Wrapper Induction: Learning (k,l)-Contextual Tree Languages Directly as Unranked Tree Automata

A (k, l)-contextual tree language can be learned from positive exam- ples only; such languages have been successfully used as wrappers for informa- tion extraction from web pages. This paper shows how to represent the wrapper as an unranked tree automaton and how to construct it directly from the exam- ples instead of using the (k, l)-forks of the examples. The former speeds up the extraction, the latter speeds up the learning.