Morphological and pseudomorphological effects in English visual word processing: How much can we attribute the statistical structure of the language?
- Patience Stevens, Carnegie Mellon University, Pittsburgh, Pennsylvania, United States
- David Plaut, Psychology, Carnegie Mellon University, Pittsburgh, Pennsylvania, United States
AbstractThe statistical structure of a given language likely drives our sensitivity to words' morphological structure. The current work begins to investigate to what degree morphological processing effects observed in visual word recognition can be attributed to statistical regularities between orthography and semantics in English, without any prior knowledge or explicitly coded processes. We trained a simple feedforward neural network on form-to-meaning mappings for words from an English educational text corpus. Over the course of training, we originally examined the network's processing times for prime-target word pairs taken from two masked primed lexical decision studies (Rastle, Davis & New, 2004; Beyersmann, Castles, & Coltheart, 2012) to determine if the network was learning similar sensitivities to those seen in human participants. Results showed no morphological sensitivity to prime-target pairs with a transparent morphological relationship (e.g., teacher - TEACH) or an opaque morphological relationship (e.g., corner - CORN). To increase power, unique prime-target pairs from a larger set of studies (10 in total) were added to the testing set. With the larger testing set, strong transparent morphological priming effects were observed, while opaque morphological priming was nonexistent. This work shows that morphological sensitivity can emerge without any explicit knowledge of morphemes or word structure, and that opaque morphological priming cannot be explained solely by feedfoward mapping of existing orthographic-semantic regularities. Preliminary work on a more dynamic and neurally-plausible model meant to better capture emerging morphological processing effects is described.