Paper

Modeling Morphological Processing in Human Magnetoencephalography

---

Abstract

In this paper, we conduct a magnetoencephalography (MEG) lexical decision experiment and computationally model morphological processing in the human brain, especially the Visual Word Form Area (VWFA) in the visual ventral stream. Five neurocomputational models of morphological processing are constructed and evaluated against human neural activities: Character Markov Model and Syllable Markov Model as "amorphous" models without morpheme units, and Morpheme Markov Model, Hidden Markov Model (HMM), and Probabilistic Context-Free Grammar (PCFG) as "morphous" models with morpheme units structured linearly or hierarchically. Our MEG experiment and computational modeling demonstrate that "morphous" models outperformed "amorphous" models, PCFG was most neurologically accurate among "morphous" models, and PCFG better explained nested words with non-local dependencies between prefixes and suffixes. These results strongly suggest that morphemes are represented in the human brain and parsed into hierarchical morphological structures.

Keywords: morphology, computational modeling, magnetoencephalography

How to Cite:

Oseki, Y. & Marantz, A., (2020) “Modeling Morphological Processing in Human Magnetoencephalography”, Society for Computation in Linguistics 3(1), 209-219. doi: https://doi.org/10.7275/7tak-9s75