Introduction

The paper reports on experiments with a population of visually grounded robotic agents capable of bootstrapping their own ontology and shared lexicon without prior design nor other forms of human intervention. The agents do so while playing a particular language game called the guessing game. We show that synonymy and ambiguity arise as emergent properties in the lexicon, due to the situated grounded character of the agent–environment interaction, but that there are also tendencies to dampen them so as to make the language more coherent and thus more optimal from the viewpoints of communicative success, cognitive complexity, and learnability.

How do words get their meanings? An answer to this question requires a theory of the origins of meanings, a theory of how forms get recruited for expressing meanings, and a theory of how associations between forms and meanings may propagate in a population. Each theory must characterize properties of a cognitive agent's architecture: components a cognitive agent needs to have, and details of how the different components coordinate their activities. More specifically, the theories should detail what kind of associative memory the agents must have for storing and acquiring form–meaning relations, what type of mechanisms they might use to categorize the environment through sensory inputs, how they might acquire a repertoire of perceptually grounded categories (an ontology), and what behaviors the agents must be capable of so as to communicate successfully through language.

To allow validation, theories of agent architecture should be formally specified and hence testable through computer simulations or even better through experiments with robotic agents interacting with real world environments through a sensory apparatus.