Abstract rule generalization for composing novel meaning recruits a frontoparietal control network

The ability to generalize previously learned knowledge to novel situations is crucial for adaptive behavior, representing a form of cognitive flexibility that is particularly relevant in language. Humans excel at combining linguistic building blocks to infer the meanings of novel compositional words, such as "un-reject-able-ish". The neural mechanisms and representations required for this ability remain unclear. To unravel these, we trained participants on a semi-artificial language in which the meanings of compositional words could be derived from known stems and unknown affixes, using abstract relational structure rules (e.g., "good-kla" which means "bad", where "-kla" reverses the meaning of the stem word "good"). According to these rules, word meaning depended on the sequential relation between the stem and the affix (i.e., pre- vs. post-stem). During fMRI, participants performed a semantic priming task, with novel compositional words as either sequential order congruent (e.g., "short-kla") or incongruent primes (e.g., "kla-short"), and real words serving as targets that were synonyms of the composed meaning of the congruent primes (e.g., "long"). Our results show that the compositional process engaged a broad temporoparietal network, while representations of composed word meaning were localized in a more circumscribed left-lateralized language network. Strikingly, newly composed meanings were decodable already at the time of the prime in a way that could not be accounted for representations of the prime words themselves. Finally, we found that the composition process recruited abstract rule representations in a bilateral frontoparietal network, in contrast to our preregistered prediction of a medial prefrontal-hippocampal network. These results support the hypothesis that people activate a bilateral frontoparietal circuitry for compositional inference and generalization in language.