Neural Finite-State Transducers: Beyond Rational Relations
Chu-Cheng
Lin
author
Hao
Zhu
author
Matthew
R
Gormley
author
Jason
Eisner
author
2019-jun
text
Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)
Association for Computational Linguistics
Minneapolis, Minnesota
conference publication
We introduce neural finite state transducers (NFSTs), a family of string transduction models defining joint and conditional probability distributions over pairs of strings. The probability of a string pair is obtained by marginalizing over all its accepting paths in a finite state transducer. In contrast to ordinary weighted FSTs, however, each path is scored using an arbitrary function such as a recurrent neural network, which breaks the usual conditional independence assumption (Markov property). NFSTs are more powerful than previous finite-state models with neural features (Rastogi et al., 2016.) We present training and inference algorithms for locally and globally normalized variants of NFSTs. In experiments on different transduction tasks, they compete favorably against seq2seq models while offering interpretable paths that correspond to hard monotonic alignments.
lin-etal-2019-neural
10.18653/v1/N19-1024
https://www.aclweb.org/anthology/N19-1024
2019-jun
272
283