Using the world wide web for learning new words in continuous speech recognition tasks: Two case studies

Abstract

In this paper, Web-based lexicon augmentation is addressed: using various strategies for Out-Of-Vocabulary (OOV) word learning, we discuss their relevance in two types of applications: broadcast news, or topic-specific corpora transcription. The Web-based OOV word learning is first tested on the French news corpus ESTER; the same approach is applied to a very specific corpus concerning surgical interventions. These tests allow us to assess the value of the OOV word learning methods proposed, emphasizing their strengths and weaknesses, regarding the particular type of application considered. 1.