The behaviorist Quine famously argued that matters of semantics couldn't be behaviorally distinguished from matters of belief, a thesis that easily generalizes to syntax. Allowing some plausibility to Quine's claim, I argue that, ironically enough, it serves as a surprisingly strong argument against current general statistical learning, even nativist Bayesian alternatives to Chomskyan proposals of an innate Universal Grammar. For if ordinary behavior is an insufficient basis for learning linguistic distinctions that a child manifestly masters within a few years, then children must be born equipped with a language processing system that automatically imposes those distinctions on the behavior they observe, largely independent of statistical hypothesis testing.