Abstract It is commonly believed that phonology, morphology, syntax, and semantics are distinct modules of language, governed by very different principles with little common ground. Nonetheless several approaches (e.g. Government Phonology, Distributed Morphology) subscribe to the idea that at least some of these language modules exhibit profound parallels. Interestingly, recent findings in computational linguistics corroborate this hypothesis: it seems that linguistic dependencies belong to a small, heavily constrained region of the so-called subregular hierarchy. If correct, this has far-reaching typological implications, sheds new light on learnability issues, and raises deep questions about the cognitive architecture of language.

The subregular hierarchy provides a system for classifying structural dependencies according to their computational complexity and expressive power. It turns out that linguistic dependencies largely cluster at the lower end of the hierarchy, thanks to the limiting effects of the locality conditions that they are subject to. This computational result furnishes answers to a variety of seemingly unrelated questions, such as:

• Why is there no first-last harmony in phonology?
• Why are some prefixes and suffixes freely iterable, but no circumfixes?
• Why does no language have a monomorphemic version of the quantifier an even number of?
• Why is syntactic movement upward rather than downward?

This is but the tip of a much larger iceberg. The predictions span phonology, morphology, syntax, and semantics, and they even intersect with psycho- and neurolinguistics. Computational linguists cannot do all of this by themselves. Without the expertise of theoretical linguists, it is impossible to fully explore the cognitive, typological, and theoretical ramifications of this “subregular conspiracy”. If we want to know how deep the rabbit hole goes, we have to go down there together.