the physical and the intellectual capacities of a man.’ We regard it as a competing virtue
of our focus on utterance-activity that it demands attending to bodies and environments.
By making utterance-activity central, we are not eschewing abstraction and theory at all.2
Rather, at least provisionally, we are suspending commitment to the view that there is a
theoretically well-motivated gulf separating language ‘proper’ from other aspects of
behaviour.
The supposed gulf between language proper and the rest of behaviour finds a suggestive
analogue in Clark’s work. Describing that gulf will help us get more specific about the
kind of extended mind thesis we are going to sketch.
A Tale of Two Clarks
We detect two quite strikingly different registers or moods in Clark (1997). On the one
hand there is a line of thinking focused on embodied, and typically mobile, cognition in
robots, animals and humans, which emphasises the ways in which traditional
expectations concerning the ‘inner’ character of cognition fail to capture the manifest
cognitive properties of both living systems and effective engineered ones. On the other
hand there are arguments and surveys of evidence centred on the cognitive advantages of
language, which also reject the view that cognitive processes are exclusively handled by
the brain (a view we call ‘cognitive internalism’) but which focuses on ‘higher level’
functions, paying less attention to embodiment and motion. ‘The Extended Mind’ is an
instance of this line of thinking.
When Clark talks about robots, indeed anything that moves, he emphasises, inter alia,
the importance of non-neural resources for controlling locomotion and other functions,
the greater efficiency and biological plausibility of ‘subsumption architectures’ (Clark
1997: 13-15, Brooks 1991) and ‘soft-assembly’ (Clark 1997: 42, Thelen and Smith 1994)
as opposed to control systems with fixed hierarchies and/or a central executive. In
addition, he combines agnosticism about the necessity of representations with
commitment to the view that if there are to be representations they had better pay their
way by being directly capable of serving control functions, rather than salvaging
outmoded intuitions about the representational nature of thinking (Clark 1997: 149-153).
This is one way of thinking about the ‘extended mind’ - an image of brains as parts of
embodied coalitions.
When he focuses on language, on the other hand, Clark urges us to relinquish the notion
that the primary or only function of language is communication, and instead think of it as
an external public and symbolic collection of resources, the exploitation of which grants
us a range of cognitive advantages. These cognitive advantages include a capacity for
self-stimulation that serves to improve control and performance at tasks (Clark 1997:
202), being able to use symbolic systems to augment memory by using non-neural
storage media (Clark 1997: 201), using labels and symbols to simplify our environments
and learning processes (Clark 1 997: 201, Clark and Thornton 1997), and simplifying
various other types of problem solving. This type of ‘extended mind’ is hooked up to a
range of external symbolic resources; language, and language enabled cognition, is
highly distributed, but does not seem especially embodied.
We are thoroughly sympathetic to both of Clark’s approaches here. We think that he’s on
the right track, or two right tracks, and drawing on the right kinds of research.
Nonetheless we think that there is an important set of questions which his account of
language does not touch on, and which we think need to be part of the type of approach
he defends. To see something of what concerns us, consider his discussion of learning
with and without labels (Clark 1993: 69-112, Clark and Thornton 1997). Whether or not
you are surprised that labelling can improve learning efficiency, or open up different
3 / 23