Spontaneous speech understanding in train timetable inquiry processing based on N-gram language models and finite state transducers

The presented paper concerns the spoken language understanding in an information retrieval dialogue system. There are described methods of use the finite state transducers for conceptual semantic parsing and meaning extraction from speaker's utterances. In this case, the main aim of understanding is an identification of important semantic constituents and their interpretation within supposed frame structure. The key problem is to create an appropriate mapping between sequence of recognized words and concept based meaning that represents real-world entities. We propose a hierarchical semantic n-gram language model for parsing of a first initiative spontaneous speech train timetable inquiry. The design, implementation and evaluation of the model in experimental understanding system are described below.