(17)a.Inhis latest movie, Tom Cruise gets framed for murder. b. In Ben’s picture of her, Rosa is riding a horse. c. So far as she knows, Sally is well-liked. Relative to the object, process-internal modifiers behave in the same way that process-external modifiers behave relative to the subject. That is, they are auto- matically construed within the object’s dominion so long as they follow it in the linear string, as in (18a) and (18b). Unlike the subject, which is the figure for the entire clause, the object is a more local figure, prominent only in relation to elements with which it is more directly interconnected, that is, participants in the core interaction described by the verb (see van Hoek 1997a: 92–94 for a more detailed account). When process-internal modifiers are preposed, they therefore escape the object’s dominion, as in (18c) and (18d) (examples 18b and 18d are from Reinhart 1983). (18) a. *I handed him the contract outside Ralph’s office. b. *Rosa tickled him with Ben’s peacock feather. c. Outside Ralph’s office, I handed him the contract. d. Rosa tickled Ben with his peacock feather. Only one type of modifier seems to straddle the two categories: temporal modifiers which take the form of a subordinate clause. As (19) illustrates, they are construed within the dominion of the object when they follow the object. (19) a. I saw John after he came back from work. b. *I saw him after John came back from work. c. Alex called Sue when she was in Chicago. d. *Alex called her when Sue was in Chicago. As Bolinger (1979) puts it, the temporal modifiers are ‘‘captured’’ by the verb phrase when they follow the verb. However, when they are preposed, they behave as process-external modifiers, escaping even the dominion of the subject. It is therefore possible to have a name in the preposed modifier which corresponds to a pronominal subject of the main clause, as in (20). (20) a. After John came home from work, he took a shower. b. When Sue was in Chicago, she visited the museum. We can assume that it is the ambiguous status of the first clause vis-a ` -vis the second that allows it to behave as a process-external modifier. On the one hand, a temporal modifier describes the temporal setting for the main clause; on the other hand, these are separate clauses with their own tensed verbs, describing events distinct from those described by the main clause, and may therefore be construed almost as separate sentences in the discourse. In a similar vein, Matthiesen and Thompson (1988) propose that the first clause be thought of as an independent ‘‘satellite’’ describing an event which is related to the ‘‘nucleus’’ provided by the second clause. 910 karen van hoek 4.3. Discontinuity at the Discourse Level The last point on the connectivity continuum is the point at which two nominals are separated by an attentional shift, in which case neither nominal is in the dominion of the other. In that case, speakers reidentify the referent with a full nominal. The nature of the relevant discontinuity varies depending on the spoken or textual genre. Fox (1987b) points out that in popular narrative a new narrative unit begins at the point at which a character takes action. Authors typically then re- identify the character, as in Fox’s example in (21). (21) She [Ripley] did not see the massive hand reaching out for her from the concealment of deep shadow. But Jones did. He yowled. Ripley spun, found herself facing the creature. [Alien,p.267] Fox (1987a) further identifies textual unit boundaries in other genres and points out that the beginning of a new textual unit coincides with reestablishing the referent by name. Tomlin (1987) has found similar patterns in spoken narratives. The continuum of conceptual connectivity thus spans the range from con- figurations in which the coreference possibilities are rigidly determined to con- figurations at the sentential and discourse level at which there is more flexibility. Under this approach, there is no need for a distinction between ‘‘syntactic’’ anaphora facts and ‘‘discourse’’ facts with separate principles for each, as in the traditional generative c-command account (see Reinhart 1983). Rather, there is a single set of considerations—prominence (including the kind of prominence contributed by precedence in the linear string) and conceptual connectivity— determining the organization of reference points and dominions, at all levels of linguistic organization. 5. Point of View Effects Traditional generative accounts give short shrift to the effects of point of view in anaphora, for several reasons. Point of view is difficult or impossible to represent in syntactic tree structures. Moreover, the judgments of point of view effects are notoriously slippery and variable. It is thus tempting to ignore them or at least relegate them to some other domain of study, outside the realm of pronominal anaphora proper. I would argue, however, that it is misguided to dismiss point of view effects. First of all, they cannot be fully separated from the ‘‘core’’ anaphora facts. As noted in section 2, restrictions on the use of names for first- and second-person reference are essentially a matter of point of view: it is anomalous for the speaker to portray pronominal anaphora 911 himself or herself or the addressee as anything other than a coviewer of the con- ceptions placed onstage in the discourse. Moreover, third-person coreference restrictions, even in the domain of ‘‘core’’ anaphora facts, frequently involve point of view considerations as well. As noted, the complement chain is a metaphorical ‘‘line of sight’’ through the clause, and a reference point is essentially a schematic viewpoint. The only thing needed to make a reference point a full-fledged point of view is a vivid sense that one is empa- thizing with a particular person or seeing through his or her eyes. Empathy is however a matter of degree. Many (though not all) speakers report that their understanding of such ‘‘core’’ disjoint reference examples as Near Luke, he saw a skunk involves a vivid sense that they are seeing through the subject’s eyes and are disconcerted by then seeing the same person at some distance away. Others report that they are looking over the subject’s shoulder. Still others have no vivid visual imagery at all. The extent to which the subject is merely a schematic reference point or a full-fledged point of view is thus a matter of degree and varies across speakers (and no doubt across contexts, as well). However, it is true that for the anaphora facts which have traditionally been considered significant, viewpoint does not have to be taken into account if one merely wishes to make the right predictions concerning coreference. Point of view needs to be considered in such cases only if one wishes to fully explicate a native speaker’s experience of the constructions. In other cases, point of view is a critical factor which must be considered in order to explain why coreference is impossible. Such data often involve slippery or variable judgments, but this is not always the case. Sentence pair (22) provides very clear judgments concerning coreference, in which the only relevant factor is point of view. (22)a.John’s worst fear is that he might have to sing in public. b. *His worst fear is that John might have to sing in public. The unacceptability of coreference in (22b) is explained very simply: his worst fear sets up the expectation that the clause which follows it will be construed as a representation of John’s thoughts. It is therefore ‘viewed’ from John’s viewpoint, thus the name is anomalous. Note that c-command cannot explain (22), as the pronoun does not c-command the name. An example which gives more variable judgments is (23). (23) a. That he might have AIDS worries John. b. %That John might have AIDS worried him. Speakers’ judgments are about evenly split as to whether coreference is acceptable in (23b), although even those who find it unacceptable find it less so than (22b). The reason for the ambiguity is that, unlike in (22b) where the phrase his worst fear explicitly sets up the idea of John’s point of view as a reference point from the beginning of the sentence, in (23b) the embedded clause that John might have AIDS comes first in the linear string. Some speakers therefore construe the full noun 912 karen van hoek phrase from the point of view of the speaker and judge coreference to be accept- able. Others take John’s implied point of view as a reference point with the em- bedded clause in its dominion, even though it comes later in the linear string— similar to the different interpretation possibilities seen for the sentences in (16) and (17) above, where speakers may take either prominence or linear order as their primary cue for determining reference point organization. In this case, some speakers feel that both (23a) and (23b) are acceptable, while some feel that the point of view effect in (23b) is too strong to allow for coreference. The fact that perspective effects are more variable than the effects of prominence and connectivity should not come as any surprise, nor is it a reason to relegate point of view effects to a separate domain of study. We have already seen that the strongest, clearest judgments of coreference possibilities are found where prominence and conceptual connectivity combine to produce a clear, unequivocal reference point/ dominion configuration. Where the cues are more subtle, there is more flexibility in interpretation, and other factors have more influence. The very fact that a full noun phrase is used in a particular position, such as the embedded clause in (23b), may in fact be construed as a signal that the embedded clause is not to be construed from the viewpoint of the experiencer him. In the core complement chain examples, such as He loves John’s mother, the use of the name John would not be sufficient to signal that one should not construe the clause within the dominion of the subject; the subject is too firmly established as a reference point to be overridden in that way. In some cases, however, point of view effects can interfere with even the some- what more established anaphora patterns. As explained in section 4.1,amodifierofa nominal is construed within the same dominions as the nominal it modifies. When the modifier is complex, however, such as a relative clause, and has an explicit marker of a perspective shift, it is sometimes possible to construe it as independent of the dominions to which it would otherwise belong. The perspective shift serves as a conceptual discontinuity, in other words. This explains contrasts such as (24). (24)a.*She joined a new organization, which paid Sally a lot more money. b. She joined a new organization, whose members all found Sally to be absolutely delightful. c. *He found a new insurance company, which promised Mark excel- lent benefits. d. He married a former dental hygienist, who clearly thinks Mark is the greatest guy on earth. A c-command account would claim that all the examples in (24) should be equally unacceptable. In fact, the shift in perspective in (24b) and (24d) changes the co- reference possibilities by interrupting the flow of attention from the subject of the main clause to the modifier of the nominal. (Not all speakers find 24b and 24d fully acceptable, but the majority consulted do, and the others report that they are markedly improved as compared with 24a and 24c.) Far from being irrelevant to the ‘‘core’’ anaphora model, viewpoint is a significant factor, albeit one which is often difficult to isolate from the others. pronominal anaphora 913 Nevertheless, these examples show that it can interact with other factors to change coreference possibilities. Although it is discussed here as a separate subsection, the contribution of viewpoint effects should be considered an integral element in the reference point model of anaphora. 6. Conclusion The model briefly summarized here shows the potential of Cognitive Linguistic constructs to provide insightful analyses of phenomena which have traditionally been considered strong evidence for the necessity of autonomous syntax. More- over, it illustrates several of the key principles of Cognitive Linguistics: the search for syntactic analyses that are grounded in meaning; the goal of unifying domains of data (such as sentence-internal and discourse-level coreference facts, as well as point of view effects) which would be subjected to piecemeal treatment under a modular account; and the goal of not only predicting grammaticality judgments but explaining why variability exists in some cases but not in others and how different ways of saying ‘‘the same thing’’ reflect different construals of meaning. The domain of pronominal anaphora is a particularly rich area for exploring some of the subtle yet significant facets of sentential and cross-sentential seman- tics in the form of reference point organization. The reference point model also has potential for explaining phenomena outside of pronominal anaphora, such as quantifier scope and constraints on the formation of Wh-questions. It is par- ticularly noteworthy that the central factors in reference point organization are prominence, conceptual connectivity, temporal sequence, and empathy or point of view—all factors which are arguably grounded in more general, nonlinguistic di- mensions of cognition. Thus, the analysis of pronominal anaphora and the true nature of so-called ‘‘c-command effects,’’ which have traditionally been among the centerpieces of autonomous syntactic theory, points toward cognitive bases of grammar which are not unique to language. The study of pronominal anaphora, and of reference point effects more generally, thus have deeper and broader im- plications for the understanding of the nature of language itself. REFERENCES Ariel, Mira. 1990. Accessing noun-phrase antecedents. London: Routledge. Bolinger, Dwight. 1979. Pronouns in discourse. In Talmy Givo ´ n, ed., Syntax and semantics, vol. 12, Syntax and discourse 289–309. New York: Academic Press. Bosch, Peter. 1983. Agreement and anaphora. New York: Academic Press. 914 karen van hoek Fauconnier, Gilles. 1985. Mental spaces: Aspects of meaning construction in natural lan- guage. Cambridge, MA: MIT Press. (2nd ed., Cambridge: Cambridge University Press, 1994) Fox, Barbara. 1987a. Anaphora in popular written English narratives. In Russell S. Tomlin, ed., Coherence and grounding in discourse 157–74. Amsterdam: John Benjamins. Fox, Barbara. 1987b. Discourse structure and anaphora. Cambridge: Cambridge Univer- sity Press. Givo ´ n, Talmy. 1989. The grammar of referential coherence as mental processing instruc- tions. Technical Report No. 89–7. Eugene: University of Oregon. Langacker, Ronald W. 1985. Observations and speculations on subjectivity. In John Hai- man, ed., Iconicity in syntax 109–50. Amsterdam: John Benjamins. Langacker, Ronald W. 1987. Foundations of cognitive grammar. Vol. 1, Theoretical pre- requisites. Stanford, CA: Stanford University Press. Langacker, Ronald W. 1991. Foundations of cognitive grammar. Vol. 2, Descriptive appli- cation. Stanford, CA: Stanford University Press. Langacker, Ronald W. 1995. Viewing in cognition and grammar. In Philip W. Davis, ed., Alternative linguistics: Descriptive and theoretical models 153–212. Amsterdam: John Benjamins. Langacker, Ronald W. 1998. Conceptualization, symbolization, and grammar. In Michael Tomasello, ed., The new psychology of language: Cognitive and functional approaches to language structure 1: 1–39. Mahwah, NJ: Lawrence Erlbaum. Matthiesen, Christian, and Sandra A. Thompson. 1988. The structure of discourse and ‘‘subordination.’’ In John Haiman and Sandra A. Thompson, eds., Clause combining in grammar and discourse 275–330. Amsterdam: John Benjamins. Reinhart, Tanya. 1983. Anaphora and semantic interpretation. Chicago: University of Chi- cago Press. Tomlin, Russell S. 1987. Linguistic reflections of cognitive events. In Russell S. Tomlin, ed., Coherence and grounding in discourse. Amsterdam: John Benjamins. van Hoek, Karen. 1997a. Anaphora and conceptual structure. Chicago: University of Chicago Press. van Hoek, Karen. 1997b. Backwards anaphora as a constructional category. Functions of Language 4: 47–82. van Hoek, Karen. 1992. Paths Through Conceptual Structure: Constraints on Pronominal Anaphora. Ph.D. dissertation, University of California at San Diego pronominal anaphora 915 chapter 35 DISCOURSE AND TEXT STRUCTURE ted sanders and wilbert spooren 1. Discourse, Text Structure, and Cognitive Linguistics The alliance between Cognitive Linguistics and the study of discourse has become stronger in the recent past. This is a natural development. On the one hand, Cog- nitive Linguistics focuses on language as an instrument for organizing, processing, and conveying information; on the other, language users communicate through discourse rather than through isolated sentences. Nevertheless, at the moment, the cognitive linguistic study of discourse is still more of a promising challenge to linguists and students of discourse, rather than a well-established part of everyday cognitive linguistic practice. We start this chapter from the assumption that the grounding of language in discourse is central to any functional account of language (Langacker 2001). Discourse is often considered a crucial notion for understand- ing human communication, or, as Graesser, Millis, and Zwaan (1997, 164) put it, ‘‘Discourse is what makes us human.’’ Consider the following example from a Dutch electronic newspaper, which we have segmented into (1a) and (1b). (1)a.Greenpeace heeft in het Zuid-Duitse Beieren een nucleair transport verstoord. ‘Greenpeace has obstructed a nuclear transport in the South German state of Bavaria.’ b. Demonstranten ketenden zich vast aan de rails.(Telegraaf, April 10, 2001) ‘Demonstrators chained themselves to the rails.’ This short electronic news item does not create any interpretation difficulties. Nevertheless, in order to understand the fragment correctly, a massive amount of inferencing has to take place. For instance, we have to infer that the n uclear transport was not disturbed by the organization Greenpeace, but by members of that organi- zation; that the protesters are members of the organization; that the nuclear transport took place by train; that the place where the protesters chained themselves to the rails is on the route that the train took; that the time at which the protesters chained themselves to the rails coincided with the time of the transport; and that the ob- struction of the transport was caused by the protesters chaining themselves to the rails. Some of these inferences are based on world knowledge, for instance that or- ganizations consist of people and that people, but not organizations, can carry out physical actions. Others are based on discourse structural characteristics. Here are two examples: (i) The phrase the rails is a definite noun phrase that functions as an anaphor with a presupposed antecedent. Since there is no explicit candidate to fulfill the role of antecedent, the noun phrase necessarily invites the inference of a referential link with transport, the most plausible interpretation being that the transport took place by a vehicle on rails, a train. (ii) People reading news texts expect to get explanations for the phenomena described. When one event in the text can be interpreted as an explanation for another, readers will infer a causal link between them. 1 In this chapter, we will focus on discourse-structural characteristics like these, which, we believe, can account for the connectedness that discourse shows when compared to a random set of sentences. Given the limited space of a chapter like this, there are many specific issues that we cannot discuss, despite the fact that they are of great interest. Thus, we will not discuss the structure of spoken discourse. Obviously, there are fundamental differences between written and spoken discourse. For instance, many connectives in written language function to express the meaning relationships— coherence relations—between segments, such as but in example (2), which expresses a relation of denial of expectation. Connectives fulfill the same function in conver- sation, but often they simultaneously function as sequential markers: for instance, they signal the move from a digression back to the main line of the conversation. This type of marker is commonly referred to by the term discourse marker (see Schiffrin 2001, who is the source of example 3). (2) The murder suspect—described by Hampshire police as ‘‘very dangerous’’— had been spotted by a British tourist on Saturday, but she only informed New York police on Tuesday afternoon after returning home and seeing his photo in the British media. (The Guardian, June 6, 2002) (3) Jack: [The rabbis preach, [‘‘Don’t intermarry’’ Freda: [But I did— [But I did say those intermarriages that we have in this country are healthy. According to Schiffrin, but in (3) performs multiple functions, including the func- tion of displaying nonalignment with Jack, realizing an action of rebuttal during an argument, and attempting to establish Freda as the current speaker. discourse and text structure 917 Clearly, connectives have multiple functions, and, clearly, these functions are related. It is an interesting research question under which conditions a connec- tive that expresses a coherence relation can also be used as sequential marker. This type of question is under investigation in the grammaticalization literature (see Hopper and Traugott 1993 : chapter 7; Traugott 1995). In this chapter we confine ourselves to the coherence relation function of connectives. Other aspects of discourse structure that are specific to spoken language in- clude prosody and the occurrence of so-called adjacency pairs, minimal pairs such as Question-Answer and Summons-Response (Sacks, Schegloff, and Jefferson 1974). These topics, too, are subject to ongoing research (see the overview in Ford, Fox, and Thompson 2002) and can be considered especially important, as they cut across linguistic subdisciplines such as grammar and the study of conversation. Still, however important and promising this research may be, we will for reasons of space not go further into it. Instead, we want to focus on the crucial characteristics spoken and writ- ten discourse have in common. After all, these characteristics are central to the linguistic study of the level at stake here, namely, that of discourse. We will use the term ‘‘discourse’’ as the more general term to refer to both spoken and written language, and we will only use ‘‘text’’ to refer to phenomena restricted to written language. Over the years, the notion of ‘‘discourse’’ has become increasingly important in linguistics—a remarkable development, considering that linguistics used to deal almost exclusively with sentences in isolation. Nowadays, the discipline includes the study of form and meaning of utterances in context, and there exist formal, func- tional, and cognitive approaches that consider the discourse level as the core object of study. There seems to be a consensus that what makes a set of utterances into genuine discourse is (primarily) their meaning rather than their form. More spe- cifically, there is a shared belief that ‘‘discoursehood’’ is based on the possibility to relate discourse segments to form a coherent message. As a result, the dividing line between cognitive linguistic approaches and more formal approaches seems to be less clear-cut than at the sentence level (Knott, Sanders, and Oberlander 2001). Still, there are large differences between formal and cognitive or functional accounts of discourse. In formal linguistics, discourse-oriented work centers on the semantic theories of Kamp (e.g., Kamp and Reyle 1993) and Heim (1982). Here, issues like anaphora and presupposition are studied in short stretches of discourses usually consisting of constructed sets of sentences. In formal computational linguistics, however, attention is increasingly turning to the interpretation and production of extended pieces of text (Lascarides and Asher 1993). This type of approach is gradually moving in the direction of cognitively and functionally inspired work, which focuses on the discourse structure of naturally occurring language (Mann and Thompson 1986; Polanyi 1988) and on the cognitive representation of discourse in the mind of the language user (Sanders, Spooren, and Noordman 1992, 1993). The central claim of this chapter is that the connectedness of discourse is a mental phenomenon. When confronted with a stretch of discourse, language users 918 ted sanders and wilbert spooren make a coherent representation of it. At the same time, discourse itself contains (more or less) overt signals that direct this interpretation process, which is in line with views of grammar as a processing instructor (Givo ´ n 1995; Kintsch 1995). Thus, our view of discourse revolves around two central notions: ‘‘mental representa- tion’’ and ‘‘overt linguistic signals.’’ The latter goes back to the Hallidayan work on cohesion (Halliday and Hasan 1976), which describes text connectedness in terms of cohesive ties such as conjunction and ellipsis. The problem with this approach is that sequences like John was happy. It was a Saturday can be coherent, even though they do not have any cohesive ties. The notion of ‘‘mental repre- sentation’’ relates to approaches like Hobbs’s (1979), who coined the phrase co- herence relation for interclausal relationships. What we inherit from this work is what we consider the best of both worlds: the attention for linguistic detail in the cohesion approach is combined with the basic insight that coherence is a cognitive phenomenon. Considering coherence as a mental phenomenon implies that it is not an in- herent property of a text under consideration. Language users establish coherence by relating the different information units in the text. The notion of coherence has a prominent place in both (text-)linguistic and psycholinguistic theories of text and discourse. Although this is not a particularly new view of coherence (see, among many others, van Dijk and Kintsch 1983; Hobbs 1990; Garnham and Oakhill 1992; Sanders, Spooren, and Noordman 1992; Gernsbacher and Givo ´ n 1995; Noordman and Vonk 1997), it is a crucial starting point for theories that aim at describing the link between the structure of a text as a linguistic object, its cognitive representa- tions, and the processes of text production and understanding. In our view, it is this type of theory, located at the intersection of linguistics and psycholinguistics, that could lead to significant progress in the field of discourse studies (T. Sanders and Spooren 2001b). Cognitive linguists have already made substantial contributions to the study of discourse. At the same time, Cognitive Linguistics can benefit from insights in discourse to further develop itself as the study of language in use (Barlow and Kemmer 2000). In the remainder of this chapter, we will discuss two types of coherence and their textual signals: (i) Referential coherence: how does reference to individuals create continuity and (as a result) coherence? The signals that we will be considering involve reference to objects and concepts; more specifically, we will consider the ways in which reference is realized: through full NPs, pronouns, zero anaphora, and so on. (ii) Relational coherence: how do coherence relations like causals and con- trastives constitute connectedness? The signals that we will be considering are connectives and lexical cue phrases. At the end of this chapter, we will reach some conclusions about the relationship between discourse/text structure and Cogni- tive Linguistics, and on the basis of our analysis of the state of the art, we will suggest some challenging issues for future research. discourse and text structure 919 . rails is on the route that the train took; that the time at which the protesters chained themselves to the rails coincided with the time of the transport; and that the ob- struction of the transport. interrupting the flow of attention from the subject of the main clause to the modifier of the nominal. (Not all speakers find 24b and 24d fully acceptable, but the majority consulted do, and the others. case neither nominal is in the dominion of the other. In that case, speakers reidentify the referent with a full nominal. The nature of the relevant discontinuity varies depending on the spoken