Neural Conversational QA: Learning to Reason v.s. Exploiting Patterns
This addresses dataset bias issues in conversational QA for researchers, though it is incremental as it focuses on dataset refinement rather than novel methods.
The paper identifies that neural models on the ShARC conversational QA task exploit spurious patterns in the dataset rather than learning to reason, and it creates a modified dataset with fewer patterns to improve model learning.
Neural Conversational QA tasks like ShARC require systems to answer questions based on the contents of a given passage. On studying recent state-of-the-art models on the ShARCQA task, we found indications that the models learn spurious clues/patterns in the dataset. Furthermore, we show that a heuristic-based program designed to exploit these patterns can have performance comparable to that of the neural models. In this paper we share our findings about four types of patterns found in the ShARC corpus and describe how neural models exploit them. Motivated by the aforementioned findings, we create and share a modified dataset that has fewer spurious patterns, consequently allowing models to learn better.