Semantic ambiguity dissociates the roles of word meaning and co-occurrence statistics in sentence processing
Year: 2026
Type: preprint
Abstract: Next-word prediction is a key component of human language processing, which manifests itself in higher processing costs (longer reading times and enhanced neural responses) to unpredictable words. These costs are captured by large language model (LLM) surprisal, a metric inferred from statistical patterns of word co-occurrence. This makes word co-occurrence statistics a plausible candidate source for human next-word prediction. In contrast, psycholinguistic models posit that human next-word prediction relies on representations of word meaning. To contrast these views, we placed words that carry multiple meanings in contexts that rendered meaning ambiguous (e.g., ‘My colleagues portrayed the head’ : body part or leader). With this, we kept word co-occurrence statistics available for prediction but introduced uncertainty into word meaning, effectively decorrelating these putative drivers of next-word prediction. Across three self-paced reading and magnetoencephalography (MEG) experiments, human processing costs for unpredictable unambiguous words were captured by LLM word surprisal, while ambiguity disrupted this relation. This shows that human next-word prediction is mechanistically linked to ambiguity resolution, diverging from LLM-like statistics-driven prediction specifically under ambiguity. We further found that a shared cortical network supports prediction and ambiguity resolution in parallel. Overall, our findings corroborate a distinctive feature of predictive language processing in humans, namely its reliance on word meaning beyond statistical patterns of word co-occurrence. Significance Statement Next-word prediction is a key principle of human language comprehension, and the performance goal of large language models (LLMs). By design, LLMs exploit statistical patterns of word co-occurrence to predict upcoming words. In contrast, the basis for human next-word prediction remains debated. To understand whether word co-occurrence statistics likewise underlie prediction in humans, we designed sentences with unpredictable ambiguous words, such as ‘My colleague portrayed the head [leader vs. body part]’. LLM predictions captured how humans processed these words when context disambiguated their meaning but not when word meaning remained ambiguous. This reveals a distinctive feature of predictive language processing in humans: It is contingent on word meaning beyond statistical regularities of word usage.
Authors: Andrey Zyryanov, Victoria Pierz, Yulia Oganian
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
Funder: Boehringer Ingelheim Fonds
OpenAlex ID: W7161403516