Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
How long is long? M100 response in Bangla tracks number of phonemes not graphemes of glyphs.
Poster B75 in Poster Session B and Reception, Thursday, October 6, 6:30 - 8:30 pm EDT, Millennium Hall
Swarnendu Moitra1, Dustin A. Chacón2,3, Linnaea Stockall1; 1Queen Mary University of London, 2New York University Abu Dhabi, 3University of Georgia
[INTRODUCTION] Electrophysiological responses demonstrate an early effect of word length (M100) and stem-to-whole word transition probability (TP; M170). These effects localize to posterior and anterior left fusiform gyrus (visual word form area) respectively. The M100 is usually taken to reflect lower level visual processing, while the M170 has been interpreted as reflecting early stages of morphological analysis. However, most research on these responses has focused on languages written in alphabetic scripts, with one character roughly corresponding to one phoneme. How universal are the M100/M170 responses, and does the M100 also reflect linguistic analysis, or a psychophysical response to visual complexity? We report on an MEG study on Bangla, a language with an abugida script. In abugidas, one character corresponds to a consonant. Vowel ligatures are written on either side of the consonant, above, or below it. Some consonants have an unwritten "implicit" vowel, allowing some consonant clusters to be represented as a single complex character. Consequently, there are several distinct ways to quantify word length, e.g. প্র(prô) consists of 1 grapheme, 2 glyphs: প(p) and ্র(r), and 3 phonemes: /p/,/r/,/o/. We ask two questions:(1) Which measure of word length best correlates with the M100 response? and (2) Is a stem-to-whole word TP-effect observed in the M170 response? Preliminary results suggest that M100 responses reflect the number of phonemes in the word, suggesting that the M100 response indexes rapid abstract linguistic analysis. [METHODS] MEG recordings were obtained from 22 Bangla speakers(18-62 years,X̄=28) as they performed a lexical decision task. Stimuli consisted of 152 morphologically complex words and 152 pseudoword fillers(not reported). Three different measures of length were calculated from indicNLP(966 million words): number of graphemes, number of glyphs (unicode derived, nchar() fn in R), and number of phonemes (native speaker annotated).TP was estimated as the log of lemma frequency to stem frequency. [RESULTS] 600ms epochs were extracted for each word. We used a two-stage regression analysis in which regressions were fit at each time-point and source point per subject for factors of TP and word length.There were six regressions– TP+word length for 3 definitions of word length in two separate time windows. Spatio-temporal cluster-based permutation tests were conducted on the one-sample t-test values derived from the beta coefficient of the regressions in left fusiform gyrus in the M100(100-130ms) and M170(170-200ms)time window. After correction for multiple comparisons, only one word length cluster was significant(p = 0.01), corresponding to number of phonemes in posterior left fusiform gyrus from 100ms-130ms. One TP cluster was significant and negatively correlated in anterior left fusiform gyrus from 170ms-200ms(p = 0.04). No other significant clusters sensitive to any other word length measures were identified. [CONCLUSION] Early visual responses to words suggest rapid analysis of abstract linguistic structure. Our results leverage the complexity of an abugida system to demonstrate that the M100 response is modulated by the number of phonemes in a writing system without a clear one-to-one phoneme-grapheme correspondence, and that the M170 reflects stem-to-word TP, consistent with results for other languages (see Wray et al 2021).
Topic Areas: Morphology, Methods