The performance of current speech recognition systems and applications will be improved considerably through incorporating more linguistic knowledge. This paper describes how this knowledge may be encapsulated for automated systems for information-seeking telephone dialogues. The data for the model is a large number of recorded dialogues. The linguistic model itself is comprehensive, and extends the established linguistic theory known as systemic functional grammar. This has the inherent capability to model, in computable forms, all levels of language, including prosodic, lexico-grammatic, semantic, pragmatic and dialogic. The model can also handle the short non-lexical utterances (such as mmm's and uh's) which fulfil useful roles in telephone dialogues. The paper describes how the data for the model can be drawn from the data analysis, and illustrates the systemic principles with examples of system networks. The paper outlines how this modelling technique can be used to support speech recognition, speech understanding and speech synthesis in automated speech response systems.
|Number of pages||5|
|Journal||National Conference Publication - Institution of Engineers, Australia|
|Issue number||92 pt 12|
|Publication status||Published - 1992|