Experimental evidence supporting the overlapping distribution of core and exempt anaphors: Re-examination of long-distance bound caki-casin in Korean

2.1 The distribution of anaphors: Core vs. exempt binding

It is well-known that natural languages possess anaphoric expressions such as reflexives and pronouns whose interpretations are constrained by syntax. Principles A and B of Chomsky’s (1980, 1981 classical Binding Theory (BT hereafter) are best known formulations of the syntactic constraints on the interpretation of such anaphoric expressions. BT defines a syntactic domain in which reflexives (also called anaphors) need to be bound by an antecedent but pronouns cannot. As such, it is designed to capture the complementary distribution between pronouns and anaphors seen in sentences such as (1) below.

However, it has been known for some time that the complementary distribution of reflexives and pronouns fails to hold, in contexts such as those shown in (2).^[1]

In order to deal with examples such as these, proponents of BT introduced revisions to the algorithm that determines the Binding Domain, so that the Binding Domain for anaphors and pronouns can diverge in some cases. Specifically, the revised BT in Chomsky (1986) accounts for the sentences in (2) by defining the Binding Domain as the smallest complete domain of a head with all its syntactic arguments (dubbed the Complete Functional Complex, or CFC) where the BT could have been satisfied, that is, the minimal BT-compatible CFC containing the anaphoric expression. Since Principle A requires anaphors to be bound, a CFC where an anaphor occurs in the highest (A)-position cannot be the BT-compatible CFC for the anaphor, while it can for a pronoun. In such cases, the Binding Domain for anaphors is extended to the next CFC that is BT-compatible. Assuming that the bracketed constituents in (2a)–(2d) are the minimal CFCs containing the anaphoric expressions, the revised BT correctly predicts that while pronouns simply need to be free within them, anaphors need to go outside them to be bound. Thus, their overlapping distribution is predicted. By contrast, in (1), the embedded clause is a BT-compatible CFC for both reflexives and pronouns, and the two must be in complementary distribution.

While the revised BT makes some progress toward explaining when anaphors and pronouns are in complementary distribution and when they need not be, it still fails to provide an account of the full range of counterexamples to BT. For example, it fails to account for the sentences in (3). In (3a), the revised BT predicts that the anaphor should be bound in the intermediate clause. Though the anaphor is not bound in this domain, the sentence is acceptable. BT predicts that (3b), where the anaphor is unbound, to be unacceptable. Similarly, in (3c), the minimal BT-compatible CFC for the anaphor is the embedded clause and yet the sentence is fine with the anaphor bound by the matrix subject.

Data such as (2) and (3) have prompted researchers to develop a comprehensive alternative to the (revised) BT that can draw a principled distinction between anaphors that are in complementary distribution with pronouns and those that are not (Pollard and Sag 1992, 1994; Reinhart and Reuland 1991, 1993, among others). The alternative approach starts from very different premises about the nature of (anaphor) binding. The most important difference is that under this approach not all anaphors are subject to syntactic BT. Only some are, and it is these anaphors that show a complementary distribution with pronouns. It thus posits a distinction between syntactically constrained core (or plain) anaphors and exempt (or logophoric) anaphors that are not constrained by syntactic BT but are licensed extra-syntactically instead. Since it typically makes the distinction between core and exempt binding on the basis of whether an anaphor has a superior co-argument, we call this alternative approach the Argument Structure-based Binding Theory (abbreviated AS-BT).^[2] In this approach, exempt anaphors are those without superior co-arguments. This can happen if the anaphor is the sole argument of a monovalent predicate, or expresses the most prominent argument of a polyvalent predicate. The anaphor in (2) expresses the sole argument of the (nominal) predicate, and hence is exempt, while that in (1) has a superior co-argument, Bill, and is predicted to be a syntactic/core anaphor that must be bound in the local (argument structure) domain. The anaphor in (3a) is also exempt, and is free to participate in discourse-sanctioned long-distance binding. The exempt anaphor in (3b) also occurs in an exempt position and is discourse-bound by the addressee, while in (3c), the anaphor is part of a coordinate object, and hence also exempt, since the argument of the verb is the entire coordinate NP and not just a part of it.

While there is some convergence between the mechanisms of how the Binding Domain for anaphors is extended in the revised BT of Chomsky (1986) and how the exempt status of an anaphor is determined in AS-BT approaches, the convergence does not extend very far. For one, in the revised BT, an anaphor in the highest position of a CFC (let’s call such a position an exempt position, though technically, the revised BT does not recognize a distinct class of exempt anaphors) needs to be bound in the next CFC that is BT-compatible. This means that the next CFC must have a binder, and that the anaphor cannot bypass that domain for an antecedent in a more superordinate domain, or be discourse-bound. On the other hand, the AS-BT makes no such predictions. An anaphor in an exempt position is released from the confines of syntactic BT, and is permitted to take long-distance antecedents in a more superordinate domain. Thus, the two theories diverge with respect to sentences such as (3a) and (3c), which show that long-distance binding of an exempt anaphor across a c-commanding DP in the immediate CFC above it is possible, supporting the prediction of AS-BT.

Another difference between the revised BT and AS-BT has to do with whether exempt anaphors can be freed from syntactic conditions on binding, specifically c-command. Proponents of the AS-BT have argued that anaphors in exempt positions can take non-c-commanding antecedents (cf. [4a]) or even discourse antecedents (cf. [3b], [4b]) under the right conditions, neither of which is predicted under the revised BT that requires an anaphor in an exempt position to be bound.

A final difference between the revised BT and AS-BT has to do with interpretive properties of anaphors that occupy exempt positions. In the former, since these are still syntactically bound anaphors, they are not expected to behave differently from anaphors bound by c-commanding DPs in a local domain in terms of their interpretive properties. In the latter, anaphors in exempt positions are not syntactically bound, but licensed extra-grammatically. Therefore, these anaphors are expected to show interpretive properties that are different from those that are syntactically bound. It has been claimed that this prediction is borne out. For example, when anaphors are syntactically bound, they cannot have split antecedents while anaphors in exempt positions allow split antecedents, as shown in the contrast between (5a) and (5b). They also show differences in contexts of ellipsis/pro-forms. When the antecedent for VP ellipsis contains a syntactically bound anaphor as in (5c), the elliptical VP preferentially receives a sloppy interpretation rather than a strict interpretation. On the other hand, both sloppy and strict readings can be obtained if the missing VP includes an anaphor that occupies an exempt position as in (5d). The ellipsis in (5d) can be construed either as ‘Bill also thought that an article written by Bill caused the uproar’ or as ‘Bill also thought that an article written by John caused the uproar’.

In sum, the AS-BT offers a principled account of the distribution of anaphors and pronouns in ways that are superior to the coverage of the revised BT. This is why the distinction between core and exempt anaphors is widely adopted in current research on anaphor binding (Cunnings and Sturt 2014; Kim et al. 2015; Koornneef 2008, 2010; Runner et al. 2002, 2006, etc.).

However, some researchers (Charnavel and Sportiche 2016; Huang and Liu 2001; Kim and Yoon 2009; Zribi-Hertz 1989) have raised questions about the additional premises held in the foundational works on core versus exempt binding. These additional premises are the following. The first premise has to do with the complementary distribution of core and exempt binding. That is, an anaphor that can be licensed by syntactic BT must be so licensed. It does not seem possible to bypass syntactic BT and license the anaphor as an exempt anaphor. If this were possible, we would not have the standard judgments that serve as the basis for the complementary distribution of anaphors and pronouns in sentences such as (1). Why would we have the judgment that an anaphor cannot be bound by the matrix subject in (1) if long-distance exempt binding remains an option for it? Therefore, the foundational works on AS-BT assume that core and exempt binding are in complementary distribution (Pollard and Sag 1992, 1994; Reinhart and Reuland 1991, 1993; Reuland 2001, 2011). Exempt binding cannot apply in a domain where core binding can. Nevertheless, theoretical and experimental research has uncovered data where this premise seems to be challenged. Consider the sentences in (6).

Under AS-BT, himself in (6a) is an exempt anaphor while himself in (6b) is a core anaphor because the former lacks a superior co-argument while the other has one, the possessor Bill. Under the assumption that core binding is in complementary distribution with exempt binding, the binding of himself by John is predicted to be well-formed in (6a) but not in (6b). However, experimental investigations of binding in NPs with possessors have found that the interpretation that the AS-BT predicts to be impossible (John = himself in 6b) is accepted by many test subjects (Asudeh and Keller 2001; Runner et al. 2002), indicating that core and exempt binding may not be mutually exclusive in this context.^[3]

Zribi-Hertz (1989) provides additional examples where reflexives are long-distance bound across intervening c-commanding subjects. In (7a) below, himself can be interpreted as referring to the matrix subject John even though there is a closer intervening subject, Paul, in the embedded finite clause that should force the anaphor to be bound locally under core binding. Interestingly, while long-distance binding of reflexives is possible in (6) and (7a), Zribi-Hertz (1989) notes that it is disallowed in (7b).

According to Zribi-Hertz (1989), the contrast between (7a) and (7b) is due to emphasis. An anaphor can bypass syntactic BT and be licensed as an exempt anaphor if such binding is motivated in discourse by factors such as emphasis.

Data like (6b) and (7a) lead us to question the assumption held in standard AS-BT that core and exempt binding are always in complementary distribution. Therefore, one of the goals of the present paper is to investigate this question further based on data from Korean.

A second additional premise that is adopted in standard AS-BT concerns the determination of the domain of core binding. As we have seen, the Binding Domain is taken to be determined by the presence of a superior co-argument. An anaphor without a superior co-argument occurs in a position exempt from syntactic BT. However, Charnavel and Sportiche (2016) argue that not all anaphors in exempt positions are licensed as exempt anaphors, which entails that the domain where core binding applies must be determined differently than in standard AS-BT. They make their argument on the basis of the distribution of inanimate anaphors in French.

If there is any consensus on what an exempt anaphor is, it is that it must be animate. This is because the antecedents of exempt anaphors must be logophoric centers (Sells 1987), animate/sentient beings whose thought, speech and internal attitudes can be assessed. Under the standard AS-BT view of how core versus exempt binding is demarcated, an inanimate anaphor is predicted not to be able to occur in an exempt position. This is because in that position it needs to be licensed as an exempt anaphor, but being inanimate, it cannot. Charnavel and Sportiche (2016) argue that this prediction is not supported, and that inanimate anaphors like elle-même ‘itself’ can occur in exempt positions in French, as seen in (9).

Elle-même is the sole argument of the predicate, sur ‘about’ in (8) (or of the noun livre, ‘book’, if the preposition is not counted). Hence, it is in an exempt position. AS-BT incorrectly predicts (8) to be ill-formed because the anaphor can only be licensed as an exempt anaphor, but it cannot, being inanimate. However, (8) is well-formed, which can only mean that the inanimate anaphor is licensed as a core anaphor. This in turn implies that the domain for core binding must be defined differently than in standard AS-BT. In (8), the core Binding Domain must be the entire clause, as assumed in BT and its successors.

Therefore, Charnavel and Sportiche (2016) posit that the domain in which core anaphors in French (best exemplified by inanimate anaphors) must be bound is defined by the SSC (Specified Subject Condition) and the TSC (Tensed-S Condition), as shown below.

The Binding Domain for elle-même in (9a) is the entire clause, where there is a c-commanding antecedent la terre ‘the earth’. The Binding Domain for elle-même in (9b) is the finite clause complement of the noun fait. As the anaphor is not bound in the Binding Domain, it is ungrammatical (due to a violation of the TSC). In (9c), the finite clause complement is again the Binding Domain for the anaphor. The sentence is out because the antecedent is outside this domain (incurring violations of both SSC and TSC). By contrast, under AS-BT, all sentences in (9) (as well as 8) are predicted to be ill-formed because elle-même occupies an exempt position but cannot be licensed as an exempt anaphor.

Though Charnavel and Sportiche (2016) focus on the issue of the determination of the core Binding Domain (in French), they also take a specific position on the first premise of standard AS-BT which holds that core and exempt binding are in complementary distribution. Recall that under AS-BT the anaphor himself in (10a) occupies an exempt position and must be licensed as an exempt anaphor since core and exempt binding are deemed to be in non-overlapping distribution.

In Charnavel and Sportiche’s (2016) approach, the Binding Domain for the anaphor qua a core anaphor is the clause and himself is predicted to be licensed as a core anaphor. However, they also assume that nothing prevents it from being licensed as an exempt anaphor as well, because they assume that the domain of core binding and exempt binding are not mutually exclusive. To the extent that the strict/sloppy reading can serve as a useful diagnostic for core versus exempt binding, the possibility of strict reading in (10a) compared to (10b) suggests that himself in (10a) can indeed be an exempt anaphor. But the fact that the sloppy (bound variable) reading is also possible would mean that core binding is still an option. And though Charnavel and Sportiche (2016) did not discuss binding of anaphors across possessors (6b), their system would predict that such binding is possible, because exempt binding (across the c-commanding possessor) is not ruled out just because core binding is possible in the same domain.^[4]

In sum, there are two issues under debate within the overall approach to anaphor binding that posits a distinction between core/syntactic binding and exempt/logophoric binding. The first has to do with the relationship between core and exempt binding: are they in complementary distribution or can they overlap? The second concerns the determination of the domain of core/syntactic binding: is it the domain of argument structure or is the characterization of the domain closer to the view of Binding Domain held in BT?

Yet another question that may be raised about exempt binding is the following: since exempt binding has been investigated mostly in languages with a simple anaphor inventory consisting of local anaphors, how widespread is it? In particular, since one of the behavioral traits of exempt binding is long-distance binding, do languages with a richer inventory of anaphors—local and long-distance—exploit the core-exempt distinction in anaphor binding, or do the long-distance anaphors in the language take over the functional space occupied by exempt anaphors in other languages with a simpler anaphor inventory, especially since it has been found that many long-distance anaphors are subject to logophoric constraints? There aren’t many studies on the core-exempt distinction in languages with both local and long-distance anaphors. Huang and Liu (2001) investigated this question with respect to Mandarin Chinese, but their study focused on the long-distance anaphor ziji, rather than the local anaphor pronoun-ziji. Their conclusion was that locally bound ziji is a core anaphor, while LD-bound ziji behaves as an exempt anaphor, which implies that ziji may not be a true long-distance anaphor, contrary to widely held assumptions. Pollard and Xue (2001) challenge this conclusion, based on LD-bound ziji that is not subject to any special logophoric conditions. To the best of our knowledge, Kim and Yoon (2009) is the only study of a language (Korean) with a rich inventory of anaphors in which the core-exempt distinction was investigated for local anaphors. It is to this study that we now turn.

2.2 Exempt anaphors in Korean

Korean has a rich inventory of anaphors. They can be used interchangeably in many contexts but each anaphor has its own properties (Cole et al. 1990; Kang 1998; Yoon 1989). Among them, the morphologically complex anaphor caki-casin is generally assumed to be a local anaphor due to its preference for local antecedents while the monomorphemic anaphor caki functions as a long-distance anaphor (LDA) because it prefers long-distance antecedents to local ones.

The question of whether the core versus exempt distinction exists in Korean, a language with both local and LDA, was not investigated until Kim and Yoon (2009). They conducted an experimental investigation to see if native speakers of Korean accept a local anaphor (caki-casin) that appears in a context where it cannot be bound locally and is forced to be LD-bound. They also examined whether the LD-bound caki-casin displays the interpretive properties attributed to exempt binding, such as sensitivity to logophoricity and allowing a greater degree of strict readings in VP ellipsis/proform contexts. The results showed that caki-casin can be coerced into LD-binding when core/local binding is not an option and that LD-bound caki-casin behaves like exempt anaphors in terms of interpretive properties, which led Kim and Yoon (2009) to conclude that the core/exempt binding distinction is independent of the existence of long-distance anaphors in a language.

However, there are shortcomings with their study which weaken their overall conclusions and call for a different study that does not suffer from these deficiencies. In order to evaluate Kim and Yoon (2009), assumptions about what constitutes the Binding Domain for core anaphors and how core and exempt binding relate to each other need to be clarified. Since Kim and Yoon (2009) did not focus explicitly on these questions, but instead on whether exempt binding of local anaphors is attested in a language with a diverse anaphor inventory, we shall reconstruct their positions in what follows, in light of the importance of these questions to the understanding of core vs. exempt binding. We turn first to the question of the determination of the Binding Domain for core binding.

AS-BT posits that a local anaphor in an exempt position can only be licensed as an exempt anaphor. However, in Korean, all kinds of anaphors (local caki-casin, neutral casin,^[6] long-distance caki) can occur in what would be considered an exempt position in AS-BT, the subject position of embedded clauses, in violation of the TSC. This is shown in (12) below.

Because all anaphors can violate the TSC, Kim and Yoon (2009) assumed that TSC-violating anaphors are not exempt but core anaphors.^[7]

With regard to the role of the Specified Subject Condition (SSC) in defining the core binding domain for local anaphors, things are not straightforward. This is so because when we place the local anaphor caki-casin in a configuration that violates the SSC, the acceptability of the structure seems to depend on the animacy of the SSC-inducing local NP, as shown below.

That is, when the SSC-inducing local NP is a possible antecedent of caki-casin (being an animate NP), the LD-binding of the anaphor appears to be blocked (cf. [13a]). However, when the local NP is not a possible antecedent of the anaphor (being an inanimate, as in [13b]), LD-binding of caki-casin appears to be sanctioned. The main objective of Kim and Yoon (2009) was to investigate sentences like (13b), which had not been noticed before in the literature, in order to determine if sentences like (13b) are indeed accepted by native speakers of Korean, and if they are, whether the LD-bound caki-casin behaves in a manner typical of core or exempt anaphors. Their experimental investigation confirmed that native speakers accepted sentences like (13b) with LD-bound caki-casin, and that the anaphor behaves in a manner typical of exempt, not core, anaphors when it is LD-bound. Kim and Yoon (2009) reasoned that caki-casin can be LD-bound in (13b) but not (13a), because while local/core binding is possible in (13a), it is impossible in principle in (13b), so that the anaphor is allowed to be licensed via LD/exempt binding.

Kim and Yoon (2009) did not fully explore the consequences of their interpretation of the contrast between (13a) and (13b) for currently debated questions in core and exempt binding, since these questions were not their primary concern. However, we can deduce that their positions on these issues are different from those held in standard AS-based approaches to exempt binding. For one, the fact that they assumed that core anaphors can violate TSC goes against the view held in AS-based approaches that requires anaphors in exempt positions (such as the TSC-violating position) to be exempt anaphors. Rather, their position is consistent with the position advanced later in Charnavel and Sportiche (2016) which holds that core anaphors may occur in exempt positions. With regard to the SSC, the conclusion that Kim and Yoon (2009) reached is consistent with the interpretation where the SSC configuration defines the Binding Domain for core anaphors like caki-casin, but when core binding is not an option (due to the lack of a potential antecedent), the anaphor is allowed to exploit exempt binding (which yields LD-binding, as in 13b). If exempt binding were impossible in core binding domains, (13b) would be judged ungrammatical. In other words, they seem to hold the view that the domains of core and exempt binding are not mutually exclusive, unlike standard AS-based approaches to exempt binding.

Now, if the reason that (13b) is acceptable is because exempt binding can apply in core binding domains, and it is the lack of availability of core binding in the local domain that allows the anaphor contained in the core binding domain to exploit LD/exempt binding, it is predicted that even in (13a), where the anaphor can be licensed under core/local binding by an animate local antecedent, LD/exempt binding should be possible, even though the acceptability of LD/exempt binding in (13a) may not be as robust as in (13b) because that core binding is possible in the embedded clause. Unfortunately, Kim and Yoon (2009) did not investigate sentences like (13a) in their study, but simply assumed that they were ungrammatical. This is a major omission, and we need to examine sentences like (13a) systematically to determine whether LD/exempt binding of caki-casin is possible even when local/core binding remains an option.

The reason Kim and Yoon (2009) did not investigate sentences such as (13a) appears to related to their choice of baseline sentences illustrating local/core binding. Sentences like (13a) with the anaphor bound by the local subject could have been the baseline sentences illustrating local/core binding. Instead of examining them, Kim and Yoon (2009) took TSC-violating sentences where the anaphor is bound by the closest animate NP (cf. 12) to exemplify local/core binding.

However, this assumption turns out to be problematic. For one, the acceptability of a sentence like (12) does not tell us anything about whether speakers are treating the TSC-violating anaphor as core or exempt.^[8] Secondly, and more problematically, when the diagnostic properties that differentiate core from exempt binding (such as strict vs. sloppy interpretations in pro-form/ellipsis contexts) were considered, Kim and Yoon (2009) found that there was essentially no difference between the TSC-violating sentences that they took to be baseline examples of core binding and sentences they took to exemplify LD/exempt binding (cf. 13b). TSC-violating caki-casin allowed a high rate of strict readings in pro-form/ellipsis contexts, just like the SSC-violating exempt anaphors in sentences such as (13b). This result, which was replicated in Kim and Yoon (2013), implies that Kim and Yoon’s (2009) study lacked genuine baseline examples of core binding. Baseline sentences exemplifying core/local binding need to be tested. Such sentences would be like (13a), but where the anaphor is bound by the local antecedent.

Another gap in Kim and Yoon’s (2009) study is the following. According to the typology of long-distance reflexives proposed in Cole et al. (2001, 2006, genuine LDAs (bound variable/anaphor type LDAs) are different from the local anaphors turned into exempt anaphors. Kim and Yoon (2009) claimed that LD-bound caki-casin is a local anaphor turned into an exempt anaphor and not a genuine LDA based on the fact that it exhibits the interpretive properties attributed to the exempt anaphors (strict reading availability; sensitivity to logophoricity). However, as Charnavel and Sportiche (2016) reminded us, using diagnostic properties to argue for exempt status is risky, as there are few, if any, properties that categorically distinguish core and exempt anaphors. For example, while LD-bound caki-casin is sensitive to logophoricity, it has long been noted in the literature (Park 1986; Yoon 1989) that genuine LDAs like caki in Korean are also sensitive to it. Therefore, a new way of establishing that the LD-binding of caki-casin differs from that of LDAs like caki needs to be sought.

To sum up, the conclusions reached in Kim and Yoon (2009) – (i) that LD/exempt binding of local anaphors is attested in Korean, a language with a diverse inventory of anaphors, and (ii) that core and exempt binding may not be in complementary distribution – need to be re-evaluated with a properly conducted follow-up study that addresses the shortcomings of the original study. This is because their study lacked genuine examples of baseline sentences showing core/local binding, due to their mistaken identification of TSC-violating sentences as exemplifying core binding, and because they failed to investigate the prediction that LD/exempt binding should be allowed even in sentences like (13a), and not just in (13b). Finally, we need to be able to establish that LD-bound caki-casin is different from typical, bound anaphor/variable type LDAs in its properties.

3.1 Research questions

The theoretically relevant questions that motivate this study are the following:

1. Can the domains of core and exempt binding overlap in Korean?

2. Is the LD-binding of exempt local anaphors different from that of genuine LDAs?

The first question was addressed in Kim and Yoon (2009), but we are investigating it with a study that does not have the design flaws of that work. The second was not addressed in Kim and Yoon (2009), but is important to answer if we are to establish that LD-bound caki-casin is a local anaphor that has been coerced into LD usage, rather than a bound variable/anaphor type LDA.

The following are the specific research questions whose answers will have implications for the answers to the questions just posed.

The first question attempts to fill a gap in the experimental design of Kim and Yoon (2009). If Kim and Yoon (2009) are correct in assuming that core and exempt binding are not in complementary distribution, we predict that caki-casin can be LD-bound even in the presence of an animate local NP, and not just with inanimate local NPs. In addition, since Kim and Yoon’s (2009) study lacked baseline examples of local/core binding because of mistaken assumptions about the status of the TSC in Korean, we compared the behavior of LD-bound caki-casin with that of the baseline sentences where it was bound by a local antecedent.

Recall that the evidence Kim and Yoon (2009) provided in support of the exempt anaphor status of LD-bound caki-casin was based largely on interpretive properties. Given that these properties are not unique to exempt anaphors, we need to look for a different way in which we can determine if LD-bound caki-casin is an exempt anaphor. This is what the second question addresses. We seek to answer this question by comparing the long-distance binding of caki-casin and caki. If LD-bound caki-casin is an exempt anaphor, and if the licensing of exempt binding involves competition with local/core binding, we expect to find reflexes of such competition. In particular, we expect the possibility of local/core binding to have an effect on the LD-binding of caki-casin. As a genuine LDA, caki will be different, since the LD-binding of caki does not involve competition with local/core binding. Since local/core binding is possible only if an NP that locally c-commands the anaphor is animate, we expect to see the effect of what we shall call animacy intervention in the LD-binding of caki-casin but not in that of caki. On the contrary, if there is no difference between the two, then we would have to revisit the conclusion reached in Kim and Yoon (2009) that LD-bound caki-casin is an exempt anaphor.

3.2 Methods