Documentation

Linglib.Phenomena.Nonliteral.Hyperbole.KaoEtAl2014

@cite{kao-etal-2014-hyperbole} @cite{kao-etal-2014-hyperbole} #

"Nonliteral understanding of number words" PNAS 111(33): 12002-12007

Experimental Design #

Participants heard price utterances ("It cost $X") about items with known typical prices (electric kettles, laptops, watches). Three experiments measured: (3a) price priors, (3b) affect priors conditional on price, (3c) listener interpretations of literal, hyperbolic, and round/sharp utterances.

The Model #

Speakers may use literally false utterances to convey affective information when the listener is uncertain about the speaker's communicative goal.

S1(u|s,a,g) ∝ exp(α · [ln L0(g(s,a)|u) - C(u)]) [Eq. 5,7]

where g composes precision (exact vs approximate) with relevance (price, affect, or both), yielding 5 distinct goals.

L1 marginalizes over goals:

L1(s,a|u) ∝ P_S(s) · P_A(a|s) · Σ_g P_G(g) · S1(u|s,a,g) [Eq. 10]

Grounding #

Price semantics is grounded in HasDegree. The literal meaning of "fifty dollars" is numeralExact 50 price — the price equals $50.

Qualitative Findings #

Any model of hyperbole should account for these 6 findings:

#FindingTheorem
1At the inferred price, the listener infers notable affecthyperbole_affect_at_modal
2Marginal notable affect > no affect for hyperbolic utteranceshyperbole_affect
3Listener infers speaker's QUD is valence, not pricehyperbole_qud
4Literal utterance → listener infers correct priceliteral_correct
5Literal utterance is not interpreted hyperbolicallyliteral_not_hyperbolic
6Sharp numbers interpreted more precisely than roundhalo_sharp_500

The 6 qualitative findings from @cite{kao-etal-2014-hyperbole} Experiments 3a–3c. Each model of hyperbole should formalize and prove all 6 findings.

  • affect_at_modal : Finding

    Hearing "$10,000" for a kettle, the listener infers notable affect at the modal price (the most likely actual price under the posterior).

  • affect_marginal : Finding

    Marginalizing over all prices, notable affect dominates for "$10,000": Σ_s P(s, notable | "$10K") > Σ_s P(s, none | "$10K").

  • qud_valence : Finding

    The listener infers the speaker's communicative goal (QUD) is to express valence/affect rather than to communicate exact price.

  • literal_correct : Finding

    Hearing "$50" for a $50 kettle, the listener infers the correct price: P($50 | "$50") > P($500 | "$50").

  • literal_not_hyperbolic : Finding

    Literal utterances are not interpreted hyperbolically: P($50 | "$50") > P($10K | "$50").

  • halo_sharp_precise : Finding

    Sharp numbers are interpreted more precisely than round numbers: P(exact match | "$501") > P(exact match | "$500").

Instances For
    Equations
    • One or more equations did not get rendered due to their size.
    Instances For

      All findings from the paper.

      Equations
      • One or more equations did not get rendered due to their size.
      Instances For

        Item types (Experiments 3a/3b): electric kettles, laptops, watches.

        Instances For
          Equations
          • One or more equations did not get rendered due to their size.
          Instances For

            Price states: round/sharp pairs {50,51,500,501,...,10000,10001}.

            Instances For
              Equations
              • One or more equations did not get rendered due to their size.
              Instances For

                Affect: no affect vs notable affect.

                Instances For
                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For

                    Communicative goals: 5 = (3 relevance × 2 precision) minus 1 collapse.

                    Instances For
                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For
                        Equations
                        • One or more equations did not get rendered due to their size.
                        Equations
                        • One or more equations did not get rendered due to their size.
                        Equations
                        • One or more equations did not get rendered due to their size.
                        Equations
                        • One or more equations did not get rendered due to their size.

                        Price prior P_S (Experiment 3a). Unnormalized.

                        Equations
                        Instances For

                          Affect prior P_A(a|s) (Experiment 3b). Unnormalized (pairs sum to 10000).

                          Equations
                          Instances For

                            C(round) = 0, C(sharp) = 1.

                            Equations
                            Instances For
                              Equations
                              • One or more equations did not get rendered due to their size.
                              Instances For

                                L0(s,a|u) = P_A(a|s) if s = u, 0 otherwise. [Eq. 9]

                                Equations
                                Instances For

                                  Sum L0 over the QUD equivalence class of w under goal q. The class is derived from project: {w' | project w' q = project w q}.

                                  Equations
                                  • One or more equations did not get rendered due to their size.
                                  Instances For

                                    @cite{kao-etal-2014-hyperbole} hyperbole model, parametric in item.

                                    Equations
                                    • One or more equations did not get rendered due to their size.
                                    Instances For

                                      At the modal price ($10K), notable affect > no affect. The speaker saying "$10K" about a kettle signals frustration.

                                      Marginal: notable affect dominates overall for "$10K". Σ_s L1(s, notable | "$10K") > Σ_s L1(s, none | "$10K").

                                      Sharp "$501" is interpreted more precisely than round "$500".

                                      Map each empirical finding to the RSA model prediction that accounts for it.

                                      Equations
                                      • One or more equations did not get rendered due to their size.
                                      Instances For

                                        The RSA model accounts for all 6 empirical findings from @cite{kao-etal-2014-hyperbole}.