This paper argues that connectionist systems are a good approach to implement a speech understanding computational model. In this direction, we propose SUM, a Speech Understanding Model, which is a software architecture based on recent neurocognitive researches. The SUM´s computational implementation applies wavelets transforms to speech signal processing and connectionist models to syntactic parsing and prosodic-semantic mapping. This approach enables a system of analysis by compensation, when the syntactic analysis does not offer a good reliable level, it is possible to evaluate prosodicsemantic analysis, such as in human speech understanding.
I. INTRODUCTIONThe studies of computational systems, which research oral human-computer communication, are commonly called Spoken Language Understanding (SLU) Systems. This kind of system is derived from two joint technologies: Automatic Speech Recognition (ASR) and Natural Language Processing (NLP). These two technologies are complementary. The natural language can help the speech recognition through information on syntax and semantics. Furthermore, the speech recognition can improve the understanding of the language with contextual information, such as the intonation of the words (prosody) [1].Current SLU systems lack technology based on biological studies and there is large interference of the designer in the system configuration. Therefore, it is necessary a study of the biological processes involved in the speech understanding, which will lead to a coherent and valid computational representation of such processes. Our work supports the possibility of computational systems integration to represent the speech understanding process, which will improve the SLU systems development.In fact, recent neurocognitive researches present new approaches to base computational representations. The neurocognitive model developed by Friederici [2] demonstrates a parallel and complementary process that is performed by syntactic and semantics analyses. In this direction, we have proposed a new computational architecture to allow a more robust approach to spoken language processing [3] [4].This work argues that it is possible to unify several computational systems to represent the speech understanding process. Thus, we propose SUM (figure 1), a Speech Understanding Model based on a neurocognitive model of auditory sentence (section III). Through SUM, we search a computational model that works by compensation: when syntactic analysis fails, semantic analysis can be successful, and vice versa.