2025/06/14 by Nathaniel Imel, Christopher Haberland, Shane Steinert-Threlkeld
#efficient communication #language of thought #learnability #semantic typology #software
paper · doi:10.7275/scil.3144
The Unnatural Language Toolkit (ULTK) is an open-source Python library for computational semantic typology research (https://clmbr.shane.st/ultk/ultk.html). ULTK's key features include unifying data structures, algorithms for generating artificial languages, and data analysis tools for related computational experiments. The language module organizes the basic data structures for constructing meaning spaces, expressions, and languages. A grammar submodule contains methods for building and enumerating expressions from custom Language of Thought (Fodor, 1975, 2008; Quilty-Dunn et al., 2022) grammars, which allows for straightforward computation of minimum length descriptions for symbolically expressible semantic representations. This approach has been used successfully in many investigations of concept learning (Feldman, 2000; Goodman et al., 2015). The second main module of ULTK, effcomm, organizes efficient communication analyses, which have become popular styles of explanation in recent functionalist accounts of semantic universals (Kemp et al., 2018). This module contains functions for defining informativity based on literal and pragmatic communicative agents and algorithms for exploring the space of artificial languages. After first elaborating on the structure of these two modules, we then provide two case studies, illustrating two major styles of explanation in computational semantic typology research: (1) an efficient communication analysis of modal semantic typology, and (2) an analysis of the relative ease of learning of monotone versus non-monotone quantifiers. ULTK’s accessible design, documentation, and open-source nature are intended to reduce barriers for researchers when implementing computational linguistic typological experiments.