Interactive Model with Structural Loss for Language-based Abductive Reasoning

Linhao Li, Ming Xu, Yongfeng Dong, Xin Li, Jianhua Tao, Qinghua Hu

Keywords: abductive natural language, abductive reasoning, BiLSTM, joint loss function.

Abstract: The abductive natural language inference task ($\alpha$NLI) is proposed to infer the most plausible explanation between the cause and the event. In the $\alpha$NLI task, two observations are given, and the most plausible hypothesis is asked to pick out from the candidates. Existing methods model the relation between each candidate hypothesis separately and penalize the inference network uniformly. In this paper, we argue that it is unnecessary to distinguish the reasoning abilities among correct hypotheses; and similarly, all wrong hypotheses contribute the same when explaining the reasons of the observations. Therefore, we propose to group instead of ranking the hypotheses and design a structural loss called "joint softmax focal loss" in this paper. Based on the observation that the hypotheses are generally semantically related, we have designed a novel interactive language model aiming at exploiting the rich interaction among competing hypotheses. We name this new model for $\alpha$NLI: Interactive Model with Structural Loss (IMSL). The experimental results show that our IMSL has achieved the highest performance on the RoBERTa-large pretrained model, with ACC and AUC results increased by about 1% and 5% respectively.

One-sentence Summary: For abductive NLI task, on one hand, we establish a way to reorganize the hypotheses groups and construct a joint loss for this; on the other hand, we propose a novel interactive language model that exploits the rich interaction among hypotheses.