Word learning and optimality

Background

Languages generally use short words for common meanings, and longer words for rare meanings. This is a form of optimality known as Zipf's Law of Abbreviation. We know how language users optimise language for a static set of meanings, but the frequencies of meanings can change quickly (for example, you make have recently switched to saying "iso" instead of "isolation" after corona virus hit). However, very little is known about how language users keep up with these changes, and whether or not social factors like "common ground" affect the ability to re-optimise this langugae use.

Research Questions / Hypotheses

We hypothesised that participants would use shorter, more ambiguous word forms to communicate more frequent objects while mapping the unique and longer word forms to the infrequent object, and would generally continue to do so after language changes. However, after language changed we hypothesised that participants with no common ground with their partner would have higher odds of optimally using language compared to those with common ground. Finally, we hypothesised that participants with no common ground with their communicative partners would have higher odds of changing their more general communicative strategies than those with common ground.

Participants

75 participants completed the study. No participants were excluded from analysis.

Methods

We replicated and extended the combined pressures condition from the artificial language learning paradigm and communication game of Kanwal et al. (2017). In this paradigm, participants had to learn to align word forms with presented objects, and communicate these objects to a (robot) communicative partner. Notably, we also included a second round of learning and communication in which perceived frequencies of words were altered, and common ground of communication partners manipulated.

Results

We used mixed-effects logistic regression models to analyse our data. We found that generally, participants would communicate optimally even after language changed (supporting hypothesis one). We also found that re-optimisation of language use was hampered by existing common ground (in line with hypothesis two). We did not find evidence to support our third hypothesis that common ground had an effect on language users’ tendencies to switch their strategies of communication more generally.

Implications

This research allows us to examine how language change occurs, and the factors that may affect this within a community. This has implications towards faciliating optimal information transfer in communities that may have struggles keeping up with language change (e.g., elderly communities adjusting to language surrounding new technological innovations). These results have been communicated in an Honours student's thesis, and may also be submitted to a journal.