Invited Talk: Processes that Shape Conversation and their Implications for Computational Linguistics

Processes that Shape Conversation and their Implications for Computational Linguistics Susan E. Brennan Department of Psychology State University of New York Stony Brook, NY, US 11794-2500 susan.brennan@sunysb.edu Abstract Experimental studies of interactive language use have shed light on the cognitive and interpersonal processes that shape conversation; corpora are the emergent products of these processes. I will survey studies that focus on under-modelled aspects of interactive language use, including the processing of spontaneous speech and disfluencies; metalinguistic displays such as hedges; interactive processes that affect choices of referring expressions; and how communication media shape conversations. The findings suggest some agendas for computational linguistics. Introduction Language is shaped not only by grammar, but also by the cognitive processing of speakers and addressees, and by the medium in which it is used. These forces have, until recently, received little attention, having been originally consigned to "performance" by Chomsky, and considered to be of secondary importance by many others. But as anyone who has listened to a tape of herself lecturing surely knows, spoken language is formally quite different from written language. And as those who have transcribed conversation are excruciatingly aware, interactive, spontaneous speech is especially messy and disfluent. This fact is rarely acknowledged by psychological theories of comprehension and production (although see Brennan & Schober, in press; Clark, 1994, 1997; Fox Tree, 1995). In fact, experimental psycholinguists still make up most of their materials, so that much of what we know about sentence processing is based on a sanitized, ideal form of language that no one actually speaks. But the field of computational linguistics has taken an interesting turn: Linguists and computational linguists who formerly used made-up sentences are now using naturally- and experimentally-generated corpora on which to base and test their theories. One of the most exciting developments since the early 1990s has been the focus on corpus data. Organized efforts such as LDC and ELRA have assembled large and varied corpora of speech and text, making them widely available to researchers and creators of natural language and speech recognition systems. Finally, Internet usage has generated huge corpora of interactive spontaneous text or "visible conversations" that little resemble edited texts. Of course, ethnographers and sociolinguists who practice conversation analysis (e.g., Sacks, Schegloff, & Jefferson, 1974; Goodwin, 1981) have known for a long time that spontaneous interaction is interesting in its own right, and that although conversation seems messy at first glance, it is actually orderly. Conversation analysts have demonstrated that speakers coordinate with each other such feats as achieving a joint focus of attention, producing closely timed turn exchanges, and finishing each another’s utterances. These demonstrations have been compelling enough to inspire researchers from psychology, linguistics, computer science, and human-computer interaction to turn their attention to naturalistic language data. But it is important to keep in mind that a corpus is, after all, only an artifact—a product that emerges from the processes that occur between and within speakers and addressees. Researchers who analyze the textual records of conversation are only overhearers, …

Content truncated to 500 words