Simulating the N400 ERP Component as Semantic Network Error: Insights from a Feature-Based Connectionist Attractor Model of Word Meaning
URL with Digital Object Identifier
The N400 ERP component is widely used in research on language and semantic memory. Although the component’s relation to semantic processing is well-established, the computational mechanisms underlying N400 generation are currently unclear (Kutas & Federmeier, 2011). We explored the mechanisms underlying the N400 by examining how a connectionist model’s performance measures covary with N400 amplitudes. We simulated seven N400 effects obtained in human empirical research. Network error was consistently in the same direction as N400 amplitudes, namely larger for low frequency words, larger for words with many features, larger for words with many orthographic neighbors, and smaller for semantically related target words as well as repeated words. Furthermore, the repetition-induced decrease was stronger for low frequency words, and for words with many semantic features. In contrast, semantic activation corresponded less well with the N400. Our results suggest an interesting relation between N400 amplitudes and semantic network error. In psychological terms, error values in connectionist models have been conceptualized as implicit prediction error, and we interpret our results as support for the idea that N400 amplitudes reflect implicit prediction error in semantic memory (McClelland, 1994).