Using story topics for language model adaptation

Abstract

The subject matter of any conversation or document can typically be described as some combination of elemental topics. We have developed a language model adaptation scheme that takes a piece of text, chooses the most similar topic clusters from a set of over 5000 elemental topics, and uses topic specific language models built from the topic clusters to rescore N-best lists. We are able to achieve a 15% reduction in perplexity and a small improvement in WER by using this adaptation. We also investigate the use of a topic tree, where the amount of training data for a specific topic can be judiciously increased in cases where the elemental topic cluster has too few word tokens to build a reliably smoothed and representative language model. Our system is able to fine-tune topic adaptation by interpolating models chosen from thousands of topics, allowing for adaptation to unique, previously unseen combinations of subjects.

Language modeling for dialog system

“…[3][4][12] also reported that using clustering technique either deteriorates or slightly improves the recognition performance. [12] suggested that better smoothing method (e.g. Kneser-Ney smoothing) need to be applied to cluster language models in order to get good performance.…”

Nonlinear interpolation of topic models for language model adaptation

“…Previous work in topic adaptation [1,3,4,5,7,10,11] has mainly focused on identifying topic-specific subsets of the training text and building language models from them. The topic language models are linearly interpolated with a general language model built from all of the training text.…”

Contributions to speech analytics based on speech recognition and topic identification

“…In these approaches, new sources of information are used to generate a context-dependent LM which is then merged with a static LM. These new sources of information may come, for instance, from text categorization systems as in [Seymore and Rosenfeld, 1997], from speaker identification systems [Nanjo and Kawahara, 2003], from linguistic analysis systems [Liu and Liu, 2008] or from the application context itself .…”

“…In this regard, there are different ways in which the interpolation weight can be selected: it can be set empirically by minimizing the perplexity in a development stage with data not seen during training [Clarkson, 1999, Tur and; it can also be estimated under some optimization algorithm, such as Expectation Maximization [Daumé et al, 2010] or Maximum A Posteriori (MAP) adaptation [Wang and Stolcke, 2007]; or it can be set dynamically depending on the current situation of the interaction (the topic of the speech, a specific speaker, etc.) [Haidar andO'Shaughnessy, 2012, Seymore andRosenfeld, 1997].…”

“…There are different ways in which the interpolation weight can be selected: it can be set empirically by minimizing the perplexity in a development stage with data not seen during training [Clarkson, 1999, Tur and; it can also be estimated under some optimization algorithm, such as Expectation Maximization [Daumé et al, 2010] or Maximum A Posteriori (MAP) adaptation [Wang and Stolcke, 2007]; or it can be set dynamically depending on a specific situation of the recognition process (related to the topic, the speaker, etc.) [Seymore and Rosenfeld, 1997]. In this work, we explore different options, including the last of the options mentioned above, to determine the interpolation weight of the models.…”