Neural models for unsupervised disambiguation in morphologically rich languages

The problem of morphological ambiguity is central to many natural language processing tasks. In particular, morphologically rich languages pose a unique challenge due to the large number of possible forms some words can take. In this work, we implement and evaluate a method for morphological disambiguation of morphologically rich languages. We use deep learning techniques to build a disambiguation model and leverage existing tools to automatically generate a training data set. We evaluate our approach on the Finnish, Russian and Spanish languages. For these languages, our method surpasses the state-of-the-art results for the tasks of part-of-speech and lemma disambiguation