Chapter 14. Conclusions From “The Computational Nature of Language Learning and Evolution” Summarized by Seok Ho-Sik

14. 1 A Summary of the Major Insights 14.1.1 Learning and Memory Learning at the individual level and evolution at the population level are related. Memoryless learner A single attractor to which the population converged from all initial conditions. Batch learner A bistable situation with two stable attractors. Asymmetric cue-based learner Two different regimes – one with a single stable attractor and one with two stable attractors. There is a qualitative difference in evolutionary behavior between small-n and large-n setting  a change in the developmental lifecycle of protohumans could result in a qualitatively different evolutionary pattern for the emergence of communal languages. © 2009 SNU CSE Biointelligence Lab

© 2009 SNU CSE Biointelligence Lab 14. 1 A Summary of the Major Insights 14.1.2 Bifurcations in the History of Language Bifurcation Appropriate explanatory constructs to account for major transitions in language. The change of language from one seemingly stable mode to another. Principle & Parameters model For frequencies a and b  a=b: no evolutionary change, a>b or a<b: one language is stable and the other is unstable. A change in frequency can make a language go from a stable to an unstable state. If the usage frequencies cross a threshold, then rapid change may come about. © 2009 SNU CSE Biointelligence Lab

© 2009 SNU CSE Biointelligence Lab 14. 1 A Summary of the Major Insights 14.1.3 Natural Selection and the Emergence of Language Coherence emerges only if the learning fidelity is high (for every possible target grammar, the learner will learn it with high confidence). Learning from parents Natural selection based on communicative fitness is necessary for the emergence of a shared linguistic system. Learning from community Natural selection is not necessary  it is not necessary to postulate mechanisms of natural selection for the emergence of language. © 2009 SNU CSE Biointelligence Lab

© 2009 SNU CSE Biointelligence Lab 14.2 Future Directions (1/2) 1. Need for a network of influences on a child  a better understanding on the effect of network topology on the dynamics of language evolution is required. 2-1. Finite population size  a stochastic process  what is the relationship between the stationary distribution of a stochastic process and the attractors of the dynamical system? 2-2. Dependency only between successive generations  what is the effect of more complicated generation structures on learning? 3. Bilingual and monolingual models of learning could have different evolutionary consequences. 4. What is the relationship between first-and second-language learning and their relative effects on language change? © 2009 SNU CSE Biointelligence Lab

© 2009 SNU CSE Biointelligence Lab 14.2 Future Directions (2/2) 5. No commitment to detail of linguistic theory or learning theory  every choice of H (human grammatical systems) and A (learning algorithm) would give rise to a potentially different population dynamics. 6. The role of fitness and natural selection is poorly understood. 7. Evolution of novel structure is poorly introduced. 8. Conditions necessary for emergence of coherence  learning fidelity must be high (H needs to be a highly restricted family). 9. Exploring some of the natural evolutionary questions in the context of animal communication. © 2009 SNU CSE Biointelligence Lab

14.2.2 Connections to Other Discipline Analogies between evolution in linguistic populations and biological populations Grammars are formal and discrete. Genes are discrete. Grammars are transmitted from one generation to the next via learning. Statistical physics and language evolution Statistical physics: deriving the characteristics of the ensemble from the statistics of individual particles. Phase transition & bifurcation, linguistic agents & particles, learning fidelity (a measure of how well a learner learns a potential target language) & temperature. A bifurcation led to the transition from incoherent regimes to coherent regimes (a majority of the population spoke a shared language). Language evolution as a particular instantiation of cultural evolution The contents of this book could also serve theories of cultural evolution. © 2009 SNU CSE Biointelligence Lab

© 2009 SNU CSE Biointelligence Lab 14.3 A Concluding Thought A historical perspective provides a deeper appreciation of why things are the way they are. “Nothing in biology makes sense except in the light of evolution” An evolutionary perspective will provide a richer understanding of the fundamental nature of human language, and more generally of communication in humans, animals, and machines. © 2009 SNU CSE Biointelligence Lab