When learners surpass their models: mathematical modeling of learning from an inconsistent source

DOI10.1007/S11538-014-9990-2MaRDI QIDQ467753zbMATH OpenOpenAlexWikidataFDO

Authors Yelena Mandelshtam, Natalia L. Komarova

Publication date 5 November 2014

Published in Bulletin of Mathematical Biology (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1402.4678, https://escholarship.org/uc/item/3ns9h89t

zbMATH Keywords

cultural transmissionlearningMarkov chainsregularization

Mathematics Subject Classification ID

Applications of Markov chains and discrete-time Markov processes on general state spaces (social mobility, learning theory, industrial processes, etc.) (60J20) Memory and learning in psychology (91E40) Linguistics (91F20)

Abstract: We present a new algorithm to model and investigate the learning process of a learner mastering a set of grammatical rules from an inconsistent source. The compelling interest of human language acquisition is that the learning succeeds in virtually every case, despite the fact that the input data are formally inadequate to explain the success of learning. Our model explains how a learner can successfully learn from or even surpass its imperfect source without possessing any additional biases or constraints about the types of patterns that exist in the language. We use the data collected by Singleton and Newport (2004) on the performance of a 7-year boy Simon, who mastered the American Sign Language (ASL) by learning it from his parents, both of whom were imperfect speakers of ASL. We show that the algorithm possesses a frequency-boosting property, whereby the frequency of the most common form of the source is increased by the learner. We also explain several key features of Simon's ASL.

Review text

In cultural (especially linguistic) transmission across generations, children tend to regularize the often inconsistent rules they face, far more than adults do. One simple means of regularization is what the authors call ``frequency boosting. Faced with contradictory language rules for instance, children tend to use the more frequent one more often than they actually encounter it, whereas adults tend to reproduce the observed frequency. In this paper, the authors develop a very simple dynamical algorithm to model how children achieve such frequency boosting. Imagine that different rules apply to the same situation (think of different pronunciations of a word, for example), and a child is given inconsistent information about which rule should be applied. Basically, the learning algorithm developed by the authors amounts to increment by a fixed number the probability to use a certain rule R whenever R is encountered, and to decrement the probability of competing rules. Through this process, the probability of the most frequent rule quickly reaches the maximum (100). The model does not explain why and how adults, on the other hand, more or less reproduce observed frequencies.

Recommendations

• Mathematical modeling of learning from an inconsistent source: a nonlinear approach
• A model of language learning with semantics and meaning-preserving corrections
• When unlearning helps
• Characterization of language learning front informant under various monotonicity constraints
• Towards a robuster interpretive parsing. Learning from overt forms in optimality theory

Cites work

• Evolution of universal grammar
• Markov processes and learning models
• Stochastic dynamic models of response time and accuracy: A foundation primer

Cited in

(3)

• Spatial evolution of regularization in learned behavior of animals
• Mathematical modeling of learning from an inconsistent source: a nonlinear approach
• Learning dishonesty

This page was built for publication: When learners surpass their models: mathematical modeling of learning from an inconsistent source

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q467753)