A unified framework for translation and understanding allowing discriminative joint decoding for multilingual speech semantic interpretation

摘要: HighlightsA comparison between the methods used for speech translation and understanding.A unified framework discriminative joint decoding multilingual semantic interpretation.The proposition is competitive with state-of-the-art techniques.The can be generalized to other components of a dialogue system. Probabilistic approaches are now widespread in most natural language processing applications selection particular approach usually depends on task at hand. Targeting interpretation context, this paper presents machine understanding. This justifies our both tasks based approach. We demonstrate that perform translation-understanding which allows combine, same process, tagging scores sentence. A cascade finite-state transducers compose understanding hypothesis graphs (1-bests, word or confusion networks). Not only techniques but also its even more attractive as it human-machine vocal interfaces (e.g. recognizer) so allow richer transmission information them.