nucleation

LIT-816: Pragmatic Language Interpretation as Probabilistic Inference

2 min read

Properties4
StatusActive — read 2026-10-09 (NOTE-606), from the authors' manuscript; worth reading as the standard statement of the rational speech act (RSA) framework: a pragmatic listener P_L(w|u) ∝ P_S(u|w)P(w) inverting a softmax speaker whose utility is log P_Lit(w|u), with a literal listener conditioning the prior on the utterance's truth. Its uncertain-RSA extension, a joint posterior over the world and the speaker's type, topic or lexicon, is what carries hyperbole, irony, metaphor, vague adjectives and embedded implicature. A review: the empirical support it cites is summarized, not shown, and its claims of near-ceiling fits rest on the cited papers. Source, with LIT-821 and LIT-809, of THEORY-172.
Tagspragmatics, linguistics, cognition, probabilistic-modeling
Cited as “paper” byNOTE-606 — Pragmatic language interpretation as probabilistic inference
Cited as “source” byTHEORY-172 — In Frank and Goodman's one-shot reference games, listeners' mean bets matched a parameter-free listener inverting a speaker who chooses among true utterances by informativeness, with a separately measured salience prior, and not salience alone; and in Bergen, Goodman and Levy's signalling games with novel symbols, listeners read a signal as excluding what the available alternative would have conveyed, THEORY-785 — Listeners interpret an utterance by inverting a model of a speaker who chooses among alternatives by informativeness, so what a fixed form conveys depends on what else the speaker could have said, as the rational speech act account of Frank, Goodman and coauthors proposes

Graph View

Noah D. Goodman and Michael C. Frank (2016), Trends in Cognitive Sciences 20(11):818–829 — DOI-10.1016/j.tics.2016.08.005

Key takeaways

  • The model. P_L(w|u) ∝ P_S(u|w)P(w); P_S(u|w) ∝ exp(α U(u; w)); U(u; w) = log P_Lit(w|u); P_Lit(w|u) ∝ δ_⟦u⟧(w) P(w). The literal listener is where conventional, compositional meaning enters; the speaker chooses among a set of alternative utterances.
  • Utility refinements (Box): cost, U = log P_Lit(w|u) − cost(u) (the manuscript prints ”+ cost(u)”); speaker uncertainty, expected utility over the speaker’s knowledge k; topic relevance (a QUD), U(u; w, t) = log Σ_{w′: t(w′)=t(w)} P_Lit(w′|u); and non-informational social goals such as politeness.
  • uRSA. P_L(w, s|u) ∝ P_S(u|w, s)P(s)P(w), with s the speaker’s type: topic, word meanings, background knowledge, discourse context. Under topic uncertainty “$1,000 kettle” is read as affect (hyperbole), and likewise irony and metaphor; under threshold uncertainty “tall” gets class-relative, borderline and sorites behaviour; under lexical uncertainty embedded implicatures appear.
  • Evidence summarized. Reference-game fits (LIT-821 and a replication), cost sensitivity (LIT-809 and a typing-speed study), scalar implicature with speaker knowledge manipulated; deeper recursion only in about 15% of participants in one more complex paradigm.
  • Stated limits. A computational-level account; how it is computed fast, how alternatives are fixed, how speakers actually produce (overspecification), and whether it scales are open.

Standing in the record

Filed on 2026-10-09 at the owner’s request, from the manuscript bibliography of 2026-10-09 (work what-survives-translation): one of the works the manuscript considered and dropped from its final reference list.

Read on 2026-10-09 (NOTE-606), from the authors’ manuscript. It is the record’s statement of the RSA framework, and with LIT-821 and LIT-809 the source of THEORY-172. Its box on language change points to iterated learning and to typological efficiency results of the kind LIT-810 later gave for colour.

Backlinks