VIDEO DESCRIPTION:

Play, download and edit the free video Unsupervised learning: computation and connectionism through modal logic with grammar as a case study.

Infants, at some stage, produce above-chance grammatical outputs. This demonstrates our innate knowledge of grammar. Without innateness, all learning is through interactions with the environment under supervision. If there be an innate language module that encodes grammaticality, then language acquisition is not completely supervised. I propose the following unsupervised learning algorithm to explain such innateness. In supervised learning, such as in a backprop connectionist network, the error of an output compared with the correct output, provided by the supervisor, is fed back to the network for connection weight modification between nodes. With an innate utility function that assigns utility to outcomes, the difference between the expected utility of an output and the actual utility of the output may be used as the error for learning without such a supervisor. At some time, some linguistic output, such as a sentence, instead of another is expected to have higher utility because the utility function assigns higher utility through grammaticality. The innate language module, with such learning algorithm, explains the non-randomness of linguistic outputs of infants. Methodologically, although Chomsky (1957/2002) assumes Markov processes to be incompatible with his generative grammar, I propose a modal logic with syntactic accessibility between linguistic objects that takes prior syntactic states into posterior ones with applications of grammar rules. A syntactic function, which constitutes the language module, that encodes such transitions, given linguistic inputs, into posterior syntactic states with grammatical outputs better than chance explains the innate knowledge of grammar. What is universal cross-culture may thus be investigated.

References
Chomsky, Noam. 1957/2002. Syntactic Structures. 2nd ed. Berlin: Mouton de Gruyter.

2021.8.26

Brain and Language 2021
Faculty of Education and Psychology, University of Jyväskylä
Jyväskylä, Finland

Download, play and edit free videos and free audios from Unsupervised learning: computation and connectionism through modal logic with grammar as a case study using RedcoolMedia.net web apps