Publication Details

A Fast Re-scoring Strategy to Capture Long-Distance Dependencies

DEORAS, A.; MIKOLOV, T.; CHURCH, K. A Fast Re-scoring Strategy to Capture Long-Distance Dependencies. Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing July 2011 Edinburgh, Scotland, UK. Edinburgh: Association for Computational Linguistics, 2011. p. 1116-1127. ISBN: 978-1-937284-11-4.
Czech title
Strategie pro rychlé reskórování se závislostmi přes dlouhé kontexty
Type
conference paper
Language
English
Authors
Deoras Anoop
Mikolov Tomáš, Ing., Ph.D.
Church Kenneth
URL
Keywords

language model, re-scoring strategy, recurrent neural network

Abstract

The paper describes novel approach to lattice rescoring with complex lanaguage models with long-distance dependencies, such as recurrent neural network language models.

Annotation

A re-scoring strategy is proposed that makes it feasible to capture more long-distance dependencies in the natural language. Two pass strategies have become popular in a number of recognition tasks such as ASR (automatic speech recognition), MT (machine translation) and OCR (optical character recognition). The first pass typically applies a weak language model (n-grams) to a lattice and the second pass applies a stronger language model to N-best lists. The stronger language model is intended to capture more longdistance dependencies. The proposed method uses RNN-LM (recurrent neural network language model), which is a long span LM, to rescore word lattices in the second pass. A hill climbing method (iterative decoding) is proposed to search over islands of confusability in the word lattice. An evaluation based on Broadcast News shows speedups of 20 over basic N-best re-scoring, and word error rate reduction of 8% (relative) on a highly competitive setup.

Published
2011
Pages
1116–1127
Proceedings
Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing July 2011 Edinburgh, Scotland, UK
ISBN
978-1-937284-11-4
Publisher
Association for Computational Linguistics
Place
Edinburgh
BibTeX
@inproceedings{BUT76392,
  author="Anoop {Deoras} and Tomáš {Mikolov} and Kenneth {Church}",
  title="A Fast Re-scoring Strategy to Capture Long-Distance Dependencies",
  booktitle="Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing July 2011 Edinburgh, Scotland, UK",
  year="2011",
  pages="1116--1127",
  publisher="Association for Computational Linguistics",
  address="Edinburgh",
  isbn="978-1-937284-11-4",
  url="https://www.fit.vut.cz/research/publication/9687/"
}
Back to top