Neural Lattice-to-Sequence Models for Uncertain Inputs
This work addresses uncertainty in inputs for sequence-to-sequence models, which is an incremental improvement for applications like speech translation.
The paper tackles the problem of error propagation from upstream models in neural sequence-to-sequence tasks by proposing a LatticeLSTM encoder that consumes word lattices with posterior probabilities. It reports consistent improvements over baselines in speech translation experiments.
The input to a neural sequence-to-sequence model is often determined by an up-stream system, e.g. a word segmenter, part of speech tagger, or speech recognizer. These up-stream models are potentially error-prone. Representing inputs through word lattices allows making this uncertainty explicit by capturing alternative sequences and their posterior probabilities in a compact form. In this work, we extend the TreeLSTM (Tai et al., 2015) into a LatticeLSTM that is able to consume word lattices, and can be used as encoder in an attentional encoder-decoder model. We integrate lattice posterior scores into this architecture by extending the TreeLSTM's child-sum and forget gates and introducing a bias term into the attention mechanism. We experiment with speech translation lattices and report consistent improvements over baselines that translate either the 1-best hypothesis or the lattice without posterior scores.