Lancaster EPrints

A distributional learning account of the acquisition of the locative alternation : corpus analysis and modeling

Twomey, Katherine and Chang, Franklin and Ambridge, Ben (2013) A distributional learning account of the acquisition of the locative alternation : corpus analysis and modeling. In: Proceedings of the 35th Annual Conference of the Cognitive Science Society :. Cognitive Science Society, Austin, TX, pp. 1498-1503. ISBN 9780976831891

[img] Microsoft Word (locatives-model-cogsci-revision-final) - Submitted Version
Download (563Kb)

    Abstract

    Early in acquisition children overgeneralize verbs to ungrammatical structures. The retreat from overgeneralization is linked to the acquisition of verb classes, the semantics of which constrain the structures in which a verb can appear (e.g., Pinker 1989; Ambridge, Pine & Rowland, 2012). How children learn these classes remains unclear. Some argue that distributional regularities in linguistic input provide sufficient evidence for verb classes to emerge and become linked to particular structures. A corpus analysis of the English locative construction (e.g., the woman sprayed water onto the wall/the wall with water) demonstrated that children have similar verb classes to adults. A correspondence analysis revealed that distributional regularities in the input could support these verb classes. Finally, a connectionist simulation was able to model early overgeneralization and retreat through distributional learning of verb classes. These results support a distributional learning account of verb semantics.

    Item Type: Contribution in Book/Report/Proceedings
    Uncontrolled Keywords: Language acquisition ; verb semantics ; distributional learning ; connectionist modelling
    Subjects:
    Departments: Faculty of Science and Technology > Psychology
    ID Code: 71111
    Deposited By: ep_importer_pure
    Deposited On: 07 Oct 2014 00:06
    Refereed?: No
    Published?: Published
    Last Modified: 19 Aug 2018 01:16
    Identification Number:
    URI: http://eprints.lancs.ac.uk/id/eprint/71111

    Actions (login required)

    View Item