Enhancing the Transformer Decoder with Transition-based Syntax
This work addresses syntactic generalization issues in machine translation, offering a novel method that is incremental in improving existing Transformer decoders.
The paper tackled the challenge of syntactic generalization in text decoders by proposing a transition-based approach for tree decoding, specifically incorporating Universal Dependencies syntax into machine translation, resulting in substantial improvements on syntactic generalization test sets and comparable performance on standard benchmarks.
Notwithstanding recent advances, syntactic generalization remains a challenge for text decoders. While some studies showed gains from incorporating source-side symbolic syntactic and semantic structure into text generation Transformers, very little work addressed the decoding of such structure. We propose a general approach for tree decoding using a transition-based approach. Examining the challenging test case of incorporating Universal Dependencies syntax into machine translation, we present substantial improvements on test sets that focus on syntactic generalization, while presenting improved or comparable performance on standard MT benchmarks. Further qualitative analysis addresses cases where syntactic generalization in the vanilla Transformer decoder is inadequate and demonstrates the advantages afforded by integrating syntactic information.