Logic-Driven Context Extension and Data Augmentation for Logical Reasoning of Text

Abstract

Logical reasoning of text requires identifying critical logical structures in the text and performing inference over them. Existing methods for logical reasoning mainly focus on contextual semantics of text while struggling to explicitly model the logical inference process. In this paper, we not only put forward a logic-driven context extension framework but also propose a logic-driven data augmentation algorithm. The former follows a three-step reasoning paradigm, and each step is respectively to extract logical expressions as elementary reasoning units, symbolically infer the implicit expressions following equivalence laws and extend the context to validate the options. The latter augments literally similar but logically different instances and incorporates contrastive learning to better capture logical information, especially logical negative and conditional relationships. We conduct experiments on two benchmark datasets, ReClor and LogiQA. The results show that our method achieves state-of-the-art performance on both datasets, and even surpasses human performance on the ReClor dataset. 1

Analytical Reasoning of Text

“…works (Gao et al, 2021;Rajpurkar et al, 2016;Welbl et al, 2018a;Yang et al, 2018a;Huang et al, 2019a;Wang et al, 2021) that examine the ability of logical reasoning. LogiQA (Liu et al, 2020b) and ReClor (Yu et al, 2020) are sourced from examination in realistic scenario and examine a range of logical reasoning skills.…”

MERIt: Meta-Path Guided Contrastive Learning for Logical Reasoning

“…Differently, Betz et al (2021) and Clark et al (2020) used synthetically generated datasets to prove that the Transformer (Vaswani et al, 2017) or pre-trained GPT-2 is able to perform complex reasoning, motivating following researchers to introduce symbolic rules into neural models. For example, Wang et al (2022) developed a context extension and data augmentation framework, which is based on the extracted logical expressions. Superior performance over its contenders can be observed on the ReClor dataset.…”

“…We evaluated our method on two challenging logical reasoning benchmarks, i.e., LogiQA and Re-Clor, with several strong baselines, including the pre-trained language models, DAGN , Focal Reasoner (Ouyang et al, 2021) and LReasoner (Wang et al, 2022). For more details, please refer to Appendix B.…”

“…For instance, and Ouyang et al (2021) first constructed a graph of different constituents and then performed implicit reasoning with graph neural networks (GNNs). Wang et al (2022) proposed LReasoner, a unified context extension and data augmentation framework based on the parsed logical expressions.…”

“…It captures both global connections between facts and the local concepts or actions inside the fact. LReasoner(Wang et al, 2022) includes a context extension framework and a data augmentation algorithm, which are all conducted based on the extracted logical expressions. This method has achieved new state-of-the-art performance on Re-Clor recently.…”