Search

The way humans generate meaning is incredibly efficient but also flexible. When you hear a sentence like “We got cash from the bank” you presumably do not think about the side of a river, even though the word “bank” has multiple meanings. In the current study we evaluate the contribution of word form and its (sentential) context to its mental representation. To this end, we developed a novel task where subjects learned to associate symbols with homonyms where the specific meaning could be derived from a constraining sentence. We collected online data from an exploratory (N=32) and a replication (N=114) sample. We quantified implicit representation of meaning using a priming task, where subjects performed orientation discrimination judgment on a sequence of the trained symbols associated with context-driven word meanings. Participants’ explicit representation of meaning was quantified using a distance-based similarity rating task. In both samples, we found significant explicit representations of context-dependent meaning. In contrast, we found a trend of implicit meaning representation only in the exploratory sample. Explicit representation could also be explained by a corpus-based computational linguistic model. The current study is a first attempt to quantify word meaning representation in a distance-based manner, linking experimentally controlled measures with previous data-driven approaches.

The ability to generalize previously learned information to novel situations is key for adaptive behavior. Here, we investigated the neural mechanisms underlying the ability to infer novel compositional word meanings using a novel behavioral paradigm. Participants were trained on compositional word meaning from an artificial language comprising extant stems (“good”) and novel affixes (“kla”). The meaning of the compositional words depended on the position of the novel affix (“goodkla = bad”, “klahorse = pony”). We then asked them to infer the meaning of novel compositional words (“whitekla =?”, “klacat =?”) which were either congruent or incongruent with the rule (“klawhite” is incongruent because a small version of “white” does not exist). During fMRI, participants performed a semantic priming task in which the novel words served as congruent or incongruent primes (“whitekla”) and their synonyms (“black”) served as targets. Our results show that people are able to generate novel compositional meanings on the fly, successfully inferring meanings of congruent versus incongruent words. Repetition suppression effects at target were greater when primed with congruent than incongruent words in the left inferior frontal gyrus, suggesting the novel meanings to be derived at this linguistic “building” hub. Analysis of congruent versus incongruent prime-related activity revealed a broad frontal-parietal network, including hippocampus, the brain area commonly associated with the generalization process of structural relationships. We argue that generative compositionality in language recruits a domain-general network shared with action planning, compositional vision and constructive relational memory, while the newly inferred meanings are represented in more language-specific regions.

The ability to generalize previously learned information to novel situations is fundamental for adaptive behavior. When seeing the word “un-reject-able-ish” for the first time, one can quickly infer its meaning by generalizing the knowledge of its constituent parts and integrating them based on certain abstract structural rules (e.g., the sequential order of the word parts). How do we generate novel, compositional meaning? What are the neuro-computational mechanisms that underlie structural inference in not only meaning generalization but also across different cognitive domains? This efficient but also flexible inferential process may leverage neural mechanisms commonly studied in the nonlinguistic domains of action planning, relational memory and model-based reinforcement learning, including medial prefrontal-hippocampal circuitry. To address these questions, we developed a novel experimental paradigm for quantifying novel structural inference for the generation of word meaning. We taught participants compositional words from an artificial language and tested them with novel words using a semantic priming task. Results from two behavioral experiments showed that participants can learn and generalize structural (sequential order) rules for inferring novel, compositional meanings on the fly. An ongoing neuroimaging study in which we combine this paradigm with fMRI adaptation will unravel the neural mechanisms of meaning composition, allow us to test the prediction that correct compositional inference can be predicted from neural activity in a medial prefrontal-hippocampal network, measured during the generation of the novel word meaning.

The hippocampal-entorhinal system represents spatial knowledge in so-called cognitive maps. Recent evidence suggests that the same principle also holds for other types of relational information, such as the transition probabilities between objects presented in a non-random order (Garvert et al. 2017). However, objects can also be characterized in terms of semantic similarities that participants learned over the course of their lifetime. Does the system forming relational maps of experienced transitions also represent those object properties in parallel cognitive maps? To explore whether prior (semantic) knowledge is simultaneously mapped in the same system, we reanalyzed the functional magnetic resonance imaging (fMRI) data by Garvert et al. (2017). We used a triplet odd-one-out task to construct a semantic distance matrix and applied fMRI adaptation analysis to show that, indeed, the degree of similarity of object representations in bilateral hippocampus decreases as a function of semantic distance between objects. Most importantly, this semantic map is precisely localized in the hippocampus, yet anatomically distinct from the original, entorhinal map which codes for newly learnt statistical regularities. This supports the idea that the hippocampal-entorhinal system forms parallel cognitive maps reflecting the embedding of a stimulus in diverse relational structures, even if neither structure is task-relevant.