Poster E26, Saturday, August 18, 3:00 – 4:45 pm, Room 2000AB

Predicting what, when, and how in the course of Japanese classifier-noun comprehension

Hiromu Sakai1, Yohei Oseki1, Naho Orita2;1Waseda University, 2Tokyo University of Science

Predicting up-coming elements is one of the key-features of language comprehension. The question remains to be answered about what types of factors influence prediction how and when in real-time comprehension processes. This study aims to address this issue by examining native speakers’ event-related potentials (ERPs) evoked by classifier-noun associations in Japanese, in which classifiers yield strong forward prediction for up-coming nouns [1]. While we replicated previous observations that predictability modulates N400 amplitude, we discovered that word frequency of classifiers is reflected in left anterior negativity (LAN). The overall results suggest that predictability influences language comprehension through the interaction of factors such as prediction strength, predictor frequency, and prediction satisfaction. Japanese is one of the languages that employ classifier systems in counting elements. For instance, “three cars” must be expressed as “san dai-no kuruma (lit. three bodies of car)” and omission or mismatch of classifiers makes the relevant expression ungrammatical. ERP responses to classifier-noun associations were examined in languages such as Chinese or Japanese. Previous researchers observed that prediction dissatisfaction modulates amplitudes of N400 [2], [3] or LAN [4]. In the present study, we selected the following three classifiers: “-tsu (piece),” “-mai (sheet),” and “-dai (body),” based on the analysis of BCCWJ (Balanced Corpus of Contemporary Written Japanese). “-tsu (piece)” is widely used with either concrete or abstract objects and imposes little semantic restrictions on the associated nouns. “-dai (body),” on the other hand, narrowly restricts semantic types of associated nouns and used only with a large machinery such as cars. “-mai” can be used with any kinds of objects so long as they have a flat shape. As one consequence, “-tsu” is by far the most frequent among classifiers and “-dai” is relatively less frequent. Another consequence is that bigram probabilities of classifier-noun associations with “-dai” is much higher compared to the ones with “-tsu”. This means that “-dai” yields the strongest prediction but “-tsu” yields the weakest prediction. These properties of classifiers allowed us to manipulate prediction strength (high, mid, low) and prediction satisfaction (satisfied, unsatisfied). Twenty native Japanese speakers participated in the experiment. ERPs elicited by nouns preceded by classifiers showed an interaction with prediction strength and prediction satisfaction. First, centro-parietal negativities in the N400 time window are inversely correlated with the prediction strength. The classifier “-tsu”, which triggers the weakest prediction for the up-coming nouns, elicited the largest N400. Second, dissatisfaction of prediction elicited significantly larger N400 with the classifier “-dai” but N400 modulation did not reach significance with other two classifiers. Instead, a significant contrast was observed in left anterior electrodes with the classifier “-tsu”. Summarizing, prediction strength and prediction satisfaction modulated N400 whereas the interaction between prediction satisfaction and predictor frequency is reflected by LAN. [1] Yoshida, M., Aoshima, S., & Phillips, C., 17th CUNY Conference on Human Sentence Processing, 2004. [2] Zhou, X., et al., Neuropsychologia 48, 1551-1562, 2010. [3] Chou, C.-J., et al., Journal of Neurolinguistics 31, 42-54, 2014. [4] Mueller, J., et al., Journal of Cognitive Neuroscience 17:8, 1229-1244, 2005.

Topic Area: Grammar: Syntax