Modeling imprecise interpretation and production using probabilistic pragmatics

This repository contains the code and visualizations associated with Waldon (2022), "A Novel Probabilistic Approach to Linguistic Imprecision", which appears in Measurements, Numerals and Scales Essays in Honour of Stephanie Solt (eds. Nicole Gotzner and Uli Sauerland).

The repository consists of the following parts:

WebPPL simulation code: model_code and model_code_morealts

Two extensionless files -- model_code and model_code_morealts -- contain WebPPL code that is executable either from R (see below) or directly in the online WebPPL interpreter (accessible at http://webppl.org/).

model_code defines the context parameters for the simulations that correspond to Model 1 of the paper, while model_code_morealts defines those of the simulations that correspond to Model 2.

Constructing speaker and listener distributions

The following commands may be appended (either in R or in the WebPPL interpreter) to the base simulation code to construct speaker and listener distributions, respectively.

impSpeaker(state,alpha,beta,halos)

• state: an intended meaning (an element of states)
• alpha: the alpha optimality parameter
• beta: the global imprecision parameter that determines the degree of "Partial-Truth" (PT) activation for the speaker. When beta = 0, impSpeaker behaves as the classic RSA pragmtic speaker would.
• halos: an array, either yeshalos or nohalos. With nohalos, the degree of PT activation is equal for all non-numeral quantifiers in the alternative set. (This corresponds to Simulation #X). nohalos corresponds to Simulation #X: only the has the potential to be used imprecisely.

pragmaticListener(utterance,alpha,beta,halos)

• utterance: an utterance (an element of utterances)
• alpha, beta, halos: defined as above. (These parameters are passed to impSpeaker within the call to pragmaticListener)

NOTE: To explore the model predictions vis a vis imprecise production of numerals (as discusssed in Footnote 11), refer to the model code in model_numerals.

R interface to WebPPL and visualization scripts: analysis.R

This file includes all the code for generating the distributions reported in the paper. It relies on MH Tessler's rwebppl package (see https://github.com/mhtess/rwebppl for more info). It also contains all the code necessary for reproducing the graphs reported in the paper.