Little Drop of Mulligatawny Soup, Miss Sophie? Automatic Speech Understanding provided by Petri Nets

Authors: Markus Huber, Ronald Römer, Matthias Wolff

Abstract:

This paper contributes to seamless automata-based integration of speech recognition and natural language understanding. We describe a new transducer between sequential utterances and partially-ordered semantic structures which can be composed with finite state machines. After a brief discussion of the limitations of previous approaches and some mathematical preliminaries, we will recapitulate the application of Petri net transducers (PNTs) to the bidirectional translation between sequences and partial orders. Then we will show how recursive hierarchical PNTs can be used to create partially-ordered semantic structures of arbitrary width and how to build a seamless speech signal-to-semantics recognition network.