Neuron Interaction Based Representation Composition for Neural Machine Translation

TL;DR

This paper introduces a novel neuron interaction-based method for neural machine translation that models pairwise neuron interactions to improve translation quality, outperforming the Transformer baseline on standard benchmarks.

Contribution

It proposes a bilinear pooling approach to model neuron interactions in NMT, enhancing linguistic representation and translation performance.

Findings

Improved translation accuracy over Transformer baseline.

Captures more syntactic and semantic information.

Effective modeling of neuron interactions enhances representations.

Abstract

Recent NLP studies reveal that substantial linguistic information can be attributed to single neurons, i.e., individual dimensions of the representation vectors. We hypothesize that modeling strong interactions among neurons helps to better capture complex information by composing the linguistic properties embedded in individual neurons. Starting from this intuition, we propose a novel approach to compose representations learned by different components in neural machine translation (e.g., multi-layer networks or multi-head attention), based on modeling strong interactions among neurons in the representation vectors. Specifically, we leverage bilinear pooling to model pairwise multiplicative interactions among individual neurons, and a low-rank approximation to make the model computationally feasible. We further propose extended bilinear pooling to incorporate first-order representations. Experiments on WMT14 English-German and English-French translation tasks show that our model consistently improves performances over the SOTA Transformer baseline. Further analyses demonstrate that our approach indeed captures more syntactic and semantic information as expected.