Traits, Genes, and Coding 389
ganized and exploited in the right way, to carry some relevant item or body of
information (e.g., that the book is heavy enough to hold the door open). Here it
seems safe to say that the elements in question represent the associated worldly
features. But now consider the outcome of simply holding my office door open.
The equivalence class of suitable objects which may perform this role will be fixed
by (roughly) the non-informational properties of being heavy enough and being
sufficiently non-obstructive with respect to passing through the doorway. Here,
where the equivalence class of different elements that could perform the function at
issue is fixed by certain non-informational physical properties of those elements,
there is simply no place for the language of representation. This suggests that
arbitrariness is necessary for representation. And if, as my architecture-related re-
flections suggest, arbitrariness and homuncularity arrive on the explanatory scene
arm in arm (conceptually speaking), then the claim that homuncularity is neces-
sary for representation looks to be concurrently established.^12 So, if I am right, the
joint presence of (i) systematic causal co-variation between the putative vehicles
of content and specific causally downstream structures, (ii) arbitrariness, and (iii)
systemic homuncularity is necessary and sufficient for coding-talk.
As my description of the machinery underlying protein synthesis indicates, that
machinery satisfies conditions (i)–(iii). (Although I have not argued explicitly that
there are appropriate causal co-variations in protein synthesis, it should be clear
enough that there are systematic causal mappings between, on the one hand, both
DNA and mRNA, and, on the other, proteins.) But how secure is the general
account of representation that I have given? Here I shall consider four objections.
First, one might object to the claim that arbitrariness is necessary for repre-
sentation, on the grounds that not all elements that we take to be representations
have that property. This is the sort of complaint that needs to be settled on
a case-by-case basis, but let’s at least consider one of the more plausible candi-
dates for positive representational status coupled with non-arbitrariness, namely
onomatopoeiaic words.^13 Since the pronunciation of such words suggests their
meaning (e.g., meow), it might seem that they cannot be arbitrary. Yet we still
think of them as representational, so they provide a counter-example to my sug-
gestion that arbitrariness is necessary for representation. However, it seems to
me that the intuition that onomatopoeiaic words cannot be arbitrary trades on a
thought that is not reflected in the concept of arbitrariness, as I have unpacked
that concept here. Many different physical sound patterns could realize the word
‘meow’ (compare the way in which native French and native English speakers pro-
nounce the word), and what fixes the equivalence class of appropriate sounds is the
informational content that they carry (roughly, this is a sound that cats make). Of
course, the class of sounds that may be legitimate physical realizers of the word
‘meow’ is presumably not infinite, but then infinite realizability is not required
(^12) In [Wheeler, 1995, chapter 10] I give independent reasons for thinking that homuncularity is
necessary for representation.
(^13) This worry was put to me by Elliott Sober (in discussion). Thanks to Phyllis McKay and
Peter Sullivan for helping me to think about the best way to repel it.