<i>Do not let the history haunt you</i> - Mitigating Compounding Errors in Conversational Question Answering

Mandya, Angrosh, O'Neill, James, Bollegala, Danushka ORCID: 0000-0003-4476-7003 and Coenen, Frans ORCID: 0000-0003-1026-6649 (2020) <i>Do not let the history haunt you</i> - Mitigating Compounding Errors in Conversational Question Answering. In: 12th International Conference on Language Resources and Evaluation, 2020-5-13 - 2020-4-15, Marseille.

Text SS_CoQA_LREC_2020.pdf - Author Accepted Manuscript Download (315kB) | Preview

Abstract

The Conversational Question Answering (CoQA) task involves answering a sequence of inter-related conversational questions about a contextual paragraph. Although existing approaches employ human-written ground-truth answers for answering conversational questions at test time, in a realistic scenario, the CoQA model will not have any access to ground-truth answers for the previous questions, compelling the model to rely upon its own previously predicted answers for answering the subsequent questions. In this paper, we find that compounding errors occur when using previously predicted answers at test time, significantly lowering the performance of CoQA systems. To solve this problem, we propose a sampling strategy that dynamically selects between target answers and model predictions during training, thereby closely simulating the situation at test time. Further, we analyse the severity of this phenomena as a function of the question type, conversation length and domain type.

Item Type:	Conference or Workshop Item (Unspecified)
Uncontrolled Keywords:	Scheduled Sampling, Conversational Question Answering, CoQA
Depositing User:	Symplectic Admin
Date Deposited:	13 May 2020 10:22
Last Modified:	19 Oct 2023 08:47
Related URLs:	Author
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3085264