Poster B39, Tuesday, August 20, 2019, 3:15 – 5:00 pm, Restaurant Hall
Modulating expectations with connectives in discourse comprehension
Linmin Zhang1,2, Linyang He3,4, Xing Tian1,2;1NYU Shanghai, 2NYU-ECNU Institute of Brain and Cognitive Science, 3Fudan University, 4Toyota Technological Institute at Chicago
Introduction. Natural language connectives (e.g., but, even so, so, because) enable humans to go beyond describing single events and reason through relations among them. Most typically, connectives address the conditional probability or causal dependence among events, modulating our expectation in processing discourse-level meanings. For example, in "Mary failed in the exam, (even so,) she celebrated", the conditional probability "P(Mary celebrated | Mary failed in the exam)" should be low according to our world knowledge. However, the use of concessive connective "even so" here reverses our expectation, improving the coherence of the whole discourse and leading to an attenuated N400 during the processing of target words (here "celebrated") (Xiang & Kuperberg, 2015). Moreover, the use of "even so" also enhances predictive processing, leading to larger N400 attenuation between conditions with this connective (e.g., Mary failed/aced in the exam, even so, she celebrated) than between those without this connective (e.g., Mary failed/aced in the exam; she celebrated) (Xiang & Kuperberg, 2015). How do other types of connectives modulate our expectation and affect predictive processing? Here we adopted a 3 by 2 design, in which the use of connectives (no connective (null) vs. but vs. so) and conditional probabilities (high vs. low) between the two events in a discourse are the two factors (sample stimuli: Mary lived far away, (null / but / so) her commuting time was (long / short)). We aimed to test whether (i) both "but" and "so" enhance predictive processing so that larger N400 attenuation can be observed for conditions with (vs. without) these connectives and (ii) compared to conditions without connectives, only "but", but not "so", reverses expectation and leads to larger N400 effects for conditions with high P(event 2 | event 1). Methods. We collected 32-channel EEG data from 16 participants (native speakers of Chinese). Each participant read 180 trials in Chinese, with a word-by-word presentation, and answered comprehension questions after 1/3 of the trials. The manipulation of conditional probabilities between the two events was on the last word of the whole discourse. N400 effects were examined between 300 to 500 ms after the onset of the last word in each trial. Results. Within this time window, we found that compared to conditions without connectives, conditions with "but" or "so" elicited larger EEG differences between the discourses with high vs. low P(event 2 | event 1), suggesting that the use of connectives enhanced predictive processing. With the use of "but", discourses with high P(event 2 | event 1) elicited larger EEG responses than those with low P(event 2|event 1), suggesting that "but" reversed participants' expectation in discourse processing. Intriguingly, the use of "so" showed the same pattern as the use of "but". Presumably, these results suggest that different types of connectives share the same mechanism in enhancing general predictive processing, but differ in modulating specific expectation in context. Conclusion. In discourse comprehension, the use of connectives robustly enhances predictive processing. Some connectives reverse expectation, while the modulating effects of others do not solely depend on conditional probabilities.
Themes: Meaning: Discourse and Pragmatics, Meaning: Combinatorial Semantics
Method: Electrophysiology (MEG/EEG/ECOG)