BRAIN IMAGING, CONNECTIONISM, AND COGNITIVE NEUROPSYCHOLOGY Max Coltheart Macquarie University, Sydney, Australia I���m still an ultra-cognitive-neuropsychologist after all these years, and so my comments on Trevor Harley���s most interesting book review will be from that perspective. Ultra-cognitive-neuropsychologists take cognitive neuropsychology to be a branch of cognitive psychology. The aim of cognitive psychology is to learn more about the mental information-processing systems that people use to carry out various cognitive activities. Some cognitive psychologists do that by studying the performance of people whose cognitive processing systems are normal. Others do it by studying people in whom some cognitive processing system is abnormal: Such investigators are cognitive neuropsychologists. Cognitive neuropsychologists are not studying the brain. Investigating the brain processes upon which carrying out cognitive activities depends is a different discipline with a different name: It is called cognitive neuroscience. Some cognitive neuropsychologists are also cognitive neuro- scientists because they do both kinds of work (Tim Shallice, Chris Frith, or Martha Farah, for example) some are just plain cognitive neuro- psychologists (John Marshall, Alfonso Caramazza, or Max Coltheart, for example). SHOULD THERE BE ANY ���NEURO��� IN COGNITIVE NEUROPSYCHOLOGY? Harley considers this question mainly in relation to research in cognitive neuroimaging. He quotes (p. 9) an unkind epithet from Uttal (2001), who ���derogatorily dubbed this emphasis on imaging ���the new phrenology���.��� I have come across the even less kind term ���chromophrenology.��� Is there more to cognitive neuroimaging research than what Harley refers to as ���tokenism��� (���a glossy magazine might print a picture of a car because it looks good and is funtodo���)?Whatdidwelearnaboutthemindfrom the Decade of the Brain? Here we need to distinguish between two questions: 1. Has cognitive neuroscience already succeeded in using, or if not might it ever (in principle, or even in practice) successfully use, data from cognitive neuroimaging to localise cognitive processes in the brain? 2. Has cognitive neuroscience already succeeded in using, or if not might it ever (in principle, or even in practice) successfully use data from cognitive neuroimaging to make theoretical decisions entirely at the cognitive level (e.g., to adjudicate between competing information-processing models of some cognitive system)? Fortunately, questions about what might happen with cognitive neuroimaging in the future are beyond both the scope and the space limitations of this article so I will largely confine myself, as did Harley, to a discussion of what has happened as of the time of writing with respect to these two questions about cognitive neuroimaging. I will consider the question of localisation first. Some of the earliest work in cognitive neuroscience had localisation of cognitive processes as an aim for example, PET research in both St Louis (Peterson, COGNITIVE NEUROPSYCHOLOGY, 2004, 21 (1), 21���25 �� 2004 Psychology Press Ltd http://www.tandf.co.uk/journals/pp/02643294.html DOI:10.1080/02643290342000159 21

Fox, Posner, & Mintun, 1989 Peterson, Fox, Posner, Mintun, & Raichle, 1988) and at the Hammersmith Hospital in London (Howard et al., 1992) was aimed at localising three components of the reading system���orthography, phonology, and semantics. The visual word form system or orthographic lexicon was localised in prestriate visual cortex by the St Louis group and in the posterior part of the left middle temporal gyrus by the London group. Phonology for reading was localised in the temporal-parietal junction by the St Louis group and in the middle part of the left superior and middle temporal gyri by the London group. Semantics was localised in prefrontal cortex by the St Louis group and in the temporal lobe by the London group. So there was simply no agreement between the two groups as to the cerebral localisa- tion of any of these components of the reading system. Even today there is no agreement as to the localisation of components of the reading system on the basis of cognitive neuroimaging research (see, e.g., Dehaene, Le Clec, Poline, Le Bihan, & Cohen, 2002 Leff, Crewes, Plant, Scott, Kennard, & Wise, 2001). This problem is not confined to reading. I don���t know of any examples in which there is current consensus as to the cerebral localisation of any module of any cognitive system on the basis of cognitive neuroimaging data. Consider face processing, for example. Much imaging work has implicated the FFA (Fusiform Face Area see, e.g., Kanwisher, Tong, & Nakayama, 1998) in the processing of faces. But which module of the face processing system is localised to the FFA is still not known: ���What clues do the present data provide about the exact operations that are carried out in the FFA? . . . the FFA may be involved in face detection but not face recognition . . . However, the present data cannot rule out the possibility that the FFA is involved in face recognition��� (Kanwisher et al., 1998, p. B9). Perhaps I am being too impatient here such consensus may be achieved in the future. But perhaps we should also take seriously the arguments of Van Orden and Paap (1997). It is common in contemporary cognitive modelling to view processing as cascaded (rather than thresholded) and as interactive (rather than purely feedforward). If a cognitive system does have these properties, that poses grave difficulties for the use of imaging to discover the cerebral localisation of cognitive modules. Suppose one believes that some cognitive system includes a sequence of three modules A to B to C, and one wants to localise one of these, say A, by imaging the brains of people as they carry out some task that requires module A. Because of the cascaded nature of the system, modules B and C will be activated if module A is, even if modules B and C are irrelevant as far as the task is concerned. And because of the interactive nature of the system, some of the activity in module A will be due to feedback from modules B and C. So it will not be the case that as the task is being performed all activity in the brain detected by imaging will reflect just the operation of module A in isolation. If so, how could one ever determine which parts of the brain activity here are specifically associated with module A? The other possible aim of cognitive neuro- imaging is to use imaging data for testing or adjudicating between cognitive models. Here the ultra-cognitive-neuropsychological position is a particularly extreme one: The assertion is that this aim is impossible to achieve in principle, because facts about the brain do not constrain the possible natures of mental information-processing systems. No amount of knowledge about the hardware of a computer will tell you anything serious about the nature of the software that the computer runs. In the same way, no facts about the activity of the brain could be used to confirm or refute some informa- tion-processing model of cognition. This is why the ultra-cognitive-neuropsychologist���s answer to the question ���Should there be any ���neuro��� in cognitive neuropsychology?��� is ���Certainly not what would be the point?���. Is this extreme position tenable? If there are already any examples from the literature of cognitive neuroimaging in which some otherwise plausible information-processing model of cogni- tion has been refuted on the basis of imaging data, then this plank of the ultra-cognitive- neuropsychological platform would have to be COLTHEART 22 COGNITIVE NEUROPSYCHOLOGY, 2004, 21 (1)