[ExI] Pure Philosophy Dispute: Are Categories Objective?

[Damien Broderick]

This is the debate over the ontological status of "natural kinds". A 
very enduring topic.

The following (derived from my dissertation, and hence not 
referencing especially recent work) will probably bore everyone silly, but hey:

==================

... concept-nodes may be linked by pathways characterised by mutable 
`weights', or differential probabilities of access. A plausible 
version is sketched by the sociologist of science, Barry Barnes 
(1983). For Barnes, two apparently conflicting features of any viable 
model of mind and memory are indisputable: first, we learn in an 
empirical context, a real physical environment; second, such learning 
`always initially occurs within a social context; to learn to 
classify is to learn to employ the classifications of some community 
or culture, and this involves interaction with competent members of 
the culture' (p. 21).
           Concept-building within these constraints--the natural and 
the social--occurs, according to Barnes, via two processes: ostension 
(just pointing at the object or process you're interested in, and 
giving it a name), and generalisation (p. 22). The trouble with 
ostension is well recognised. How do you know what she's pointed at, 
if you don't known what she's pointing at? Still, for Barnes it 
remains the indispensable bottom line. `A potentially infinite series 
of questions [concerning usage...] terminates in actual situations 
only because ostensibly given indications of usage lead out of the 
morass.' These indications, he assumes, are grounded in hard-wired 
determinants: `we possess an incompletely understood perceptual and 
cognitive apparatus with at least some rudimentary inherent 
properties for making learning possible'. Linguist Steven Pinker, 
too, argues that we swiftly home in on the meaning of unknown words 
when we see them applied `because we are not open-minded logicians 
but happily blinkered humans, innately constrained to make only 
certain kinds of guesses... about how the world and its occupants 
work' (Pinker, 1994, pp. 153-4). As Barnes remarks emphatically (p. 38)

           We should not shrink from admitting that cognitively we 
operate as inductive learning machines (Hesse, 1974). This crude 
formulation stresses that basic inductive propensities are inherent 
in our characteristics as organisms.... We are congenitally inductive.

Generalisation controls the way terms are linked together. `They are 
what make us regard a form of culture as a body of knowledge rather 
than a mere taxonomy' (p. 23). These links take the form of what 
Barnes terms a Hesse net (ibid.; see Hesse, 1974). Each 
generalisation has an associated (and constantly updated) 
`probability' (p. 24). `Under every concept stands a number of 
specific instances thereof. These instances I shall call the tension 
of the associated concept' (ibid.). (Presumably there is an `elastic 
force' pun here, which is somewhat regrettable in what is basically a 
cybernetic metaphor.) In a clarifying footnote, he comments

           I use the term `tension' in deliberate allusion to 
`extension' as used in philosophical semantics. In the extension of a 
term are thought to be included all the entities to which it properly 
applies, or of which it is true. In the tension of a term are 
included only past instances of use--a finite number of instances. To 
talk merely of the tension of a term is to accept that its future 
proper usage is indeterminate. To talk of the extension is to imply 
that future proper usage is determined already.

So although Barnes is prepared to ground his theory of 
concept-formation in a real world susceptible of ostension, he is 
adamant that social determination controls the categories within 
which concepts are placed. Yet, unlike Saussurean analysts, Barnes 
insists--correctly, in my view--on a dimension of similarity as well 
as one of difference, though one lacking any coercive or 
`essentialist' implication (p. 26):

           An assertion of resemblance... involves asserting that 
similarities outweigh differences. But there is no scale for the 
weighting of similarity against difference given in the nature of 
external reality, or inherent in the nature of the mind. An agent 
could as well assert insufficient resemblance and withhold 
application of the concept as far as reality or reason is concerned. 
It follows that the tension of a term such as `dog' is an 
insufficient determinant of its subsequent usage. All applications of 
`dog' involve the contingent judgment that similarity outweighs 
difference in that case....

And judgment is, of course, a socially situated act.

           Knowledge in Context

Barnes proceeds to several implications: delocalisation (to know a 
goose, it also helps to know a swan); hence, there are no 
free-floating `atomic' concepts (p. 29); the application of a term is 
a judgment, as we have noted, since `the tension of a term represents 
a conventional relationship of sameness between the instances within 
it', and this can always be revised (pp. 30 1); proper usage is 
agreed usage, so that a creature might be at one time deemed a moth, 
at another a butterfly: `Cases such as these are sometimes thought to 
result from an inadequate knowledge of the "real meanings" of terms 
themselves; and occasionally the achievement of
consensus in these cases is conceived as a "discovery" of the "real 
meaning". But such consensus merely marks the successful negotiation 
of an extension of usage'; and equivalence, which is to say that 
`different Hesse nets are always equivalent' (p. 33), since 
`"Reality" does not mind how we cluster it; "reality" is simply the 
massively complex array of unverbalized information which we cluster. 
This suggests that different nets stand equivalently in relation to 
"reality" or to the physical environment', and also `as far as the 
possibility of "rational justification" is concerned'(p. 33). In 
short, `alternative classifications are conventions between which 
neither "reality" nor "pure reason" can discriminate. Accepted 
systems of classification are institutions which are socially 
sustained. (p. 33)'
           In my own view, this strong relativist position is surely 
inconsistent with an implacable universe warranting ostension. Barnes 
offers in support of his case the instance of Karam animal taxonomy, 
which places cassowaries (a kind of flightless bird) in the special 
taxon kobtiy, outside that of flying beasts like other birds and bats 
(pp 34 37), and compares that categorisation with the zoological 
taxonomy used in an advanced industrial Hesse net:

           How can the pattern of either net distort reality? Rather, 
reality provides the information incorporated in both nets; it has no 
preference for the one or the other. (p. 35)

However bracing this might be as a corrective to imperialistic 
anthropology, it is nonsense if taken literally. DNA sequences, for 
example, are not `randomly' or `purely culturally' associated with 
the genomes of each taxa, but contain clear natural-historical 
markers endorsing the phylogenetic claims of one over the other--that 
is, the history of their natural selection. (These deep links might 
be of no interest or use to humans, of course, and for most of 
history they have been altogether inaccessible, but they remain coded 
as the DNA `text' or `recipe': an almost indelible inscription).
           Animals and plants, whose phenotypes are the expression of 
the interaction between environment and coded genotype, are the 
`naturally-chunked' perceptual fields, or `natural kinds', which 
humans are prone to tag with lexemes (whatever further totemic or 
commercial significance they may be given). This perspective--that 
human language, prior to the legitimate claims of cultural 
relativity, is founded in its capacity for adaptation to an 
indefinitely complex interacting universe--gives the lie to Barnes's 
easy assertion: `"Reality" does not mind how we cluster it.' Reality 
might not mind, but finding the correct clustering certainly matters.
           The relativist view has been canvassed by John Dean, who 
found within our own botanical science two rival taxonomies for the 
plant Gilia inconspicua, and noted that both `are built upon 
perceptible, systematizable, stable distinctions between individual 
plants. In this sense the natural order sustains both taxonomies; 
neither can be said to be erroneous' (Dean, 1979, p. 226; see 
especially his taxonomic discussion, pp. 211-28). This view does not 
convince me that `reality does not mind how we classify it'; it 
simply reminds us that the reality we notate on our low-dimensional 
grids is multidimensional. Reality is not, however, utterly or even 
very indeterminate: it would be very strange to classify Gilia 
inconspicua as a variety of possum or igneous rock, or to attempt to 
breed it in the wild with an elephant. True, one might throw it in 
with anything imaginable for, say, totemic purposes, but that is a 
different point entirely. Ironically, the arch-conventionalist Pierre 
Duhem looked to the emergence of naturally-chunked classification: 
`The more a theory is perfected, the more we apprehend that the 
logical order in which it arranges experimental laws is the 
reflection of an ontological order' (cited Lakatos, 1978, p. 21).
           Taken together, these converging models from artificial 
intelligence research and the sociology of scientific knowledge offer 
a useful springboard to the further examination of semiosis: the ways 
in which humans recognise, construct and manipulate logics and 
contexts in the service of signification.