The evolutionary language game formulates the origin of language as a game-theoretic problem. It was first described in ["The Evolutionary Language Game" (Nowak, Plotkin, Krakauer)] (https://www.sciencedirect.com/science/article/abs/pii/S0022519399909815).
The game models a communicative event as an attempt by one agent A to convey accurately a world-state W to a second agent B. Agent A encodes W as a message, M, which is then received by B. B's interpretation of M results in an internal representation of a world-state, W'. M may be received in degenerate form, i.e. transmission errors can occur. The success of a communicative act is measured as the proximity of W to W'. The game assumes altruism, i.e. that a successful communicative act benefits both agents, so both agents are rewarded for successful communication.
This model implements a version of the evolutionary language game described in Nowak et al..
python generate_fire.py basic_model \
--individuals 10 --signals 6 --meanings 5 --k 5 --terminal "iterations(100)"The population size will be ten. The languages engendered will apply a "lexicon" of six signals to a semantic space of five discrete meanings. Association matrices will be inferred with k=5; this means that, for each meaning, a new member of the population will sample its parent's responses to ech object in the semantic space five times. Learning will terminate after one hundred iterations.
TODO