Hierarchical clustering analysis of reading aloud data: a new technique for evaluating the performance of computational models

摘要: DRC (Coltheart et al., 2001) and CDP++ (Perry, Zorzi, & Ziegler, 2010) are two of the most successful models reading aloud. These differ primarily in how their sublexical systems convert letter strings into phonological codes. adopts a set grapheme-to-phoneme conversion rules (GPCs) while uses simple trained network that has been exposed to combination spellings pronunciations known words. Thus far debate between fixed learned associations largely emphasized reaction time experiments, error rates dyslexias, item-level variance from large-scale databases. Recently, Pritchard, Coltheart, Palethorpe, Castles (2012) examined models’ nonword new way. They compared responses produced by those 45 skilled readers. Their item-by-item analysis is informative, but leaves open some questions can be addressed with different technique. Using hierarchical clustering techniques, we first subject data identify if there classes subjects similar each other overall response profiles. We found indeed groups for certain consonant clusters. also tested possibility modeling one well, subjects. does not fit any human reader’s pattern very fits readers as well or better than reader.