Simplifications made early in learning can reshape language complexity: an experimental test of the Linguistic Niche Hypothesis

Kenny SmithPsyArXiv 2024 · DOI

Abstract: Languages spoken in larger populations seem to be relatively simple. One possible explanation is that this is a consequence of the simplifying influence of non-native speakers: adult learners tend to reduce complexity during learning, and large languages tend to have a higher proportion of non-native speakers. This Linguistic Niche Hypothesis, that languages adapt to their social niche, receives some statistical support from typological studies which show negative correlations between population size or number of non-native speakers and morphological complexity. Here I report an experimental test of this hypothesis, using two artificial language learning experiments to explore the impact of simplifications made by non-native-like early learners on morphological complexity. These experiments show that the presence of non-native-like early learners in a population can lead to the simplification of that language's morphology as a result of inter-generational language transmission, providing experimental support for the Linguistic Niche Hypothesis.

Introduction

Why is English morphologically simpler than Icelandic or many small-scale indigenous languages? Linguists have long observed that languages spoken by larger populations with many non-native speakers tend to possess simpler grammar and fewer irregular forms. This observation forms the core of the Linguistic Niche Hypothesis (LNH), which posits that languages adapt to the cognitive constraints of their speakers. Specifically, adult learners, who often struggle with complex grammar, simplify the language, and these simplifications are then passed on to subsequent generations.

However, previous experimental attempts to demonstrate this have faced a challenge: when the simplified outputs of multiple learners are aggregated, the resulting language often remains complex because different individuals make different errors. This study investigates the specific conditions—population size and the ratio of non-native-like input—under which individual simplifications successfully reshape a language.

Methods

The study utilized a custom-designed artificial language describing scenes (e.g., "one dog bouncing"). The language featured stems for numbers (won, sum), animals (snap, grol, twit), and motions (woosh, boing, loop), with a complex system of suffixes that redundantly marked number and animal type, including irregular exceptions.

Experiment 1: The Baseline

94 participants learned the miniature language over 8 rounds. This experiment aimed to track how internal "grammar" evolved from early-stage learning to mastery. To quantify complexity, multinomial regression models determined how many semantic features (number, animal, motion) a participant’s suffix choices tracked.

Experiment 2: The Transmission

522 new participants were trained on the output produced by participants in Experiment 1. Two variables were manipulated:

  • Population Size: Input was drawn from either 2 sources ("Small") or 8 sources ("Large").
  • Data Composition: The input was either "No Early" (purely complex/accurate), "Mixed" (50% early-simplified, 50% late), or "All Early" (entirely simplified).

Results

Learning Dynamics and Simplification

Experiment 1 confirmed that learners naturally simplify complex rules during the early stages of learning. As shown in Figure 4, complexity scores for quantifiers, nouns, and verbs are significantly lower in early rounds (e.g., Round 2) compared to the mastery achieved by Round 8. This "early learning" phase serves as a proxy for non-native speakers, who similarly tend to regularize and simplify complex morphology.

Complexity of internal grammar across learning rounds in Experiment 1, showing early simplification followed by mastery.

The Impact of Population Size

Experiment 2 revealed a critical interaction between population size and data composition. In small populations, learners were often able to reconstruct the complex system even when faced with "Mixed" input. However, in large populations, the results were different.

As illustrated in Figure 5, participants in the Large Population/Mixed condition produced significantly simpler languages than those in the No Early condition ($p = .005$). When a learner in a large population receives inconsistent, simplified input from 8 different sources, they are more likely to abandon complex rules entirely and converge on a simpler, more regular grammar. This suggests that the social niche of a large, diverse population acts as a catalyst for language simplification.

Complexity outcomes in large populations based on input data composition, highlighting the significant simplification in the 'Mixed' condition.

Discussion

These findings provide experimental support for the Linguistic Niche Hypothesis. Simplification is not a guaranteed outcome of having non-native speakers; rather, it depends on a "tipping point" where simplified input is frequent enough to outweigh complex, correct input.

While early-stage learners are a useful proxy, future work should involve actual non-native speakers whose "fossilized" errors may differ from the transient errors of fast learners. Additionally, extending this study over multiple generations through iterated learning would likely amplify the observed effects, further clarifying how social niches reshape the languages we speak.

Summary updated

Abstract

Languages spoken in larger populations seem to be relatively simple. One possible explanation is that this is a consequence of the simplifying influence of non-native speakers: adult learners tend to reduce complexity during learning, and large languages tend to have a higher proportion of non-native speakers. This Linguistic Niche Hypothesis, that languages adapt to their social niche, receives some statistical support from typological studies which show negative correlations between population size or number of non-native speakers and morphological complexity. Here I report an experimental test of this hypothesis, using two artificial language learning experiments to explore the impact of simplifications made by non-native-like early learners on morphological complexity. These experiments show that the presence of non-native-like early learners in a population can lead to the simplification of that language's morphology as a result of inter-generational language transmission, providing experimental support for the Linguistic Niche Hypothesis.
BibTeX
@misc{smith2024simplifications,
  author = {Kenny Smith},
  title = {{Simplifications made early in learning can reshape language complexity: an experimental test of the Linguistic Niche Hypothesis}},
  year = {2024},
  doi = {10.31234/osf.io/2kzd4},
  howpublished = {PsyArXiv},
}
Figure