The Role of Intrinsic Motivation in Artificial Language Emergence: a Case Study on Colour

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1646–1656, Osaka, Japan, December 11-17 2016. The Role of Intrinsic Motivation in Artificial Language Emergence: a Case Study on Colour Miquel Cornudella1,2 Thierry Poibeau2 Remi van Trijp1 1Sony Computer Science Laboratory Paris 6 rue Amyot, 75005 Paris, France {cornudella,remi}@csl.sony.fr 2Laboratoire LATTICE-CNRS ENS & U. Paris 3 & CNRS PSL and USPC 1 rue Maurice Arnoux, 92120 Montrouge, France thierry.poibeau@ens.fr Abstract Human languages have multiple strategies that allow us to discriminate objects in a vast variety of contexts. Colours have been extensively studied from this point of view. In particular, previous research in artificial language evolution has shown how artificial languages may emerge based on specific strategies to distinguish colours. Still, it has not been shown how several strategies of diverse complexity can be autonomously managed by artificial agents . We propose an intrinsic motivation system that allows agents in a population to create a shared artificial language and progressively increase its expressive power. Our results show that with such a system agents successfully regulate their language development, which indicates a relation between population size and consistency in the emergent communicative systems. 1 Introduction Over the past two decades, language evolution studies have attracted the attention of researchers work- ing on domains such as biology, anthropology, artificial life or linguistics. This multitude of perspectives provides a rich variety of techniques on how to address this issue, including including agent-based mod- elling, which consists in studying the emergence and evolution of artificial languages, i.e. human-like communicative systems, in a population of artificial agents through recurrent peer-to-peer interactions (Smith et al., 2003; Steels, 2012). Results using this approach have shed light on the emergence of spa- tial relations (Spranger, 2013), case systems (van Trijp, 2013), colour categories (Bleys, 2010) or syntax (Garcia-Casademont and Steels, 2016). In most of these experiments the control of the complexity relies on the experimenter, who carefully selects the stages of the experiment, constraining the language development. Insights from different models dealing with complexity come from research in AI and robotics, where a number of studies where a number of studies tried to specify how agents regulate the complexity of their actions in an autonomous way. Several models have been proposed, including error reduction (Andry et al., 2001), prediction (Marshall et al., 2004), interest (Merrick and Maher, 2009) or curiosity (Oudeyer et al., 2007). These systems have been deeply inspired by psychological studies on the role of motivation (see Graham (1996) for an overview). According to Ryan and Deci (2000), motivation can be defined as “to be moved to do something”. Psychologists further distinguish two types of motivation, depending on the reasons to perform an action: extrinsic, when the interest relies on the outcome of the action, or intrinsic, when the action itself results inherently enjoyable. Both types of motivation can incite to take part in the same activity. For example, a tennis player can work on improving her slice shots because she wants to …

Content truncated to 500 words