Fact-driven Logical Reasoning

Siru Ouyang, Zhuosheng Zhang, hai zhao

Keywords: logical reasoning, machine reading comprehension, language understanding

Abstract: Logical reasoning deeply relies on accurate, clearly presented clue forms which are usually modeled as entity-like knowledge in existing studies. However, in real hierarchical reasoning motivated machine reading comprehension (MRC), such one-side modeling are insufficient for those indispensable local complete facts or events when only "global" knowledge is really paid attention to. Thus, in view of language being a complete knowledge/clue carrier, we propose a general formalism to support representing logic units by extracting backbone constituents of the sentence such as the subject-verb-object formed "facts", covering both global and local knowledge pieces that are necessary as the basis for logical reasoning. Beyond building the ad-hoc graphs, we propose a more general and convenient fact-driven approach to construct a supergraph on top of our newly defined fact units, and enhance the supergraph with further explicit guidance of local question and option interactions. Experiments on two challenging logical reasoning MRC benchmarks show that our proposed model, \textsc{Focal Reasoner}, outperforms the baseline models dramatically.

Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.

Community Implementations: [ 2 code implementations](https://www.catalyzex.com/paper/arxiv:2105.10334/code)