Integration of supra-lexical linguistic models with speech recognition using shallow parsing and finite state transducers

Mou, Xiaolong; Seneff, Stephanie; Zue, Victor W.

doi:10.21437/icslp.2002-395

Cited by 1 publication

(2 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This too proved to be impractical except for very small domains. A more promising approach was to select a subset of the categories in the NL grammar as classes in a class n-gram, and to expand those classes using RTN's, as exemplified by [4,11]. While this approach gives a performance that compares favorable to that of a standard class n-gram, it too suffers computationally, in part because our FST technology for speech recognition has been optimized for performance on traditional class n-gram's.…”

Section: Introductionmentioning

confidence: 99%

“…This paper details our work in developing a technique which satisfies all of the required constraints: computational efficiency, ease of maintenance, and high performance. It is similar to the approach in [4,11], except that the RTN is precompiled into an automatically generated set of multi-word units, such that the resulting n-gram is very similar to a standard class n-gram. An important difference is that there is the opportunity to define multiple classes for a single word.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Automatic induction of n-gram language models from a natural language grammar

Seneff

Wang

Hazen

2003

8th European Conference on Speech Communication and Technology (Eurospeech 2003)

Self Cite

View full text Add to dashboard Cite

This paper details our work in developing a technique which can automatically generate class n-gram language models from natural language (NL) grammars in dialogue systems. The procedure eliminates the need for double maintenance of the recognizer language model and NL grammar. The resulting language model adopts the standard class n-gram framework for computational efficiency. Moreover, both the n-gram classes and training sentences are enhanced with semantic/syntactic tags defined in the NL grammar, such that the trained language model preserves the distinctive statistics associated with different word senses. We have applied this approach in several different domains and languages, and have evaluated it on our most mature dialogue systems to assess its competitiveness with preexisting n-gram language models. The speech recognition performances with the new language model are in fact the best we have achieved in both the JUPITER weather domain and the MERCURY flight reservation domain.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%