Textual Entailment with Structured Attentions and Composition
This work addresses the problem of improving textual entailment accuracy for NLP researchers by introducing a novel method that integrates syntax and recursion into attention models, though it is incremental as it builds on prior attention-based approaches.
The paper tackles the limitation of existing attention models in textual entailment by ignoring syntax and recursion, proposing a structured attention model that operates on tree nodes and composes subtree-level entailment relations, resulting in significant improvements in accuracy.
Deep learning techniques are increasingly popular in the textual entailment task, overcoming the fragility of traditional discrete models with hard alignments and logics. In particular, the recently proposed attention models (Rocktäschel et al., 2015; Wang and Jiang, 2015) achieves state-of-the-art accuracy by computing soft word alignments between the premise and hypothesis sentences. However, there remains a major limitation: this line of work completely ignores syntax and recursion, which is helpful in many traditional efforts. We show that it is beneficial to extend the attention model to tree nodes between premise and hypothesis. More importantly, this subtree-level attention reveals information about entailment relation. We study the recursive composition of this subtree-level entailment relation, which can be viewed as a soft version of the Natural Logic framework (MacCartney and Manning, 2009). Experiments show that our structured attention and entailment composition model can correctly identify and infer entailment relations from the bottom up, and bring significant improvements in accuracy.