Slide Slam S8 Sandbox Series
An MEG Investigation of Category Ambiguous Words’ Number of Roots and the Homonymy/Polysemy Distinction
Kyra Wilson1, Alec Marantz1,2; 1New York University Abu Dhabi, 2New York University
[INTRODUCTION] A pervasive feature of language is lexical ambiguity, where words have many meanings (homonyms) and/or senses (polysemes). English ambiguity extends to syntactic category; stems can serve as nouns or verbs (the/to clash). Some theories treat category ambiguity like derivational morphology--(null) categorizing affixes follow roots (Hypothesis 1). Lexicalist accounts treat noun-verb ambiguous words as distinct lemmas (Hypothesis 2). Further, if polysemy and homonymy are representationally different, Hypothesis 1 would assign one root to polysemous noun-verb pairs (clash) but two to homonyms (duck) (1A). If the difference is a matter of degree, both types share a single root (1B). Some previous work suggests homonyms and polysemes are distinct, citing evidence for a facilitatory polysemy and inhibitory homonymy effect. This study (in preliminary stages due to COVID) finds evidence for 1B: one root for noun-verb ambiguous stems across homonymy and polysemy. [PREDICTIONS] Previous single word recognition MEG studies showed an effect of subcategorization entropy in ATL at 200-300ms PSO (M250). Hypothesis 1 predicts that form-based noun-verb entropy (higher entropy = noun and verb forms have more equal frequencies) coincides with subcategorization and derivational family entropy effects, all reflecting uncertainty over syntactic category continuations from the stem. 1A predicts number of meanings (homonymy) and senses (polysemy) have opposite effects on M350 (=N400m) responses, while 1B predicts the same effect direction. [METHODS] During a visual lexical decision task, we recorded brain activity from 11 adults using magnetoencephalography (MEG). The stimuli were 631 monomorphemic, monosyllabic, noun-verb ambiguous English words having a lexical decision accuracy > 80% plus 631 matched non-words. [RESULTS] We identified an earlier and later peak corresponding to the M250 and M350 in average activity across all items and subjects. For each trial within each subject, we averaged 50ms of activity centered on the group average peaks in corresponding ROIs. These averages were used in linear mixed effects models for further analysis. [M250] The anterior MTG ROI showed an effect of derivational entropy (estimate=-1.601, p=0.002) and noun-verb entropy (estimate=-2.985, p=0.032). [M350] At 360ms there was a negative peak with two larger, more distributed spatial components in anterior STG and MTG. In STG, there was an effect of log word frequency (estimate=-1.933, p<0.001) and number of meanings (estimate=-1.587, p=0.016). In MTG, there was an effect of log word frequency (estimate=-2.264, p<0.001) and number of senses (estimate=-0.133, p=0.038). Crucially, there was no noun-verb entropy and number of meanings interaction. [CONCLUSION] We find support for Hypothesis 1B: noun-verb ambiguous words are represented as a single item and categorized later with a null affix. We find early derivational and noun-verb entropy effects around 250ms (both reflecting potential affixation of category to a stem), and a later word frequency effect around 360ms indicating continuing lexical access. Number of senses and meanings had inhibitory effects, and there was no noun-verb entropy and number of meanings interaction, suggesting homonyms and polysemes are not represented differently.