Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Reliance on sentence context during naturalistic listening predicts word reading skill and word reading gains
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Daniel Kleinman1, Luca Campanelli2, Brittany Lee3, Julie Van Dyke1,4, Christian Brodbeck5, Nicole Landi1,4; 1Yale University, 2New York Medical College, 3Chapman University, 4University of Connecticut, 5McMaster University
Individuals with decoding-based reading disability (also known as dyslexia) often exhibit deficits in phonological processing (Shaywitz, 1996), including deficits in speech sound processing (Schulte-Körne & Bruder, 2010; Vellutino et al., 2004) and in low-level auditory processing more generally (Giraud & Ramus, 2013; Lehongre et al., 2011). Many children with reading disability also have higher-level language difficulties (Adlof & Hogan, 2018), including reduced sensitivity to predictability and less efficient anticipatory processing (Huettig & Brouwer, 2015; Mani & Huettig, 2014; see Hestvik et al., 2022 for evidence in children with DLD). Here, we used naturalistic stimuli (audiobooks) and assessed how well each individual’s EEG “tracked” content as it accrued during story comprehension across multiple linguistic levels, as indexed by linguistic surprisal and entropy derived from a sentence context model. Then, we asked whether this content-tracking measure was related to contemporaneous and/or future reading ability. METHOD: Participants (n=85) were students with reading disability attending schools specializing in remediating language- and reading-based disabilities (at study enrollment, mean age=10.2 years, SD=1.5 years, range=7.5–14.2 years). To measure word reading ability, participants were administered the Woodcock-Johnson Letter-Word ID subtest, a timed test of word decoding (baseline M=48, SD=11, range=23–65). This test was administered again after 6 months; gains were measured as change in score. At the first timepoint, participants also listened to audiobook excerpts for 12 minutes while their EEG was recorded. A sentence (lexical 5-gram) context model was used to quantify word surprisal, cohort entropy, and phoneme entropy for all words and phonemes in the excerpts, based on the four preceding words and the partial current word. Then, time-lagged regression analyses were used to measure how much each participant’s EEG covaried with this time series of linguistic probabilities (see Brodbeck et al., 2022 for more details). RESULTS: Participants who relied more on sentence context to generate and evaluate predictions about upcoming words and phonemes during the listening task at the first timepoint had significantly greater contemporaneous (baseline) reading scores (r2=6%, p=.023). They also made significantly greater gains in reading scores over the next 6 months (Δr2=8%, p=.002), after controlling for baseline reading scores. Descriptively, EEG covariation with linguistic probability was driven by a centroparietal negativity to phonemes in more contextually surprising words from 250 to 400 ms – consistent with an auditory N400 effect. Both relationships were smaller but remained significant when also controlling for acoustic predictors, sublexical context, and lexical context (contemporaneous: r2=5%, p=.048; gains: Δr2=4%, p=.047). CONCLUSIONS: The present study provides further evidence that neural responses to continuous speech can index individual differences in predictive processing (cf. Gillis et al., 2023; Keshavarzi et al., 2022), while also showing that these differences can identify listeners who are on the verge of making reading gains beyond the extent reflected in current reading scores. Speculatively, leveraging sentence contexts to generate predictions about upcoming speech in real-time may allow a listener to derive more benefit from linguistic input, enabling faster progress in reading-related skills.
Topic Areas: Reading, Speech Perception