Investigating the Benefits of Free-Form Rationales

Anonymous

Abstract: Free-form rationales aim to aid model interpretability by supplying the background knowledge that can help understand model decisions. Crowdsourced rationales are provided for commonsense QA instances in popular datasets such as CoS-E and ECQA, but their utility remains under-investigated. We present human studies which show that ECQA rationales indeed provide additional information to understand a decision, while 70% of CoS-E rationales do not. Inspired by this finding, we ask: can the additional context provided by free-form rationales benefit models, similar to human users? We investigate the utility of rationales as an additional source of supervision, by varying the quantity and quality of rationales during training. After controlling for instances where rationales leak the correct answer, we find that incorporating only 5% of rationales during training can boost model performance by 16.89%. Moreover, we also show that rationale quality matters: compared to crowdsourced rationales, T5-generated rationales provide not only much weaker supervision to models, but are also not helpful for human users in aiding model interpretability.

Paper Type: long