Abstract:

Words within a language share more sounds than would be expected by chance, and this similarity cannot be explained by phonological rules alone (Dautriche et al. 2017; Mahowald et al. 2018). This is somewhat puzzling, since language comprehension is easier when words are more distinct. However, language production is easier when words are more alike. We use techniques from language evolution research to test the hypothesis that, in this instance, pressures from production outweigh those from comprehension. We model language transmission in an agent-based exemplar model, where the output of one generation becomes the input to the next (Kirby et al. 2015; Smith et al. 2003). We find that words become more alike under two assumptions about the production process: first, that errors tend to replace lower frequency sounds with higher frequency ones, and second, that retrieval tends to favour words that resemble others the agent knows. As these effects accumulate over generations – modulated by inferences about what the listener is likely to understand – initially random lexicons start to exhibit natural-language-like properties. Overall, this work sheds light on how languages balance competing pressures through a process of cultural evolution.

Speaker's social media:

Twitter: @mandolinguist