Tree-structured Attention with Hierarchical Accumulation
This addresses the challenge of efficiently incorporating hierarchical linguistic structures into NLP models, which is incremental as it builds on existing Transformer and Tree-LSTM approaches.
The paper tackled the problem of encoding hierarchical structures like constituency trees into efficient self-attention models, bridging the gap between sequence-based Transformers and tree-based models. The result was a method that outperformed state-of-the-art methods in multiple translation tasks, achieving improvements in IWSLT and WMT'14 English-German translation, and also enhanced performance on text classification tasks.
Incorporating hierarchical structures like constituency trees has been shown to be effective for various natural language processing (NLP) tasks. However, it is evident that state-of-the-art (SOTA) sequence-based models like the Transformer struggle to encode such structures inherently. On the other hand, dedicated models like the Tree-LSTM, while explicitly modeling hierarchical structures, do not perform as efficiently as the Transformer. In this paper, we attempt to bridge this gap with "Hierarchical Accumulation" to encode parse tree structures into self-attention at constant time complexity. Our approach outperforms SOTA methods in four IWSLT translation tasks and the WMT'14 English-German translation task. It also yields improvements over Transformer and Tree-LSTM on three text classification tasks. We further demonstrate that using hierarchical priors can compensate for data shortage, and that our model prefers phrase-level attentions over token-level attentions.