Robust information extraction from automatically generated speech transcriptions

摘要: This paper describes a robust system for information extraction (IE) from spoken language data. The extends previous hidden Markov model (HMM) work in IE, using state topology designed explicit modeling of variable-length phrases and class-based statistical smoothing to produce state-of-the-art performance wide range speech error rates. Experiments on broadcast news data show that the performs well with temporal source differences In addition, strategies integrating word-level confidence estimates into are introduced, showing improved by generic token incorrectly recognized words training low test