Simulated annealing: theory and applications
Simulated annealing: theory and applications
On smoothing techniques for bigram-based natural language modelling
ICASSP '91 Proceedings of the Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference
Hi-index | 0.00 |
Continuous speech recognition on very large vocabularies can be improved in theory using a language model specifying the a priori conditional probability of finding a word given a word sequence. As this seems utopic to implement in practice, more realistic solutions have been proposed, as the determination of n-gram word models [3] or of n-gram class models [4], limiting the length of the word sequence to n items. We built a bigram class model which gives the probability of a word class given its predecessor class. A stochastic method, namely simulated annealing, is used to automatically classify the words of large text corpora. We present here a first validation of the use of simulated annealing in language modelling. Results are presented' using respectively a French corpus of 40000 words and a German corpus of 100000 words and a comparison with another method of statistical clustering is exhibited.