Generalizable Neuro-symbolic Systems for Commonsense Question Answering

TL;DR

This paper discusses neuro-symbolic models that combine neural language models with knowledge graphs to improve domain generalizability and robustness in commonsense question answering, supported by quantitative and qualitative evaluations.

Contribution

It introduces methods for integrating neural models with knowledge graphs and characterizes when this approach is most effective.

Findings

Enhanced domain generalizability in QA tasks

Robustness improvements demonstrated across datasets

Insights into effective neuro-symbolic integration

Abstract

This chapter illustrates how suitable neuro-symbolic models for language understanding can enable domain generalizability and robustness in downstream tasks. Different methods for integrating neural language models and knowledge graphs are discussed. The situations in which this combination is most appropriate are characterized, including quantitative evaluation and qualitative error analysis on a variety of commonsense question answering benchmark datasets.