ILM
research: replication with different bias
Batali (1998). RNN learners. Large static population.
Feature vector semantics. Convergence on partially compositional morphology.
Kirby (2000). Heuristic CFG inducers. Small population with
turnover. Feature-vector semantics. Emergence of perfect compositionality.
Batali (2002). Symbolic instance-based learner. Static
population. Simple predicates. Emergence of recursion, passivisation,
reflexives.
Teal & Taylor (1999). MDL FSN inducer. Minimal pop.
No semantics.
Emergence of compressed grammars.
Zuidema (2001). CFG inducer. No semantics. Stable
languages outside mathematically predicted bounds on richness of
input.
Tonkes & Wiles (2000). RNN learners.
Various populations.
Real-valued semantics.
Emergence of regular
structured language.
Kirby et al (2002). Associative nets. Minimal population.
Feature-vector semantics. Conditions for emergent structure
CURRENT WORK: Mathematical characterisation of
stability conditions. Coevolution of bias, meaning space & syntax. Population
dynamics.