Semantic processing in autism during speech-in-music listening: insights from congruency and surprisal-based N400 analyses

Understanding speech in background music is a common real-world challenge, particularly when vocals compete for linguistic processing resources. This study examined how the presence and intelligibility of sung lyrics influence semantic processing in autistic and non-autistic adults. Twenty-nine participants per group performed a sentence acceptability judgement task while EEG was recorded. Sentences ended with either semantically congruent or incongruent words and were presented alongside instrumental, Simlish (phonologically English-like but unintelligible), or English-lyric versions of the same songs. To examine semantic processing, we analysed the N400 using two complementary approaches: a categorical congruency contrast, indexing the neural cost of processing semantic anomalies, and a continuous lexical surprisal measure, capturing graded sensitivity to word predictability. In non-autistic participants, both analyses showed largest N400 responses in the instrumental condition, attenuated responses in vocal conditions, and reduced behavioural accuracy as lyrics became more intelligible. Autistic participants showed lower accuracy and a reduced N400 effect relative to non-autistic participants, particularly in the instrumental music condition. In addition, they exhibited no behavioural difference between the English and Simlish vocal conditions, suggesting that changes in lyric intelligibility did not affect accuracy. By combining ecologically valid speech-in-music masking with dual analytic approaches, this study provides the first neurophysiological evidence of these semantic processing differences in autism and demonstrates how integrating categorical and probabilistic measures can yield a richer and more nuanced account of speech processing in complex auditory environments.