[PS-2.85] Updating mid-sentence predictions using information from negating elements in variably predictable contexts

Darley, E. , Kent, C. & Kazanina, N.

University of Bristol

Comprehending an unfolding sentence involves making predictions about upcoming material, a task that may be more difficult when negating elements (e.g. "not") are encountered. A logic-based approach (e.g. Wason & Johnson-Laird, 1972) holds that conceptualising a negation requires first understanding the negated proposition (a two-step process), so when this embedded proposition is not yet available in full, comprehenders may be unable to update predictions taking into account the negation. However, the contextual properties of the sentence can also influence the ease of making predictions. In two computer mouse-tracking experiments we used episodically-presented visual contexts alongside related sets of affirmative and negative sentences to investigate how the predictability of a critical word in a sentence affects the extent to which listeners can update incrementally on the information contained in the negating element. To tap into participants' mid-sentence predictions, we recorded mouse trajectories while they performed a truth-value judgement task and a sentence completion task in highly predictable and less predictable contexts. Large deviations towards the wrong answer (indicating evidence of a two-step process) occurred more often for negatives than affirmatives, while the pattern obtained by manipulating predictability could be partly explained by a strategy for minimising resource consumption during prediction.