Coreferentiality Considered from a Cognitive Perspective

Abstract

We propose an explanation of the observations of Leddon and Lidz that the predictions of binding theory are not always borne out by the facts. More specifically their participants did not always interpret bound pronouns in line with government and binding principles. Our analysis is based on a paper by Buffart and Jacobs where they recognized structures and substructures in languages in accordance with Focus theory. In the theory, every element in a structure, and thus an anaphor as well, is bound to it. In the absence of a reference within the main- or substructure, an anaphor may refer to an element in the related sub- or main structure. We show how preference works in case of duality of interpretations.

1. Introduction

Personal pronouns like “she” behave differently from reflexive pronouns like “herself.” A sentence like She sees the sun is fine, but a sentence like Herself sees the sun is not. A pronoun like “she” can occur independently in a sentence whereas a pronoun like “herself” cannot. Personal pronouns are also called free and reflexive pronouns are called bound pronouns. The distribution of bound pronouns is syntactically determined by the principles of binding (cf. [1]).

In this paper, we will especially consider the behavior of bound pronouns and concentrate on cases that seem exceptional from a binding-theoretical perspective. We apply Focus theory to offer an alternative account for the exceptional behavior of bound pronouns.

This paper is structured as follows. In Section 1, we will briefly sketch the syntactic view on bound pronouns and we discuss the exceptional cases reported on by Leddon and Lidz [2]. After that, Section 2 introduces the key elements of Focus theory [3, 4]. Section 3 is devoted to an application of the theory on the observed problematic cases, and, finally, Section 4 summarizes the main results of this paper.

2. Background

2.1 Coreference

Reflexive pronouns like “himself” or “herself,” in contrast to non-reflexive personal pronouns like “he” or “she,” cannot occur without an antecedent, the person to which they refer, in a sentence. Let us next consider the sentences in (1), where a coreferential interpretation is indicated by the same subscript number.

1. (1a) My brother₁ likes himself₁ very much.

2. (1b) *Himself₁ likes my brother₁ very much.

3. (1c) *John₁ thinks that Bill likes himself₁ very much.

4. (1d) John thinks that Bill₁ likes himself₁ very much.

The reason why (1a) is fine, but (1b) is unacceptable is that the reflexive pronoun must be preceded by its antecedent (1a) and cannot be followed by it as in sentence (1b). Sentence (1c) shows that the antecedent cannot be too far away from the reflexive pronoun. In (1c), the reflexive pronoun is in the subordinate clause and the antecedent is in a different clause, the main clause. In (1d), both the antecedent and the reflexive are located within the same clause, the embedded or subordinate clause. In more technical terms, (cf. among many others in [1]) the conditions on coreference can be stated as a binding relation between the antecedent and the reflexive pronoun or anaphor within the same clause. Let us simplify it here by saying that the reflexive has as its antecedent the closest possible preceding subject.

Let us next turn to the sentence in (2).

1. (2) Which pictures of himself₁ does my brother₁ like most?

   At first sight, the sentence in (2) should be unacceptable, given the fact that the reflexive pronoun is not preceded by its antecedent, but followed by it. The standard explanation in the syntactic literature is that the sentence in (2) is derived from a sentence where the reflexive pronoun is preceded by its antecedent, as in (3).

2. (3) My brother₁ likes those/which pictures of himself₁ most.

   The sentence in (2) is thus derived from a sentence like (3) by movement of the wh-phrase to the initial position as indicated in (4). In order to judge a sentence like (2) to be grammatical or acceptable, a speaker thus has to reconstruct (3) as the underlying structure of (2) and interpret the antecedent reflexive pronoun relation in the underlying form.

3. (4) [Which pictures of himself₁] my brother₁ likes [which pictures of himself₁] most?

2.2 The problem

Leddon and Lidz [2] have drawn attention to the sentences (5) and (6), which are problematic to the above view and which show conflicting results.

1. (5) Bill₁ knew which pictures of himself_1/2 John₂ liked.

2. (6) Bill₁ knew how proud of himself*_1/2 John₂ was.

In (5), “himself” can be interpreted as either referring to Bill or to John, whereas in (6) a coreferential interpretation between “Bill” and “himself” is not allowed. In (5), the moved part from the subordinate close (CP, for complementizer phrase) is an argument, the direct object (NP) of the verb “liked” and in (6), the moved part is the predicate (AP) of the verb “was.” The underlying structures, the structures that need to be reconstructed in the process of evaluating the surface sentences in (5) and (6) are given in (7) and (8), respectively.

1. (7) Bill knew [CP John liked [NP which pictures of himself].

2. (8) Bill knew [CP John was [AP how/very proud of himself].

If the relation between antecedent and reflexive pronoun is always interpreted in the underlying structure, sentence (6) should behave as sentence (7), that is a coreferential interpretation of “Bill” and “himself” should be excluded in both (5) and (6). Leddon and Lidz [2] concluded on the basis of three experiments with children and adults that both children and adults in the case of a moved predicate do reconstruct the underlying form and thus are unanimous in judging sentences like (6). When it comes to sentences like (5) they conclude that reconstruction is optional both for adults and for children, but that: “Children demonstrate a clear preference for the surface structure interpretation when licensed by their grammar.” They explain their findings by assuming different parsing mechanisms: “Specifically, it is likely that children initially assign the simplest possible parse to a given sentence, interpreting items where they are pronounced, or in other words, assign a surface structure reading.” This seems, however, not to explain why adults still allow for a surface reading when it comes to moved arguments. In the remainder of this paper, we would like to explore an alternative explanation for the difference between the two types of moved elements. In the next section, we will briefly introduce the Focus theory developed by Buffart [3].

3. Theory

Focus theory starts from the observation that human processing is sequential, at least in the first 60 milliseconds. This means that there is a sequence of “elements” to be processed. Between the elements, relations can exist. For example, in vision it can be a similar color, a similar angle, or a similar shape/figure; in music, one can think of a similar pitch distance, pitch duration, or rhythm; in language, a relation between a subject and a verb, or between an adjective and a noun and so forth. Some of these relations are relations between neighboring elements (which are called neighbor relations), and others are relations between non-neighboring elements (jump relations). The basic assumption of the theory is, that in cognition, an element can only be involved in one type of relation at the same time. If this holds for each element of the sequence to be processed, the set of relations is called an interpretation of the sequence. This does not mean that an element cannot be involved in both a jump relation and in a neighbor relation, but it means that in an interpretation it can only be involved in one of both. Thus, a sequence can have more than one interpretation. The system that handles the sequences in this way is called Focus.

In fact, the statement is a little bit more complicated, since it depends on the mode of the relation. In a sequence of visual elements, for example, one could compare their shapes or one could compare their colors; in a sentence, one can compare parts or words with respect to the grammar or with respect to the meaning; in music, it differs when comparing melodic aspects of pitches or beats. In language, the distinction between grammar and meaning is difficult since they affect each other.

The assumption that an element is involved in only one type of relation at the same time leads to a mathematical theorem (see in [3]) according to which each sequence with a length of maximal seven elements can be described by maximal two interpretations. In vision the existence of two interpretations is well known, It is called duality or complementarity. Moreover, the discussion on the number seven has a long history in psychology and psycholinguistics. In almost all publications one generally assumes that the number seven is related to the short-term memory (STM) capacity or working memory (WM). In Focus theory, it expresses the capacity of Focus. In ref. [4], it is shown that it can explain the results of STM-capacity experiments, which seem to support the idea that this capacity is four.

Systems with a Focus capacity of four elements or less qualitatively differ from those with a capacity of five, six, or seven elements. We [5] assumed that the latter systems reflect the capacities of the standard population. This implies that the structure of the communication within such a population must be based on a Focus capacity of five. For sure, there may exist communication methods, that require a minimum capacity of six or seven, but that cannot be the case for languages since language is used by everyone.

In ref. [5], we applied this theory on structure in cognition to linguistic phenomena especially syntactic embedding in sentences and phonological interacting processes. Karlsson [6, 7] and Christiansen and Charter [8] showed that to understand center-embedded sentences their depth of embeddedness is limited. Universal Grammar (UG) cannot explain these results since the depth of embedding only plays a role in center-embeddedness, but not in right- or left-embeddedness, whereas from a purely generative perspective, they are equally complex. Focus theory has a different starting point. As we argued [5], structural grammar is a consequence of the Focus-restrictions. They not only generate grammar, but they also account for the restrictions on embeddedness. The theoretical calculations cover the experimental findings. In order to explain restrictions, one often (e.g. [7]) argues that these might be due to WM or STM without being able to substantiate it by a calculation.

We will use Focus theory to explain co-referentiality since it generates grammar and correctly predicts the experimental findings on center-embedding and STM-capacity. A more extensive description can be found in the paper mentioned [5]. Since duality and the restricted capacity of five up to seven are two sides of the same coin the theory offers two ways to look at experimental data. One way is based on Focus capacity, which we applied in [5] to restrictions on embeddedness. The other way is based on the duality or complementarity of interpretations, which we applied to consonant shifts and which we will use in this paper to provide an alternative explanation for coreferentiality.

4. Application to coreference

4.1 Duality and transposition

Sentence 9b is a question sentence that can be generated from or is related to sentence 9a. Figure 1 shows this in a simplified grammar notation. In a similar way, one can generate a so-called “in situ” question sentence (9d), that is a question where the question word is not moved, from sentence (9c). In a language like French, an “in situ” question is quite normal: Le chien a chassé quel chat? In English or Dutch, a similar question requires a specific intonation pattern, and the noun “which cat” in a question like (9d) is normally transposed to the front of the sentence, and the noun “the dog” is replaced after the verb (Figure 2), so that the surface structure of this question sentence is the same as the surface structure of the question sentence (9b) generated by (9a). However, their meanings differ; that is, expression (9b) can be interpreted in two ways; “which cat” can be the subject or the object.

1. (9a) The cat chases the dog.

2. (9b) Which cat chases the dog?

3. (9c) The dog chases the cat.

4. (9d) The dog chases which cat?

Figure 1.

The cat chases the dog.

Figure 2.

The dog chases the cat.

The relative strength of the two interpretations can be influenced by context as can be seen in the question sentences (9e) and (9f). But, the duality is still there.

1. (9e) Which aggressive cat chases the dog?

2. (9f) Which cat chases the aggressive dog?

4.2 Duality and subordinate clauses

Let us have a look at transpositions in more complex sentences, that contain a main and a subordinate clause. As can be seen in (10), “himself” refers to the subject within the subordinate clause. This is in accordance with ref. [5]. A general rule of their approach implies that all elements have only internal relations, so that “himself” refers to an item of the (subordinate) clause to which “himself” belongs, provided that there is such an item. Nevertheless, in the examples in (10) “himself” always refers to the subject in the subordinate clause.

In general, the free nouns “he,” “her,” “she,” and “him” can refer to everybody, even to those, who are not mentioned. If we restrict ourselves to subjects that are mentioned, then in (10c,d) “he” in the second clause refers to James and so does “himself” since it must refer to “he”. In (10e,f), it is not clear whether the “he” in the subordinate clause refers to Bill or James. It depends to whom “He” in the main clause of the second sentence refers. If (main clause) “He” refers to James then (subclause) “he” refers to Bill and the other way around. Since Bill is already the subject in the first sentence, Bill is by preference, but not necessarily, the subject in the main clause of the second sentence. In (10 g,h), there is no context so “he” can only refer to Bill.

1. (10a) Bill knows that John is proud of himself.

2. (10b) Bill knows how proud of himself John is.

3. (10c) Bill meets James. Bill knows that he is proud of himself.

4. (10d) Bill meets James. Bill knows how proud of himself he is.

5. (10e) Bill meets James. He knows that he is proud of himself.

6. (10f) Bill meets James. He knows how proud of himself he is.

7. (10 g) Bill knows that he is proud of himself.

8. (10 h) Bill knows how proud of himself he is.

In (11), sentences are given, where “himself” refers to Bill in the main clause, even when “himself” belongs to the subordinate clause. In (11a,b), “himself” belongs to the main clause. If one interprets the subordinate clause in (11c) to begin with “the pictures of himself” then “himself” belongs to the subordinate clause. However, considering the type of sentences in (11a,b), one may assume that the subordinate clause begins with “that are.” In that case “himself” belongs to the main clause as well. In (11d,e), it belongs to the subordinate clause. However, there is no item in the subordinate clause to which “himself” can refer. So the reflexive pronoun needs to find its antecedent in the main clause. Notice that this reference precedes “himself.”

1. (11a) Bill finds these pictures of himself.

2. (11b) Bill knows these pictures of himself.

3. (11c) Bill recognizes the pictures of himself that are red.

4. (11d) Bill knows which pictures of himself are red.

5. (11e) Bill knows that those pictures of himself are red.

If we compare (11b) with (11d) and (11e), then in (11b) the object of “knows” is “pictures,” whereas in (11d) and (11e), the object of “knows” is no longer the NP, but the entire subordinate clause.

The sentences in (12) show another complexity. In accordance with the analysis of the sentences in (11), “himself” in both (12a) and (12b) can refer to Bill. This similarly holds for (12c), where “which pictures” is subject instead of object. That the application of the analysis of (11) applies here as well, is supported by the sentences (12d,e,f). As in the case of (9b), “which cat chases the dog?”, the sentences in (12b,c) have dual interpretations, since here “himself” can also refer to John. Although the difference between (12a) and (12d), between (12b) and (12e), and between (12c) and (12f) is only a name, the sentences (12b,c) show duality whereas the sentences (12e,f) have only one interpretation (see also Figures 3 and 4). Sentences (12b,c) are like the Leddon and Lidz sentence (5), and in (12b) one could say that in one interpretation “pictures” is the object of “knows,” whereas in the other interpretation, the object of “knows” is the entire subordinate clause “John likes which pictures of himself.”

1. (12a) Bill knows the pictures of himself that John likes.

2. (12b) Bill knows which pictures of himself John likes.

3. (12c) Bill knows which pictures of himself please John.

4. (12d) Bill knows the pictures of himself that Mary likes.

5. (12e) Bill knows which pictures of himself Mary likes.

6. (12f) Bill knows which pictures of himself please Mary.

Figure 3.

Reference into the main clause.

Figure 4.

Duality.

In sentence (12 g), “himself” refers to John, but in sentence (12 h), there is duality again, although “himself” belongs to the subordinate clause. If one replaces “please John” with “are red,” the reference of “himself” to Bill is clear. In all those sentences, the continuation of the sentence is uncertain up to and including “pictures of himself.” In a lot of cases, “Bill” is the only possible reference. In other words, “himself” seeks a reference that precedes it. If, in case of a subordinate clause, there is no reference in this part preceding it, a preceding element in the main clause is a candidate.

1. (12 g) Bill knows that John likes those pictures of himself.

2. (12 h) Bill knows that those pictures of himself please John.

In the case of duality, the question of preference for one of both references arises.

4.3 Preference

The role of him and her differs from the function of himself and herself. As he or she in a sentence may refer freely to nouns in a context so do him and her. Himself and herself must refer to an element in the sentence. This can be seen in the sentences in (13). In (13a) and (13b), “himself” and “him” belong to the main clause, but there is no reference possible to the subject of the main clause. Therefore himself in (13a) will refer to John in the subordinate clause. If we would add “Ann meets James” the “him” in (13b) could easily refer to James, but “himself” in (13a,c) could not.

1. (13a) You know the pictures of himself that John likes.

2. (13b) You know the pictures of him that John likes.

3. (13c) You know which pictures of himself John likes.

In (14), there are similar considerations. In (14b), “him” refers to Bill since there is no other context. But if we introduce James here, as above, by adding “Ann meets James” before the sentence, “him” will be attached to James, but “himself” in (14a) will not. Sentence (14c) is not acceptable since in principle “himself” will refer to the subject in the subordinate clause, but there is no available reference there. Since “him” is a free pronoun, it does not encounter this problem in (14d). In (14e), the dog rises its tail, and that is allowed.

1. (14a) Bill knows which pictures of himself she likes.

2. (14b) Bill knows which pictures of him she likes.

3. (14c) Bill knows how proud of himself she is.

4. (14d) Bill knows how proud of him she is.

5. (14e) Bill knows how proud of himself the dog is.

Especially in the case of duality, but also if the sentence or the context does not give a clear hold, the question of preference for some solution bubbles up. In case of duality without context information, the reading direction may play a role. The interpretation of (9b), for example, tends more to (9a) than to (9c). The influence of reading direction is not restricted to language. It is a general phenomenon. The expressions (15b,c,e) are equally preferred interpretations of the sequence in (15a) if one can lock out the influence of the reading direction [9, 10]. A left–right reading direction supports interpretation (15b) and a right–left reading direction interpretation (15c).

1. (15a) abaabaaa

2. (15b) (aba)(aba)aa

3. (15c) (ab)a(ab)(aaa)

4. (15d) a(baa)(baa)a

A language has only one given reading direction (generally left-to-right or right-to-left). But language is in the first place a spoken, not a written, means of communication. Besides context, intonation and accent may direct to the interpretation intended by the sender. This neutralizes the linearity of the surface structure.

Since the sentence part “Bill knows which pictures of himself….” can be finished in several ways, as, for example, in (11d), (12c,f), and (14a), there will be a natural tendency to attach “himself” to Bill and to interpret “pictures” as the object of “knows,” although the other interpretation with the object of “knows” as the entire subordinate clause “John likes which pictures of himself” is the only meaning allowed if only reconstruction had its way.

This is a crucial difference in the two sentences (5) and (6) above, repeated here as such.

1. (5) Bill knew which pictures of himself John liked.

2. (6) Bill knew how proud of himself John was.

The sentence part “Bill knew how proud of himself” cannot occur in isolation and cannot be finished in several ways, contrary to the “Bill knows which pictures of himself” part. In order to interpret the former part, the interpreter needs to have further information and needs to have the subject of the subclause. Only if that subject is “he,” a coreferential interpretation is allowed, whereas any other subject will exclude that.

The comparison of (16a) and (16b) brings the preferential duality phenomenon to awareness. In (16b), one experiences the necessity of a correction between himself and the subject of the subordinate clause since the author of the sentence can only mean self-portraits of Rembrandt, but not in (16a) where the more reading-supported link is between himself and Bill. The comparison of (16c) and (16d) delivers a similar phenomenon.

1. (16a) Bill knows which portraits of himself have been painted by his brother.

2. (16b) Bill knows which portraits of himself have been painted by Rembrandt.

3. (16c) Bill knows which portraits of himself have been stolen by the thief.

4. (16d) Bill knows which portraits of himself have been stolen by his brother.

The structure of the sentences in (16) is however the same in all cases:

1. (16) Bill knows [CP Rembrandt/brother painted [NP which pictures of himself].

2. (16) Bill knows [CP thief/brother stole [NP which pictures of himself].

The preferred interpretation is thus not guided by competence, based on reconstruction, but created by performance in action as high-lighted by Focus theory.

5. Discussion

In the foregoing analysis, we used the principles as formulated by ref. [5]. They argued that due to a filter with a restricted capacity in the communication between a human being and its environment, humans are forced to create substructures. The maximum capacity for languages is five. Apart from the free elements, every element is in principle bound to its substructure. Free elements make connections with elements in the context, in general through memory.

The only rule we used to explain coreference is that “….self” refers to the subject within the (sub)structure at hand. Sometimes this rule can be broken, but only when there is no fitting element within this part. Then, using a free element is often better than using an anaphor. In the case of duality, there are two solutions, which only means that two different meanings cover the same surface structure. This happens in examples (9b) and (12b,c,h). Moreover we argued that in the latter sentences there will be a preference for a solution with an attachment to the subject in the main clause, since reading the sentence until and inclusive the word “himself” there are many continuations possible which only have an attachment to an element of the main clause. Thus creating a “natural” relationship between both. A predicate at the beginning of a subordinate clause implies a transposition within this substructure and the attachment is to the subject of this subordinate clause by definition.

We formulated a more satisfactory explanation for the findings of Leddon and Lidz based on the assumption [5] that humans built substructures within a longer sequence due to the fact that sentences pass Focus, which has a maximum capacity of 5, 6, or 7 elements. In principle, these substructures are closed so that only relations within these structures can exist. Free nouns can escape this embedding. In the surface structure, duality in meaning can occur. Our analysis explains the preference in case of duality and the preference for some sentences above others, and so it explains the observations of Leddon and Lidz [2]. It looks like in language also, the competence allows more than one interpretation. In these cases, competence looks more like online interaction with the environs.

6. Conclusion

A cognitive psychological alternative explanation is proposed for unexpected exceptional linguistic behavior from a binding theory perspective.