Technical report on Conversational Question Answering

TL;DR

This work presents a RoBERTa-based conversational QA framework incorporating rationale tagging, adversarial training, virtual adversarial training, and knowledge distillation to enhance robustness and performance on CoQA. It introduces targeted multi-task learning, perturbation-based regularization, and teacher-student guidance, complemented by post-processing and GA-driven ensembling. The approach achieves a new state-of-the-art single-model $F1$ of $90.4$ on CoQA and $90.7$ with ensembling, while providing upper-bound analyses that reveal limited headroom for extractive methods. The study also analyzes data characteristics and error types, outlining practical future directions for hybrid extractive-generative systems and knowledge-enriched QA.

Abstract

Conversational Question Answering is a challenging task since it requires understanding of conversational history. In this project, we propose a new system RoBERTa + AT +KD, which involves rationale tagging multi-task, adversarial training, knowledge distillation and a linguistic post-process strategy. Our single model achieves 90.4(F1) on the CoQA test set without data augmentation, outperforming the current state-of-the-art single model by 2.6% F1.

Figures (1)

• Figure 1: F1 score of best and worst individual of each generation.