Reading comprehension involves distinct cognitive operators for establishing simple and composite discourse referents in a mental model

Reading comprehension involves constructing a cumulative mental model that represents unfolding information. Within this model, the mind establishes discourse referents—entities that can be picked out and referred to, such as ‘sponges’ and ‘steel wool’ in ‘He bought sponges of steel wool…’. Though the composite entity ‘sponges of steel wool’ is also referenceable (‘…and glued them together.’), it remains unclear whether the mind establishes such composite referents in the mental model, and whether that is different than establishing simple referents. Here, we answer this using a sentence level-naturalistic reading paradigm with an embedded categorical manipulation. Participants (n=43) read 72 five-sentence English stories, where the fourth, critical sentence established three simple referents (‘wool, sponges and steel’; simple3), two simple referents (‘steel wool and sponges’; simple2), or two simple referents forming a composite (‘sponges of steel wool’; composite). We predicted that increased referential load–either due to additional composite or simple referents–should significantly increase reading times (RTs) beyond syntactic, semantic and lexical factors. Linear regression analyses revealed longer RTs for additional simple (RTsimple3>RTsimple2) and composite (RTcomposite>RTsimple2) referents, only during the critical sentences. Bayesian modeling of all sentence RTs further showed that a model distinguishing between composite and simple referents best fit RT distributions, outperforming models that tracked referents indiscriminately, tracked only simple referents, or ignored referents altogether. Overall, we find evidence for distinct cognitive operators for establishing composite and simple referents in the mental model. This paves the way for future work investigating hierarchical structure-building within the mental model during comprehension.