Sequential data is found in many different places, among others in language, music, DNA, and financial data. These types of data can be structured in meaningful ways. In the first part of this talk, I will describe the Alignment-Based Learning (ABL) framework that can be used to structure plain data automatically. This framework is based on Harris's (1955) idea of substitutability. ABL extends the idea by adding statistics to solve some of the problems and ambiguities of Harris's approach.

The second part of the talk consists of an overview of applications of the ABL framework. I will describe how I have applied ABL to the problems of finding phrase structure in natural language and selecting expected answer types in questions answering systems. Then I explain how I think it is possible to create a machine translation system using ABL and what problems arise when one tries to learn structure in music.