A morphosyntactic inductive bias in artificial language learning

Itamar Kastner, Tal Linzen

Research output: Chapter in Book/Report/Conference proceedingChapter (peer-reviewed)peer-review

Abstract

A learner’s job is to look at the data and find patterns. But in any given dataset, a large number of patterns can be found if the hypothesis space is not constrained. This is particularly true for language acquisition, where the learner has in principle infinitely many hypotheses to pursue, all of which match the data. Polar questions in English are formed by fronting the auxiliary before the subject: Is the man who is tall in the other room?When acquiring language, children seem to gradually acquire this rule, but they do not adopt other hypotheses (Crain & Nakayama 1987). For instance, they do not conclude that the sixth word of the sentence should be fronted, even though the auxiliary is starts off as the sixth word. The hypothesis space explored by the learner appears to be constrained in many parts of the grammar. It is therefore important to study which biases a human learner might bring with them when making generalizations over the data; not all patterns in the input are necessarily learned as such.
Original languageEnglish
Title of host publicationProceedings of The 48th meeting of the North East Linguistics Society (NELS 48)
EditorsSherry Hucklebridge, Max Nelson
PublisherGLSA Publications
Number of pages10
Volume2
ISBN (Print)9781727605815
Publication statusPublished - 23 Oct 2018

Fingerprint Dive into the research topics of 'A morphosyntactic inductive bias in artificial language learning'. Together they form a unique fingerprint.

Cite this