The neural architecture of compositional generalization: how do we infer the meaning of “un-forget-able-ish”

The ability to generalize previously learned knowledge to novel situations is crucial for adaptive behavior, a cognitive flexibility well illustrated in language. We excel at combining linguistic building blocks to infer the meaning of novel compositional words, such as “un-forget-able-ish”. In the current study, we investigated whether compositional inference in language recruits a medial prefrontal-hippocampal network, previously implicated in compositional processes in the domains of relational memory, action planning and vision (Baram et al., 2021; Barron et al., 2020; Schwartenbeck et al., 2023). To this end, we trained 30 healthy participants on an artificial language where meanings of compositional words could be derived from known stems and unknown affixes, using abstract affixation rules (e.g., “good-kla” which means something 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., “white-kla”) or incongruent primes (e.g., “kla-white”), and synonyms of the composed meaning as targets (“black”).
We analyzed the data using univariate fMRI repetition suppression and multivariate representational similarity analysis.