Discourse can be viewed in many ways. A text is concocted in the mind of one person. It is produced and exists in the world as something that can be examined independently. It is interpreted, perhaps in different ways, by each member of the speaker's audience, and takes on a life in each listener's mind. The text carries meaning--it bears some quite detailed relation to the world. In fact, it carries more than one meaning: the speaker means one thing by it, each of the listeners might take it to mean something quite different, and it has a meaning independent of its occasion which is imposed by the language in which it is framed. By exchanging texts, people increase their knowledge and increase the portion of their knowledge that they share, thereby binding themselves more tightly together, increasing for example their possibilities for joint action. The texts people exchange take their place in a long line of similar texts exchanged by similar people, and this discourse is the principal means by which social organization is constituted and given continuous life.

In this inquiry we will confine ourselves primarily to one view of discourse--its interpretation. We will investigate the means by which a listener transforms the text, as something out in the world, into something in his^1.1 own mind. That is, we will study the psychology of discourse, rather than the philosophy, the sociology, or the history.

A speaker's production of a text is also a psychological process, but this volume will be for the most part concerned with interpretation. The two processes certainly access the same knowledge, and they surely overlap importantly in their subprocesses. For example, in production we frequently ask ourselves how we can say something in a way that our listener will be able to interpret it, and in interpretation we frequently ask ourselves what the speaker is really trying to say. But in this volume the focus will be on interpretation, with production viewed, when at all, as a process of constructing the utterance in a way that will produce the right interpretation.

Much of this study will consider a single listener interpreting a single text produced by a single speaker. But in addition we will examine some multi-participant discourses. Something very different happens when more than one speaker is involved than when the discourse is largely under the control of a single speaker. With more than one speaker, there are likely to be more than one set of intentions or goals for the discourse. These intentions will come into conflict and the conflicts will be resolved in complex and unpredictable ways. To study multi-participant discourse, we need all the machinery we need for single-speaker discourse, plus a way of talking about and analyzing the participants' conflicting plans. Nevertheless, each participant must interpret the contributions of the others, and although often at a much finer grain, the interpretation processes are the same as in interpreting single-speaker discourse.

In discourse, information is communicated from one person to another by some means. The communication can be verbal or nonverbal; it is usually a mixture of both. Much of the work in the AI tradition is based on the notions of the logical representation of knowledge and language and of the propositional content of sentences. In nonverbal communication, including intonation, it is not clear a priori to what extent these notions apply. What is the propositional content of a gesture? In this book we focus primarily on verbal communication, but nonverbal communication will be addressed in a very limited way. In any case, we can be sure that verbal communication can be meaningfully investigated in isolation. The success of written communication guarantees this: we are able to understand discourse in the absence of its nonverbal component.

Communication may also be spoken or written, and it may vary along a large number of other dimensions. These distinctions, however, will not be significant in our inquiry. Most of the specimens examined in this book are written texts, and there is perhaps a bias toward written discourse. Written texts have a number of attractive features for this inquiry. A written text is under the complete control of the author; the reader does not have a chance to intervene in the course of the writing to redirect it (although a writer may imagine her reader's responses as she writes). Except for illustrations, it is entirely verbal. The reader's interpretation takes place in isolation from the production of the text, and thus is a separable phenomenon amenable to independent investigation. But this study is decidedly not restricted to written discourse. There are certainly numerous differences between written discourse and the verbal component of spoken discourse, as research has shown (e.g., Tannen, 1982). But the analysis of a wide variety of materials, both spoken and written, have indicated that these differences do not prove especially significant for the principal focus of this book, the processes for deploying knowledge in interpretation.

Among the other dimensions along which discourses can be classified are the spatial or temporal proximity of the participants, the degree of formality, the amount of prior rehearsal, and the extent of the participants' shared knowledge (cf. Rubin, 1978). The place of a discourse on each of these dimensions has many influences on the shape of the discourse--lexical choice, fullness of the definite descriptions, and so on. Many of the differences arise out of the differences in shared knowledge each of these conditions imposes and differences in how exacting or forgiving the listener should be. But once the extent of shared knowledge and the exactitude of the interpretation are fixed, it is likely that the procedures that use this shared knowledge to achieve an interpretation do not differ significantly from condition to condition. For this reason, these distinctions do not play a role in the present investigation.

To summarize, the phenomenon under investigation will be the processes involved in interpreting stretches of verbal communication produced by a single speaker, sometimes in dialogue with other participants. It is obvious that a theory of this phenomenon will play an important role in any more complete theory of human discourse.

This investigation is organized around seven ``target texts''. These are

1. the first two sentences of Edward Sapir's Language;
2. the first paragraph of Carson McCullers' The Ballad of the Sad Cafe;
3. the first two paragraphs in an article from the San Jose Mercury News business section;
4. two paragraphs from a Science magazine article on AIDS;
5. two brief military equipment failure reports;
6. Shakespeare's 64th sonnet;
7. the transcript of about a minute and a half of a three-person decision-making meeting.

These target texts are given in an appendix. In this book I hold myself responsible for the representational issues, the syntactic constructions, the underlying commonsense knowledge, and the interpretation problems presented by these texts. The aim of the diversity of the texts is to force generality in the theoretical framework. If all of the phenomena in these texts can be handled in a uniform fashion, that is a good argument for the broad applicability of the theory.

Each of these texts comes from a larger corpus that will be used at several points in the investigation as ``target corpora''. They are

1. the first two paragraphs of Edward Sapir's Language;
2. the first 14,000 words of Carson McCullers' The Ballad of the Sad Cafe;
3. about 30,000 words from the San Jose Mercury News business section;
4. four Science magazine articles on AIDS (nearly 22,000 words);
5. thirty-six brief military equipment failure reports;
6. all 154 of Shakespeare's sonnets (nearly 18,000 words);
7. the transcripts of about 68 minutes of five three-person decision-making meetings (nearly 14,000 words).

In addition, the lyrics of the one hundred most popular country-and-western songs of the 1980s are used as a target corpus. Again this diversity forces generality.

In addition, I will make use of various naturally occurring fragments of discourse, as well as made-up linguistic examples, where they illustrate the problem being discussed. But they will not play the same forcing role as the target texts and the target corpora.