuming Deg to be the functional projection that normally serves Adjectives in Bangla (Dasgupta 1989a), we take Participle structures to consist of a DegP whose Deg head takes a CP-shielded IP complement whose Infl head is specified [+N, +V]. The shield CP’s head C is where Pro lands in such structures.

Adverbial Participles are a heterogeneous lot. At the dependent end of the spectrum, we treat the Conjunctive Participle and the Infinitival as positional variants (adjunct and complement respectively) of a CP whose non-finite (specifically, [-N, -V]) C head is invariantly anaphoric and non-pronominal. The anaphoric character of this C makes it dependent on whatever goes on upstairs. This explains why, for instance, there is a minimal contrast between the unambiguous Infinitive in (63) and the ambiguous Gerund in (64):

(63) Projitaa[[i]] Pro[[i]] bitxkel posaak porte pachondo karen naa

‘Projita doesn’t like to wear bizarre clothes’

(64) Projitaa[[i]] Pro[[ij]] bitxkel posaak paraa pachondo karen naa

‘Projita[[i]] doesn’t like herself[[i]] / others[[j]] wearing bizarre clothes’

Then there are the independent adverbial participles, like the Conditional Participle in (65), which gives its subject referential independence (even if it is a Pro subject) and Nominative Case (this becomes clear with an overt subject like /aami/ ‘I’):

(65) Pro[[i]]/ aami bitxkel posaak porle Projitaa[[j]] taarxiye deben

‘If one wears / I wear bizarre clothes, Projita will drive one/ me away’

Does it bother us that overt Case-marked nominals can alternate with Pro in Gerunds and Conditional Participles? Does this compel us to imagine Pro as Case-bearing part of the time or all the time? We don’t see this as a Bangla-particular worry, or as an urgent issue; it has long been clear that the relative Pro in English infinitivals looks as Caseless as the relative Pro in English finites looks Case-endowed, and we have left the problem cheerfully unsolved in Anglophone linguistics. When people get around to the problem, Bangla’s turn will also come.

What we do need to explain is why English allows a matrix-initial infinitival construction as in (66) which, unlike its gerund counterpart in (67), is unmatched in Bangla:

(66) English: Pro to err is human

Bangla: *Pro bhul korte maanuser sabhaab

lit.: Pro errors to-make people’s nature

(67) English: Pro making mistakes is human nature

Bangla: Pro bhul karaa maanuser sabhaab

lit.: Pro errors making people’s nature

This difference seems to reflect the parameter that makes infinitival questions possible in English and available in the same matrix-initial position, as in (68), and completely stars this construction in Bangla in any position:

(68) English: Which boss to work for is not clear

Bangla: *Kon moniber jonne kaaj korte (etxaa) spastxo nae

lit.: which boss for work to-do (it) clear isn’t

Whatever the parameter value is, it concretely makes English allow a wh-phrase in an infinitival CP to move to Spec of CP, which Bangla doesn’t. Once wh is preposed to that Spec, the CP counts as an honorary nominal, and can even complement a preposition:

(69) We talked about which boss to work for

Our suggestion is that Bangla, whose K-word interrogatives and J-word relatives sharply contrast in morphological form, has a correspondingly rigid syntactic distinction between interrogative CPs and sequential relative DPs as postulated above. In contrast, English uses the same Wh-word set for both purposes. Perhaps there is some leakage allowed in English across the relative /interrogative boundary. Such conversion of interrogative content into relative format, if that is what the parameter setting lets English but not Bangla do, gives us an interrogative DP, to whose Specifier a wh-phrase can move.

It then also follows that an English infinitival within which Pro moves to arbitrary C (which, you recall, means working in association with the D head of a shield DP) can occupy matrix topic position, while an infinitival relative whose Pro moves to Spec of CP can even be a matrix subject. Bangla allows neither possibility, since for one thing it lacks wh-preposing to Spec of CP, and for another, it cannot assign plus pronominal to an infinitival or conjunctive C, as seen above.

That the Bangla conjunctive participle and infinitival constructions have a non-pronominal anaphoric C is a stipulation. But one is still depending on stipulations to understand the fact that in the much better studied language English Gerunds can and Adverbial Participles cannot have an arbitrary Pro subject. The depth of our understanding of nonfinites leaves a lot to be desired.

What has been said so far glosses over the differences between relative and complement clauses. It also leaves it unclear how the present analysis accounts for the parallel, cited at the outset (see the discussion surrounding (5)-(14)), between the distribution of control Pro in “governed and ungoverned” control clauses (all of them complement clauses, given the traditional understanding of the notions of complement and relative clause) and that of relative pronouns in, likewise, headed and unheaded relative clauses.

That we have been able to get away with this is a comment, though, on what treatment the phenomena do and don’t compel. This analysis, framed without reference to complement clauses, can be readily extended to cover Bangla complement clauses, both the regular kind as in (70) and the topicalized complement clause as in (71):

(70) aamraa jaani [je tumi aasbe naa]

we know that you come-Fut Neg

‘We know that you won’t come’

(71) [tumi je aasbe naa], aamraa se kathaa jaani

you that come-Fut Neg, we that fact know

‘That you won’t come, we know (this)’

To perform this extension, consider subtrees (72)a, b for the complement clause bracketed in (70) and (71):

(72)a. CP[C[Prt/Agr] IP]

b. DP[D CP[XP C’[C[Prt/Agr] IP]]]

Subtree schema (72) allows us to choose either to leave C a phonologically null bearer of Agr as in the cases considered earlier, or to insert a Particle Prt (here /je/ ‘that’) under C as in (70) and (71).

Descriptively speaking, if C contains Prt, there are two choices. Choice (70) is to move Agr from C to the superordinate head Infl. The evidence for such movement includes the fact that this Infl must be finite. Such a complement clause is in this sense controlled by the matrix. Choice (71) is to extract XP, here /tumi/ ‘you’, from the IP and move it to Spec of CP. This constituent now controls Agr in C. Consequently this (topicalized) type of complement clause is only connected to the sequel “matrix” by the loose device of having an argument, /se kathaa/ ‘that fact’ in (71), provide referential follow-up to the topicalized complement clause. This is syntactically optional; the two clauses are loosely connected, along the lines of the corresponding case of the Sequential RC. As in that case, the sequel can be omitted entirely, as in (73):

(73) tumi je aasbe naa!

you Prt come-Fut Neg

‘[It’s just] that you won’t come!’

In other words, Agr in C seeks no CP-external matrix controller in (71) and (73), exemplifying the independent CP construction where IP-internal XP moves to Spec of CP. In (70), Agr in C is dependent on external control from an antecedent, the higher Infl. (Recall that the indexing of C per se has no referential significance.)

This basic description looks similar not only to our two types of Relative Clause handled above, but to the classical bifurcation in control clauses familiar from Manzini 1983 and Borer 1989. In their work, a dependent control CP has anaphoric Agr in C bound by a CP-external antecedent. In contrast, an apparently self-insulating (but on this analysis D-shielded) independent CP allows its Pro to have coreferential freedom, choosing either arbitrary reference or coreference with a constituent in a position from which binding is not expected. Such arbitrary (co-)reference is mediated by the D site.

Whether Agr in C seeks a CP-external antecedent does not depend on whether it heads a complement CP. Control into an adjunct CP is fine, in the case of classical control:

(74) They went outside [Pro to smoke]

Likewise, an adjunct CP in Bangla with the structure (72a) permits CP-external control of C-Agr:

(75) eato chaatro eato golmaal koreche je sabaai cotxeche

‘So many students have made so much trouble that everybody is annoyed’

It is not the complement or adjunct status of the CP, then, that determines whether C-Agr is going to have a CP-external matrix antecedent. Agr in C is anaphoric and normally non-pronominal. It therefore naturally seeks a binder, with consequences that need to be explored case by case. The task of the present paper is to make some of these cases concrete and more generally establish the continuability of the enterprise, not to attempt a comprehensive survey. The parametric specifications of English, including wh-movement and the general availability of a DP shield with a pronominal anaphor headed CP, can be understood as enabling matrix-initial control clauses to arbitrarily refer or corefer with the freedom described by Manzini.

This is our account of the parallel between the left-right asymmetry in Bangla relatives and the left-right asymmetric behaviour of control clauses in English.

Other constructions parallel to (72b) have been studied in Dasgupta 1980. We now take it that the topicalizing particle /to/ is a C Particle specified as pronominal and non-anaphoric. For the interrogative Particle /ki/ in C that occurs in alternative questions like (76) and (77) –

(76) tumi ki aasbe?

‘Will you come?’

(77) tumi ki aasbe naa thaakbe?

‘Will you come or stay?’

– one needs to construct a proposal that handles the distinction between the interrogative ‘or’ word /naa/ of (77) and the declarative ‘or’ word /baa/. We presume that an account with that property would also explain why option (72a) is not available for /ki/ – why one cannot say (78) but must resort to (79):

(78) *Raam jiges korlo ki tumi aasbe

‘Ram asked whether you would come’

(79) Raam jiges korlo tumi aasbe ki (aasbe) naa

‘Ram asked whether you would come or not (come)’

We can do better for constituent questions, where the account postulates a simple CP with Agr in the head C and no DP shield giving the Agr an arbitrary referential closure. Quantificational cases LF-move to C or its Spec, while discourse linked cases stay in situ and are unselectively bound by Agr in C. If Agr in C stays as a pronominal anaphor unclosed by any D and uncontrolled by any element c-commanding it, then the pragmatics ensures that this construction invites the inference that the speaker wants some other part of the discourse, typically a reply by a listener, to fill that cognitive gap. Such a gambit is normally called a question.

Acknowledgements

This work, having grown over more than two decades, owes a lot to many conversations, institutions, and persons. If I single out exchanges with Dennis Perzanowski, Ray Dougherty, Robert Fiengo, Joan Bachenko, Mark Baltin, Keith Allan, Geoff Millar, Paul Modini, Alice Davison, Luigi Rizzi, K.A. Jayaseelan, R. Amritavalli, P. Madhavan, Madhu Gurtu, Kashi Wali, Richie Kayne, Norbert Hornstein, Noam Chomsky, Ashok Kelkar, Michael Halliday, Rajendra Singh, Alan Ford, Gautam Sengupta, Howard Lasnik, Josef Bayer, Wim van der Wurff, Hans Henrich Hock, Udaya Narayana Singh, George Lakoff, Ronald Langacker, Clinton Seely, Mark Aronoff, Josef Bayer, Henk van Riemsdijk, Riny Huybregts, Brendan Gillon, Mark Baker, Lisa Travis, Denis Bouchard, Peter Hook, James Gair, M.T. Hanybabu, Lyn Ohira, the late Lewis Levine and the late James D. McCawley, this is not to imply that other associates have not been part of the work in progress that this presents an interim report about (I have edited out the names of my own students and chaatrosthaaniyos on the grounds that one does not thank within the production circle). The usual disclaimers apply.

References

Baker, C.L. 1970. Notes on the description of English questions: the role of an abstract question morpheme. Foundations of Language 6.192-219.

Borer,H. 1989. Anaphoric Agr. O. Jaeggli, K. Safir (eds.) The Null Subject Parameter. Dordrecht: Reidel.

Chomsky, N. 1981. Lectures on government and binding. Dordrecht: Foris.

Chomsky, N. 1982. Some concepts and consequences of the theory of government and binding. Cambridge, MA: MIT P.

Chomsky, N. 1986. Barriers. Cambridge, MA: MIT P.

Cinque, G. 1982. On the theory of relative clauses and markedness. The Linguistic Review 1.247-94.

Dasgupta, Probal. 1980. Questions and relative and complement clauses in a Bangla grammar. New York University doctoral dissertation.

Dasgupta, Probal. 1989a. Degree words in Esperanto and categories in Universal Grammar. In: Klaus Schubert (ed.) Interlinguistics: aspects of the science of planned languages. Berlin /New York: Mouton de Gruyter. 231-247.

Dasgupta, Probal. 1989b. Projective syntax: theory and applications. Pune: Deccan College Postgraduate and Research Institute.

Dasgupta, Probal. 1990 . The ubiquitous complementizer. (Circulated as a 1990 University of Hyderabad ms.; forthcoming soon in a festschrift not yet nameable).

Dasgupta, Probal. 1992. Pronominality and deixis in Bangla. Linguistic Analysis 22:1-2.61-77.

Dasgupta, Probal. 1994. Verbless and nonfinite clauses in Bangla. PILC Journal of Dravidic Studies 4:1.57-67.

Dasgupta, Probal. 1996a. Baanglaa saapekkho sarbonaam (in Bangla: 'Bangla relative pronouns'). Saahitto Potrikaa 40:1.5-24. [Dhaka: Dhaka University.]

Dasgupta, Probal. 1996b. Remarks on subjunctivity. In Shivendra K. Verma, Dilip Singh, eds. Perspectives on language in society. Delhi: Kalinga. 72-89.

Dasgupta, Probal. 1997. Postlexical morphology in parametric grammar. In M. Hariprasad et al. (ed.) Phases and interfaces in morphology, Volume One. Hyderabad: CIEFL Publications. 36-52.

Dasgupta, Probal. 1998. Knowing the word /trikkhe/: against purism in the study of language.Rajendra Singh (ed.) The Yearbook of South Asian Languages and Linguistics 1998. New Delhi: Sage.

Dasgupta, Probal. 2000. Tesniere indicators and Indian languages. Rajendra Singh (ed.) The Yearbook of South Asian Languages and Linguistics 2000. New Delhi: Sage.

Dasgupta, Probal; Ford, Alan; Singh, Rajendra. 2000. After etymology: towards a substantivist linguistics. Muenchen: Lincom Europa.

Davison, A. 1986. Binding relations in relative clauses. Chicago Linguistic Society 22.154-65.

Davison, A. 1987. Wh-movement in Hindi-Urdu relative clauses. Studies in the Linguistic Sciences 17.

Davison, A. 1988. Operator binding, gaps, and pronouns. Linguistics 26.181-214.

Ford, Alan; Singh, Rajendra; Martohardjono, Gita. 1997. Pace Panini. New York: Peter Lang.

Hendrick, Randall. 1982. Construing relative pronouns. Linguistic Analysis 9.205-224.

Kachru, Y. 1980. Aspects of Hindi grammar. Delhi: Manohar.

Koopman, H.; Sportiche, D. 1982. Variables and the bijection principle. The Linguistic Review 2.139-160.

Manzini, M.R. 1983. On control and control theory. Linguistic Inquiry 14.421-446.

Pesetsky, D. 1982. Paths and categories. Cambridge, MA: MIT Ph.D. dissertation.

Pesetsky, D. 1987. Wh in situ: movement and unselective binding. E. Reuland & A. ter Meulen (eds.) The representation of (in)definiteness. Cambridge, MA: MIT P.

Rothstein, S. 1983. The syntactic form of predication. Cambridge, MA: MIT Ph.D. dissertation.

Safir, K. 1986. Relatives in a theory of binding and levels. Linguistic Inquiry 17.663-89.

Srivastava, V. 1990. The syntax and semantics of correlatives. Natural Language and Linguistic Theory 8.

Subbarao, K.V. 1984. Complementation in Hindi syntax. Delhi: Academic Publications.