Pan More Gold from the Sand: Refining Open-domain Dialogue Training with Noisy Self-Retrieval Generation. (arXiv:2201.11367v1 [cs.CL])

Web: http://arxiv.org/abs/2201.11367

Real human conversation data are complicated, heterogeneous, and noisy, from
whom building open-domain dialogue systems remains a challenging task. In fact,
such dialogue data can still contain a wealth of information and knowledge,
however, they are not fully explored. In this paper, we show existing
open-domain dialogue generation methods by memorizing context-response paired
data with causal or encode-decode language models underutilize the training
data. Different from current approaches, using external knowledge, we explore a
retrieval-generation training framework that can increase …

arxiv open retrieval training