Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Expectation management cues modulate sentence processing
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Benjamin Menashe1, Michal Ben-Shachar1; 1Gonda Multidisciplinary Brain Research Center, Bar-Ilan University
During sentence processing, humans continuously use context to form expectations about upcoming linguistic stimuli. Encountering a semantic violation of expectations elicits known behavioral and neural processing costs. Natural language enables expectation management via mirative markers, linguistic elements that encode an upcoming violation of expectations (e.g., surprisingly, remarkably). Currently, empirical evidence regarding the cognitive processing effects of miratives is inconclusive, as observed by conflicting results from two previous electrophysiological studies (Rasenberg et al., 2020; Xiang & Kuperberg, 2015). Several hypotheses remain: miratives lead to a preemptive adjustment of expectations and attenuate the processing cost of surprise (strong modulation), miratives do not alter expectations (no modulation), or miratives result in a more complex timeline, such that the processing cost of surprise emerges initially but is quickly suppressed (dynamic modulation). In a series of experiments, we show that mirative markers modify subsequent expectations, such that immediate predictability effects initially emerge but are suppressed soon after, congruent with dynamic modulation. First, offline sentence completion experiments reveal that mirative markers significantly increase overall response entropy (shown in both Hebrew, N = 80, and English, N = 40). In complementary analyses of Large Language Models, mirative markers similarly increased the entropy of next-token distributions and, furthermore, significantly shifted sentence-embedding representations along a latent axis of expectedness. Together, these offline studies demonstrate that mirative markers modify expectations for upcoming sentential events, in line with linguistic theory (Simeonova, 2015; Zanuttini & Portner, 2003). Next, to investigate how mirative markers affect human online sentence processing, we conducted three self-paced reading experiments (N = 60 each, using different stimulus sets). Hebrew speaking adults read sentences containing expected and unexpected target words, preceded by either neutral or mirative markers. Linear mixed-effects models fitted to participant RT revealed conflicting results: predictability effects were significantly attenuated by miratives in one study, but completely unaffected by miratives in the other two studies, at either the target or post-target “spillover” regions. To resolve these inconsistencies and probe less controlled indices of processing cost, we conducted a simultaneous eye-tracking and EEG experiment (N = 30). Participants were presented with sentences (presented in rapid serial visual presentation) containing expected and unexpected target words preceded by either neutral or mirative markers. Online processing cost was quantified by oculomotor inhibition (OMI) duration and by the N400 amplitude. Linear mixed-effects models revealed predictability effects at the target word, specifically longer OMI duration and enhanced N400 amplitude, without a documented effect of mirative presence. However, mirative presence significantly attenuated the OMI processing cost at the “spillover” region. These findings support the dynamic modulation hypothesis: while initial processing appears encapsulated and unaffected by explicit expectation management, mirative markers are subsequently integrated to resolve expectation violations and reduce late processing costs. Taken together, these studies demonstrate that expectation management facilitates efficient communication, at least to some extent. Readers utilize expectation management cues to reduce late processing costs, which ultimately allows for easier integration of unexpected messages.
Topic Areas: Meaning: Discourse and Pragmatics, Reading