First International Conference on Spoken Language Processing (ICSLP 90)
This paper deals with the integration of different levels of knowledge in a continous speech recognition framework. A system has been developed to carry out recognition and understanding processes in parallel. We present a method which allows the system to dynamically compute weights, taking semantic knowledge into account during the recognition process, which is based on a dynamic time warping (IjTW) algorithm. The weights change during the course of an utterance recognition process, according to the interpretation of partial recognized sentences. Nested into the DTW algorithm, a natural language processor (NLP) computes semantic weights. It builds an interpretation using the joining operation of the conceptual graph model. The syntax describes the joining rules of conceptual graphs. Then, we show how a set of joining rules can be represented by a conceptual grammar from which a parser can be generated. The parsing process evaluates an interpretation of a partially recognized sentence according to a criterion using the overlapping rate of all the conceptual graphs which participate into the interpretation, in order to return a semantic weight.
Bibliographic reference. Bornerand, S. / Neel, Francoise / Sabah, G. (1990): "Semantic weights derived from syntax-directed understanding in DTW-based spoken language processing", In ICSLP-1990, 913-916.