Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Semantic Seeds Shape the Neural Dynamics of Distributional Learning
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Abigail Laver1, Albert E. Kim2, John C. Trueswell1; 1University of Pennsylvania, 2University of Colorado, Boulder
Language learners often encounter long stretches of unfamiliar words, punctuated by a few familiar meaningful items. Familiar words may act as semantic seeds, shaping the neural dynamics of distributional learning and the semantic inferences learners draw about the otherwise meaningless unfamiliar words. We tested this using EEG and behavioral measures in a novel Seeded Distributional Learning Paradigm, in which initial vocabulary was manipulated before participants encountered an artificial language. In a Seed Phase, Semantic Seed participants learned meanings for a few pseudowords referring to animates, inanimates, and events. Non-Semantic Seed participants were exposed to these same word-forms without meanings. In the Distributional Exposure Phase, participants encountered an ~35-pseudoword Subject-Object-Verb artificial language containing seed and unknown nonseed words arranged into ~150 four-sentence dialogues. The grammar defined three distributional categories, each seeded by a few words with animate, inanimate, or event meanings, respectively. Exposure contained no referent world; thus, for nonseed words, the only learning cues were distributional. Across three behavioral studies, 144 participants completed either Semantic or Non-Semantic Seeding and then the Exposure Phase. In a Test Phase, learning was assessed with a Word-Meaning Selection Task, where participants guessed nonseed words’ meanings, choosing among animate, inanimate, and event referents. Semantic Seed, but not Non-Semantic Seed, participants successfully inferred nonseed meanings, reliably preferring referents consistent with their distributional categories: nonseed words in categories containing animate, inanimate, or event seeds were preferentially assigned animate, inanimate, or event meanings, respectively. A Sentence-Familiarity Task assessed distributional learning during the Test Phase. Participants reliably rated novel grammatical sentences as more familiar than ungrammatical sentences, with no effect size differences between Seed groups. Thus, semantic seeds did not increase end-state sensitivity to distributional structure; rather, meaningful words within a distributional category shaped inferences about otherwise meaningless words in that category. To examine how semantic seeds affect learning as it unfolds, a fourth study used EEG to track neural responses during exposure. Target enrollment is 64, 32 per Seed group; currently N=24, 12 per group. The same Seeding and Exposure paradigm was presented via Rapid Serial Visual Presentation. ERPs were quantified for nonseed words during exposure while participants read dialogues. We examined occipito-temporal N170 and centro-parietal N400 responses as indices of visual word-form processing and emerging meaning-related sensitivity, respectively. N170 amplitude decreased over exposure, p=.013, consistent with increased efficiency in processing novel nonseed word-forms. This reduction was larger for Semantic than Non-Semantic Seed participants, p=.031, suggesting that semantic grounding modulated novel word-form processing during learning. N400 amplitude decreased over exposure, p<.001, and trended toward reduced amplitude for Semantic Seed participants, p=.063, suggesting facilitation of meaning-related representations. Behavioral measures were also collected and will test whether exposure-related neural changes predict semantic generalization, while neural decoding will assess whether category structure emerges in the EEG signal over time. Together, these results suggest that sparse semantic grounding changes both the neural trajectory and semantic consequences of distributional learning: meaningful words within a distributional category shape how learners process novel word-forms and what they infer about otherwise meaningless words in that category.
Topic Areas: Language Development/Acquisition, Meaning: Lexical Semantics