Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Influence of sentence information load on predictive processing: An ERP study
Poster Session A, Friday, September 12, 11:00 am - 12:30 pm, Field House
This poster is part of the Sandbox Series.
Yu Lu1, Simon Fischer-Baum1, Randi Martin1; 1Rice University
Listeners make predictions about upcoming content using previous information in the sentence, which contributes to language processing efficiency (Ryskin & Nieuwland, 2023). Capacity limitations on human cognitive resources mean that we may not always be able to keep all of the information in mind that might be relevant to making predictions. The goal of the current study is to examine the impact of information load on sentence final prediction. Consider these two sentences: (a) “I wanted to give him a phone call, but I could not remember his number.” (b) “I wanted to give him a phone call when I was at work yesterday, but I could not remember his number.” Based on the cloze task, we know that both sentences lead to the prediction of “number,” given the context of giving a phone call, but not when that context is missing, as the most common completion of the final clause is “I could not remember his name.” Sentence (b) involves an intervening middle clause, which increases the information load and cognitive demand involved in maintaining that critical initial context that drives prediction of “number”. The hypothesis being tested here is that with the intervening clause the listener is more likely to fail to maintain the key contextual information in the first clause in a working memory (WM) system, and is therefore more likely to generate a prediction based on adjacent context. Comparing predictive processing following sentence contexts constructed like (a) versus (b) allows us to examine whether the sentence information load and WM demand involved in processing the context impact prediction. The experiment uses EEG recording during a sentence comprehension task, and the key measure is the N400 event-related potential (ERP) as a measure of prediction. Participants read sentences structured as (a) and (b), and the sentences are followed by an expected word (number), a word predicted by adjacent context (name), or an unexpected word (jacket). The N400 has a strong linear relation with the word’s predictability and is sensitive to the degree of semantic prediction violation (Federmeier et al., 2007; Federmeier & Kutas, 1999). Thus, for (a) and (b) sentences, one should expect a small N400 following the expected word, a medium N400 following the word predicted only by the adjacent context, and a large N400 following the unexpected word. However, the key question is whether the presence of the intervening clause in (b) sentences impacts the sentence-final prediction relative to sentence (a). We hypothesized that if the intervening clause in type (b) sentences leads to higher WM demand and biases the comprehender towards a prediction biased by the adjacent information, then one should also expect a bigger N400 following the expected word “number” and a smaller N400 following the less expected word “name” for sentence (b) than for sentence (a). By designing sentences with varying information load and examining the impact on sentence-final prediction-related effect, the current experiment aims to enrich our understanding of the cognitive mechanisms of predictive processing in language.
Topic Areas: Meaning: Lexical Semantics,