The importance of high-quality input for WSD: an application-oriented comparison of part-of-speech taggers