Boosting trees for clause splitting

We present a system for the CoNLL-2001 shared task: the clause splitting problem. Our approach consists in decomposing the clause splitting problem into a combination of binary "simple" decisions, which we solve with the AdaBoost learning algorithm. The whole problem is decomposed in two levels, with two chained decisions per level. The first level corresponds to parts 1 and 2 presented in the introductory document for the task. The second level corresponds to the part 3, which we decompose in two decisions and a combination procedure.