A rational model of linguistic accommodation and its potential role in language simplification

Stella Frank, Kenny SmithProceedings of the 12th International Conference on the Evolution of Language (Evolang12) 2018 · DOI

Abstract

Linguistic complexity often correlates with population history; languages with many adult learners tend to be morphosyntactically simpler than those in isolated communities. While traditional theories attribute this to adult learning deficiencies, this paper proposes a "rational accommodation" mechanism. Through Bayesian modeling, we demonstrate that native speakers who simplify their speech to be understood by learners can inadvertently drive language regularization, reducing entropy and "locking in" simpler variants for future generations.

Introduction

A long-standing observation in linguistics is that "esoteric" languages spoken in small, isolated communities often possess greater morphosyntactic complexity than "exoteric" languages like English or Swahili, which have large numbers of non-native speakers. The traditional "deficiency hypothesis" suggests this is due to adults' inability to master complex grammar. However, Frank and Smith propose that the communicative helpfulness of native speakers—specifically their tendency to accommodate learners—may be the primary engine of simplification.

Methods

The authors developed a computational model involving two rational Bayesian agents: a Native Speaker (NS) and a Non-native Speaker (NNS). Language is represented as a probability distribution over linguistic variants (e.g., different ways to form a past tense). The NNS is modeled as an agent with significantly less exposure to data ($|D|$) than the NS.

Crucially, the NS employs a "Theory of Mind" parameter for accommodation ($p$). Instead of simply using their own complex distribution, the NS reasons about what the NNS likely knows based on their limited exposure. Because the NS doesn't know exactly which words the learner has heard, they integrate over all possible datasets the learner might have encountered. The model measures "regularization" through entropy reduction; a language simplifies as its most frequent variants become more dominant, making the system more predictable.

Key Results

Simulations using a simulated sample language with a baseline entropy of 1.96 demonstrated that accommodation significantly impacts linguistic evolution.

Distributions of entropy across different levels of learner data (|D|) and accommodation probabilities (p).

As shown in Figure 1, when accommodation is absent ($p=0$), the NNS eventually converges to the "true" complex language (blue peaks near 1.96). However, with high accommodation ($p=1$), the resulting language exhibits significantly lower entropy. This effect is most dramatic when the NNS has seen very little data ($|D|=1$), where entropy shifts toward 1.0, indicating a highly regularized system. This suggests a "lock-in" effect: by accommodating a learner's initial bias toward frequent items, native speakers reinforce that skew, preventing the learner from ever encountering rarer, more complex variants.

Discussion and Future Work

This model suggests that simplification emerges not from "errors," but from rational agents optimizing for successful communication. In complex languages, low-frequency variants are the first to go because learners are unlikely to have seen them, and accommodating native speakers stop using them, leading to their eventual extinction.

While the model is a simplified abstraction of human psychology and linguistic structure, it provides a theoretical framework for understanding how social interaction drives language change. The authors acknowledge that the model assumes perfect Bayesian rationality and does not yet account for hierarchical syntax or social prestige. Future work aims to validate these predictions through empirical laboratory studies with human participants to see if native speakers actually accommodate in the specific ways the model predicts.

Summary updated

BibTeX
@inproceedings{frank2018rational,
  author = {Stella Frank and Kenny Smith},
  title = {{A rational model of linguistic accommodation and its potential role in language simplification}},
  booktitle = {Proceedings of the 12th International Conference on the Evolution of Language (Evolang12)},
  year = {2018},
  doi = {10.12775/3991-1.031},
}
Figure