Tell Me Everything You Know: A Conversation Update System for the Rational Speech Acts Framework

What it is about

RSA models normally simulate one utterance; this paper makes them run over many turns. The common ground is a probability distribution over candidate worlds (here persons with features, from the Mutual Friends task), each participant also has private beliefs, and the common ground replaces the world prior inside the RSA speaker and listener. After each turn the pragmatic listener's posterior is added to the prior state with a learning rate (0.2 in the simulations); separate common grounds per participant let misunderstanding and divergence appear, and the learning rate can be raised when a listener's belief entropy is high (an unsure listener learns more). Speakers choose what to say by sampling from their beliefs; two fixes make this cooperative: a belief threshold with a null utterance (a speaker without a firm belief passes instead of contradicting), and a difference-based sampler that prefers the world whose probability the speaker would raise most over the common ground (no repeating what is already shared). There are only illustrative simulations, no data evaluation.

Ideas

• I1. Keep a common ground apart from the SLM's own memory in the dialogue layer (world/src/talk.rs, world/src/mind.rs): the set of frames already said in this conversation by either side. Use it as the paper's non-redundancy filter: an answer or explanation does not restate a frame that is already in the common ground, and it states first the remembered frame the other side has not yet heard. Deterministic: frames, not probabilities. Measure by counting repeated frames in the replies of a scripted dialogue test before and after.
• I2. Implement the threshold-and-pass rule for the student: when no remembered frame supports an answer above a fixed score margin, the student says it does not know instead of guessing (the «null utterance»). world knowledge diagnose already separates «no option supported (guess)» (29.4% on the Torah course); report the accuracy the student would have on the answered questions if it passed on those, so the trade-off is visible.
• I3. Track per-character belief sets in tales the same way (who saw or was told which frame), as a cheap step towards false-belief questions («Where will Sally look for the ball?»): a Say or perceive frame adds its content to the hearer's set; the narrator's frames go to the reader only. Test on a few hand-written false-belief stories first.
• I4. If prag/src/rsa.rs is extended to several turns, update the prior with exact rational arithmetic and a rational learning rate (for example 1/5), so multi-turn games keep the module's exact, reproducible tables.

For Mova

38. Dialogue, discourse markers and reported speech §3 (grounding and common ground), 33-implicature-and-presupposition.md §9 (RSA). Code: world/src/talk.rs, world/src/mind.rs, world/src/knowledge.rs (pass instead of guess), prag/src/rsa.rs. Measures: repeated frames in replies; accuracy on answered questions vs share passed; hand-written false-belief tests.

Topics: Speech act theory

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.7275/j6z3-9k74

@inproceedings{anderson2021tell,
  title = {{Tell Me Everything You Know: A Conversation Update System for the Rational Speech Acts Framework}},
  author = {Anderson, Carolyn Jane},
  booktitle = {Proceedings of the Society for Computation in Linguistics 2021},
  year = {2021},
  doi = {10.7275/j6z3-9k74},
  url = {https://doi.org/10.7275/j6z3-9k74}
}

References

27 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Carolyn Jane Anderson and Brian W. Dillon. 2019. Guess Who’s Coming (and Who’s Going): Bringing Perspective to the Rational Speech Acts Framework. In Proceedings of the Society for Computation in Linguistics, volume 2. find
2. Leon Bergen, Noah D. Goodman, and Roger Levy. 2012. That’s what she (could have) said: how alternative utterances affect language use. In Proceedings of the Annual Conference of the Cognitive Science Society, volume 34. find
3. Leon Bergen, Roger Levy, and Noah D. Goodman. 2016. Pragmatic Reasoning through Semantic Inference. Semantics & Pragmatics, 9(20). find
4. M. Bohn, Michael Henry Tessler, and Michael C. Frank. 2019. Integrating Common Ground and Informativeness in Pragmatic Word Learning. In Cognitive Science Society. find
5. Thomas Brochhagen, Michael Franke, and Robert van Rooij. 2016. Learning biases may prevent lexicalization of pragmatic inferences: a case study combining Alternatively, if the goal is perspective alignment, they should seek to minimize the Kullback–Leibler divergence of the Common Ground and their beliefs. iterated (Bayesian) learning and functional selection. In Cognitive Science Society. find
6. Reuben Cohn-Gordon and Leon Bergen. 2019. Verbal Irony, Pretense, and the Common Ground. Stanford. find
7. Reuben Cohn-Gordon and Ciyang Qing. 2018. Modeling “non-literal” social meaning with Bayesian pragmatics. In Sinn und Bedeutung. find
8. Elizabeth Coppock. 2018. Outlook-based semantics. Linguistics and Philosophy, 41:125–164. find
9. Judith Degen, Michael Henry Tessler, and Noah D. Goodman. 2015. Wonky worlds: Listeners revise world knowledge when utterances are odd. In Cognitive Science Society. find
10. Michael C. Frank and Noah D. Goodman. 2012. Predicting pragmatic reasoning in language games. Science, 6084(336):998–998. find
11. Martı́n Fuchs. 2020. On the Synchrony and Diachrony of the Spanish Imperfective Domain: Contextual Modulation and Semantic Change. Ms. New Haven, CT. find
12. Noah D. Goodman and Daniel Lassiter. 2014. Probabilistic Semantics and Pragmatics: Uncertainty in Language and Thought. In Handbook of Contemporary Semantics. find
13. Noah D. Goodman and Andreas Stuhlmüller. 2013. Knowledge and Implicature: Modeling Language Understanding as Social Cognition. Topics in Cognitive Science, 5. find
14. Paul Grice. 1975. Logic and conversation. In P. Cole and J. Morgan, editors, Syntax and semantics, 3, pages 41–58. Academic Press, New York. find
15. Robert X. D. Hawkins, Andreas Stuhlmüller, Judith Degen, and Noah D. Goodman. 2015. Why do you ask? Good questions provoke informative answers. In Cognitive Science Society. find
16. He He, Anusha Balakrishnan, Mihail Eric, and Percy Liang. 2017. Learning symmetric collaborative dialogue agents with dynamic knowledge graph embeddings. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1766–1776, Vancouver, Canada. Association for Computational Linguistics. find
17. Justine T. Kao, Jean Y. Wu, Leon Bergen, and Noah D. Goodman. 2014. Nonliteral understanding of number words. Proceedings of the National Academy of Sciences, 111(33). find our card
18. David Lewis. 1979. Scorekeeping in a language game. Journal of Philosophical Logic, 8:339–359. find
19. Christopher Potts, Daniel Lassiter, Roger Levy, and Michael C. Frank. 2016. Embedded implicatures as pragmatic inferences undercompositional lexical uncertainty. Journal of Semantics, 33(4). find
20. Ciyang Qing and Michael C. Franke. 2015. Variations on a Bayesian theme: Comparing Bayesian models of referential reasoning. In H Zeevat and HC Schmitz, editors, Bayesian Natural Language Semantics and Pragmatics, volume 2 of Language, Cognition, and Mind. Spring. find
21. Ciyang Qing, Noah D. Goodman, and Daniel Lassiter. 2016. A rational speech-act model of projective content. Proceedings of the Cognitive Science Society, 38. find
22. Craige Roberts. 1996. Information structure in discourse: Towards an integrated formal theory of pragmatics. Semantics & Pragmatics, 5:1–69. find our card
23. S Schuster and Judith Degen. 2020. I know what you’re probably going to say: Listener adaptation to variable use of uncertainty expressions. Cognition. find
24. Nathaniel J. Smith, Noah D. Goodman, and Michael C. Frank. 2013. Learning and using language via recursive pragmatic reasoning about other agents. In Advances in Neural Information Processing Systems, volume 26. find
25. Robert Stalnaker. 2002. Common Ground. Linguistics and Philosophy, 25:701–721. find
26. Erica J. Yoon, Michael Henry Tessler, Noah D. Goodman, and Michael C. Frank. 2016. Talking with tact: Polite language as a balance between kindness and informativity. In Proceedings of the Cognitive Science Society, volume 38. find
27. Erica J. Yoon, Michael Henry Tessler, Noah D. Goodman, and Michael C. Frank. 2017. ”I won’t lie, it wasn’t amazing”: Modeling polite indirect speech. In Cognitive Science Society. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/anderson-2021-tell-me-everything-you-know/