University of Cambridge > Talks.cam > Cambridge University Linguistic Society > Towards a model of morphological processing grounded in principles of discriminative learning

Towards a model of morphological processing grounded in principles of discriminative learning

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Alison Biggs.

Most research on morphological processing takes for granted that morphemes (or exponents) exist, and that they mediate lexical access in speech production and language comprehension. It is only in subsymbolic connectionist approaches that the theoretical concept of morphemes (or exponents) plays no role. However, after a series of promising studies promoting the triangle model around the beginning of the present century, reports of advances in this area have not been forthcoming. As a consequence, the scholarly discourse has drifted towards a view of morphological processing in which units such as morphemes (or exponents) are part and parcel of interactive activation networks.

In this presentation, I will present a new computational model for lexical processing that takes inspiration on the one hand from word and paradigm morphology, and on the other hand from learning theory in psychology. This new model is strictly a-morphous, in the sense that there are no units representing stems, morphemes, or exponents. This does not mean that the model would be committed to subsymbolic representations. To the contrary, it’s elementary units are symbolic, and range from units for melodies in articulatory scores to orthographic n-grams (typically letter bigrams or trigrams), and from lexeme nodes (nodes mediating between form and meaning for both content words and grammatical features such as plurality or tense) to units representing snippets of acoustic information (operationalized with diphones or demi-syllables). These different sets of units participate in a range of networks, the weights of which are estimated with the Rescorla-Wagner equations from learning theory in psychology. Complementing networks that learn the mapping from form to lexemes during listening and reading are are networks learning the linear order of form units (n-graphs or n-phones) for combinations of lexemes. How this model works will be illustrated by taking a closer look at how it deals with inflectional morphology. Examples from Dutch and English will be discussed, as well as examples from fusional and agglutinative inflectional paradigms.

This talk is part of the Cambridge University Linguistic Society series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2020 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity