An investigation of grammar design in natural-language speech-recognition.

Date of Award


Publication Type

Master Thesis

Degree Name



Computer Science

First Advisor

Frost, R.


Computer Science.



Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.


With the growing interest and demand for human-machine interaction, much work concerning speech-recognition has been carried out over the past three decades. Although a variety of approaches have been proposed to address speech-recognition issues, such as stochastic (statistical) techniques, grammar-based techniques, techniques integrated with linguistic features, and other approaches, recognition accuracy and robustness remain among the major problems that need to be addressed. At the state of the art, most commercial speech products are constructed using grammar-based speech-recognition technology. In this thesis, we investigate a number of features involved in grammar design in natural-language speech-recognition technology. We hypothesize that: with the same domain, a semantic grammar, which directly encodes some semantic constraints into the recognition grammar, achieves better accuracy, but less robustness; a syntactic grammar defines a language with a larger size, thereby it has better robustness, but less accuracy; a word-sequence grammar, which includes neither semantics nor syntax, defines the largest language, therefore, is the most robust, but has very poor recognition accuracy. In this Master's thesis, we claim that proper grammar design can achieve the appropriate compromise between recognition accuracy and robustness. The thesis has been proven by experiments using the IBM Voice-Server SDK, which consists of a VoiceXML browser, IBM ViaVoice Speech Recognition and Text-To-Speech (TTS) engines, sample applications, and other tools for developing and testing VoiceXML applications. The experimental grammars are written in the Java Speech Grammar Format (JSGF), and the testing applications are written in VoiceXML. The tentative experimental results suggest that grammar design is a good area for further study. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis2003 .S555. Source: Masters Abstracts International, Volume: 43-01, page: 0244. Adviser: Richard A. Frost. Thesis (M.Sc.)--University of Windsor (Canada), 2004.