Johni has no covering over himi. Johni ignored the oil on himi.
University of Pittsburgh and University of Quebec at Montreal
tenny+@pitt.edu, tenny@linguist.kollar.com
1. Introduction
It is well-known that the classical structure-based binding theory of Chomsky (1981), although it makes sense of a large body of data, fails to account for certain instances of pronominal and anaphoric coreference. Since Chomsky (1981), attempts have been made to incorporate two other elements besides syntactic structure into explanations of coreference facts across languages; these two elements are thematic structure and logophoricity or discourse factors (see Reinhart and Reuland, 1993 and 1991; Pollard and Sag, 1992; Sells, 1987). Much of this work has been motivated by detailed study of long-distance anaphora, as the well-known violations of classical binding theory which must be accounted for. The challenges posed by long-distance anaphora for syntactic theory have been a rich source of discussion in the recent literature (see Koster and Reuland, 1991; Cole and Wang, 1996). Less attention has been paid to the complementary violation of classical binding theory: short-distance pronouns (SDP), or pronouns that take an antecedent within a local domain where Condition B of Chomsky (1981) should rule them out, as in the well-known example:
(1) Johni found a snake near himi.
Short-distance pronouns have been noted in the literature as early as Lees and Klima (1963), who recorded the examples in (2) below. (Indices are not included in the original examples; these have been added here for the sake of exposition.)
(2) The meni found a smokescreen around themi.
Short distance pronouns have mostly been studied in the context of the breakdown of the complementary distribution of anaphora and pronouns in certain syntactic environments. A notable exception in the recent literature are Reinhart and Reuland (1993), who addressed examples of SDP such as the following in a systematic way:
(3) Maxi saw a gun near himi.
(4) Maxi likes jokes about himi.
In (3) the SDPs appear inside a prepositional phrase (PP) representing a goal or location. In (4) the SDPs are embedded within a noun phrase (NP) argument of the verb. In this paper I will focus on SDPs like those in (3).
In brief, this paper argues that short distance pronouns occur logophorically only in certain types of logophoric contexts (section 2). The proposal is made in this paper, that the property of referential independence in NPs falls under the grammatical system governing logophoricity and point of view. It is proposed that referential independence crucially requires the availability of an independent point of view (section 3). A distinction is made between intensional and extensional deixis, where only intensional deixis interacts with referential independence (section 4). This explains why logophoric environments of locational deixis permit SDP, while other kinds of logophoric environments do not. Finally section 5 returns to the theme of this conference, by examining how these ideas can inform the debate about thematic structure versus geometric structure in binding. The facts discussed in this paper lead to the conclusion that some kind of reference to thematic or discourse structure is necessary. Whether or not reference to geometric or syntactic structure is also necessary is not indicated by these data, and remains to be seen.
Much of the discussion in this paper is quite cursory, due to the limitations of the conference format. I will not discuss in this paper how the ideas above may be extended to the types of examples in (4).
2. The distribution of short-distance pronouns in English
This section illustrates the distribution of short-distance pronouns (SDP) in English. There are three conditions on the distribution of short distance pronouns:
(i) Short distance pronouns and their antecedents
(ii) Short distance pronouns are logophoric.
(iii) Short distance pronouns involve locational deixis.
The first two conditions have been observed in some form and discussed in the literature to varying degrees, although not always with specific reference to SDPs. The third condition has not previously been proposed to the best of my knowledge.
Short distance pronouns are found where the complementarity of pronouns and anaphors breaks down within a local binding domain. It has become apparent that SDP are possible only where the pronoun and its antecedent are not coarguments, or arguments of the same predicate (Partee and Bach, 1981; Farmer 1984; Farmer and Harnish, 1987; Reinhart and Reuland, 1991 and 1993). In (1) through (3) above the SDP is an adjunct, not an argument of the verb, and in (4) above the SDP is embedded within an argument of the verb and consequently is not itself directly an argument of the verb. In all these examples, therefore, the SDPs and their coindexed antecedents are non-coarguments.
Reinhart and Reuland (1991) and (1993) (and similarly Pollard and Sag, 1992) argue that the scope of structural binding theory should be limited to coargument domains. (Reinhart and Reuland refer specifically to semantic coarguments.) This approach restricts the range of data that should be submissable to the binding theory, and makes a distinction between the phenomena that submit to a structural explanation, and the phenomena that are outside of structural binding theory. Reinhart and Reuland (1991) and (1993) further maintain that this second category of binding facts outside the realm of structural binding theory must be explained as instances of logophoric binding. The term logophoric, as used here, involves some reference to an individual who holds the salient perspective or point of view (usually the source of communication or the center of consciousness) in the sentence or discourse. Logophoric environments are most often found in the context of verbs of saying or reporting. The term was originally used by Clements (1975) with respect to Ewe, who applied it to pronouns in Ewe that took as their antecedents, the individual whose point of view or state of mind was being reported. The concept can be extended to include coreference with experiencers of mental states (Sells, 1987) or empathy (Kuno and Kaburaki, 1977). Reinhart and Reuland (1993) also include the emphatic or focus use of discourse anaphors under the label of logophoricity, distinguishing two types of logophors: the 'focus logophors' and the point-of-view logophors (which they label 'perspective logophors'). In this paper the term logophor refers only to perspective logophors in the sense of Reinhart and Reuland (1993).
All the examples of SDP presented above in (1) through (3) are logophoric,[1] in the sense that the pronoun represents the point of view or the perspective from which the situation described by the prepositional phrase (PP) is observed by the pronoun's antecedent. There is a deictic element in the interpretation of PPs such as near him, over him, under him, or towards him, since the location described by the PP depends on the location of the antecedent NP, which in these sentences can be determined only in the context of the utterance. I will refer to this as locational deixis in this paper. (The examples in (4) also involve a point of view, in the sense that the NP containing the SDP stands for a representation which can be perceived by the antecedent.)
The two conditions on SDP that Reinhart and Reuland provide -- that SDPs and their antecedents must be semantic non-coarguments and that SDPs must be logophoric -- are met in all the examples above involving locational deixis. Furthermore, these conditions correctly predict that SDPs will not be possible in non-logophoric contexts, even when they and their antecedents are non-coarguments. The following examples illustrate several non-logophoric adjuncts, including benefactive (5a), accompaniment (5b), and causal adjuncts (5c). SDP are not possible here, as predicted. (The corresponding anaphors are included in these examples to demonstrate the contrast.)
(5) a. Maxi bought a gun for *himi/himselfi.
The prediction that SDPs will not be possible for coarguments in logophoric contexts is also supported. SDPs are prohibited for true dative arguments, as diagnosed by the possibility of dative shifting as in (6a) versus (6b).[2] This generalization is quite robust.[3] This is true even if the dative-shiftable arguments occur in a potentially logophoric environment. In (6) the antecedent of the dative pronoun is reporting information, the classic logophoric environment. In (7) the antecedent is an experiencer of a mental state, also a potentially logophoric environment.
(6) a. Maxi told a story to *himi/himselfi.
(7) Maxi fears *himi/himselfi.
Having demonstrated the validity of the non-coargument condition and the logophoricity condition on SDPs, I will next show that it is still inadequate in capturing the relevant generalizations on the distribution of SDPs. The two conditions are necessary but not sufficient to account for certain patterns in the distribution of SDPs. Only certain types of logophoric environments permit SDPs. Sells (1987) following Kamp (1984) employs a set of three discourse roles in his account of logophoricity: SOURCE, SELF, and PIVOT. SELF refers to the one whose internal mental state or point of view is being reported, SOURCE refers to the one who does the reporting, and PIVOT refers to the one who represents a physical locus -- a location in fact -- to which deictic elements refer. The PIVOT represents quite literally a very physical point of view. All the examples of SDP in (1) through (3) have antecedents associated with the PIVOT discourse role, since they provide an anchor point for some sort of locational deixis. As Sells has argued, logophoric binding is not a unitary phenomenon, but must be understood in terms of several different aspects of discourse structure having to do with point of view. Yet SDPs seem to invoke only one aspect of logophoricity, that of locational deixis (or in Sells' terms, that involving the PIVOT discourse role). Non-coarguments in other kinds of logophoric contexts do not give rise to SDPs. The following examples with reportative or state-of-mind predicates are somewhat awkward without a strong context, but the pronouns are clearly less acceptable than the anaphora. (8a), (8b) and (8e) involve verbs of saying ('talk', 'recount'); (8c) involves a verb of psychological state ('fear'); and (8d) involves a subordinate purpose clause (which Clements (1975) showed can induce a logophoric context in Ewe).
(8) a. Maxi talked all day about *himi/himselfi to his friends.
In each case the SDP refers to Max, who is the source of communication (8a-b, 8e), or the locus of the mental state: a state of fear as in (8c) or a state of knowledge of an inner purpose as in (8d). There is no explanation for why we do not find SDPs in these kinds of logophoric contexts other than locational deixis.
It is also clear that locational deixis provides a productive environment for SDPs. The examples in (1) through (3) are not simply idiosyncratic or idiomatically determined, but part of a systematic and productive pattern. Numerous examples can be adduced of short distance pronouns in contexts of locational deixis. Some examples are listed in (9):
(9) Janei saw/ heard/ felt/ sensed the cat near herI/
These data support the generalization that it is locational deixis that induces a context in which SDP are possible. We have to examine locative and directional PPs for a special property having to do with locational deixis, which allows for SDP. We will look first at the nature of referential independence and its relation to point-of-view.
3. Referential independence and the difference between pronouns and anaphors
A fundamental difference between pronouns and anaphors lies in their ability to refer; anaphors are referentially defective (Chomsky, 1986; Keenan, 1987), while pronouns have the capacity for referential independence (Reinhart and Reuland, 1993). Reinhart and Reuland represent the property of referential independence with a feature +R distinguishing pronouns from anaphors (the feature dating back to Chomsky (1981)'s R-expressions). I follow Reinhart and Reuland here but take the further step of proposing that what the feature +R for referential independence in fact means is that, crucially, there is an independent point of view available through which the referent of the NP can get its identity. (To put it crudely: there is someone who thinks the referent of the NP is other than itself.) I assume that any NP bearing the feature +R in fact is represented grammatically as viewed from outside by someone. I assume further that a (grammatical) point of view is always present, for every NP bearing the feature +R; the default point of view being that of an omniscient observer. Anaphors, which lack the feature +R, are viewed from the inside by their antecedent (if their antecedent can perceive); or else they are viewed as inside of, or part of, their antecedent by an omniscient observer. The lack of referential independence means that there is no independent point of view available, to look at the anaphor as referring to something with a distinct identity of its own. By default the anaphor has to be viewed from inside some other existing point of view; the anaphor gets this from its antecedent when it gets its reference from its antecedent. True logophors share with anaphors the lack of the feature +R, so they must also get their reference (and an independent point of view) from elsewhere. For a NP to get reference means (at least in part) that it gets connected up to some independent point of view.
There are three important theoretical consequences of this idea. First, the grammatical organization of reference is subsumed within the grammatical sub-system dealing with logophoricity and point of view; it is in fact a part of that system. Second, the distinction between pronouns and anaphors in their capacity for referential independence is a more complex and substantive distinction than is fully expressed by a feature account, which simply invokes the plus or minus values of the feature R. Third, anaphors and true logophors (lexical or morphological items that can only take logophoric coreference, as in some West African languages) are unified into a natural class of NPs that must connect to some independent point of view.
There is a practical consequence of this idea as well; it says that the reference of a NP is always (potentially) relative to some point of view. The following example helps to illustrate this point:
(10) Maxi said hisi brother would deliver the goods.
There is room for ambiguity in the interpretation of (10), even given the coindexing above. Suppose Max believes his brother to be Ralph, but Max is mistaken and his brother is in fact Paul. From Max's point of view the referent of 'his brother' is Ralph,while from the point of view of an omniscient observer, the referent of 'his brother' is Paul. The truth conditions for this sentence depend on the point of view that is invoked -- either Max's or an omniscient observer's. Ambiguity that depends on point of view is grammaticalized in some West African languages (Clements, 1975). It is not surprising that it should exist in English as a grammatical ambiguity, even though it is not marked morphologically; or that the default point of view of an omniscient observer (which we take for granted as something that does not need to be specified) is a default value for something which is always specified by the semantics of sentence-level grammar.
Extending this approach to logophors, true logophors share with anaphors the lack of referential independence or an independent point of view in their specification. It follows furthermore that pronouns, by virtue of their inherent specification for +R, cannot behave as true logophors, which (being unspecified for R) adopt wholesale the point of view associated with another NP.
It is important to clarify however, that NPs specified for +R can be dependent on the point of view of another referentially independent NP, as in the case of the reading of (10) in which the referent of 'his brother' is keyed to Max's point of view. However, this type of association between NPs is not the same as that of a true logophor with its antecedent, where the logophor can only be identified with an antecedent that supplies its point of view.[4]
If in general pronouns cannot be logophors, the question formulated in section 2 becomes: why are SDP permitted in the case of locational deixis? A more careful distinction of locational deixis from other kinds of logophoricity is called for.
For the answer to this question I will look in the next section at a two-part division of the grammatical system of deixis: into extensional and intensional deixis. The answer will lie in the fact that the property of referential independence is connectedwith the intensional deictic system, while locational deixis is connected with the extensional system.
4. Extensional and intensional deixis
An analysis of how the grammar of a natural language organizes and permits reference to the context of an utterance is susceptible to more than one approach. There is little disagreement, however, about what the deictic elements are that may enter into the grammar of a sentence; these are time, place, and the persons involved. I follow Fillmore (1971) in assuming that an utterance is associated with a 'Center', which consists of a specification of these three contextual elements:
(11) Center = [participants (speaker and addressee), time, place]
However, I argue here that deixis in natural language should be understood as consisting of two separate systems: the intensional and the extensional deictic systems. Of these three deictic elements, the persons involved belong to the intensional deictic system, while time and place or location belong to the extensional system. In terms of the Sells/Kamp discourse roles, the SOURCE and the SELF roles belong to the intensional system, while the PIVOT role (which is the center that provides deictic reference for location) belongs to the extensional system.
The intensional roles, the SOURCE and SELF roles, have to do with point of view as it is invoked in referential independence. Under the approach I am taking here, grammar relativizes the referential identity of a NP to a mind; and a SOURCE or a SELF role, at the very least, makes reference to a mind. The PIVOT role does not interact with point of view as invoked in referential independence, because the referential independence of a NP (as organized by grammar) does not depend on, or alter with, the time and place of utterance of that NP. Intensional roles have to do with minds: with states of mind, or with truth as relativized to minds. The PIVOT role has to do with the perception of something as located in time and space. It is extensional because the same truth of perception can obtain across minds. For example, Bill and Sam might disagree that a fellow was really Jim, but they would agree about whether that fellow was standing next to Sam or next to Bill.
This division of the deixis system into an intensional and an extensional component dovetails with certain other ideas to be found in the literature. Culicover and Wilkins (1989) make a distinction between intensional and extensional thematic roles. The aspectual roles of Tenny (1994) may be understood as representing the extensional part of the thematic role system. The extensional/intensional distinction also makes it easier to understand why the PIVOT role is distinguished from the SOURCE and SELF roles in other respects. Although Sells groups all three roles together in his discourse role hierarchy, he has no explanation for why the PIVOT role is always at the bottom of the hierarchy. Although they all refer to certain kinds of points of view, the PIVOT role and the SOURCE and SELF roles belong to different systems.
The statement that pronouns cannot be logophors, must be revised as follows: pronouns cannot be logophors, in contexts of intensional logophoricity. Pronouns cannot be logophoric in the sense of being linked with a SOURCE or a SELF role, because they already have an independent reference and point of view, which would conflict with that received from these roles. They can, however, be linked with a PIVOT role, which does not supply the point of view internal to an independent mind.
5. Two approaches: structural binding versus discourse roles.
In this section I will briefly compare two approaches to the binding facts discussed in sections 2, 3 and 4, in the context of the theme of this conference. The first approach is modeled after Enc (1987)'s treatment of tense and temporal reference, which employs some geometric/structural notions; the second approach is modeled after Sells' (1987) treatment of logophoric phenomena, and employs discourse roles, which are more closely related to thematic structure. These two approaches to binding-type phenomena each have something to contribute to an analysis of the material presented here. Because of the limitations of the conference setting, only the briefest of sketches is possible here.
Enc (1987) analyzes sequence-of-tense phenomena by treating tense as a referential expression which must be bound (or anchored), and which is syntactically present in Infl. She lays out four anchoring principles and conditions, which I summarize as follows, in (12):
(12) Every tense must be anchored. It may be anchored either by being bound directly by a higher tense, or through its local Comp's being anchored. The Comp may be anchored either by being bound within its governing category, or by denoting the speech time.
Omitting details, the structural notions that Enc employs are: governing category, government, and the categories Comp and Infl. Using these structural elements she is able to derive the correct sequence-of-tense effects for matrix clauses, relative clauses and complement clauses.
The similarities between the sequence-of-tense phenomena addressed by Enc and the phenomena described in this paper are striking enough to call for similar treatment. Some of the ideas proposed in this paper could be easily cast in a system modeled on Enc's treatment of tense. Reference is parallel to tense in several ways. A sketch of such a treatment follows, in (13):
(13) A referentially independent NP is associated with a point of view (just as a tense is associated with an interval). All referentially independent NPs must be anchored, either to some other referentially independent NP (bearing a discourse role), or else to the default point of view -- the omniscient observer.
In spite of the similarity of this approach to Enc's, no structural notions appear in this account. The question remains: what relevance, if any, does structure have for this kind of anchoring of reference?
The structure reflected in the difference between matrix clauses, complement clauses and relative clauses has consequences for sequence-of-tense phenomena -- but do not figure in the binding of SDP, by definition. Logophoric binding in general does not appear to be highly sensitive to syntax (see references previously cited). However, there is some evidence of the involvement of Comp or clausal-type in the licensing of logophoric binding in Icelandic (see Sells 1987 and references therein) and in Ewe (Clements 1975). Short distance pronouns do not supply any evidence for oragainst this in English. Furthermore, the account in (13) has had to make reference to discourse roles, which do not submit to a geometric/structural analysis.
Sells (1987) analyzes various kinds of logophoric phenomena in terms of the three discourse roles already mentioned. Logophoric expressions are linked to some other NP by virtue of being linked with one of these discourse roles. The option of being linked to an omniscient observer is provided by the option of having the SOURCE or SELF role external to the sentence. Sells makes relatively little use of geometric or structural notions such as c-command. Some structure is provided by virtue of a hierarchy of discourse roles, which constrains the possible combinations of discourse roles internal and external to sentence. The ideas presented in this paper are sketched in an approach modeled on Sells' in (14):
(14) Discourse roles are divided into intensional and extensional roles. Short distance pronouns can only refer to extensional discourse roles (PIVOTs). Pronouns in general, and any referentially independent NP, cannot logophorically bind with an intensional role (a SOURCE or a SELF). Every referentially independent NP is linked to a SOURCE or a SELF role, or to an external speaker.
Examination of short distance pronouns in English leads to the conclusion that some kind of reference to thematic or discourse structure is necessary. SDP in English do not require reference to geometric or syntactic structure; whether this is necessary for other kinds of logophoric binding remains to be seen.
Notes
[1]The use of the term 'logophoric' to apply to these examples broadens the sense in which Clements (1975) originally used it. However, it is used this way implicitly by Reinhart and Reuland (1993) and explicitly by Sells (1987).
[2]Benefactives such as (5a) also fall under this account, if the possibility of dative shifting is taken as the criterion for argument-hood.
[3]Note that it is not the obligatoriness of the NP that diagnoses argument-status here, as Max told a story is perfectly grammatical.
[4]Another difference between the reading of (10) which is keyed to Max's point of view, and the impossible logophoric readings of SDPs, is that the relevant reading of (10) is included in the normal range of possible readings for a pronoun in any case, which is not true forany logophoric reading of a SDP. The approach I am taking predicts that more differences should emerge between the logophoric-like binding of the regular pronoun in (10) and the binding of true logophors with their antecedents. This is a fruitful question for further research.
References
Chomsky, N. (1981). Lectures on Government and Binding. Foris. Dordrecht.
Chomsky, N. (1986). Barriers. MIT Press. Cambridge, Mass.
Clements, G. N. (1975). "The Logophoric Pronoun in Ewe: Its Role in Discourse." Journal of West African Languages 2, 141-177.
Cole, P, and C. Wang. (1996). "Antecedents and Blockers of Long-Distance Reflexives: The Case of Chinese Ziji." Linguistic Inquiry 27(3): 357-390.
Culicover, P.W. and W. Wilkins. (1989). "Control, PRO, and the Projection Principle." Language 62:1. 120-153.
Enç, M. (1987). "Anchoring Conditions for Tense." Linguistic Inquiry 18 (4):633-657.
Farmer, A. (1984). Modularity in Syntax: A Study of English and Japanese. MIT Press. Cambridge, Mass.
Farmer, A. and M. Harnish. (1987). "Communicative reference with pronouns." In M. Papi and J. Verschueren, eds., The pragmatic perspective. Benjamins. Amsterdam.
Fillmore, C. (1971). Santa Cruz Lectures on Deixis. Indiana University Linguistics Club.
Kamp, J.A.W. (1984). "A Theory of Truth and Semantic Representation." In J. Groenendijk, T. Janssen, and M. Stockhof, eds., Truth, Interpretation and Information. Foris. Dordrecht. 1-41.
Keenan, E. (1987). "On semantics and the binding theory." To appear in J. Hawkins, ed., Explaining language universals. Blackwell. Oxford.
Koster, J., and E. Reuland, eds. (1991). Long-Distance Anaphora. Cambridge University Press. Cambridge.
Kuno S., and E. Kaburaki. (1977). "Empathy and Syntax." Linguistic Inquiry 8, 627-672.
Lees, R., and E. S. Klima. (1963). "Rules for English Pronominalization." Language 39 (1):17-28.
Partee, B. and E. Bach. (1981). "Quantification, pronouns, and VP-anaphora." In Formal methods in the study of language. Mathematisch Centrum, Amsterdam University.
Pollard, C., and I. Sag. (1992). "Anaphors in English and the Scope of Binding Theory." Linguistic Inquiry 23 (2):261-303.
Reinhart, T., and E. Reuland. (1993). "Reflexivity." Linguistic Inquiry 24 (4):657-720.
Reinhart, T., and E.Reuland. (1991). "Anaphors and logophors: an argument structure perspective." In J. Koster and E. Reuland, eds., Long-Distance Anaphora. Cambridge University Press. Cambridge.
Sells, P. (1987). "Aspects of Logophoricity." Linguistic Inquiry 18 (3):445-479.
Tenny, C.L. (1994). Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer.