Conditional entropy as an evaluation metrics of morphological analyses: comparing stem-based and word-based approaches
This talk is given by Dr Matteo Pellegrini as part of his visiting IAS Fellowship at the University of Surrey. Hosted by the Surrey Linguistics Circle.
In recent years, detailed analyses of predictability in inflectional systems have been proposed, taking as basic units either stems and endings, or unsegmented word forms. The aim of my talk is comparing the efficiency of stem-based and word-based approaches in a principled way, using Italian verb inflection as a case study.
I will first review stem-based accounts of the structure of Italian verb paradigms proposed in previous work, and present a word-based account of the same data. Results obtained with different approaches will then be compared in detail, and conditional entropy will be used as an evaluation metrics of the complexity of different analyses.
