Actuation without production bias

TL;DR

This paper interrogates the actuation problem in sound change by asking whether production-bias dynamics are unique and whether forces beyond production bias can drive population-level change. It extends a two-bias learning framework to include subpopulation contact and social weighting (across variants, groups, and individuals) and evaluates their capacity to produce stable variation and abrupt change using a multi-teacher learning setup and a coarticulated contextual vowel variant. The results show that dynamics are not unique across forces; some social-weight configurations can yield rapid, nonlinear shifts, while others fail to produce change, highlighting the role of population structure and bias-correlation patterns. Overall, the findings suggest actuation can occur without production bias and that similar population trajectories can emerge from different external drivers, offering a partial solution to the non-phonologization problem in language change.

Abstract

Phonetic production bias is the external force most commonly invoked in computational models of sound change, despite the fact that it is not responsible for all, or even most, sound changes. Furthermore, the existence of production bias alone cannot account for how changes do or do not propagate throughout a speech community. While many other factors have been invoked by (socio)phoneticians, including but not limited to contact (between subpopulations) and differences in social evaluation (of variants, groups, or individuals), these are not typically modeled in computational simulations of sound change. In this paper, we consider whether production biases have a unique dynamics in terms of how they impact the population-level spread of change in a setting where agents learn from multiple teachers. We show that, while the dynamics conditioned by production bias are not unique, it is not the case that all perturbing forces have the same dynamics: in particular, if social weight is a function of individual teachers and the correlation between a teacher's social weight and the extent to which they realize a production bias is weak, change is unlikely to propagate. Nevertheless, it remains the case that changes initiated from different sources may display a similar dynamics. A more nuanced understanding of how population structure interacts with individual biases can thus provide a (partial) solution to the `non-phonologization problem'.

Figures (11)

• Figure 1: Lexicon and learning parameters. $\mu_a$ and $\mu_i$ are the means of normally-distributed vowel categories V$_1$ (/a/) and V$_2$ (/i/). $c$ is the mean of the normally-distributed contextual variant, V$_{12}$. $\lambda$ represents the strength of the bias favouring coarticulated variants.
• Figure 2: Prior distribution over $c$, for values between the means of V$_2$ and V$_1$ ($\mu_i = 530$, $\mu_a = 730$). The parameter $a$ controls the strength of the categoricity bias, with values nearer to 0 corresponding to a greater preference for values of $c$ near either endpoint.
• Figure 3: Two types of population structure considered in models in kirby2013modelkirby2015bias: (a) Single-teacher scenario, where each learner in generation $t+1$ receives all her data from a single teacher in generation $t$. (b) Multiple-teacher scenario. Each data point comes from a random teacher, each chosen uniformly at random (with replacement) from teachers in generation $t$.
• Figure 4: Example of the distribution of $c$ in the population at time $t$ (probability density function $P(C^t)$). The population starts with minimal coarticulation at $t=0$ ($c \sim N(720, 10^2)$) and ends with full coarticulation by $t=100$. All agents have weak categoricity bias ($a=0.02$) and strong production bias ($\lambda = 2$). Left panel: shading is proportional to $P(C^t)$. Right panel: solid line and shading show the mean $\pm$ 2 standard deviations of $P(C^t)$.
• Figure 5: Mean value of $c$ in the population in its stable state, starting from a population of agents with a minimally coarticulated /ai/ variant ($c \sim N(\mu_a - 10, 10^2)$) with production bias $\lambda$ and categoricity bias $a$ (smaller $a$ = stronger bias). The final mean changes non-linearly as $\lambda$ and $a$ are varied. Red and dark blue correspond to full/no coarticulation of $V_{12}$, respectively.