Multi-layer Representation Fusion for Neural Machine Translation

Abstract

Neural machine translation systems require a number of stacked layers for deep models. But the prediction depends on the sentence representation of the top-most layer with no access to low-level representations. This makes it more difficult to train the model and poses a risk of information loss to prediction. In this paper, we propose a multi-layer representation fusion (MLRF) approach to fusing stacked layers. In particular, we design three fusion functions to learn a better representation from the stack. Experimental results show that our approach yields improvements of 0.92 and 0.56 BLEU points over the strong Transformer baseline on IWSLT German-English and NIST Chinese-English MT tasks respectively. The result is new state-of-the-art in German-English translation.

COLING 2018

References in corpus (3)

• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• A Structured Self-attentive Sentence Embedding
• An Actor-Critic Algorithm for Sequence Prediction

Cited by in corpus (12)

• Learning Deep Transformer Models for Machine Translation
• Improving Image Captioning by Leveraging Intra- and Inter-layer Global Representation in Transformer Network
• Dynamic Layer Aggregation for Neural Machine Translation with Routing-by-Agreement
• Understanding and Improving Encoder Layer Fusion in Sequence-to-Sequence Learning
• Acquiring Knowledge from Pre-trained Model to Neural Machine Translation
• Exploiting Sentential Context for Neural Machine Translation
• Graph-to-Sequence Neural Machine Translation
• Balancing Cost and Benefit with Tied-Multi Transformers
• Learning Architectures from an Extended Search Space for Language Modeling
• Learning Source Phrase Representations for Neural Machine Translation
• GRET: Global Representation Enhanced Transformer
• Layer-Wise Multi-View Learning for Neural Machine Translation