Distributional Encroachment and Its Consequences for Morphological Learning

We describe a common but neglected pattern of linguistic exceptions, which involve “distributional encroachment.” This occurs when the distribution of allomorphs is determined by phonological context, but a few exceptional forms take the “wrong” allomorph. For learning algorithms, this can complicate the task of identifying distributions. We present an algorithm for learning allomorph distributions, then show how it can be modified to handle distributional encroachment.