Expression/induction models of language evolution: dimensions and issues

• Published in: LINGUISTIC EVOLUTION THROUGH LANGUAGE ACQUISITION: FORMAL AND COMPUTATIONAL MODELS, Briscoe, Ted [Ed], Cambridge, UK: Cambridge U Press, 2002, pp 301-344 pp. 301 - 344
• Main Author: Hurford, James R.
• Format: Book Chapter
• Language: English
• Published: United Kingdom Cambridge University Press 08.08.2002
• Subjects: Computational Linguistics; Computer Modeling and Simulation; Language acquisition; Learning; Origin of Language; Syntax Semantics Relationship
• ISBN: 9780521662994; 9780521078931; 0521662990; 0521078938
• DOI: 10.1017/CBO9780511486524.010

A comparison of five expression/induction models simulating the evolution of syntactic systems, proposed respectively by John Batali (1998), Batali (2002), Hurford (2000), Simon Kirby (2000), & Kirby (2002), is argued to show that the emergence of systems with recursion & compositionality in computational models is facilitated by five factors: (1) the provision of predefined hierarchical semantic representations; (2) an invention algorithm & (3) a learning algorithm biased towards conformity to the compositionality principle; (4) a strong semantic bottleneck effect, enabling learning from examples in a small subset of semantic space; & (5) a production bottleneck effect that favors rules of increased generality through a feedback loop. In the framework of these models, which treat language as an interaction between internalized grammars & their expression in interaction, a hybrid model is suggested that weakens the bottleneck effects of (4) & (5), uses flat semantic structures, removes the compositional bias of (2), & experimentally varies the predisposition toward compositionality of the learning algorithm. 1 Table, 2 Figures, 40 References. J. Hitchcock