Segmental Recurrent Neural Networks

TL;DR

Segmental Recurrent Neural Networks (SRNNs) are introduced to jointly model input segmentations and labelings, improving accuracy in handwriting recognition and Chinese word segmentation tasks by explicitly representing segments.

Contribution

The paper presents SRNNs, a novel neural network architecture that explicitly models segments and their labels, integrating bidirectional RNNs with semi-Markov CRFs for improved sequence labeling.

Findings

SRNNs outperform models without explicit segment representation in accuracy.

Effective in handwriting recognition and Chinese word segmentation.

Supports both fully and partially supervised training.

Abstract

We introduce segmental recurrent neural networks (SRNNs) which define, given an input sequence, a joint probability distribution over segmentations of the input and labelings of the segments. Representations of the input segments (i.e., contiguous subsequences of the input) are computed by encoding their constituent tokens using bidirectional recurrent neural nets, and these "segment embeddings" are used to define compatibility scores with output labels. These local compatibility scores are integrated using a global semi-Markov conditional random field. Both fully supervised training -- in which segment boundaries and labels are observed -- as well as partially supervised training -- in which segment boundaries are latent -- are straightforward. Experiments on handwriting recognition and joint Chinese word segmentation/POS tagging show that, compared to models that do not explicitly represent segments such as BIO tagging schemes and connectionist temporal classification (CTC), SRNNs obtain substantially higher accuracies.