We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CL

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Partner Personas Generation for Diverse Dialogue Generation

Abstract: Incorporating personas information allows diverse and engaging responses in dialogue response generation. Unfortunately, prior works have primarily focused on self personas and have overlooked the value of partner personas. Moreover, in practical applications, the availability of ground truth partner personas is often not the case. This paper attempts to tackle these issues by offering a novel framework that leverages automatic partner personas generation to enhance the succeeding dialogue generation. We incorporate reinforcement learning with a dedicatedly designed critic network for reward judgement. Experimental results from both automatic and human evaluation demonstrate a) Our framework is capable of generating relevant, informative and coherent partner personas, even compared to the ground truth partner personas. b) Generated partner personas enhance the succeeding response generation, thus surpassing our baselines and comparison model when partner personas are missing during the inference stage. c) Our framework generates responses that are more informative and engaging than our baseline conditioned on the ground truth partner personas during inference. d) Our dedicatedly designed critic network reinforces our framework effectively. Finally, our framework gives better explainability and reduces the demands for external databases for partner personas.
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2111.13833 [cs.CL]
  (or arXiv:2111.13833v1 [cs.CL] for this version)

Submission history

From: Hongyuan Lu [view email]
[v1] Sat, 27 Nov 2021 06:54:16 GMT (421kb,D)

Link back to: arXiv, form interface, contact.