Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Automatic processing of relational structure in language: A frequency-tagging EEG study of Chinese compounds
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
Fengyun Hou1, Alexander Anderson1, Nina Kazanina1,2; 1School of Psychological Science, University of Bristol, 2Department of Basic Neuroscience, University of Geneva
Language is a primary means of exchanging information about the world, and much of the exchange concerns relations between entities in the world. On a par with lexical means for expressing relations, such as prepositions (e.g., lamp ON the bench, lesson AFTER the break), relations can also be expressed morphologically either using overt affixation (possession: Ivy’s cat) or null morphemes (raincoat ‘a coat FOR rain’, snowball ‘ball MADE-OF snow’). Previous studies argued that relational structure is accessed routinely, as shown by facilitatory priming effects between compounds sharing the same semantic relation, e.g. raincoat [coat FOR rain] facilitates rainboots [boots for rain] relative to rainwater [water ORIGINATING FROM rain] (Gagné & Spalding, 2009; Raffray et al, 2007). However, this finding was confounded by the fact that compounds sharing semantic relation frequently also contained semantically related second constituents (e.g., coat & boots are semantically closer than coats & rain), hence the facilitation couldn’t be definitively attributed to the shared semantic relation. More definitive demonstration of automatic extraction of relational structure is needed; we address this question via a frequency-tagging EEG study of Chinese compounds. Twenty native Chinese speakers viewed 60-second-long sequences of Chinese compounds presented at a regular rate of 166.7ms/compound (i.e. frequency-tagged @6Hz). Each sequence consisted of four base stimuli alternating with a fifth oddball (@1.2 Hz). Base compounds all featured the FUNCTION relation between constituents within each XY compound interpretable as ‘Y for X; e.g. 酒瓶 (/jiǔpíng/, wine bottle – ‘bottle FOR wine’), the second constituent 瓶 (/píng/, bottle) specified the function of the first constituent 酒 (/jiǔ/, wine). Oddball stimuli varied across five conditions: a) POSSESSION (‘Y of X’): 蛋 (/dàn/, egg) + 壳 (/ké/, shell) → 蛋壳 (/dànké/, eggshell, ‘shell of egg’); b) COORDINATION (‘X and Y’): 花(/huā/, flower) + 草 (/cǎo/, grass) →花草 (/huācǎo/, ‘flowers and grass’); c) ACTION (‘X act on Y’): 弹 (/tán/, play) + 琴 (/qín/, instrument) →弹琴 (/tánqín/, ‘play[action] instrument [object]’); d) PSEUDO-COMPOUND (‘PC’): 头 (/tóu/, "head") + 柴 (/chái/, "firewood") →头柴 (/tóuchái/, illicit compound); e) FUNCTION (‘Y for X’): 药 (/yào/, medicine) + 箱 (/xiāng/, box) →药箱 (/yàoxiāng/, medicine box, ‘box for medicine'). If semantic relations are extracted automatically, we expected that a spectral peak would appear at the rate of oddballs (i.e. 1.2 Hz and its harmonics) in conditions (a-d) in which the semantic relation within the oddball stimuli was different from that in the base stimuli. This prediction was confirmed by the results of non-parametric cluster-based permutation tests: the power at the oddball rate 1.2 Hz and its harmonics was significantly higher than in the neighboring bins in conditions (a-d), but not in condition (e) in which base and oddball stimuli shared the same relation. These results demonstrate that participants automatically extracted the detailed semantic relations between constituent morphemes in compounds, providing the first neural demonstration of extraction of latent relational structures in language. We place these results into a larger theoretical framework that considers the role of relations across cognitive domains, including vision and memory (Hafri & Firestone, 2021).
Topic Areas: Meaning: Lexical Semantics,