D. Goodine, S. Sene#, L. Hirschman, and M. Phillips, "Full Integration of Speech and Language Understanding in the MIT Spoken Language System," Proc. Eurospeech, 845--848, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of the other word hypotheses are allowed to proceed. Therefore, any theory that completes is guaranteed to parse. Researchers are beginning to find that such a tightly coupled integration strategy can achieve higher performance than an N best interface, often with a considerably smaller stack size [15, 13, 44, 25]. The future is likely to see increasing instances of systems making use of linguistic analysis at early stages in the recognition process. 2.3. Spoken Language Generation On the output side, a conversational interface must be able to convey the information to the user in natural sounding ....

Goodine, D., Sene#, S., Hirschman, L., Phillips, M. "Full Integration of Speech and Language Understanding in the MIT Spoken Language System," Proc. Eurospeech, 845--848, 1991.

....of the vocabulary used 4 i.e. prefix probabilities Lattice Parsing for Speech Recognition Several solutions have been studied for tight coupling: 1. to use a non probabilistic linguistic model to generate word transition list and assign the probabilities by the usual N gram language methods (Goodine et al. 1991; Hauenstein Weber, 1994) 2. to use the linguistic model to smooth estimate N gram probabilities of words, providing a direct transfer to usual language model. The advantage of such an approach is that it requires almost no modification of the acoustic decoder. However, the integration of long ....

GOODINE D., SENEFF S., HIRSCHMAN L. & PHILLIPS M. (1991). Full integration of speech and language understanding in the MIT spoken language system. In Proc. of Eurospeech'91, p. 24--26, Genova (Italy).

....including Pitcher (1989) Future Work In the past, the fields of speech recognition and natural language processing have been kept fairly distinct. A number of researchers have called for more use of natural languagebackend information by the recognizer (Moore et al. 1989; Seneff et al. 1992; Goodine et al. 1991). Nonetheless, despite the trend in recent systems to tighten the coupling between the recognizer and the backend, the truly tightly coupled approach passing syntax andsemantic based word transition probabilities from the backend to the recognizer is quite difficult and has not been ....

GOODINE, D., S. SENEFF, L. HIRSCHMAN, & M. PHILLIPS. 1991. Full integration of speech and language understanding in the MIT spoken language system. In Proceedings of Eurospeech 91, 24---26, Genova, Italy.

....grammar) for reprocessing the original input. Furthermore it is worth notice that when hooked into MINDS II, SOUL uses additional knowledge of inferred speaker intentions (plans, goals) to further augment semantic interpretation of the spoken input. MIT: SUMMIT TINA [Phillips et al. 1991, Goodine et al. 1991, Seneff 1989, Zue et al. 1991] MIT s (Massachusetts Institute of Technology s) speech understanding system is composed of the phone based speaker independent continuous speech recogniser SUMMIT and the NLS component TINA. The system has been coupled both to the Air Travel Information System ATIS ....

....fully and correctly recognised sentence. In this case, performance scores evaluated by the NISTsoftware tell little about the overall performance of the system. In 1991 a new simple metric for computing overall performance of speech understanding systems was introduced within the DARPA community [Goodine et al. 1991, p. 846] This metric presupposes a test database with a number of sampled sentences and an equal number of answers from the system. The answers are classified into three groups: Correct answers, incorrect answers, and No answers. The latter group might in a specific dialogue system be expressed ....

[Article contains additional citation context not shown here]

Goodine, D., S. Seneff, L. Hirschman, M. Philipps : "Full Integration of Speech and Language Understanding in the MIT Spoken Language System". in proc. Eurospeech 2 (1991), pp. 845-848.

....is extra work to do. 6 Conclusion One effect of close integration of syntax and acoustics was that word error rates became essentially meaningless due to the fact that utterances were either recognized correctly or not recognized at all. This has also been observed in Goodine et al. 91 [3]. One might argue that this sacrifices flexibility compared to an approach were acoustic decoding is done like a filter before any linguistic analysis begins. However, measuring word accuracy in unparsable word sequences does not make sense in a speech understanding system which relies on parsing ....

David Goodine, Stephanie Seneff, Lynette Hirschman, and Michael Phillips. Full integration of speech and language understanding in the MIT spoken language system. In Proc. Eurospeech 1991, pages 845--848, 1991.

....from using discoursecontext information in the LM. 1. TIGHT COUPLING A number of researchers have proposed ways to use naturallanguage backend information in the speech recognition process. Moore et al. 1989) used a unification based CFG to generate word transitions for a Viterbi recognizer. Goodine et al. 1991) describe a system which uses the CFG based TINA parser to predict next words for the SUMMIT speech recognizer, Kita Ward (1991) used a CFG to filter bigram follow sets for the Sphinx recognizer. Hauenstein Weber (1994) also used a unification based CFG to filter bigram follow sets. In all ....

GOODINE, DAVID, STEPHANIE SENEFF, LYNETTE HIRSCHMAN, & MICHAEL PHILLIPS. 1991. Full integration of speech and language understanding in the MIT spoken language system. In Proceedings of Eurospeech 91, 24---26, Genova, Italy.

....from MIT. The SUMMIT speech recognition system and the TINA language understanding system can be run in several configurations. In the most tightly coupled configuration, TINA s parser is called interactively during the recognizer s search phase to prune impossible theories from the search space [Goodine 91] The TINA language understanding system currently includes only a fairly traditional context free grammar and does not yet incorporate dialogue level knowledge. Still, this architecture is promising because of its potential for incorporating multiple higher level knowledge sources into the ....

David Goodine, Stephanie Seneff, Lynette Hirschman, and Michael Phillips, "Full Integration of Speech and Language Understanding in the MIT Spoken Language System," Second European Conference on Speech Communication and Technology (Sept. 1991).

....from using discourse context information in the LM. 1. TIGHT COUPLING A number of researchers haveproposedways to use natural languagebackend information in the speech recognition process. Moore et al. 1989) used a unification based CFG to generate word transitions for a Viterbi recognizer. Goodine et al. 1991) describe a system which uses the CFG based TINA parser to predict next words for the SUMMIT speech recognizer, Kita Ward (1991) used a CFG to filter bigram follow sets for the Sphinx recognizer. Hauenstein Weber (1994) also used a unification based CFG to filter bigram follow sets. In all ....

GOODINE, DAVID, STEPHANIE SENEFF, LYNETTEHIRSCHMAN, &MICHAEL PHILLIPS. 1991. Full integration of speech and language understanding in the MIT spoken language system. In Proceedings of Eurospeech 91, 24---26, Genova, Italy.

....recognition results. In several approaches statistical information derived from corpora has been used. In recent work it is attempted to make extensive use of linguistic knowledge in speech recognition by automatic transformations [10] and a closer interaction between recognition and understanding [5]. However, up to now only linguistic knowledge has been applied. In contrast, we enhance the knowledge we use for building up the language models by dynamically inferred information from a non linguistic context and use it to improve speech recognition results. The source for the knowledge is the ....

David Goodine, Stephanie Seneff, Lynette Hirschmann, Michael Philips, Full Integration of Speech and Language Understanding in the MIT Spoken Language System. Proc. European Conf. on Speech Technology, 1991, pp. 845--848.

No context found.

GOODINE, DAVID, STEPHANIE SENEFF, LYNETTE HIRSCHMAN, & MICHAEL PHILLIPS. 1991. Full integration of speech and language understanding in the MIT spoken language system. In Proceedings of Eurospeech 91, 24---26, Genova, Italy.

No context found.

D. Goodine, S. Sene#, L. Hirschman, and M. Phillips, "Full Integration of Speech and Language Understanding in the MIT Spoken Language System," Proc. Eurospeech, 845--848, 1991.

No context found.

D. Goodine, S. Seneff, L. Hirschman, and M. Phillips. Full integration of speech and language understanding in the mit spoken language system. In Proc. European Conf. on Speech Communication and Technology, pages 845--848, 1991.

No context found.

D. Goodine, S. Sene#, L. Hirschmann, and M. Philips. Full Integration of Speech and Language Understanding in the MIT Spoken Language System. In Proc. European Conf. on Speech Communication and Technology, pages 845#848, 1991.

No context found.

Goodine, D., Sene#, S., Hirschman, L., Phillips, M. "Full Integration of Speech and Language Understanding in the MIT Spoken Language System," Proc. Eurospeech, 845--848, 1991.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC