Is There a Universal Ontology?

Cathy & Cathy faced off on the "Yes" side of this question against the "No" team of ontological skeptics. Here is something of what they said, and I heard.

Team Yes began with some chosen definitions. What "universal" means is what a normal -- or typical -- human would encounter in its environment over the course of development. The most obvious example of this is the world itself to which we are all exposed.

"Ontology" is a theory of the world and its workings. Such a theory need not be formalized, but merely a set of consistent concepts which begin to be accessible from the moment we are born. For example, children can have a sense of grammar long before they gain a formal sense of what grammar is; we can come to ontology in a similarly informal, but effective, way.

Xu's experiment with two sets of children whose languages made different mass/count distinctions found that these children extended novel words for objects in different ways depending on the degree to which their language grammaticized the distinction between objects and substances. Underlying that difference, though, was the universal and early distinction made by all children between the "wordly" properties of object and substance, though their languages may have led them to express the details of this distinction differently.

Additionally, 10-month-olds have a working sortal concept of "object" as shown by their differing attention to toys which move out of and back into sight in ways consistent with indivisible, solid, persistent objects. Though they perceive this concept before they can completely differentiate objects such as a toy duck and a ball, they clearly are establishing a basic spatio-temporal ontology: principles which apply to all objects.

There's good reason to believe that this sortal concept can be acquired and used by very young (pre-productive) children, because Smith has shown that even a connectionist network of no great sophistication can be trained to make the same kind of differentiation. This fundamental distinction may not be a complete "theory" of ontology, but it could certainly serve as the basis for one.

No took the floor with a different interpretation of "ontology". It's not sufficient to credit a few basic principles such as those mentioned with the role of an "ontology". A universal ontology implies that everyone's conceptual system is centered around common basic categories of objects and their relation to each other. This creates a hierarchy of distinctions (such as objects divided into animate and inanimate, animate divided into people and animals, and so forth). This conceptual system (which need not be innate) must be shared by all people.

To successfully argue "yes" in this question, then, it must be shown that such a hierarchy not only exists, but pervades our conceptual system. It's tempting to posit such a pervasive pre-linguistic categorization of the world, for it would help tremendously in the acquisition of words for things in the world. Nevertheless, there is no evidence for such a universal, cognitively pervasive system, nor is there a need to posit one to render a satisfactory explanation of early word learning. Smith's artificial model, which comes to make a shape-based object distinction for itself, serves as a demonstration of this, since it succeeded in its task without any pre-existing cognitive model of objects. Instead, this important behavior can emerge under the influence of the language itself which reinforces such a distinction. This apparently universal ontological basis, then, is not in the universal environment -- the world -- but simply expressed in the particular language to which one is exposed.

Yes rallied by conceding that no innate categorical system was being proposed; it is certainly sufficient, and indeed probable, that such a system is built during development. This ontology need not serve any linguistic purpose, but only express some observable evidence. In this respect, Smith's model supports just such a conclusion, for it assembled a categorical distinction by doing (at least) what any child must do: attending to regularities in its perceptions.

No: But it is necessary for a proposed universal ontology to serve a purpose. If what we share is simply "attendance to regularities", this hardly constitutes a universal ontology. This statistical attention is just what networks do, and networks don't have a conceptual structure at all which can be pervaded by any universal categorical system. Why postulate a complex ontological system when clearly simple statistical principles can serve? Therefore a universal ontology must accord some tangible benefit in order to make it worth proposing at all.

There's no evidence, of course, that such a benefit exists. Instead, a universal ontology is just another manifestation of the desire for universal simplifications which dismiss individual variety conferred by developmental differences. If this "universal ontology" has any real use, it must be pre-linguistic or even innate.

Yes: Nonsense! A universal ontology can indeed be a product of development: 10-month-olds, who are by any account pre-linguistic, clearly give evidence of just such a development (and plenty of time to have developed it in). The tasks that a 10-month-old is called upon to do simply don't rise above what can be accomplished by a network, so it's no surprise that they can give similar results. But that doesn't mean that the simple processes which drive a network can also model the greater cognitive/linguistic sophistication of a two-year-old, or an adult. Something with more explanatory power is needed to account for this sophistication, and an ontological system provides just that.

No: Well, then, shouldn't the "simple" network's learned shape bias have been expressed as part of the child's universal ontology quite early?

Yes: Not necessarily. Not every distinction is part of a fundamental ontology.

No: But we know that perceptual saliencies alone can exert considerable influence on the categorization of words (this refers to the "glittery ball" experiment in which some properties of an object could be selectively enhanced, and cause the object to be treated in a different linguistic way by the child). Such a simple principle, with broad effects, seems just as powerful as a universal ontology. Indeed, that same experiment shows that any ontological structure can have its boundaries broken by the influence of the environment. How can a "universal" ontology explain the wide varieties of categorization and the way fine conceptual distinctions are made? If the object/substance difference can be conceived differently in different cultures, can it really be said that we are pervaded by a universal system?

Linda Smith weighs in on the Yes side: But a universal ontology might well exist as a post-linguistic phenomenon, derived from the individual, the world and even language -- even though it's a more interesting claim if it's pre-linguistic, that doesn't disqualify a post-linguistic existence.

No: How are you going to establish that this ontology is in fact universal without examining everyone in existence for it?

Linda (apparently changing sides): And what's the evidence that the simple discriminations talked about so far can be called ontological? How do we know that they are deeply conceptually founded? At best they reveal a behavior in certain situations, not a fact about conceptualization.

Alex (rising to the defense of the Yes side): An ontology may not propose the content of categories, but simply their structure -- like a specific hierarchy waiting to be filled with individual information. Then a universal structure would prevail, while accommodating individual variation.

No: But if such as hierarchy, regardless of its form, has to be extracted from world post-linguistically, how would it help one learn words at all? An ontological system must confer a unique advantage, or else not be worth postulating at all.

And so it was that the "No"s had the last word, or at least the last note taken.