fication proved to be a much more sensitive test for
detecting category effects (albeit all living deficits) than
picture naming or naming-to-description (although
there were also differences across pathologies). This
suggests that the direction of category effects some-
times seems to depend upon which test is chosen as the
reference test (a prospect that has not been previously
entertained). In this context, it is worth noting that pa-
tients who show living disorders tend to be agnosic, and
therefore, tested with picture naming; however, sev-
eral nonliving cases have been aphasic, and thus, were
not tested with picture naming, but with tasks such
as picture-name matching (see Laws, 2004). Hence, it
is common for different category effects to rely upon
different testing procedures; and as such, again the
existing literature may be prone to some of the issues
raised here.
Dissociations and Paradoxical Dissociations
Dissociations often form the basis for speculations about
cognitive architecture and modularity especially when
they are doubly dissociated between patients. The cur-
rent study shows, however, that dissociations can occur
within a patient. Within-patient double dissociations
across tasks (e.g., a living on Task A and a nonliving
on Task B) that are believed to have some critical
processing stage in common, raise questions about the
double-dissociation methodology in single-case studies
and the interpretation of category effects per se. At a
theoretical level, many models assume that deficits in
semantics will have ‘‘knock-on’’ effects for naming; and
so, such models have difficulty accounting for paradox-
ical dissociations at the level of semantics and naming.
Paradoxical double dissociations pose problems for
double dissociations at a variety of levels of compari-
son including: across tasks (as described here), within
tasks, and patients (Laws, Gale, et al., 2005); and of
course, across patients and across tasks (the typical
approach in category specificity and cognitive neuro-
psychology more generally). Given that paradoxical
dissociations arise, how might we distinguish a paradox-
ical dissociation from a real double dissociation (i.e.,
one that might be used to ground theories of cognition
or ‘‘carve cognition at its modular joints’’)?
How should paradoxical dissociations be interpreted?
Of course, it might be argued that paradoxical dissocia-
tions are simply unreliable. Indeed, because we did not
retest patients, we have no way of confirming whether
paradoxical dissociations are reliable. Nonetheless, the
reliability of paradoxical dissociations has to be viewed
alongside the fact that reliability is hardly ever examined
for dissociations in single-case studies. Indeed, follow-up
analyses of the same patient by same or other research
groups are rare and sometimes contradictory (Laws,
1998). Hence, it is crucial for future studies to examine
the reliability of all dissociations.
We must also consider the possibility that paradoxical
double dissociations ref lect confounding variables. It
might be argued, for example, that f luctuations in
attention could impact differentially over the test session
and potentially affect one category more than the other.
This is unlikely because it would require that the
confound interacts highly selectively with category. Liv-
ing and nonliving stimuli (on all tests) were randomly
intermixed when presented, so a factor such as attention
f luctuation would have to impact only when items from
one of the two categories were presented. This seems
even more implausible in cases when we consider
paradoxical dissociations (i.e., in the opposite direction
on a second test). Consider the case of the HSE patient
MF (see Figure 1), who showed a classical double
dissociation across tasks. His picture naming was below
the 1% for living things (and normal for nonliving
things); and below the 1% for nonliving on feature
verification (but normal for living things). In this con-
text, we would argue that the dissociations are robust to
any artifacts of this kind.
Another potential confound concerns the possibility
that the dissociations reported here (whether consistent
or paradoxical) are chance findings emerging from quite
noisy patient data, and that multiple analyses might
increase the likelihood of spurious outcomes. Indeed,
typical statistical/methodological approaches may well
be prone to producing spurious and chance findings in
case study analyses. Nevertheless, Monte Carlo simula-
Figure 1. A classical paradoxical double dissociation between living
and nonliving things within one HSE patient (MF). Note: Patient MF
displays a classical double dissociation across category (i.e., impaired
picture naming for living things, but normal nonliving thing naming).
On feature verification, he shows normal living and impaired nonliving
performance. Classical double dissociations (often with weaker
evidence than here) typically provide the strongest evidence for the
separation of cognitive processed (or architecture).
Laws and Sartori 1457