Learning Inflectional Classes
Authors:
- 1 October 2006
- journal article
- Published by Taylor & Francis in Language Learning and Development
- Vol. 2 (4) , 219-250
Abstract
We describe several experiments in computational modeling that require a system to infer the existence of inflectional classes, which are morphological classes that play a role in neither syntax nor semantics. A feed-forward connectionist model will utilize patterns of activation (rather than units) in its hidden layer to encode such abstract morphological patterns. We compare this with discussion of similar problems by linguists (Aronoff, 1994; Haspelmath, 2002) and psycholinguists (MacWhinney, Leinbach, Taraban, & McDonald, 1989) and draw conclusions regarding the relationship between the goals and psycholinguistics studies of language acquisition.Keywords
This publication has 21 references indexed in Scilit:
- An algorithm for the unsupervised learning of morphologyNatural Language Engineering, 2006
- Unsupervised segmentation of words using prior distributions of morph length and frequency2003
- Gauss-Newton approximation to Bayesian learning2002
- Unsupervised Learning of the Morphology of a Natural LanguageComputational Linguistics, 2001
- An Efficient, Probabilistically Sound Algorithm for Segmentation and Word DiscoveryMachine Learning, 1999
- Computation of Conditional Probability Statistics by 8-Month-Old InfantsPsychological Science, 1998
- Bayesian InterpolationNeural Computation, 1992
- Modeling Brain Function1989
- Language learning: Cues or rules?Journal of Memory and Language, 1989
- Syntactic Structures1957