The Recent History of Dialogue Processing
|
|
|
The Recent History of Dialogue Processing |
|
Discourse and dialogue processing has had a relatively short history. On the one hand, discourse analysis and conversational analysis have been considered research areas in linguistics only as recently as the early 1970's (short list of texts). Discourse analysis focuses on the speech act while conversational analysis views discourse as a social interaction. On the other hand, computing in the humanities has been prominent only since the mid-1970's (The Association for Computers and the Humanities (ACH) was founded in 1973). Early computing efforts in the humanities concerned the analysis and generation of text.
As more recent advances have been made in Natural Language Processing (speech recognition, tagging, parsing, etc.), and with the increasing availability of computing hardware resources including processing speed and memory, there has been an increasing emphasis on the role of discourse and dialogue processing in natural language systems. NLP researchers recognize a parallel between larger textual units (supra-sentential) and dialogue units. The idea that contextual information in language extends beyond sentence and utterance boundaries is important in understanding language as it is really used. Discourse and dialogue processing in text is an important aspect of research in diverse disciplines such as Computer-Aided Language Learning (CALL), text generation, information extraction, text summarization, corpus analysis, and the automatic translation of texts.
Though discourse and dialogue processing is not limited to speech, large research efforts have been spent towards developing spoken dialogue systems. In 1984, DARPA began to concentrate funding in speech recognition towards the development of large vocabulary continuous recognition (LVCSR) . It is the capability of continuous speech recognition that has engendered a shift towards research of human-computer interaction (HCI) in spoken language systems. The types of issues that discourse and dialogue processing in spoken dialogue systems address are broad questions of how to determine if a conversational error has occurred, how to form a system response, and how to interact with the user to repair conversational errors.
In a very abstract sense, what we are concerned with is not only what is said in conversation, but the when and why of speaker participation. Basic problems this community addresses are: what is the information content in sequences of utterances, and what context is relevant for determining meaning and felicity of utterances and larger dialogue units. More fundamentally, we are attempting to resolve such questions as: what is an utterance and how do we segment dialogue. Current applications and research areas for discourse and dialogue processing include spoken dialogue systems, multimedia environments, intelligent agent communication, automatic generation of text, acoustic markers of discourse segmentation, machine translation, discourse and planning, shared beliefs in discourse, task-oriented discourse, collaborative discourse and shared plans, discourse annotation, dialogue segmentation, and coreference.