Verificación de tema en sistemas de diálogo mediante la aplicación de un test de hipótesis bayesiano

  1. Pérez-Piñar López, David
  2. García Mateo, Carmen
Revista:
Procesamiento del lenguaje natural

ISSN: 1135-5948

Ano de publicación: 2005

Número: 35

Páxinas: 221-227

Tipo: Artigo

Outras publicacións en: Procesamiento del lenguaje natural

Resumo

One of the most difficult aspects of any dialogue system is the identification of user intention and recognition errors. In this paper, we present a novel approach to the task of deciding what the user has said and what she wants to do next. We use statistic classifiers driven by confidence measures which are generated in parallel by several topic-adapted speech recognizers. This approach has shown to be especially suited for difficult topics, such as proper names or confirmations. Recognition performance is greatly enhanced through the use of this architecture, which helps also in the identification of user intention and user-initiated topic change