Top-down string-to-dependency Neural Machine Translation

2026-03-29

AI Summary

This work addresses the limited generalization capability of neural machine translation systems when handling unseen or rare long sentences. To this end, the authors propose a novel syntactic decoder that, for the first time, integrates a top-down, left-to-right dependency tree generation mechanism into the decoding process. The approach seamlessly combines attention mechanisms with target-side dependency structures within a standard encoder–decoder framework, enabling direct mapping from source-language strings to target-language dependency trees. Experimental results demonstrate that the proposed model substantially enhances the modeling of long-range structural dependencies and achieves superior generalization performance and translation quality on unseen long-sentence translation tasks compared to conventional sequence-to-sequence approaches.

Abstract

Most of modern neural machine translation (NMT) models are based on an encoder-decoder framework with an attention mechanism. While they perform well on standard datasets, they can have trouble in translation of long inputs that are rare or unseen during training. Incorporating target syntax is one approach to dealing with such length-related problems. We propose a novel syntactic decoder that generates a target-language dependency tree in a top-down, left-to-right order. Experiments show that the proposed top-down string-to-tree decoding generalizes better than conventional sequence-to-sequence decoding in translating long inputs that are not observed in the training data.

Problem

• neural machine translation
• long inputs
• unseen data
• translation generalization
• syntactic decoding

Innovation

• top-down decoding
• dependency tree
• syntactic NMT
• string-to-tree translation
• neural machine translation