Documentation

Linglib.Theories.Pragmatics.RSA.Extensions.InformationTheory.Basic

Information-Theoretic Foundations of RSA #

@cite{cover-thomas-2006} @cite{frank-goodman-2012} @cite{zaslavsky-hu-levy-2020}

Re-exports domain-agnostic information-theoretic primitives from Core.InformationTheory into the RSA.InformationTheory namespace, and adds RSA-specific definitions (RSALevel).

The ℚ-based RSA evaluation infrastructure (RSAScenarioQ, RSA.Eval, RSA.Q) has been removed. This file retains the re-exported information-theoretic definitions (entropy, KL divergence, log2Approx) that are used by other modules (e.g., ArgumentativeStrength), but all RSAScenarioQ-dependent computations and dynamics have been removed.

Original Results (To Be Re-derived with RSAConfig) #

  1. RSA as alternating maximization: RSA optimizes G_α = H_S(U|M) + α·E[V_L]
  2. G_α monotonicity: G_α(S_t, L_t) ≤ G_α(S_{t+1}, L_{t+1})
  3. Critical α = 1: Phase transition at α = 1
  4. Utility non-monotonicity: E[V_L] can decrease even as G_α increases

RSA iteration level.

Track the depth of pragmatic reasoning:

  • L0 = literal listener
  • S1 = pragmatic speaker (responds to L0)
  • L1 = pragmatic listener (responds to S1)
  • S2 = second-order speaker (responds to L1)
  • etc.
Instances For
    Equations
    • One or more equations did not get rendered due to their size.
    Instances For