In this project, we investigate Data-Oriented Transformation, 
based on the existing formalisms of Data-Oriented Parsing, synchronous
grammar, and the Goodman Reduction. The goal is to define and evaluate a general
formalism for learning arbitrary transformations composed of movement and insertions
based on annotated exemplars.  Parsing a sentence should result in a pair of trees
corresponding to the original and a transformed sentence.
This stochastic approach contrasts with purely rule-based formalisms such as
transformational grammar, in which abstract representations are manipulated
with a priori specified operations. The test case will be declarative versus
interrogative questions.