How to Evaluate the Next System: Automatic Dialogue Evaluation from the Perspective of Continual Learning
This work addresses the need for lifelong and low-cost evaluation in open-domain dialogue research, though it is incremental as it builds on existing neural evaluators with a novel adaptation method.
The paper tackles the problem of evaluation bias in automatic dialogue evaluation by proposing a continual learning approach that fine-tunes existing neural evaluators to adapt to new dialogue systems, achieving comparable performance to reconstructing new evaluators while using significantly lower resources.
Automatic dialogue evaluation plays a crucial role in open-domain dialogue research. Previous works train neural networks with limited annotation for conducting automatic dialogue evaluation, which would naturally affect the evaluation fairness as dialogue systems close to the scope of training corpus would have more preference than the other ones. In this paper, we study alleviating this problem from the perspective of continual learning: given an existing neural dialogue evaluator and the next system to be evaluated, we fine-tune the learned neural evaluator by selectively forgetting/updating its parameters, to jointly fit dialogue systems have been and will be evaluated. Our motivation is to seek for a lifelong and low-cost automatic evaluation for dialogue systems, rather than to reconstruct the evaluator over and over again. Experimental results show that our continual evaluator achieves comparable performance with reconstructing new evaluators, while requires significantly lower resources.