Can You Repeat That? Using Word Repetition to Improve Spoken Term Detection

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1316–1325, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Can You Repeat That? Using Word Repetition to Improve Spoken Term Detection Jonathan Wintrode and Sanjeev Khudanpur Center for Language and Speech Processing Johns Hopkins University jcwintr@cs.jhu.edu , khudanpur@jhu.edu Abstract We aim to improve spoken term detec- tion performance by incorporating con- textual information beyond traditional N- gram language models. Instead of taking a broad view of topic context in spoken doc- uments, variability of word co-occurrence statistics across corpora leads us to fo- cus instead the on phenomenon of word repetition within single documents. We show that given the detection of one in- stance of a term we are more likely to find additional instances of that term in the same document. We leverage this bursti- ness of keywords by taking the most con- fident keyword hypothesis in each docu- ment and interpolating with lower scor- ing hits. We then develop a principled approach to select interpolation weights using only the ASR training data. Us- ing this re-weighting approach we demon- strate consistent improvement in the term detection performance across all five lan- guages in the BABEL program. 1 Introduction The spoken term detection task arises as a key sub- task in applying NLP applications to spoken con- tent. Tasks like topic identification and named- entity detection require transforming a continu- ous acoustic signal into a stream of discrete to- kens which can then be handled by NLP and other statistical machine learning techniques. Given a small vocabulary of interest (1000-2000 words or multi-word terms) the aim of the term detection task is to enumerate occurrences of the keywords within a target corpus. Spoken term detection con- verts the raw acoustics into time-marked keyword occurrences, which may subsequently be fed (e.g. as a bag-of-terms) to standard NLP algorithms. Although spoken term detection does not re- quire the use of word-based automatic speech recognition (ASR), it is closely related. If we had perfectly accurate ASR in the language of the corpus, term detection is reduced to an exact string matching task. The word error rate (WER) and term detection performance are clearly corre- lated. Given resource constraints, domain, chan- nel, and vocabulary limitations, particularly for languages other than English, the errorful token stream makes term detection a non-trivial task. In order to improve detection performance, and restricting ourselves to an existing ASR system or systems at our disposal, we focus on leverag- ing broad document context around detection hy- potheses. ASR systems traditionally use N-gram language models to incorporate prior knowledge of word occurrence patterns into prediction of the next word in the token stream. N-gram mod- els cannot, however, capture complex linguistic or topical phenomena that occur outside the typical 3-5 word scope of the model. Yet, though many language models more sophisticated than N-grams have been proposed, N-grams are empirically hard to beat in terms of WER. We consider term detection rather than …

Content truncated to 500 words