Making sense of reference

The distinction between sense and reference has been foundational in the study of linguistic meaning since Frege’s classic On sense and reference (1892). More than a century later, the neural correlates of reference remain opaque. Meanwhile in computational psychiatry, a burgeoning field uses computational neural language models (NLMs) to predict psychosis from spontaneous speech. Clear behavioural evidence has emerged that patients with psychosis navigate the semantic space differently when retrieving words and building them into grammatical structures, causing shifts in patterns that are detectable through computational semantic similarity and probabilistic metrics. We propose the new concept that (i) grammar plays a specific role in mapping concepts onto referents so as to construct a cognitive map, which (ii) can be tracked in real time at a millisecond level, (iii) involves a cortical-hippocampal loop, and (iv) is a neural mechanism vulnerable in psychosis. Using MEG with optically pumped magnetometers (OPM), we will implement a naturalistic story comprehension paradigm in which the construction of reference through grammar will be tracked and regressed on computational semantic metrics. We aim to generate value through a novel sustainable, theoretically and computationally grounded framework for the study of language in psychosis, which will enable a transition to the larger goal of fully exploiting spontaneous speech as a rich and multidimensional data source for clinical practice.