Third International Conference on Spoken Language Processing (ICSLP 94)
In this paper, we present a Continuous Speech Understanding (CSU) system directed by semantics, in which all the required knowledge sources are automatically learnt from training data. In particular, we use an inductive learning technique in order to obtain structural models both at the acoustic-phonetic level and the semantic level. The system which we propose assumes that understanding is the ultimate goal of the system performance. Therefore, the search should mainly be constrained by the semantic relations rather than by the word relations of language, allowing for a relaxed syntax. Preliminary experiments have been carried out with a semantic constrained task consisting of the understanding of queries to a database with information about Spanish geography in natural language, using two different system architectures.
Bibliographic reference. Prieto, Natividad / Sanchis, Emilio / Palmero, Luis (1994): "Continuous speech understanding based on automatic learning of acoustic and semantic models", In ICSLP-1994, 2175-2178.