When Language Evolution Meets Multimodality: Current Status and Challenges Toward Multimodal Computational Models

Computational models can be considered human-designed computing models inspired bythe processes observed in the natural world, which allow simulating and understanding these processes.Computational modelling is notably applied to simulate the behaviour and long-term dynamics of humanLanguage. The research effort made so far in computational modelling of language evolution considerspredominantly one modality by arguing for a unimodal origin of Language. This article extends this paradigmto a new perspective that integrates into its structure and learning algorithms principles from multimodalcommunication. This article gives an overview of the current language evolution models. It discusses thekey challenges towards multimodal language evolution modelling by envisioning a conceptual frameworkto design the multimodal grounding and the language learning processes, as well as their realisation througha multi-agent multimodal referential game. This framework is valuable for many researchers working onlanguage evolution to reveal the key questions they should address and integrate for pursuing a holisticvision that combines all modalities in a multimodal language evolution model.