A Critical Analysis of EmpatheticDialogues as a Corpus for Empathetic Engagement

Alok Debnath, Owen Conlan

Abstract: This paper aims to analyze the content and relevance of one of the most popular contemporary training corpora for empathetic conversational agents: EmpatheticDialogues [23]. We provide a detailed qualitative breakdown of the corpus including the corpus creation methodology and point out some critical shortcomings of the corpus. Given the significance of the corpus as the only one of its kind at the moment, we also provide a quantitative comparison of EmpatheticDialogues to other contemporary small-talk corpora including DailyDialog and Persona-Chat, including conversation length, the ratio of conversant interaction, lexical choice, etc. With this analysis, we discuss the merit and implications of indicating a specific small-talk dialogue corpus is more empathetic than other small-talk corpora. Finally, we provide a new lens for developing conversational agents with empathetic engagement capabilities by augmenting existing dialogue datasets.