Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Linguistic structure dictates the neural encoding of animacy in human language
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Simone Krogh1, Liina Pylkkänen1; 1New York University
Not even the most basic word meanings are impermeable to linguistic structure—the contribution of HOUSE to BEACH HOUSE, GINGERBREAD HOUSE, and HOUSE KEY differs despite being closely related. While we have some understanding of how the brain grasps complex meanings built from simpler parts (Friederici2002TiCS; Bemis&Pylkkänen2011JNeuroscience; PallierEtAl2011PNAS), we know far less about how linguistic structure reshapes the meanings of the words themselves. Here, we used animacy decodability as a proxy to investigate how phrase-level syntax and semantics change the neural encoding of words. Our syntactic manipulation utilized reversible, zero-frequency noun-noun phrases to test the impact of a word’s functional role: Is the neural representation of TIGER the same in VALLEY TIGER (head noun anchoring the phrasal concept) and TIGER VALLEY (modifier whose animacy feature is irrelevant; Kamp&Partee1995Cognition)? Our semantic manipulation contrasted phrases with and without sense-shifting adjectives to examine the impact of linguistic context: Is TIGER in PLASTIC TIGER and ARABIAN TIGER represented as the same sentient animal, or does the phrase-level inanimacy of PLASTIC TIGER overrule word-level animacy (Partee2010PresuppositionsAndDiscourse)? Our phrasal stimuli were presented word-by-word during magnetoencephalography (MEG) recordings (N = 29) and served as test data for decoding analyses. Training data came from separate blocks of ANIMAL/LOCATION single-word trials (TIGER, VALLEY, FOREST, LOBSTER, etc.). This cross-condition decoding approach tested whether isolated words and words embedded in phrases share stable, discriminable neural patterns associated with animacy. To investigate whether results reflected the specific animal/location contrast or animacy more generally, a subset of participants (N = 21) also saw a mixed group of ANIMATE/INANIMATE nouns (BANJO, DENTIST, CANOE, VAMPIRE, etc.). Our decoding analyses employed temporal generalization at the sensor level (King&Dehaene2014TiCS), and significant temporal clusters were probed with source-localized spatial decoding. Group-level accuracy was determined with cluster-based permutation tests (Maris&Oostenveld2007JNeuroscience). Our results proved remarkably consistent: If word-level animacy was incongruent with the structural context, it disappeared from the word’s neural representation. Specifically, for our syntactic manipulation, classifiers trained on single-word ANIMAL/LOCATION trials successfully discriminated words like TIGER and VALLEY when functioning as head nouns (train times: ~220–380ms; test times: ~210–400ms), but not when those same words functioned as modifiers. Such feature-extrapolated animacy localized to the right anterior temporal lobe (ATL) whereas complementary, lexically grounded univariate analyses recruited the left ATL, supporting a potential hemispheric dissociation between non-verbal (right ATL; LeeEtAl2002Epilepsia; Vaz2004Seizure) and verbal (left ATL; LeeEtAl2002Epilepsia; RiceEtAl2015CerebralCortex) semantic memory. Remarkably, our semantic manipulation showed that even head noun animacy can be overruled: While the phrase-level inanimacy of PLASTIC TIGER was robustly encoded in early (train times: ~100–310ms; test times: ~210–400ms) and late (train times: ~260–380ms; test times: ~420–700ms) clusters, word-level animacy for TIGER was undetectable. These phrase-level effects localized to the left ventromedial prefrontal cortex and right inferior frontal gyrus. Results from both experimental manipulations largely replicated with classifiers trained on ANIMATE/INANIMATE trials, indicating that the neural signals captured reflexes of abstract animacy rather than mere superordinate categories. Together, our findings demonstrate how linguistic structure rapidly reshapes word meanings, allowing phrase-level meanings to override fundamental word-level features such as animacy.
Topic Areas: Syntax and Combinatorial Semantics, Meaning: Lexical Semantics