Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: On the Relation between Quality-Diversity Evaluation and Distribution-Fitting Goal in Text Generation
(Submitted on 3 Jul 2020 (v1), last revised 19 Aug 2020 (this version, v2))
Abstract: The goal of text generation models is to fit the underlying real probability distribution of text. For performance evaluation, quality and diversity metrics are usually applied. However, it is still not clear to what extend can the quality-diversity evaluation reflect the distribution-fitting goal. In this paper, we try to reveal such relation in a theoretical approach. We prove that under certain conditions, a linear combination of quality and diversity constitutes a divergence metric between the generated distribution and the real distribution. We also show that the commonly used BLEU/Self-BLEU metric pair fails to match any divergence metric, thus propose CR/NRR as a substitute for quality/diversity metric pair.
Submission history
From: Jianing Li [view email][v1] Fri, 3 Jul 2020 04:06:59 GMT (998kb,D)
[v2] Wed, 19 Aug 2020 03:37:59 GMT (1020kb,D)
Link back to: arXiv, form interface, contact.