Creole Simplicity and Compositionality of the Languages Arising in a Plain Emergent Language Contact Model

Abstract

This paper aims to analyze creole simplicity, the tendency of languages to be simplified when distinct languages are brought into contact, and compositionality, the property of languages that combination of words makes languages express tremendously large numbers of matters, of the languages arising in a plain emergent language contact model. Previous work succeeds in reproducing creole simplicity by the simulation of the language contact phenomena with emergent languages. However, their models are too intricate to ascertain whether the language contact itself contributes to the results or whether their intricate network architectures and game settings accidentally create the results. Therefore, we first design a maximally plain model and ensure that the languages arising in the plain model still show creole simplicity by a few metrics. Moreover, we visualize how language contact affects compositionality, which reveals that 1) compositionality jumps up immediately after training begins; 2) pre-contact and post-contact languages have different ways of changing; 3) post-contact languages have higher compositionality in the end.