Discriminative language modeling with conditional random fields and the perceptron algorithm Export

@inproceedings{citeulike:3662892, abstract = {This paper describes discriminative language modeling for a large vocabulary speech recognition task. We contrast two parameter estimation methods: the perceptron algorithm, and a method based on conditional random fields (CRFs). The models are encoded as deterministic weighted finite state automata, and are applied by intersecting the automata with word-lattices that are the output from a baseline recognizer. The perceptron algorithm has the benefit of automatically selecting a relatively small feature set in just a couple of passes over the training data. However, using the feature set output from the perceptron algorithm (initialized with their weights), CRF training provides an additional 0.5\% reduction in word error rate, for a total 1.8\% absolute reduction from the baseline of 39.2\%.}, address = {Morristown, NJ, USA}, author = {Roark, Brian and Saraclar, Murat and Collins, Michael and Johnson, Mark}, booktitle = {ACL '04: Proceedings of the 42nd Annual Meeting on Association for Computational Linguistics}, citeulike-article-id = {3662892}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1218955.1218962}, citeulike-linkout-1 = {http://dx.doi.org/10.3115/1218955.1218962}, doi = {10.3115/1218955.1218962}, keywords = {discriminative-training, language-modeling, perceptron}, location = {Barcelona, Spain}, posted-at = {2008-11-23 02:24:50}, priority = {2}, publisher = {Association for Computational Linguistics}, title = {Discriminative language modeling with conditional random fields and the perceptron algorithm}, url = {http://dx.doi.org/10.3115/1218955.1218962}, year = {2004} }

TY - CONF ID - citeulike:3662892 L3 - citeulike-article-id:3662892 N2 - This paper describes discriminative language modeling for a large vocabulary speech recognition task. We contrast two parameter estimation methods: the perceptron algorithm, and a method based on conditional random fields (CRFs). The models are encoded as deterministic weighted finite state automata, and are applied by intersecting the automata with word-lattices that are the output from a baseline recognizer. The perceptron algorithm has the benefit of automatically selecting a relatively small feature set in just a couple of passes over the training data. However, using the feature set output from the perceptron algorithm (initialized with their weights), CRF training provides an additional 0.5% reduction in word error rate, for a total 1.8% absolute reduction from the baseline of 39.2%. CY - Barcelona, Spain AD - Morristown, NJ, USA TI - Discriminative language modeling with conditional random fields and the perceptron algorithm PB - Association for Computational Linguistics T2 - ACL '04: Proceedings of the 42nd Annual Meeting on Association for Computational Linguistics KW - discriminative-training KW - language-modeling KW - perceptron AU - Roark, Brian AU - Saraclar, Murat AU - Collins, Michael AU - Johnson, Mark PY - 2004/// UR - http://dx.doi.org/10.3115/1218955.1218962 ER -