Tim O’Donnell was in Leiden last week for the The Comparative Biology of Language Learning workshop, held at the Lorentz Center April 3–7. He gave a talk Thursday, title and abstract below:

Bayesian Program Learning of Morphophonological Rules
Both children and linguists confront a similar problem of inference:
given utterances produced by speakers, together with aspects of the
meaning of those utterances, discover the grammatical principles that
relate form to meaning. We study this abstract computational problem
within the domain of morphophonology, contributing a computational
model that learns phenomena from many natural languages and
generalizes in humanlike ways from data used in behavioral studies of
artificial grammar learning.
Our work draws on two analogies. The child-as-linguist analogy holds
that both children and linguists must solve the same abstract
inductive reasoning problem, even though the nature of the input data
and underlying mental algorithms are surely different in precise
detail. Accordingly we isolate the problem of learning
morphophonological systems, and show that a single solution to this
problem can capture both linguistic analyses from natural languages
and infant rule learning of artificial languages. We adopt the
framework of “Bayesian Program learning” (BPL) – in which learning is
formulated a synthesizing a program which compactly describes the
input data. This learning-as-programming analogy lets us exploit
recent techniques from the field of program synthesis to induce
morphophonological rules from data. While child-as-linguist poses the
computational problem, learning-as-programming offers a solution.