## Dynamic Use Of Syntactical Knowledge In Continuous Speech Recognition (1993)

Venue: | in Proceedings Third European Conference on Speech Communication and Technology |

Citations: | 5 - 0 self |

### BibTeX

@INPROCEEDINGS{Dupont93dynamicuse,

author = {Pierre Dupont},

title = {Dynamic Use Of Syntactical Knowledge In Continuous Speech Recognition},

booktitle = {in Proceedings Third European Conference on Speech Communication and Technology},

year = {1993},

pages = {1959--1962}

}

### OpenURL

### Abstract

The control of continuous speech recognition by a context-free based language model requires a parsing process which may overload the acoustic decoding algorithm. We present a new approach to integrate such a language model in the search process. This approach extends the beam search Viterbi algorithm. In our case, the pruning technique not only selects the most likely acoustic hypotheses but also governs the dynamic expansion of a network structure. This algorithm is general enough to cope with the self-embedded recursivity of context-free languages and it favourably compares with other parsing techniques applied to spoken inputs. We present results which show that the syntactical knowledge may be efficiently included at the frame level of an acoustic decoding algorithm. Keywords: Continuous Speech Recognition, Context-Free Language Models, Beam Search Viterbi Algorithm. 1. INTRODUCTION Many continuous speech understanding systems use different language models for recognition, on the ...

### Citations

652 | An efficient context-free parsing algorithm
- Earley
- 1970
(Show Context)
Citation Context ...nchronous Viterbi search [8, 9, 10]. The LR parsing approach suffers from the drawback that it only covers a subclass of context-free grammars [7]. Chart parsing techniques, like the Earley algorithm =-=[11]-=- or the CYK algorithm [12] have a time complexity which is cubic in the length of the input sentence when handling incorrect data [9]. The beam search strategy is furthermore difficult to apply with t... |

374 |
The estimation of stochastic context-free grammars using the inside-outside algorithm. Computer Speech and Language
- Lari, Young
- 1990
(Show Context)
Citation Context ... Moreover, the SCFGs could bridge the gap between language models used for recognition and those used for understanding. The problem of estimating the rule probabilities has been already investigated =-=[2, 3, 4]-=-. In this paper, we cope with the integration of such a language model into a continuous speech recognizer. The processing of general context-free grammars requires a recursive parsing algorithm which... |

203 |
The Theory of Parsing
- Aho, Ullman
- 1972
(Show Context)
Citation Context ...ORMALIZATION A left-recursive grammar would cause the dynamic expansion to fail in a recursive loop. To avoid this problem, the language model has to be translated in Greibach Normal Form. Definition =-=[7]: A -=-CFG G = (N, S, P, S) is said to be in Greibach Normal Form (GNF) if G is e-free and each non-e-production in P is of the form A --> aa with a � S and a � N*. Theorem: Any context-free language may... |

153 |
Recognition and parsing of context-free languages in time n 3
- Younger
- 1967
(Show Context)
Citation Context ...8, 9, 10]. The LR parsing approach suffers from the drawback that it only covers a subclass of context-free grammars [7]. Chart parsing techniques, like the Earley algorithm [11] or the CYK algorithm =-=[12]-=- have a time complexity which is cubic in the length of the input sentence when handling incorrect data [9]. The beam search strategy is furthermore difficult to apply with those techniques in the cas... |

112 |
Basic Methods of Probabilistic Context Free Grammars
- Jelinek, Lafferty, et al.
- 1990
(Show Context)
Citation Context ... Moreover, the SCFGs could bridge the gap between language models used for recognition and those used for understanding. The problem of estimating the rule probabilities has been already investigated =-=[2, 3, 4]-=-. In this paper, we cope with the integration of such a language model into a continuous speech recognizer. The processing of general context-free grammars requires a recursive parsing algorithm which... |

59 |
Word graphs: an efficient interface between continuous-speech recognition and language understanding
- Oerder, Ney
- 1993
(Show Context)
Citation Context ...rd lattice (or N-best word sequences), obtained in a first step without accurate language modelling. Word graphs have been recently proposed as another interface between recognition and understanding =-=[5]-=-. Correction procedures may also be applied to recover the uttered sentence from a garbled word sequence. However, all these two-step approaches do not take full advantage of the language modelling si... |

28 |
Stochastic Grammars and Pattern Recognition
- Ney
- 1992
(Show Context)
Citation Context ...ed. They are of great interest in virtue of their ability to represent linguistic constraints in a stochastic formalism. This offers a unified framework for acoustic and syntactic modelling of speech =-=[1]-=-. Moreover, the SCFGs could bridge the gap between language models used for recognition and those used for understanding. The problem of estimating the rule probabilities has been already investigated... |

28 |
Dynamic Programming Parsing for Context-Free Grammars in Continuous Speech Recognition
- Ney
- 1991
(Show Context)
Citation Context ... when the underlying grammar is expressed in GNF. 5. COMPARISON WITH SIMILAR APPROACHES Several other attempts have been made at introducing a parsing algorithm into a time-synchronous Viterbi search =-=[8, 9, 10]-=-. The LR parsing approach suffers from the drawback that it only covers a subclass of context-free grammars [7]. Chart parsing techniques, like the Earley algorithm [11] or the CYK algorithm [12] have... |

7 |
Incorporating LR parsing into SPHINX
- WARD
- 1991
(Show Context)
Citation Context ... when the underlying grammar is expressed in GNF. 5. COMPARISON WITH SIMILAR APPROACHES Several other attempts have been made at introducing a parsing algorithm into a time-synchronous Viterbi search =-=[8, 9, 10]-=-. The LR parsing approach suffers from the drawback that it only covers a subclass of context-free grammars [7]. Chart parsing techniques, like the Earley algorithm [11] or the CYK algorithm [12] have... |

6 |
Some results on stochastic language modelling
- Mori, Kuhn
- 1991
(Show Context)
Citation Context ... Moreover, the SCFGs could bridge the gap between language models used for recognition and those used for understanding. The problem of estimating the rule probabilities has been already investigated =-=[2, 3, 4]-=-. In this paper, we cope with the integration of such a language model into a continuous speech recognizer. The processing of general context-free grammars requires a recursive parsing algorithm which... |

2 |
On the Modelization of Allophones
- Jouvet, Bartkova, et al.
- 1991
(Show Context)
Citation Context ...sentation embodies the acoustic models, the phonological and lexical knowledge. It corresponds to a network of HMMs being associated with each word. In the PHIL90 continuous speech recognition system =-=[6]-=-, the linguistic knowledge is included in the stochastic network representation. As the language model constraints (or assigns probabilities to) possible word sequences, word models are accordingly li... |

2 |
A Unification-Grammar-directed One-Pass Search Algorithm for Parsing Spoken Language
- Okada
- 1991
(Show Context)
Citation Context ... when the underlying grammar is expressed in GNF. 5. COMPARISON WITH SIMILAR APPROACHES Several other attempts have been made at introducing a parsing algorithm into a time-synchronous Viterbi search =-=[8, 9, 10]-=-. The LR parsing approach suffers from the drawback that it only covers a subclass of context-free grammars [7]. Chart parsing techniques, like the Earley algorithm [11] or the CYK algorithm [12] have... |

2 |
SpeakerIndependent Spelling Recognition over the Telephone
- Jouvet, Lain�, et al.
- 1993
(Show Context)
Citation Context ...s the results obtained on a test database of 1,641 city name spellings. The first line refers to unconstrained letter sequence recognition. The second line refers to a two-step approach (described in =-=[14]-=-) whichsconsists in correcting the garbled letter sequence from a known list of 5,000 city names. The most likely city name is computed by using a discrete HMM, given the recognized letter sequence. T... |

1 |
Flexible Use of
- Ward, Young
- 1993
(Show Context)
Citation Context ...s not affect the parsing complexity and, on the contrary, may help to better prune the search. Note that the proposed approach can be used for applying semantic constraints on spontaneous speech (see =-=[13]-=-, for example). Indeed, the general formalism of context-free grammars is not bound down to represent purely syntactical knowledge. Here again, we would take advantage of the MTN representation and of... |