Slide Slam I10
Modulation of the N400 by morphological composition and lexical access: An ERP study of Japanese derived nouns
Jun Nakajima1, Shinri Ohta1; 1Kyushu University, Fukuoka, Japan
Previous neurolinguistic studies reported that lexical access and morphological composition elicited cortical activation around 300-400 ms after the word onset (M350/N400) in the middle and superior temporal gyri. Here, we examined the processing of the two types of Japanese nouns derived from adjectives (derived -sa noun: ama-sa, sweetness, derived -mi noun: ama-mi, sweetness). While the -sa suffix is extremely productive, -mi suffix attaches to 30 adjectives. Previous studies proposed that the native speakers decomposed the derived -sa nouns into morphemes, while they memorized the derived -mi nouns and cannot decompose into morphemes, suggesting that only the former nouns show the modulation of the cortical activation M350/N400 reflecting transition probability (TP) between the adjectival stem and suffix. In the present study, we examined whether these two types of derived nouns were processed by the common neural mechanism or two distinct systems. We recruited 20 right-handed native speakers of Japanese (10 males, 22.8±4.5 yrs.). We used 25 items for each of the derived -sa nouns, derived -mi nouns, non-derived -sa nouns (ookami, wolf), and non-derived -sa nouns (nagisa, beach), as well as the same number of nonwords (total 200 items). The participants performed a visual lexical decision task. Because we were interested in the derived nouns, we targeted the derived -sa nouns and -mi nouns for the following EEG analyses. We used a 64-electrode EEG system (Nihon-Kohden) for the EEG data collection and MNE-Python for the data analysis (Gramfort et al., 2013). Since our primary target was the N400, we selected left centro-parietal electrodes (P5, CP3, CP5) for the region of interest. The analysis time window was restricted to 300-400 ms after the word onset. A two-way repeated-measures analysis of variance (rANOVA) (suffix × derivedness) for the accuracy showed the significant main effects of suffix and derivedness, as well as the significant interaction (suffix: p = .0007; derivedness: p < .0001; interaction: p < .0001). A two-way rANOVA for the RTs also showed the main effect of derivedness and the significant interaction were significant (derivedness: p = .0026, interaction: =.0010), but the main effect of suffix was not (p = .70). These results suggest that the derived nouns had lower processing loads. For the ERP data, we examined whether logarithmic LogTP between the adjectival stem and suffix, as well as the word length and the ultima (the last syllable of a word, -sa or -mi), modulated the N400 using a linear mixed-effect model (lme4 and lmerTest packages on R). We included random intercepts for the subject and stimulus to account for participant-specific variability and item-specific idiosyncrasies. We found that the model with the word length and LogTP showed the lowest AIC, indicating that both the word length and LogTP between morphemes modulated the N400. Moreover, we found that the model with the word length and token frequency showed the second-lowest AIC, suggesting that the lexical access modulated the N400 as well. These results demonstrated that both types of Japanese-derived nouns have the same processing stages, i.e., decomposition and recombination of morphemes.