Abstract:

Language exhibits systematicity in the way a set of discrete units, such as words or syllables, are used and reused to form larger linguistic sequences. This kind of systematicity is reflected by cross-linguistic regularities in the distributional statistics of language: Syllable transitions are more probable within words than at the boundary between words, and the frequency distributions of words follow a Zipfian power law. Both the statistical coherence of words and their skewed frequency distribution can aid language learning. However, it is still unclear how they arise in language in the first place. Here, we investigate whether their emergence may be driven by learnability pressures during cultural transmission. We conducted a non-linguistic iterated learning study in which participants observed and reproduced sets of color sequences that were produced by a previous participant. We show that during transmission, initially random sets of sequences evolve to have statistically coherent units with a skewed frequency distribution, and that these properties make them easier to reproduce. In doing so, we show how cultural transmission can shape the statistical structure of language and how learning pressures alone, in the absence of meaning or communication, can explain the presence of two cross-linguistic properties of language.

Speaker's social media: