CODAH: An Adversarially Authored Question-Answer Dataset for Common Sense

TL;DR

CODAH introduces an adversarially-authored commonsense QA dataset that extends SWAG to stress-test modern QA models beyond conventional benchmarks. By training annotators to target model weaknesses and rewarding those who fool the models pre- and post-fine-tuning, CODAH yields 2,801 questions that reveal substantial gaps between human performance (95.3%) and state-of-the-art models (roughly mid-60s). The study analyzes category-specific difficulties (notably quantitative reasoning and negation) and discusses artifacts and dataset-size effects, arguing for aggregating diverse datasets and updating data collection as models improve. Overall, CODAH provides a robust, complementary benchmark to SWAG that highlights directions for making commonsense QA systems more robust and human-like.

Abstract

Commonsense reasoning is a critical AI capability, but it is difficult to construct challenging datasets that test common sense. Recent neural question answering systems, based on large pre-trained models of language, have already achieved near-human-level performance on commonsense knowledge benchmarks. These systems do not possess human-level common sense, but are able to exploit limitations of the datasets to achieve human-level scores. We introduce the CODAH dataset, an adversarially-constructed evaluation dataset for testing common sense. CODAH forms a challenging extension to the recently-proposed SWAG dataset, which tests commonsense knowledge using sentence-completion questions that describe situations observed in video. To produce a more difficult dataset, we introduce a novel procedure for question acquisition in which workers author questions designed to target weaknesses of state-of-the-art neural question answering systems. Workers are rewarded for submissions that models fail to answer correctly both before and after fine-tuning (in cross-validation). We create 2.8k questions via this procedure and evaluate the performance of multiple state-of-the-art question answering systems on our dataset. We observe a significant gap between human performance, which is 95.3%, and the performance of the best baseline accuracy of 67.5% by the BERT-Large model.