Switch-reference phenomena as evidence for structure-based approaches to binding theory

1. Introduction

Binding theory originates with the attempt to account for the interpretations of ordinary and reflexive pronouns. However, binding theory has also been extended to the analysis of a range of data beyond those that originally motivated it. Switch-reference, a grammatical phenomenon found in many Native American and Papuan languages, is one such domain.

In this paper, I will argue that switch-reference (hereafter SR) is based upon the configurational properties of the sentence in which it appears; it is best explained through appeal to notions like A'-status and command rather than notions like argument or agency.

I will argue that this conclusion for SR systems has implications for the structure of binding theory in general. The interpretation of the pronominal data that originally motivated binding theory has become more complex with the recognition of interactions between purely structural notions like command and more semantic notions like agency and logophoricity.

In my view, SR systems present a more pristinely structural system of binding relationships than that found in pronominal binding, and for this reason examination of SR is important for distinguishing structural from non-structural effects in binding theory.

2. Basic assumptions

A typical instance of SR is shown in the following Choctaw examples:

(1) John-at abiika-haatokoo-sh ik-iiy-o-tok.
John-NM sick-because-SS III-go:L-NEG-PT

`Because John(i) was sick, he(i) didn't go.'


(2) John-at abiika-haatokoo-n ik-iiy-o-tok.
John-NM sick-because-DS III-go:L-NEG-PT

`Because John(i) was sick, he(j) didn't go.'

In (1), the complementizer of the subordinate clause includes a SR marker indicating that the two clauses have the same subject (SS), while in (2) the subordinate clause is marked to show different subject (DS).

Finer (1984, 1985) analysed SR as an instance of A'-binding, treating the SS marker as an anaphor that must be bound by the Infl/Comp of the matrix clause, and the DS marker as a pronominal that must be free of this same Infl/Comp. [1] Since Infl (and according to Finer's assumptions, Comp) bears the index of the subject of the matrix clause, the observed distribution follows.

A slightly modified version of Finer's assumptions is shown in the following figure.

In the preceding examples, SR markers appeared on the Comp of an adverbial (or adjunct) clause. SR markers also appear on complement clauses, as seen in cases like the following:

(3) John-at [pro pisachokma-kat] ikhanah.
John-NM goodlooking-COMP:SS know

'John(i) knows that he(i) is goodlooking.'

(4) John-at [pro pisachokma-ka] ikhanah.
John-NM goodlooking-COMP:DS know

'John(i) knows that he(j) is goodlooking.'

The analysis of switch-reference as an instance of binding requires a few assumptions about the operation of binding which should be made explicit. More detailed argumentation on these points may be found in Broadwell (1990). I depart from Finer's assumptions about switch-reference in regarding the Choctaw different-subject markers as instances of disjoint anaphors rather than pronominals. The motivation for this is discussed in section 4 below.

2.1 M-command vs. c-command

I have taken the Infl of the matrix clause to be the the element binding the SR marker. The motivation for this assumption is the idea that SR markers need an antecedent in an A'-position.

Infl m-commands Comp because the first maximal projection set that dominates Infl also dominates Comp. However, the use of m-command in binding theory requires some discussion. It has sometimes been suggested that binding theory only uses the strict definition of c-command (involving the first branching node), due to examples like the following.

(5) [the city]'s(i) destruction t(i)

If we assume that the city is in the [Spec, NP] position and its trace is in the complement position, then the trace m-commands its antecedent, creating a Condition C violation. However, if the structure of (5) involves movement to [Spec, DP] then this objection is no longer valid.

Another possibility is that the subject itself is an A'-position. If the subject originates in a VP-internal position, as many current theories assume, then the [Spec, IP] might count as an A'-position, and direct binding of the SR marker by the subject would be possible while maintaining a stricter definition of c-command.

2.2 The index of the SR marker

In many approaches to pronominal binding, anaphors are freely assigned an index and binding theory then determines the grammatical coindexings. A binding-theoretic approach to switch-reference will only work if the SR marker in Comp bears the index of the subject of that clause. I'll assume that the SR marker receives this index via Spec-Head agreement with Infl (or Agr), and that Infl is incorporated into Comp, bringing this index with it. This approach is compatible with the idea of free indexing for NPs (or DPs), but assumes that functional elements like Infl and Comp receive referential indices through operations like agreement and incorporation.

2.3 Governors in binding domains

The classic definition of governing category (Chomsky 1981) says that the minimal binding domain for an anaphor must include the anaphor, an accessible SUBJECT, and an item governing the anaphor. However, under my assumptions the binding domain for a SR marker must be the smallest maximal projection containing the SR marker and a potential A'-antecedent. Since SR markers canonically occur in ungoverned positions, including governor in the definition of the binding domain would yield the wrong results.

However, the original motivation for stipulating the presence of a governor in the definition of binding domain was somewhat questionable. The inclusion was intended to derive the distribution of PRO from the binding theory (in what is generally called the PRO theorem). Without discussing the controversy in any detail, it is sufficient to say here that there are several plausible alternatives to the binding-theoretic account of the distribution of PRO. I take the inclusion of governor in the definition of binding domains to be unnecessary, and the fact that SR markers appear in ungoverned positions is unproblematic.

3. The configurational nature of switch-reference

By virtue of being A'-anaphors, SR markers need not be bound by arguments; they are sensitive only to A'-status and command relationships. In particular, there is little evidence to support the view that SR is sensitive to notions like argument or agency.

In the Western Muskogean languages Choctaw and Chickasaw, SR interacts with a rule of Possessor Raising (PR) in a way that shows that the antecedent of a SR marker need not be an argument of the matrix clause. Possessor raising extracts the possessor of the subject of an intransitive verb and adjoins it to the sentence; the raised possessor then receives nominative case. (6) shows a Chickasaw sentence without PR; (7) shows the same sentence after PR has applied. The Chickasaw data come from Munro and Gordon (1982) and Carden, Gordon, and Munro (1982).

(6) John im-ofi'-at illi-h.
John III-dog-NM die-TNS

`John's dog died.'


(7) John-at ofi'(-at) im-illi-h.
John-NM dog(-NM) III-die-TNS

`John's dog died.'

I suggest that sentences like (7) have a structure like that shown in the following figure:

Carden, Gordon and Munro (1982) give arguments that NP1 and NP2 form a constituent in (6) but not in (7).

One such argument comes from adverb placement. Adverbs may intervene between a raised possessor and the possessed noun, but they may not intervene between a non-raised possessor and the possessed noun:

(8) Oblaashaash [John im-ofi'-at] illi-tok.
yesterday John III-dog-NM die-PT

`Yesterday Jan's dog died.'

*John oblaashaah im-ofi'-at illi-tok.

(9) Oblaashaash [John-at] [ofi'-at] im-illi-tok
yesterday John-NM dog-NM III-die-PT

OK John-at oblaashaash ofi'-at im-illi-tok.
John-NM yesterday dog-NM III-die-PT

Sentences with both PR and SR are important for showing that binding of the SR marker is not limited to arguments of the matrix clause. Consider the following example:

(10) John-at ofi'-at im-ambiika-tok
John-NM dog-NM III-sick-PT

[sa-kisili-tokat]
1sII-bite-when:SS

`John(i)'s dog(j) was sick when he(i)/it(j) bit me.'

John is not an argument of the matrix clause. However, it is among the potential antecedents of the SS marker in the subordinate clause, by virtue of appearing in an A'-position.

Crucially, if PR had not applied in (10), the interpretation of the sentence would be different:

(11) John im-ofi'-at abiika-tok
John III-dog-NM sick-PT

[sa-kisili-tokat]
1sII-bite-when:SS

`John(i)'s dog(j) was sick when it(j)/*he(i) bit me.'

Because the possessor John is still contained within the subject, it is unable to c-command the SR marker in the adverbial clause, and the subject John's dog is the only possible antecedent for the SS marker.

Sentences like (10) have a structure like that shown in figure 3.

Sentences like these are problematic for approaches that treat SR data through an appeal to semantic notions like argument or agent.

Stirling (1993) is one such treatment. She gives a treatment of SR in which it is treated as "a kind of clause-level agreement, which normally marks the clause it occurs in as syntactically and semantically dependent, and indicates whether there is continuity or discontinuity between the eventuality described by the marked clause and that described by the controlling clause." (p. 123)

Stirling's approach is framed in Unification Categorial Grammar, combined with Discourse Representation Theory. Within this approach, each clause is associated with a `structured eventuality index' which contains three parameters: the Protagonist (defined as the agentive subject), the Actuality (realis or irrealis), and the Location. SS constrains the matrix and embedded clauses to agree in their eventuality parameters; DS indicates disagreement in at least one of the eventuality parameters.

However, Stirling's claim that SS signals coreferentiality of agents cannot be maintained, since John is not the agent, or even an argument, of the matrix clause. In this case a semantic approach to switch-reference fails, but the structurally based approach makes the correct prediction.

In general, approaches to anaphora which treat the reflexive as a valency-decreasing operation on predicates (Keenan 1988) or a function imposing a identity restriction on arguments of a predicate (Reinhart and Reuland 1993) do not extend easily to SR systems. SR is canonically a relationship between NPs associated with different predicates, and the Chickasaw evidence shows that SR markers need not even be bound by arguments.

In contrast to the predicate-based approaches to anaphora, a purely structural account in terms of command relationships extends naturally to the SR data.

4. Implications of SR for the notion 'disjoint anaphor'

The interaction of SR marking with the system of possessor raising also provides evidence for the notion of disjoint anaphors. The term 'disjoint anaphor' was first introduced by Saxon (1984) in the description of the Athapaskan language Dogrib.

Consider the following examples from Dogrib:

(12) John ye-hk'e ha.
John 3DIS-shoot FUT

`John(i) is going to shoot him(j, *i).'

/ye-/, as a disjoint anaphor, must have an index distinct from that of the NP which c-commands it. It differs from a pronominal in occurring only in positions where a counterindexed antecedent is available. Thus the following instance in subject position is ungrammatical:

(13) *?ekaani ye-enda.
thus 3DIS-live

`He lives this way.'

We may implement the condition on disjoint anaphors as follows: A disjoint anaphor bearing the index i must be c-commanded by an NP bearing the index j, where i does not equal j.

There is good reason to think that the Choctaw and Chickasaw DS marker is a disjoint anaphor, rather than a pronominal.

Consider the following example:

(14) John-at ofi'-at im-ambiika-tok
John-NM dog-NM III-sick-PT

[sa-kisili-toka]
1SII-bite-when:DS

`John(i)'s dog(j) was sick when he(i)/it(j) bit me.'

This sentence has a structure like the following:

If the DS marker were a pronominal, it would have to be free of all c-commanding antecedents. However, a disjoint anaphor only needs to be c-commanded by one NP with a distinct index. Since the pro subject of the embedded clause may be either `dog' or `John', the DS marker must be a disjoint anaphor.

5. Conclusion

The binding-theoretic account of switch-reference marking sketched here succeeds in accurately describing a wide range of complex phenomena in Choctaw and Chickasaw. The account, however, relies on a purely structural notion of binding, in which only the geometric configuration of the sentence is relevant to determining the SR possibilities. Thematic notions like agency and argumenthood do not seem to play any significant role in the binding of SR markers in Choctaw and Chickasaw.

These facts argue that structural notions like command are an indispensable element in the explanation of the full range of facts associated with binding theory. While there is some evidence that the binding of arguments is affected by thematic considerations in other languages, the switch-reference data suggest that structurally-based accounts of binding cannot be eliminated from syntactic theory.

[1]. Finer treats Infl/Comp as the joint head of S. In what follows, I use Infl as an abbreviation for some number of functional projections (presumably including Tns and Agr) when the specific content of these projections is not relevant.

6. Bibliography

Broadwell, George A. 1990. Extending the binding theory: A Muskogean case study. UCLA Ph.D. thesis. [distributed as UCLA Dissertations in Linguistics #11.]

Carden, Guy; Gordon, Lynn; and Pamela Munro. 1982. Raising rules and the projection principle. ms.

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

Finer, Daniel. 1984. The formal grammar of switch- reference. UMass Ph.D. thesis.

Finer, Daniel. 1985. The syntax of switch- reference. Linguistic Inquiry 16:35.

Keenan, Edward. 1988. On semantics and the binding theory. in Hawkin, John, ed. Explaining language universals. Oxford: Blackwell.

Munro, Pamela, and Lynn Gordon. 1982. 'Syntactic relations in Western Muskogean: a typological perspective,' Language 58:81.

Reinhard, Tanya and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24:657-720.

Saxon, Leslie. 1984. 'Disjoint anaphora and the binding theory,' in Proceedings of the West Coast Conference on Formal Linguistics 3.

Stirling, Leslie. 1993. Switch-reference and discourse representation theory. Cambridge: Cambridge University Press.

In the Choctaw examples, orthographic symbols have their standard phonetic values, except that <sh>=phonetic symbol [], <ch>= [], and <lh>=[] (a voiceless lateral fricative). Underlining shows vowel nasalization.

The following abbreviations are used in the glosses: acc=accusative, caus=causative, dis=disjoint ds=different subject, foc=focus, fut=future, irr=irrealis, l=l-grade (a special form of the verb stem used with the negative), neg=negative, nm=nominative, pot=potential, pl=plural, pres = present, pt=past, S=subject, ss=same subject, tns=tense. I, II, III indicate three different agreement sets for Choctaw, as discussed in Munro and Gordon (1982).

If you wish to make a public comment about this paper; please send your message to: linconf@tamvm1.tamu.edu

Discussion of this paper is available at the Discussion Web Site at: