Lexically Constrained Decoding for Sequence Generation Using Grid Beam Search

TL;DR

The paper introduces Grid Beam Search (GBS), a flexible decoding algorithm that incorporates lexical constraints into sequence generation, improving translation quality in interactive and domain adaptation scenarios without modifying models.

Contribution

We propose GBS, a novel decoding algorithm that allows the inclusion of lexical constraints in sequence generation without altering model training or parameters.

Findings

GBS improves translation quality in interactive scenarios.

GBS achieves significant gains in domain adaptation.

The method is compatible with any sequence-generating model.

Abstract

We present Grid Beam Search (GBS), an algorithm which extends beam search to allow the inclusion of pre-specified lexical constraints. The algorithm can be used with any model that generates a sequence $ \mathbf{\hat{y}} = \{y_{0}\ldots y_{T}\} $, by maximizing $ p(\mathbf{y} | \mathbf{x}) = \prod\limits_{t}p(y_{t} | \mathbf{x}; \{y_{0} \ldots y_{t-1}\}) $. Lexical constraints take the form of phrases or words that must be present in the output sequence. This is a very general way to incorporate additional knowledge into a model's output without requiring any modification of the model parameters or training data. We demonstrate the feasibility and flexibility of Lexically Constrained Decoding by conducting experiments on Neural Interactive-Predictive Translation, as well as Domain Adaptation for Neural Machine Translation. Experiments show that GBS can provide large improvements in translation quality in interactive scenarios, and that, even without any user input, GBS can be used to achieve significant gains in performance in domain adaptation scenarios.