Work Package 3
Compositionality and contextuality
The flexibility and expressiveness of human language are based on a neurocognitive infrastructure that allows endless new combinations of words and contexts to co-determine the interpretation of what otherwise would be an inherently ambiguous signal ("the infinite use of finite means"; Wilhelm von Humboldt). WP 3 focuses on compositionality and contextuality of language which are core aspects of human language.
A universal characteristic of language is that speaking and understanding is more than just a concatenation of the individual words that are retrieved from memory. Operations are in place that dynamically construct new structures from the building blocks that have been acquired during development and for which long-term memory representations have been formed in the mental lexicon. These combinatorial operations occur at the levels of phonology, syntax and semantics, and they are responsible for the compositionality of language. The question of how such combinatorial operations are instantiated in the brain constitutes the binding problem for language, the key research theme for this WP. In addition, language context beyond the word-, phrase- and sentence-level, as well as extralinguistic information co-determine the construction of the current utterance (in speaking), and the reconstruction of a situation model (in comprehension). This is what we refer to as the contextuality of language. Despite the general agreement that compositionality and contextuality are universal features of language and key to its expressive power, it remains unresolved how to best characterize the underlying mechanisms formally and computationally, as well as in terms of the processing architecture and its neurobiological implementation.
One of the main challenges for this WP is to unify abstract computational accounts of compositionality and contextuality, that have been formalized in recent years at the mathematical level, with implementations of binding that have neural plausibility. It is still a largely open question how the mechanism of binding is realized in distributed cortical networks along different time scales. Regarding the temporal configuration, current theories on neural processing point to transient synchrony of neuronal spiking as the neural correlate of information transmission between brain areas. In this WP we will pursue this direction, and investigate different spike-time-dependent mechanisms for flexible binding that will then be incorporated in a neurally plausible model of language processing.
Another big challenge is to understand the extent to which different brain areas are involved in binding. Although still poorly understood, prefrontal cortex (Broca's area and adjacent cortex) seems to play an important role, in interaction with regions in temporal cortex and elsewhere. Crucial information will be gained by high-field (7 Tesla) imaging that allows a resolution at the level of cortical layers, and by virtual lesions (through TMS) of the binding operations at different time slices.