Backoff estimation for English verb subcategorization acquisition based on diathesis alternation

This paper proposes a new filtering method with diathesis alternation as heuristic information, which improved the performance of Korhonen's English SCF acquisition system remarkably, with the precision increased to 91.18%, recall unchanged, the absolute F-measure increased by 3.96, and the relative F-measure by 13.72, thus making the acquired lexicon much more practical for further manual proofreading and other NLP uses.