Morphology and Neural Networks
A Little Morphology
- "Regular" morphology
- Affixation: suffixation, prefixation, circumfixation, infixation
- Deletion
- Mutation
- Templates
- Reduplication
- "Irregular" morphology
- Morphophonology: phonological alternation in morphemes
Morphological Tasks
- Comprehension: mapping forms to meanings
- Production: mapping meanings to forms
- Mapping forms to forms
Neural Networks and Morphological Tasks
- Feedforward neural networks
- Supervised, error-driven learning
- Representing "meanings"
- Representing forms of different lengths
The Rumelhart and McClelland Model
- Exhibited U-shaped behavior
- Represented forms of different lengths using contextualized phonemes
(Wickelphones)
- Discontinuity in training set during training
- The Pinker and Prince attack
- Improvements on R & M (debate)
Learning to Recognize and Produce Words (Gasser)
- Production and comprehension in a network
- Representing words of different lengths in a recurrent network
- Relating production and comprehension through intermediate
phonological representations
- Learning regular "rules" of different types (all but
reduplication)
- Transfer from one "rule" to another
Last updated: 15 April 1997
URL: http://www.indiana.edu/~gasser/L700/morphology.html
Comments:
gasser@salsa.cs.indiana.edu
Copyright 1997,
The Trustees of
Indiana University