Bridging Psycholinguistics and Dialog System Research: Theoretical and Applied Issues

 

Bridging Psycholinguistics and Dialog System Research: Theoretical and Applied Issues

Ricketts 211

Now is an exciting time to be in the position of bridging the fields of Psycholingistics and Dialog System Research. Both communities are increasingly focused on many of the same theoretical issues, and there is the potential for a powerful emergent synergy between the two. On the psycholinguistics side, the prevailing view of human language comprehension is one in which language processing and processing of aspects of "the world" (visual context, embodied constraints, social cues, etc) are interleaved. Based on this class of models there is a trend toward studies investigating language use in more complicated real-world problem-solving settings, which may require new tools for tracking and manipulating aspects of the context dynamically and in real-time. Dialog systems offer a potential solution. On the Dialog system side, the prevailing model for systems is an encapsulated model in which modules (speech recognition, parsing, reasoning, etc.) process the input sequentially, utterance-by-utterance. Real-time understanding in the general case is still an open problem. Psycholinguistic models of continuous language processing suggest a promising new approach that is just beginning to be explored in the dialog system community. In my talk I will discuss both of these theoretical issues, illustrated by my own behavioral and system design research. In addition to the theoretical developments I will discuss, bridging the two fields can address applied issues in dialog system design. For example, Psycholinguistics offers powerful tools of comparison and measurement that can be applied to system evaluation. In my talk I will describe my ongoing research investigating the consequences of interface design decisions on users' levels of cognitive load, as a function of the complexity of the domain. The study compares a "natural" approach to generation and understanding of referring expressions with the popular "standardized" approach. I use a classic tool from cognitive psychology, the dual-task paradigm, to compare the two design approaches. The results may have important implications for how future systems are designed, and may therefore result in tangible benefits for real communities of users. -- Ellen Campana's research focuses on how people communicate with others during complicated problem-solving tasks. Her research combines the experimental methods of cognitive psychology with dialog system implementation and evaluation. The specific goals of her work are 1) to improve our theoretical understanding of the mechanisms of human language production and comprehension in context, and 2) to improve human-computer interaction for real communities of users. She is currently a Ph.D. candidate at the University of Rochester, Departments of Brain & Cognitive Science and Computer Science.
Add to calendar
Share|