Much of the early history of human neuroscience revolved around the study of the damaged brain – associating brain damage at particular locations with characteristic profiles of behavioural impairment. Damage-deficit analyses are attractive because they support causal inferences about what particular brain regions do in the undamaged brain. One of the earliest beneficiaries of this kind of work was the study of language: the ‘Neurological Model of Language’ (NML), which associates damage in key regions with characteristic profiles of acquired language deficit (aphasia), is based on studies conducted over 150 years ago. But though the NML is still popular today, the predictions it makes are far too coarse (and inaccurate) to be useful. Indeed, despite that promising early progress, neuroscience plays very little role in the care that today’s aphasic patients receive.

​

In this talk, I will discuss some recent efforts to try to bridge that gap, drawing on large samples of structured data from patients with acquired brain damage. I describe two streams of work, aiming: (a) to make accurate predictions about language deficits and recovery given brain damage, and (b) to identify the ‘critical regions’ where damage drives particular impairments. Through this work, I show how strongly predictive – and so potentially clinically relevant – results can be achieved, and that the pursuit of ever-better prediction can itself drive more purely theoretical insights into the structure of the language system. Building on these results, and despite some methodological challenges which remain to be addressed, I conclude by arguing that a new, predictive neurological model of language is increasingly within reach.