syntactically categorizing light elements as computationally classifier elements; and why we have classifiers at all

In neoconstructionist models, syntactic categorization is arrived at through the addition of light elements to categorically unspecified Roots or Root complexes. The effect of these elements is strikingly similar to nominal classifiers, which explicitly constrain a lexical element to nominal syntax and interpretation. Comparably, light verbs also , and indeed many light verb systems are described as verbal classifiers---predicate classifiers, that is, and not to be confused with the nominal classifiers that can incorporate into verbs.

So who cares? Anything that systematically changes and/or reduces something else to a new set of properties (particularly in constraining its semantic range) can be loosely splattered with the "classifier" label, to no usefully precise effect. Combine [green] with [book] creates an entity that belongs to the class of [green things]. True, but no real help.

But.

There's something more to this. We need to step back for a moment and ask why languages so recurrently develop classifier systems of the more familiar, unambiguous type. I think it is because they present a pretty optimal solution to problems both of lexical retrieval and of syntactic combination.

In lexical retrieval, classifiers optimize a lexical search algorithm. If you have a lexicon of 1000 distinct lexical items divided into 10 classes (for sake of argument, all evenly divided into 100 members each), then at the first sign of a classifier, the search algorithm can immediately ignore 900 out of 1000 possibilities. That's pretty sweet, but add to it also the neuropsychological component to boot: obviously through association, a classifier is going to have a rich priming effect for whatever its classifiee is. That's a double whammy of lexical retrieval optimization.

And for the computational system, just manipulating relations between these elements, reduction of the full lexicon to a few categories means that said system can contentedly process the same set of five or so kinds of boxes all day long, contentedly ignoring any complications from the richer semantic contents inside each. This is why stereotypes are so pervasive to human cognition: because they're really really efficient, provided they're actually accurate.

So that's the value of a classifier system in two directions. And you can see from the above fairly easily that the features of a classifier system that facilitatate syntactic computation/manipulation could be written again for syntactic categories with no trouble.

The idea, then, is not to equate syntactic categories with classifiers, but to simply say that they are the same kind of system, operating at different scales of complexity in syntactic representation, but always driven by the same basic information-theoretical and biosystemic properties. And in so doing, making for a rather simple mechanism for grabbing lexical items, putting them together, and reading off the resulting relations that hold between them.


[It's worth noting that this lexical access optimization effect of classifiers is rich enough that it arose not just in the spoken Sinitic languages (among others), but also in the graphic lexicon of Chinese. The vast bulk of the lexicon of Chinese characters consist of a phonological element (a pre-existing character, notionally familiar for its sound, though now in practice often a fair bit far afield from the user's actual spoken form) accompanied by an additional character/component that acts as a semantic diacritic.

In short, learning for the first time and/or recalling a previously learned Chinese graphic lexeme, you see a phonological cue (usually not terribly worse than the incomplete/inconsistent phonological information supplied by your average English alphabetic spelling), and then a semantic classifier that allows you to radically narrow your search space to [THING WITH THIS RANGE OF MEANING] that has [THIS ROUGH PHONOLOGICAL FORM]. So again, a classifier system optimizes lexical retrieval: our Encyclopedia can be and usually is orgranized into thematic chapters with nice bold headings.

Interestingly, adaptation after adaptation, innovation after innovation, and error after error over the many years of use of Chinese characters shows that the phonetic component is perhaps the most robust, salient aspect of the system (thank you, John DeFrancis, among others): nonstandard usages regularly borrow familiar characters strictly for their phonological similarity, ignoring the original semantics also encoded with them. A triumph of form over content, if you will, and a highlight of the tendency for this cognitive system (like many) to happily manhandle any "real" material into an abstracted symbol. Just as in spoken languages, classifiers aren't entirely crucial, at least at the purely lexical retrieval level; our spoken and graphic lexicons can get by without them just fine. They just help.

Not so sure if our syntactic systems could, though.]