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


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(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: Measuring the `I don't know' Problem through the Lens of Gricean Quantity

Abstract: We consider the intrinsic evaluation of neural generative dialog models through the lens of Grices Maxims of Conversation (1975). Based on the maxim of Quantity (be informative), we propose Relative Utterance Quantity (RUQ) to diagnose the `I don't know' problem. The RUQ diagnostic compares the model score of a generic response to that of the reference response. We find that for reasonable baseline models, `I don't know' is preferred over the reference more than half the time, but this can be mitigated with hyperparameter tuning.
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2010.12786 [cs.CL]
  (or arXiv:2010.12786v1 [cs.CL] for this version)

Submission history

From: Huda Khayrallah [view email]
[v1] Sat, 24 Oct 2020 05:16:36 GMT (56kb,D)

Link back to: arXiv, form interface, contact.