Conversational perspectives in a communication aid for non-vocal people.

Abstract

A computer-based text storage and retrieval system (TALK) has recently been developed as a conversation aid for non-vocal people. To facilitate the rapid retrieval (and synthesised speech output) of stored content, potential utterances on each conversational topic are organised according to three intersecting perspectives. The perspectives are "person' (me, you), "time' (past, present, future), and "issue' (what, when, where, who, how, why). Although the system has been shown capable of supporting free-flowing conversation with relatively short pauses preceding the user's turns at speech, the perspective organisation is probably less than ideal for some topics. An experiment was conducted to see whether, over a number of conversational topics, storage and retrieval of potential utterances would be faster with the existing, uniform, set of perspectives or with different sets of perspectives tailored to each topic. At the storage stage, the "uniform' set of perspectives facilitated more rapid classification of potential utterances than the "tailored' sets. At the retrieval stage, however, repetition of the classification decisions was more accurate (without being any slower) for the tailored sets of perspectives, suggesting that retrieval of stored utterances during conversation may be faster when the perspective organisation is tailored to topics.