Linguistics 431/631: Connectionist language modeling
Ben Bergen
Meeting 7: More complex morphology
October 3, 2006
MacWhinney & Leinbach
MacWhinney & Leinbach respond to all of Pinker and Prince's criticisms in a new connectionist model of the English past tense
á
Phonological coding, e.g. bet
á 6949 verb forms, presented according to their frequency in the Brown corpus
á Structure
á After 24,000 epochs, produced past tenses with 97% accuracy, and resolved many of P&PÕs concerns with R&MÕs model
á But still had some problems
o Did not memorize frequent irregulars early
o Could not deal with homophones
o DidnÕt do well with low frequency irregular compounds (like underwent)
o DidnÕt do well with derived verbs, like flied
á The overall problem Ð the model (like R&M) has no representation for lexical items
ConComp
In other work, MacWhinney has described another model, ConComp, which is a connectionist model of determiners and nouns (in German) that includes lexical representations
á There are a number of structured connectionist models in which lexical items are represented
á ConComp has lexical representations be learned
á The task is learning to produce the right forms of German nouns and their determiners
o Determiners
o Nouns
á Structure
á The model has three major parts
o CatNet learns lexical categories, which are associations between particular articles and particular cases and numbers
o ArtNet selects an appropriate article for a word with a particular phonology, case, number, and lexical category
o StemNet inflects the noun