chapter 34 PRONOMINAL ANAPHORA karen van hoek 1. Introduction One of the fundamental beliefs guiding the cognitive linguistic enterprise is that grammar is meaningful. Rather than being conceived as a separate module or sub- theory of language, grammar is viewed as the conventionalized patterns by which complex meanings are expressed. Grammatical phenomena should therefore be fully characterizable in terms of meaningful linguistic units, rather than requiring that we posit special theoretical constructs for their explication. The theory of Cognitive Grammar developed by Langacker (1987, 1991), ar- guably the most fully developed grammatical theory within the field of Cognitive Linguistics, assumes that there are in fact only three kinds of linguistic units: se- mantic, phonological, and symbolic, where a symbolic unit is a bipolar unit con- sisting of a semantic unit paired with a phonological unit, similar to a Saussurean ‘‘sign.’’ Cognitive Grammar takes the position that syntactic phenomena can be fully characterized using only these three kinds of units, without requiring a special vocabulary or special constructs for the description of syntax. An ideal test case for these claims is the classic problem of pronominal anaph- ora: the principles governing the circumstances under which a pronoun (such as he, she, it) and a name or descriptive noun phrase (Sally, the green car, that guy over there) can be interpreted as referring to the same person or thing. The principles of pronominal coreference have been the focus of intensive study in Generative Linguistics since the late 1960s. The most widely accepted models within the generative tradition are based on the notion of c-command, a theoretical construct which does not satisfy the criteria for inclusion in a cognitive linguistic approach, for reasons explained below. The mystery of pronominal anaphora can be illustrated quite simply. The sen- tences in (1a) and (1b) allow for an interpretation in which the pronoun corefers with the name. (I use the term corefer to mean ‘picks out the conception of the same person or thing’, without any implication that there must be a real-world referent with which the nominal is associated.) Though they differ only slightly from the first pair, the sentences in (1c) and (1d) require a different interpretation, in which the pronoun refers to someone else. Italics are used to indicate coreference; the as- terisk indicates that the sentence is unacceptable under the relevant reading (under a noncoreferential interpretation, each of the sentences is perfectly fine). (1) a. Near him, Luke saw a skunk. b. His mother says John is a wonderful human being. c. *Near Luke, he saw a skunk. d. *He says John’s mother is a wonderful human being. The common-sense idea that the name should be mentioned before a pronoun can be used to ‘‘refer back’’ will obviously not explain these facts. In (1a) and (1b), the pronoun precedes the name, and in ( 1 c) it does not. Observations such as these have led generativists to develop explanations based on constructs of autonomous syntax, such as c-command. Greatly simplified, the c-command analysis (Reinhart 1983) states that certain geometric configurations of nominals—as defined on syntactic tree structures—rule out coreference. If the first branching node above the pronoun in the tree structure also dominates the full noun phrase (the name or descriptive phrase), then the pronoun is said to ‘‘c- command’’ the full noun phrase, and coreference is ruled out. As illustrated in figure 34.1—a highly simplified tree structure for (1c)—the branching node S dominating the pronoun he also dominates the prepositional phrase near Luke. Coreference between the pronoun and name is therefore ruled out. From a cognitive linguistic standpoint, there are numerous problems with the c-command approach. The theory of Cognitive Grammar takes as one of its guid- ing principles the Content Requirement, which effectively rules out any analysis analogous to c-command. (2) The Content Requirement The only structures permitted in the grammar of a language are (1) phonological, semantic, or symbolic structures that actually occur in linguistic expressions; (2) schemas for such structures; and (3) categorizing relationships involving the elements in (1) and (2). (Langacker 1987: 488) The Content Requirement essentially says that only linguistic structures which have either meaning or sound, or both, are permitted in a grammar. Schemas are pronominal anaphora 891 templates or patterns which generalize across more specific exemplars and which speakers can use as templates for the construction of new expressions. They also constitute semantic, phonological, or symbolic units, but are more schematic, that is, less detailed than individual specific expressions. As the tree structures on which the c-command relationship is defined rep- resent neither meaning nor pronunciation—they are instead purely syntactic constructs—they are not allowable in a Cognitive Grammar analysis. It should also be noted that c-command does not offer a grounded explanation for the facts—that is, it does not explain the coreference patterns in terms of more general notions, but merely stipulates that certain geometric configurations are ruled out. It also offers no account of the observable similarities between sentence-internal and cross- sentential coreference patterns, nor does it provide any way of dealing with the ef- fects of phenomena, such as point of view, that are not represented on generative syntactic tree structures. Instead, it offers a rule which governs only a small range of the relevant data and which cannot be modified to apply more generally, given that it is defined on tree structures which are posited to exist only within the ‘‘syntax module’’ of a speaker’s linguistic knowledge. The goal of a Cognitive Grammar analysis is to provide a grounded explanation of the facts: one which explicates grammatical patterns in terms of the meanings which they convey and which is built on more general properties and principles of language and, ideally, of cognition in general. We can begin by briefly looking at facets of meaning which provide the basis for an explanation. Figure 34.1. Tree structure for Near Luke, he saw a skunk 892 karen van hoek 2. Nominal Semantics A number of researchers have noted that full noun phrases and pronouns simply do not convey the same nuances, even when they refer to the same person or thing. Accessibility Theory (Givo ´ n 1989; Ariel 1990) posits that different nominal forms signal different degrees of ‘‘accessibility’’ of a referent, where accessibility means something like ‘the ease with which the conception can be brought into conscious awareness’. A full noun phrase indicates that the person or thing it refers to requires relatively more effort to access, either because the addressee is not currently think- ing about the person or because the person is entirely unfamiliar. A pronoun accesses a notion that is relatively easily retrieved, such as the conception of a person already under discussion or physically present. (There are finer gradations among the different nominal forms and crosslinguistic variation; see Givo ´ n 1989; Ariel 1990.) Accessibility Theory provides a way of grounding the coreference facts in the mental experience of the speaker (or conceptualizer, as the same principles apply to private thoughts) and addressee. The issue is thus not a matter of abstract geometric relationships between nodes and nominals, but of the mental models the speakers construct and the cues they give to indicate the status of a referent relative to the current context. The notion of accessibility can also be thought of in terms of the corollary notion of conceptual distance. Something which is more accessible is conceptu- ally closer to the speaker and addressee than something which is less accessible. This idea may be made clearer if we think about a typical discourse in terms of Langacker’s (1985) metaphor of the Stage Model. The speaker and addressee are analogous to an audience watching a play; the conceptions which the speaker places in the center of awareness are, metaphorically speaking, put on ‘‘stage’’ to be viewed by the ‘‘audience.’’ In Langacker’s terms, the audience is construed sub- jectively, meaning that they are the viewers, rather than that-which-is-viewed. The ‘‘onstage’’ conceptions are viewed objectively, as the center of attention. Some forms of reference mark a clear distinction between the offstage, sub- jective viewers and the onstage focus of attention. A full noun phrase (such as Jim or the three-legged dog) focuses the speaker and addressee’s attention on a conception and renders it fully objective. It thereby portrays the referent as relatively distant from the speaker and addressee. This can be contrasted with the way a pronoun portrays its referent. A pronoun places onstage a conception that is identifiable only through its association with an offstage participant. The pronouns I and you, for example, focus attention on conceptions identified as the discourse partici- pants themselves. This is indicated in figure 34.2a by the dotted line—indicating correspondence—between the onstage conception (represented by a circle) and the speaker (represented by the circle labeled S). The person described by the pronoun I plays a dual role as both the viewer and the object of viewing. pronominal anaphora 893 The pronoun blurs the distinction between the onstage and offstage, or ob- jective and subjective, regions, thereby portraying its referent as conceptually closer to the speaker and addressee than does a full noun phrase. This is especially clear in the case of first-person reference. Under most circumstances, it is anomalous for the speaker to refer to himself or herself by name. The name Joe Smith places the conception of Joe Smith fully onstage. If Joe Smith is actually the speaker, referring to Joe Smith by name is usually bizarre, inasmuch as it implies that he can see himself ‘from the outside’, as he would see another person. Only a shift in point of view can make sense of such usage. For this reason, reference to the speaker via a name may be used as a signal of a point of view shift. In ( 3), from an interview with Oliver North, North’s use of his own name implies that he is taking an external perspective and thus underscores the fact that he is describing ‘‘Ollie North’’ within the reporters’ conception of reality, not his own self-perception. (3) While reporters were talking about how Ollie North sodomized goats on the south lawn of the White House, or how Ollie North was selling White House china to fund the Contras. (Oliver North, interview in TV Guide, December 28, 1991) There are similar restrictions on the use of a name to refer to the addressee. Generally, using a name rather than you is permissible only when the speaker wants to get the person’s attention (since the addressee is not yet paying attention to what the speaker is saying, the addressee is presumably not considered part of the offstage ‘‘audience’’). A name may also be used to reassure the addressee that she or he has the speaker’s full attention or sympathy (as in I know how you must feel, Figure 34.2a. Semantic representation of pronouns: The pronoun I 894 karen van hoek Bob). In the latter case, the addressee’s role as coviewer of the onstage conception is de-emphasized to give prominence to the addressee’s role as object of the speaker’s attention. In the absence of one of these specialized contexts, reference to the speaker or addressee via a name or descriptive phrase, rather than a pronoun, will be judged as bizarre and inappropriate. Full noun phrases imply that the referent is construed as distant, ‘‘held at arm’s length’’ from the offstage audience. Except in the special circumstances described above, the speaker and addressee cannot be construed that way. The judgments in such cases will easily be as strong as the judgments of un- acceptability for (1c) and (1d), yet in such cases there is no way to explain speakers’ judgments in terms of prohibited relationships between nominals or illegal geo- metric configurations within tree structures. Although third-person pronouns such as he, she, they, and so forth do not refer to participants in the discourse, they nevertheless place onstage a conception which is identified by its relationship with a central offstage element. In this case, the offstage element is the conception of a person or thing which is physically present or has already been discussed and thus is established as part of the shared discourse world of the speaker and addressee. Thus, even a third-person pronoun portrays its referent as conceptually close to the discourse participants by being shared knowledge understood by both. This is represented in figure 34.2b, where the person the speaker and addressee are discussing is indicated by the circle labeled X, which corresponds to the onstage profile of the pronoun (indicated by the dot- ted line). In the case of third-person reference, use of a name is not as highly restricted as it is in the case of first- and second-person reference. Since the person or thing being talked about is not a coviewer of the onstage conception, there is more flexibility with regards to construing him, her, or it as part of the shared offstage world. Nevertheless, the difference in implied conceptual distance between pro- nouns and full noun phrases is robust enough that speakers frequently choose forms of reference to convey their attitude toward the person being spoken of. Speakers frequently use full names when they wish to express ridicule or disap- proval of a person, even in contexts in which a pronoun could have been used with no loss of clarity (see van Hoek 1997a: 39–42). Full names or noun phrases are also used when the speaker disagrees with something that another speaker has just said about the referent (as in Wasn’t that Tom I saw going out the back door? – No, Tom hasn’t been here in weeks) (see Fox 1987b). The conceptual distancing implied by the name signals that the speaker is holding the other person’s idea ‘‘at arm’s length’’ rather than accepting it and building on it. Conceptual distance is merely a different way of thinking about accessibility. If a conception is construed as highly accessible, it means that it is known to both speaker and addressee, that it is part of the shared world of things which both are aware of and can easily call into the center of attention. It is therefore conceptu- ally closer than something which is less accessible and thus farther away from the shared offstage world. The twin concepts of accessibility and conceptual distance pronominal anaphora 895 are simply two sides of the same coin. Both ways of thinking about it are useful, however, for they help us to understand the ways that speakers actually use the distinctions between different nominal forms to convey nuances. In addition to its role in conveying nuances of the speaker’s attitude, the ac- cessibility distinction between pronouns and noun phrases is the basis for the pat- terns of acceptable and anomalous coreference. A full noun phrase is appropriately used only in a context in which its referent is not highly accessible. If a pronoun is used, thereby establishing that the referent is highly accessible, a full noun phrase placed in the immediate context defined by the pronoun would be anomalous. The juxtaposition of the pronoun and full noun phrase would result in contradictory signals concerning the accessibility of the referent; one way for speakers to resolve the conflict is to assume that the pronoun must refer to someone else. A simple example is given in (4). (4)*He loves John’s mother. The pronoun he indicates that its referent is accessible; the name indicates the opposite. If speakers are required to read the two as coreferential, they judge the sentence to be anomalous (i.e., unacceptable). Thinking of it in terms of con- ceptual distance, the placement of the pronoun and name here is roughly equiv- alent to portraying the person as very close to the speaker and addressee (via the pronoun) and then portraying the same person as simultaneously distant (via the name). The two views of the referent are incoherent. Note that this is not because an autonomous principle of pronominal distri- bution, such as c-command, has been violated. The judgment of anomaly comes Figure 34.2b. Semantic representation of pronouns: A third-person pronoun I 896 karen van hoek directly from the meanings of the pronoun and the name and the way that they are juxtaposed in the sentence. No additional principles are needed. We do, however, need to know how to tell which nominal is construed as part of the context within which the other one must be interpreted, or in other words, how do we know which vantage point to ‘‘look from’’ in order to decide whether the views of the referent are congruent. The examples in (1) made it clear that linear order alone will not be sufficient. In fact, while linear order does play a role, its contribution is relatively weak compared to the more central factors described in the next section. 3. Conceptual Reference Points The relevant notion of context is defined in terms of the notion of a conceptual reference point, a notion first introduced in Langacker’s (1991) analysis of pos- sessive constructions. A reference point is any element which is taken as a starting point from which to access other conceptions. It is similar to an abstract point of view. An element may be selected as a reference point because it is especially prominent (the relevant notions of prominence are defined below) or because it comes before other elements in the linear string. The conceptions which are accessed from the perspective of the reference point and construed in relationship to it are said to be in its dominion. Relative to the elements in its dominion, the reference point functions as a central but sub- jectively construed part of the background: it is a key part of the conceptual ‘‘lens’’ through which the conceptualizer ‘‘views’’ the dominion. It is not itself in the center of attention. The reference point/dominion configuration thus has a dynamic quality to it: On one level, the reference point is selected because it stands out as prominent relative to the surrounding context (even if only by virtue of coming earlier in the phonological string). Within the dominion, however, the reference point ‘‘fades into the background,’’ exerting a critical influence on the construal of the conceptions in the center of attention but not itself occupying the center of attention. A sentence or stretch of discourse may have any number of reference point/dominion configurations, such that the conceptualizer—the speaker or addressee—mentally moves through the conceptions, shifting attention from reference point to dominion and on to other reference points and their associ- ated dominions. Common configurations of reference points and dominions may become established as conventional, entrenched facets of the meaning of certain grammatical patterns. For example, the status of the subject of a clause as a reference point (relative to the predicate) is no doubt entrenched as part of the subject- predicate relationship. pronominal anaphora 897 The principles governing coreference can thus be stated very simply: a. A pronoun must appear in the dominion of a corresponding reference point (i.e., a conception of the person or thing the pronoun refers to). This is simply part of the meaning of a pronoun—correspondence between the conception it places onstage and an offstage conception which is highly accessible, i.e., functioning as a reference point. b. A full nominal—a name or descriptive phrase—cannot appear in the dominion of a corresponding reference point, as this would result in a semantic conflict between the meaning of a full nominal (which includes low accessibility and conceptual distance from the referent) and the con- text in which the full nominal appears. Note that these are not special principles which must be independently stated as abstract grammatical rules; rather, they are simply part of the meanings of nouns and pronouns. 3.1. Factors Determining Reference Point Selection The factors determining reference point selection are intuitively plausible: an ele- ment tends to be selected as a reference point if it is prominent and/or if it comes earlier in the linear string. That is, speakers tend to begin with those elements which stand out or which they encounter earlier in the string, and they interpret other parts of the phrase, sentence, or discourse within the contexts set up by those prominent or prior elements. Once a reference point has been selected, its do- minion does not extend forever, but generally encompasses only those conceptions which are felt to be connected with it, in a sense to be made clearer below. Thus, there are three facets of the reference point/dominion configuration which we need to explore in more detail: a. Prominence. The most important factor in reference point selection is prominence. Cognitive Grammar posits two different dimensions or facets of prominence: the profile/base distinction and Figure/Ground alignment. Both of these play important roles in determining reference point orga- nization. b. Linear Order. Linear precedence is a kind of prominence in its own right, but here I treat it as separate from the notions of prominence mentioned above, as linear order seems to be a weaker factor. Its effects are most visible only when other factors do not clearly determine a particular construal vis- a ` -vis reference point selection and the extent of a dominion. c. Connectivity. The dominion of a reference point extends to include those elements with which it is conceptually connected, as for example when there is a verb or preposition describing a relationship between the ele- ments. When there is no such overt connectivity—as for example when two 898 karen van hoek nominals are contained in the same paragraph, or even in the same sen- tence, but with no overtly specified connection between them—speakers tend to construe the two nominals as relatively independent from one another and may therefore not assume that one nominal must be in the dominion of the other. 3.2. Prominence The notions of prominence which are of central importance in the reference point model were not developed specially to explain the pronominal anaphora facts; rather, they are long-established constructs of Cognitive Grammar (see Langacker 1987, 1991, 1998; this volume, chapter 17). These are Figure/Ground asymmetry and the profile/base distinction. Figure/Ground Asymmetry In keeping with its central goal of grounding linguistic phenomena in more widely attested cognitive facilities, Cognitive Grammar characterizes grammatical relations in terms of the well-known distinction between Figure and Ground. Every rela- tional expression—be it a verb, an adjective, or a preposition—imposes an asym- metric construal on the participants in the relation. One of the participants stands out as the Figure, while the rest of the relational conception—including any other participants—is construed as the Ground. The distinction between above and below, for example, is a matter of Figure/ Ground alignment. Both describe a configuration involving two things in vertically oriented space. Above picks out the upper element as Figure, as in The lamp is above the table, while below picks out the lower element, as in The table is below the lamp. The technical term for the Figure is trajector, while the other participant in the relation (if there is one) is termed the landmark. These are labeled tr and lm in the diagrams, respectively. Figure 34.3. Two relations differing only in Figure/Ground alignment pronominal anaphora 899 . out the lower element, as in The table is below the lamp. The technical term for the Figure is trajector, while the other participant in the relation (if there is one) is termed the landmark. These. in the relation. One of the participants stands out as the Figure, while the rest of the relational conception—including any other participants—is construed as the Ground. The distinction between. special theoretical constructs for their explication. The theory of Cognitive Grammar developed by Langacker (1987, 1991), ar- guably the most fully developed grammatical theory within the field of Cognitive Linguistics,