Section 19.1.1.2 has shown that indefinite noun phrases are generally used to introduce new entities (e.g. persons or things) into a given domain of discourse, as illustrated in (9a), while definite noun phrases are normally used to refer to entities that are already part of that domain, as in (9b). Proper nouns (such as Jan or Napoleon Bonaparte) are probably a special case in that their reference is determined in a more rigid way by the common ground of a larger speech community.

In order to enable the language user to keep track of the discourse entities, we can assume that each entity in the discourse domain is assigned a unique referential index. This index can be seen as a file card on which all available information about the entity in question is recorded; cf. Heim (1982/1983). By using an indefinite noun phrase such as a man, we introduce a new card and add it to the file. Proper nouns refer to file cards that can exist independently of discourse, while definite noun phrases generally refer to discourse-bound file cards. However, the ambiguity of example (10a) shows that in the latter case it is not always clear beforehand (i.e. out of context) which card should be selected; the definite noun phrase de sukkelthe dope is used as an epithet and can be used to attribute the property of being a dope to either Jan or Peter. The ambiguity of (10a) thus suggests that definite descriptions can in principle be randomly assigned a referential index, i.e. randomly associated with an existing file card. Assuming that the referential indices i and j have been assigned to the proper nouns Jan and Peter respectively, the interpretation of the epithet de sukkel depends on whether it is assigned to i or j: in the former case it is interpreted as coreferential with the noun phrase Jan, as in (10b), and in the latter case with Peter, as in (10c). Of course, de sukkel could also be assigned some other index, but for the sake of simplicity we ignore this possibility here.

In principle, there are no semantic restrictions on the indices that a language user can assign to definite noun phrases. However, this does not mean that anything goes, since there seem to be several restrictions on the referential relations between noun phrases. For our present purpose, the most important restriction is binding condition C: although the epithet de sukkel can be coreferential with noun phrases that occur in an earlier (here coordinated) sentence, as in (10a), it cannot be coreferential with a c-commanding noun phrase. This is illustrated by (11): the epithet cannot be coreferential with the c-commanding proper noun Jan, regardless of whether the latter is part of the former’s anaphoric domain (i.e. its minimal clause), but must refer to some other person in the discourse domain.

It remains to be seen whether the prohibition of coreferentiality illustrated in (11) is syntactic in nature. For example, a pragmatic account might be more appropriate for (11a), since the intended message can also be expressed by the reflexive zichzelf in (12a). The reason for this is that the use of zichzelf would be preferred by Grice’s (1975/1989) maxim of manner, according to which ambiguity should be avoided, because binding condition A excludes some of the readings available for accidental coreference. A similar pragmatic reasoning can be used to explain the fact that coreference is blocked in (11b): the referential pronoun in (12b) may be preferred because it imposes stricter restrictions on its antecedent due to its person, number, and gender features. We return to this issue briefly in Section 23.3, sub IB.

If coreference in the examples in (11) is indeed blocked by their competitors in (12), we might just as well conclude that binding condition C is not a syntactic condition and should therefore be eliminated from binding theory; cf. Reinhart (1983). In fact, the rules regulating accidental coreference may be completely outside the reach of core syntax and part of an entirely different module of grammar. Take the examples in (13), which show that referential and demonstrative personal pronouns can also be accidentally coreferential with an antecedent in a preceding sentence. These examples also show that there are non-syntactic constraints on accidental coreference; while a referential personal pronoun can refer to either the subject or the object of the preceding sentence, the (unstressed) demonstrative can refer only to the object. We will not discuss the nature of this restriction here, but refer the reader to Section 19.2.3.2, sub IIA, where it will be argued that it is related to the information structure of the discourse.

Yet another non-syntactic restriction on coreference may be related to information packaging. Consider the short fragment in (14), which could be the beginning of a report; as before, coreference is indicated by assigning the same referential index to the relevant noun phrases while counter-indexing implies non-coreference.

The first sentence sets up the domain of discourse. Part of this domain is a meeting and, presumably, a chairman as well as some participants in this meeting (which are inferences based on our knowledge of the world; cf. Section 19.1.1.2, sub IIIC). Since the chairman is part of this domain, we can refer to this entity with a definite noun phrase. From then on, this entity can be referred to by a referential personal pronoun, as is the case in the third sentence of (14). Now consider the fragment in (15), which differs from (14) in that the referential expression de voorzitter and the pronoun have swapped places.

This text is certainly not felicitous if we interpret the pronoun in the second sentence as coreferential with the noun phrase de voorzitter (as indicated by the asterisk before the index i attached to the pronoun hij)—despite the fact that this reading would be imposed by our knowledge of the world. Since we have seen in (14) that the pronoun can appear as the subject of the third sentence, the infelicity of the coreference reading of (15) should be attributed to the fact that the full noun phrase follows the pronoun. We can account for this by postulating the following (provisional) discourse rule: referential expressions do not refer to a discourse topic that has already been unambiguously established (apart from epithets, which may provide additional information).

There are many interesting questions and problems concerning the discourse rule we have just introduced. Consider the examples in (16); the contrast in acceptability between the two examples in the coreferential reading is to be expected on the basis of what we have seen so far: coreference is allowed in (16b) both by binding condition B and by the discourse rule, while it is blocked in (16b) both as a (pragmatic) condition C effect and by the discourse rule. Note that, as always, disjoint reference of the noun phrase and the pronoun is of course possible in both cases, but this will be ignored here, as it is irrelevant for our present purpose.

Something quite surprising happens, however, when the adverbial clause is in the clause-initial position. First, consider the counterpart of (16a) in (17); these examples show that topicalization of the clause does not affect the acceptability judgment, despite the fact that it violates the discourse rule. This would be consistent with the more general observation that A'-movements such as topicalization do not affect the binding relations found in neutral subject-initial clauses because they do not affect the grammatical functions (i.e. the c-command relations) involved. One might conclude from this that the discourse rule is not operational within clauses, which would be expected on the hypothesis proposed above: binding theory applies to the output of syntax (i.e. sentences), while the discourse rule applies only to larger units.

Next, consider the counterpart of (16b) in (18), which poses a problem for the conclusion just reached in that topicalization of the adverbial clause (16b) does affect the acceptability judgment: while (16b) is unacceptable, (18) is perfectly fine; cf. Reinhart (1983:§3) for a wider range of cases.

The acceptability contrast between (16b) and (18) suggests that the discourse rule is operational within sentences and can in fact override binding condition C effects. This is to be expected, given our earlier conclusion that binding condition C is not syntactic but pragmatic in nature. We should immediately add that this does not solve all problems because we usually do not find this effect when an object clause is topicalized, as in example (19b), which is in stark contrast with (18).

There may be several solutions for the observed effects of topicalization of object and adverbial clauses on coreference. For example, it may be that arguments and adjuncts differ in that the former must be interpreted in their original argument position (so-called reconstruction, which is made possible by the copy theory of movement introduced in Chomsky 1995a), whereas adjuncts can also be interpreted in their surface position. Another possibility is to assume that adjuncts differ from arguments in that they can be base-generated in a clause-initial position, in which case the definite noun phrase de voorzitter in (18) would not be structurally c-commanded by the subject; cf. the discussion in Ernst (2002:10) on so-called framing adverbials. We will not digress further on this issue, and conclude with a final problem for the discourse rule under discussion. Consider the examples in (20). It seems that coreference is allowed in (20) even though the pronoun het precedes the referential noun phrase zijn boek. It may be that the discourse rule is overruled by the fact that we are dealing with a kind of semantic “subordination” in the sense that (20a) is more or less equivalent to (20b), with a topicalized concessive clause.

What we can take away from the discussion in this subsection is that accidental coreference is not a rule of syntax, but of some other language module devoted to the organization of the discourse. We have also seen that binding condition C, which prohibits the binding of referential expressions, is probably not a rule of syntax. Note, however, that we will retain the notion of binding condition C effect as a descriptive term for referring to the relevant facts, since they are ubiquitous in the output of syntax and continue to play an important role in syntactic argumentation.

Subsection A dealt with accidental coreference; the examples in (13) have shown that referential personal pronouns can be accidentally coreferential with a noun phrase in the preceding sentence. However, they can also enter into the syntactic relation of referential dependency which we have called binding and which will henceforth be indicated by italics, as in (21). Recall from Section 19.2.1.1. sub IIC, that bound pronouns are usually weak (i.e. phonetically reduced), although this is often not reflected in the orthography. This is especially true for the masculine subject pronoun hijhe, which will play a prominent role in what follows; the orthographic form hij is used for both the strong form, pronounced [hεi], and the weak form, pronounced [i].

Note in passing that English prefers to use the pronoun they in contexts like (21b) when the antecedent is everyone; the pronoun they is then used deictically to refer to the contextually determined group as a whole; cf. Partee (1978) and Higgenbotham (1980). The deictic pronoun ze can also occasionally be used in this context in Dutch, although the use of the pronoun hij as a bound variable seems to be preferred.

We will now look more closely at the examples in (21). It could perhaps be argued that in (21a) we are also dealing with accidental coreference: the noun phrase Jan and the referential pronoun hij simply happen to refer to the same entity, viz. “Jan”. There are several reasons not to adopt this view. A first reason is that we have seen that accidental coreference between the two definite descriptions is not possible in examples like (22a). Since there is no a priori reason to assume that accidental coreference would be any different in the case of referential pronouns, it seems reasonable to reject the analysis of (22b) as accidental coreference. Since this conclusion carries over to (21a), we will give this example a binding analysis similar to (21b&c).

Example (21b) cannot be a case of accidental coreference for the simple reason that the quantified noun phrase iedereen is not referential in nature: it does not refer to any particular entity (or set of entities) in the domain of discourse, but rather acts as a universal operator in the sense of predicate calculus, and the referential pronoun seems to behave as a logical variable bound by it; cf. the semantic representation in (23a). Something similar holds for (21c), where we have an interrogative pronoun, so that its interpretation is as given in (23b) with the interrogative operator ?x. Thus, the referential pronoun is not coreferential with the quantified or interrogative pronoun, but functions as a variable that is bound by it; cf. Section 19.2.1.1. sub IIC, for further discussion. For convenience, we will use here the formal expression commonly used in generative grammar: “for all/which x, x a person, holds that ...”

Extending the analysis of referential pronouns as bound variables to cases such as (21a) requires that definite expressions such as de jongen should be analyzed as quantified expressions. This is the core intuition underlying the generalized quantifier theory, which was also adopted in our discussion of the meaning of the definite article in Section 19.1.1. Of course, the same should hold for proper nouns, which may be more controversial, but we will not elaborate on that here.

That it is important to keep the notions of accidental coreference and binding carefully apart is also clear from the examples in (24), which differ from those in (21) in that the intended antecedents are embedded in the subject. Since they do not c-command the referential pronoun from this position, we correctly predict that the examples in (24b&c) are unacceptable under the intended interpretation. In (24a) the antecedent is the proper noun Jan, so we can account for the acceptability of this example by saying that we are dealing with accidental coreference. The distinction between coreference and binding thus provides a plausible explanation for the difference in acceptability between (24a) and (24b&c).

Note that the examples in (24) are all acceptable if the whole subject [NP x’s vader] is interpreted as the antecedent of the referential pronoun. This means that a sentence like (24a) is ambiguous in the sense that both the subject noun phrase Jans vader and the embedded noun phrase Jan can function as the antecedent of the referential pronoun hij. The former interpretation can be regarded as the unmarked option, since it is available in virtually any context, whereas the latter interpretation is only available when Jan is an established discourse topic. Note also that the judgments on (25) are the same as those on (24), which are less telling because the noun phrases de vader van iedereen and de vader van wie are decidedly odd in any context (apart from the echo interpretation of (25c)).

That a quantified expression can only be the antecedent of a referential pronoun if it correctly binds it implies that this relation can only be established within the confines of a single (complex) sentence: the referential pronoun hij can be interpreted as a variable bound by an operator in (26a), while such an interpretation is excluded in (26b). Note, however, that some sort of coreference can be expressed by using the plural referential pronoun zijthey in (26c): this pronoun does not act as a bound variable, but refers to the contextually determined group of people as a whole. We represent this interpretation by associating both noun phrases (iedereen and zij) with the same set of referential indices {i,j,...,n}.

Binding condition A in (8a) prohibits anaphors from being accidentally coreferential with another noun phrase: they must be syntactically bound by an antecedent. That binding is obligatory is illustrated for the third-person reflexive zichzelf in the (a)-examples in (27): it can only occur if a third-person antecedent is syntactically present. Similarly, the (b)-examples show that the reciprocal elkaar normally requires the presence of a plural antecedent; if no suitable plural antecedent is present, as in the primed example, the result is unacceptable.

That binding is involved is also clear from the fact that anaphors do not only take referential noun phrases as antecedents, but also quantified and interrogative noun phrases: since such noun phrases are not referential in nature, accidental coreference cannot be involved in (28b-c); we are dealing with bound variable readings.

Accidental coreference cannot be involved in (28), which is also clear from the examples in (29): the intended antecedents have the desired person and number features but do not c-command the anaphor, which results in unacceptability.

The examples in (30) show that definite noun phrases and referential personal pronouns can take a split antecedent from the preceding discourse, which we have indicated by placing the relevant referential indices within accolades.

A reciprocal pronoun needs a plural antecedent. The examples in (31a-b) with the ditransitive verb voorstellento introduce show that the antecedent of the reciprocal indirect object can be either the subject or the direct object, which is indicated by italics: (31a) means that Jan introduced some guests to Peter and that Peter introduced some other guests to Jan, while example (31b) means that Jan and Peter had the duty of introducing the guests to the other guests. What is crucially not possible is that the antecedent is split, as shown by the unacceptability of (31c).

This suggests that binding differs from accidental coreference in that it imposes a uniqueness restriction on the antecedent: the antecedent must be a single constituent. This is an important finding, because Koster (1987:§1.2) presents uniqueness of the antecedent as one of the core ingredients of syntactic relations in general. That binding, but not coreference, is subject to the uniqueness constraint thus provides important support for the conclusion in subsection A that only binding is syntactic in nature: accidental coreference is regulated by some other module of the grammar.

There are apparent counterexamples to the uniqueness restriction on the antecedent but these appear mainly in a well-defined syntactic context, namely in constructions with an infinitival clause such as (32).

The assumption that referential dependency is established directly is problematic for the simple reason that the presumed split antecedent is not part of the minimal clause containing the reciprocal; if such infinitival clauses are like finite clauses in that they can constitute a local domain in the sense of classical binding theory, this would lead to a violation of binding condition A. To solve this problem, it is often assumed that the interpretation of the reciprocal is mediated by a phonetically empty noun phrase, called PRO, which acts as the subject of the infinitival clause. According to such an analysis, the reciprocal is bound by PRO, which in turn is interpreted as coreferential with both the subject and the object of the finite clause. This analysis is represented schematically in (33): elkaar is bound by PRO, as indicated by the use of italics; PRO in turn is interpreted as coreferential with the noun phrases Jan and Peter, as indicated by the set of referential indices {i,j} assigned to it. Now the reciprocal has a plural antecedent (viz. PRO) and its plurality requirement is satisfied. We refer the reader to Section V5.2 for a detailed discussion of the interpretation of the implied subject PRO of (om +) te-infinitivals.

One potential problem with this analysis is that PRO is c-commanded by the constituent parts of the split antecedent, and therefore might be expected to be bound by them. However, referential pronouns also allow split antecedents in similar configurations.

The fact that accidental coreference is allowed in (33) and (34) should therefore be seen as a direct consequence of the uniqueness restriction on binding.