A neurophysiological alignment metric reveals a supportive role of unexpected stimuli in naturalistic language comprehension

Abstract

Individuals use language every day to communicate their thoughts, emotions and expectations to those around them. Communication also incorporates receiving continuous and complex streams of information, not all of which is easily predictable. The memory literature suggests that this unexpected information may be processed more effectively when an individual possesses an adequate schema of the local context. We propose that, in language comprehension, we can determine the presence of a relevant scheme via an alignment metric, which compares an individuals brain response to language (N400 amplitudes) with an expected response as determined by the probabilistic contingencies of the current context (surprisal). When alignment is higher, an individuals internal model of the world is assumed to closely mirror the environment, constituting an effective schema. The present study sought to explore how individuals process unexpected information and whether successful comprehension is related to the alignment of internal predictive models to the external environment. It was hypothesised that higher alignment would support better comprehension of unexpected information than lower alignment. 40 participants (27F; mean age: 24.6 years [SD: 5.94]) listened to 12 short stories (3 genres, audio presentation) while their electroencephalogram (EEG) was recorded. Comprehension was tested via 6 multiple choice questions per story. A comprehension linear mixed-effects model (LMM) was computed for two regions of interest and two predictability metrics (N400 amplitudes and surprisal). A functional role of unexpectedness was demonstrated in all four models, where greater unexpectedness was related to higher comprehension scores. In one model we observed a significant interaction between alignment and average N400 amplitudes, indicating that comprehension was better for more unexpected stories when there was lower alignment to expected contextual probabilities. Our results suggest that individuals can utilise unexpected information to support language comprehension and that there are conditions where successful integration of this information is more likely. Contrary to our expectations, lower alignment to the current probabilistic conditions led to a greater capacity to utilise unexpected information. Future research should explore whether lower alignment reflects an individuals openness to integrate new information into existing schemas and how this impacts comprehension of unexpected language.