Search Abstracts | Symposia | Slide Sessions | Poster Sessions
From N400 Amplitude to Decodable Meaning
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Gayane Ghazaryan1, Aino Saranpää1, Tiina Lindh-Knuutila1, Marijn van Vliet1, Riitta Salmelin1; 1Aalto University
The N400 component, a response sensitive to semantic context, has become a widely used marker of semantic processing during language comprehension. A more recently developed complementary approach to study semantics in the brain is to use multivariate pattern analysis to perform neural decoding of semantic vectors from brain activity. In the present study, we investigated the relationship between N400 modulation and semantic decoding performance using magnetoencephalography (MEG) responses in a controlled priming experiment. Participants (n=25), native speakers of Finnish, read word triplets in which target words appeared after two primes. The semantic relatedness between the primes and the target was systematically varied (highly related, moderately related, unrelated) based on their distance in a word2vec embedding space. This design created graded contextual support while preserving a repeated target word structure suitable for decoding. We combined conventional analyses of evoked responses in the N400 window with zero-shot semantic decoding using word2vec word embeddings to ask whether conditions that attenuate or enhance the N400 also differ in how reliably target word meaning can be recovered from neural activity. We further examined the timing of semantic decodability relative to the N400 window (300-500 ms), its generalization across relatedness contexts, and its expression in source-localized MEG estimates. Semantic relatedness modulated the evoked response in the expected direction: targets preceded by unrelated primes produced larger N400 responses than targets preceded by moderately or highly related primes. At the same time, semantic information about target words was decodable from the whole-brain MEG signal. When decoding was examined separately for each relatedness condition, decoding performance varied across both conditions and time: target word representations were most robustly expressed when contextual support was weakest, and the N400 response was largest. However, the temporal decoding pattern showed that the N400 window alone did not capture the full relationship between contextual support and semantic representation. Highly related targets were decoded above chance only in the pre-N400 window (100-300 ms), moderately related targets only in the N400 window (300-500 ms), and unrelated targets across pre-N400 (100-300 ms) , N400 (300-500 ms), and post-N400 (500-700 ms) windows. Together, the findings of this study suggest that contextual support shapes semantic processing in two complementary ways. It not only modulates the N400 response; it also changes the temporal dynamics of semantic representations, with strongly supported targets showing earlier decodability and weakly supported targets showing stronger and more sustained decodability.
Topic Areas: Meaning: Lexical Semantics, Computational Approaches