WORD SET PROBABILITY BOOSTING FOR IMPROVED SPONTANEOUS DIALOG RECOGNITION

Based on the observation that the unpredictable nature of conversation al speech makes it almost impossible to reliably model sequential word constraints, the notion of word set error criteria is proposed for im proved recognition of spontaneous dialogs. The single-pass adaptive bo osting (AB) algorithm enables the language model weights to be tuned u sing the word set error criteria. In the two-pass version of the algor ithm, the basic idea is to predict a set of words based on some a prio ri information, and perform a rescoring pass wherein the probabilities of the words in the predicted word set are amplified or boosted in so me manner. An adaptive gradient descent procedure for tuning the word boosting factor is formulated, which enables the boost factors to be i ncrementally adjusted to maximize accuracy of the speech recognition s ystem outputs on held-out training data using the word set error crite ria. Two novel models which predict the required word sets are present ed: i) utterance triggers, which capture within-utterance long-distanc e word interdependencies, and ii) dialog triggers, which capture local temporal dialog-oriented word relations. The proposed trigger and ada ptive boosting (TAB) algorithm, and the single-pass adaptive boosting (AB) algorithm are experimentally tested on a subset of the TRAINS-93 spontaneous dialogs and the TRAINS-95 semispontaneous corpus, and resu lts summarized.