Conceptual structure is shaped by competing pressures for simplicity and informativeness

Jon W Carr, Kenny Smith, Jennifer Culbertson, Simon KirbyProceedings of the 12th International Conference on the Evolution of Language (Evolang12) 2018 · DOI

Abstract

Languages are shaped by competing pressures from learning and communication. Learning favors simple languages, while communication favors informative ones, giving rise to the simplicity-informativeness tradeoff. Languages that evolve under this tradeoff are both maximally simple (learnable) and maximally informative (communicatively useful). However, recent research suggested that artificial languages expressing spatial relationships became more informative under learning pressure alone—a surprising result, as informativeness is typically driven by communication. This paper resolves this contradiction by demonstrating that "informativeness" in such contexts is often a side-effect of convexity—the tendency for learners to group similar meanings together. True linguistic optimality requires the dual pressures of both learning and active communication.

The Simplicity-Informativeness Tradeoff

In the study of language evolution, the "bottleneck" of passing language from one person to another (learning pressure) favors simplicity: fewer words and broader categories. Conversely, the need to distinguish between specific concepts (communication pressure) favors informativeness: more specific words to reduce ambiguity.

This paper investigates why some studies show languages becoming more informative under learning pressure alone. The authors hypothesize that the confusion stems from how informativeness is measured. They break it down into two components:

  1. Expressivity: The number of distinct words in a language.
  2. Convexity: The degree to which a word refers to a tightly clustered, "logical" set of meanings.

When the number of words is fixed, learning pressure forces the language to become more convex to be learnable. This increased convexity can be mistaken for increased informativeness, even in the absence of communicative pressure.

Experimental Design

The researchers conducted behavioral experiments using a 2D "meaning space" where stimuli varied by size and angle.

The three category systems used in Experiment 1 to test learning ease: Angle-only, Size-only, and the complex Angle & Size system.

In Experiment 1 (Learning Ease), participants were trained on one of three predefined category systems (shown above): one distinguishing only angle, one distinguishing only size, and one distinguishing both (the most "informative" but complex system). In Experiment 2 (Iterated Learning), a "telephone game" design was used where the output of one participant became the training data for the next, simulating cultural transmission over 12 generations.

Key Results and Insights

Convexity as Simplicity

The results showed that humans find it much easier to learn categories that are "convex"—meaning they pick out a continuous, tightly-packed region of the meaning space. Learning pressure doesn't just delete information; it reorganizes it into these easier-to-digest clusters. If a language is prevented from losing words (fixed expressivity), it must become more convex to remain learnable, which creates the illusion of increasing informativeness.

Evolutionary Convergence

In the iterated learning experiment, languages naturally drifted toward simplicity. Almost all "language chains" converged on systems with fewer than four categories, significantly reducing expressivity over time.

A representative chain from the iterated learning experiment, showing the transition from a random initial state (0) to a simplified, structured language over 12 generations.

As seen in the figure above, the transmission error decreased significantly as the systems became more convex and simpler. By generation 12, the random noise of the initial state is replaced by clear, structured blocks of color, representing a more learnable (though less expressive) conceptual structure.

Conclusion

The study confirms that while learning makes a language structured and convex, it does not inherently make it informative or useful for communication. Without the pressure to actually convey specific ideas to another person, languages naturally drift toward having as few words as possible. True linguistic optimality—the balance between being easy to learn and useful to speak—requires the simultaneous presence of both learning and communication pressures.

Summary updated

Abstract

Languages are shaped by competing pressures from learning and communication. Learning favours simple languages, while communication favours informative ones, giving rise to the simplicity-informativeness tradeoff. Languages that evolve under this tradeoff are both maximally simple (learnable) and maximally informative (communicatively useful). This has been shown in natural language and in experimental settings. For example, In a separate line of experimental work, However, a recent iterated learning experiment by Carstensen, Xu, Smith, and Regier (2015) showed that artificial languages expressing spatial relationships tended to become more informative when subjected to a pressure from learning. This is a surprising result given the previous work briefly reviewed above, which says that informativeness is driven by the pressure from communication, not from learning. One potential explanation for this result lies in their measure of informativeness, communicative cost, which is sensitive to (a) the number of words that the language is comprised of (expressivity) and (b) the extent to which similar meanings are expressed by the same word (which we will term convexity). In their experiment expressivity was fixed at four words. As a result, the reduction in communicative cost they found must be due to categories evolving to become more convex, i.e. picking out increasingly tightly-clustered sets of meanings.
BibTeX
@inproceedings{carr2018conceptual,
  author = {Jon W Carr and Kenny Smith and Jennifer Culbertson and Simon Kirby},
  title = {{Conceptual structure is shaped by competing pressures for simplicity and informativeness}},
  booktitle = {Proceedings of the 12th International Conference on the Evolution of Language (Evolang12)},
  year = {2018},
  doi = {10.12775/3991-1.013},
}
Figure