Duck and drake clusters

Fri 27 Aug 2010 12:16 PM

The following post about homeostatic property clusters (HPCs) is pretty long, so I've split it into several sections. Here's the very short version: Ereshefsky and Matthen argue that the HPC approach to natural kinds fetishizes similarity and is undone by polymorphism. I argue that it's not, and that the HPC approach is really about looking for causal structure.

[crossposted at It's Only a Theory]


Richard Boyd* has argued that many natural kinds should be understood as homeostatic property clusters (HPCs). The proposal is offered as an alternative to the view that a natural kind must have an essence, a set of conditions that are necessary and sufficient for membership. After Wittgenstein, it is common to say that a kind can have a cluster of properties rather than an essence. Cluster concepts are notoriously wooly-headed. What an HPC adds to the cluster is that there is a causal process that produces the properties and is responsible for their being clustered.

As an example, consider mallards. Members of the species typically look like ducks, walk like ducks, and quack like ducks. A one-legged mallard will not walk like a duck, a mute mallard will not quack, and so on, but they are nonetheless still mallards. So the properties are a cluster rather than an essence. Yet the cluster is sufficiently stable as to support induction. The cluster of duck properties is maintained in an individual duck by its physiology, literally by its homeostatic processes. It is propagated beyond individuals by duck reproduction. Mommy ducks and daddy ducks spawn clusters of duck properties in the form of ducklings.

The example may be complicated by noting that there is no such thing as a daddy duck. Male mallards are drakes.** Aside from the point about archaic nomenclature, one must also admit that ducks and drakes are rather different in many respects. Marc Ereshefsky and Mohan Matthen*** argue that the HPC approach fails precisely because of differences like these.

Enter E&M

Ereshefsky and Matthen (E&M) observe that the tradition of thinking about natural kinds has been about similarity. If kinds have essences, then all members of the kind are similar in that specifiable way. If kinds have property clusters, then members of the kind are roughly and probably similar.

As E&M note, identifying kinds cannot be done ex nihilo. One must start with some description of the relevant properties and their respective importance. E&M call such a description a morphospace.

They then reconstruct what someone who wants to identify HPCs must do. It goes like this: 1. Notice some clusters in the morphospace; i.e. similarities. 2. Look for the mechanisms responsible for those clusters. 3. Rejigger the morphospace in light of the mechanisms and start again.

The rejiggering is necessary because you ultimately want to group together specimens that would not initially look similar. For example, imagine one starts by making brief observations of caterpillars and moths. The two occupy different bits of the initial morphospace. Yet one wants to count them as members of the same species, because such caterpillars grow up to be such moths who lay eggs for such caterpillars.

Trying to understand this in terms of property clusters and similarity is awkward. The caterpillar crawls and the moth flies. If one characterizes the causal process as homeostatically maintaining a single property, then one has several unappealing options:

Option A: Rejigger the properties that are included in the morphospace. That is, say that the cluster includes the disjunctive property of crawling-or-flying.

As E&M note, this reply threatens to make similarity vacuous. Moreover, it would lose track of the fact that there are two separate clusters. Members of the species aren't just crawling-or-flying and cocoon-weaving-or-egg-laying. The crawling and the egg laying are never found together.

Option B: Rejigger the weighting of properties in the morphospace. Say that the cluster includes only crawling or flying, whichever is natural, and that the other is a kind of deformity.

This would miss the point, because both are stages in the butterfly life cycle. A caterpillar's not flying is different than a one-legged duck's not walking.

Option C. Adjust the kinds so that you do not need to rejigger the morphospace. Say that the caterpillar and the butterfly are members of different HPCs.

This would mean that the species is not an HPC after all, which is just to give up the Boydian idea.

None of A, B, or C are acceptable. So if we accept E&M's description of the workflow, the flow from 1 to 3 above, then the HPC view collapses.

Boyd often insists that natural kinds are ones that support induction, and straight-rule induction does require that members of kinds share similar properties. So E&M's reading is plausible. I think there's an alternate reading, however, according to which the HPC view does not collapse.

Another approach

Diagnosing an HPC leads us to identify causal processes. Once we have identified the causal process, we don't need to characterize the kind in terms of similarity at all. Rather, it is characterized in terms of the causal process itself.

So, rather than E&M's reconstruction, I offer the following: 1. Notice clusters of similarities. 2. Look for the mechanisms responsible for those clusters. 3. Identify natural kinds in terms of the presence or absence of those mechanisms.

The caterpillar and the butterfly are members of the same HPC because the causal process that produces the crawling and cocoon weaving cluster of properties early in its life later produces the flying and egg laying.

However, E&M argue that such a strategy won't help with sexual dimorphism. They say, "Sexual dimorphism ... is due to males and females having different chromosomes and different developmental processes. There is no theoretically meaningful similarity under which the variation between the males and females ... can be subsumed."****

For a single individual mallard, this is entirely correct. The casual processes that make a duck continue to walk and quack also give her female colouration and behaviour. Conversely, the specific causal processes that make a drake walk and quack also give him male features. Yet the causal process responsible for the mallard-cluster's persistence are not just the metabolisms of individual mallards. New mallards appear in the world. That causal process is mallard reproduction, and it necessarily requires both the ducks and the drakes.

Because the HPC approach leads us (at my step 3) to define the natural kind relative to the causal process - and because the causal process of duckiness requires both ducks and drakes - the natural kind involves both duck features and drake features. This is not by way of a rejiggered morphospace in which we pretend that ducky-or-drakey is a feature, but instead directly from the casual processes that the approach identifies.

I think my proposal is a more accurate reading of Boyd. Despite Boyd's talk of projectibility and induction aside, he does not think that scientific inference is simply a matter of applying the straight rule. It is also a matter of providing explanations, and the explanatory machinery of the HPC approach bottoms out in causal processes. Boyd says, for example, that the definition of an HPC kind "is thus a historical process like, e.g., the second world war or the emergence of capitalism."***** The best candidate for the historical process of sustained mallardhood is not the life of one mallard but the whole lineage. The lineage does not include ducks or drakes accidentally, but involves both integrally. So the HPC makes sense of the dimorphism.

So what?

There are two possibilities which I will mention but not address.

First, it may be that the suggestions I have made would be welcomed by E&M. I don't have a good grasp of the alternative that they call Population Structure Theory. Perhaps HPC as I understand it and PST look fairly similar. The disagreement then would just be one of emphasis, that mentioning 'property clusters' suggests that similarity is paramount.

Second, there may be some examples of what E&M call deep polymorphism that cannot be handled by HPC as I sketch it above. I don't see a problem with any of the other examples they mention in their paper, but I have not thought through all of them.

* All over the place. Imagine the usual flurry of citations here.

** 'Mallard', in its etymology, originally referred to a wild drake.

*** 'Taxonomy, Polymorphism, and History: An Introduction to Population Structure Theory.' Philosophy of Science, 72 (January 2005) pp. 121.

**** The ellipses leave out that they are talking about mammalian species, but I'll continue with the example of ducks. I don't think anything turns on this point.

***** Here he's discussing the example of Procyon lotor (raccoons). p. 71, 'Kinds at the "Workmanship of Men": Realism, Constructivism, and Natural Kinds.' It's in a German proceedings volume with a long title. You can Google it.