Abstract

Frequency-(in)dependent Regularization in Language Production and Cultural Transmission

Authors
  • Emily Morgan (University of California, Davis)
  • Roger P. Levy (Massachusetts Institute of Technology)

Abstract

Binomial expressions are more regularized--i.e. their ordering preferences (e.g. “bread and butter” vs. “butter and bread”) are more extreme—-the higher their frequency. Although standard iterated-learning models of language evolution can encode overall regularization biases, the stationary distributions in these standard models do not exhibit a relationship between expression frequency and regularization. We show that introducing a frequency-INdependent regularization bias into the data-generation stage of a 2-Alternative Iterated Learning Model yields frequency-dependent regularization in the stationary distribution. We also show that this model accounts for the distribution of binomial ordering preferences in corpus data.

Keywords: binomial expression, regularization, iterated learning, cultural transmission, language evolution

How to Cite:

Morgan, E. & Levy, R. P., (2020) “Frequency-(in)dependent Regularization in Language Production and Cultural Transmission”, Society for Computation in Linguistics 3(1), 466-467. doi: https://doi.org/10.7275/e377-x565

Downloads:
Download PDF

118 Views

35 Downloads

Published on
01 Jan 2020