HuLaPos 2.0 - Decoding morphology

Abstract

In this paper, a language-independent morphological annotation tool is presented that is based on the Moses SMT toolkit. Taking Hungarian as an example, we demonstrate that the algorithm performs very well for morphologically rich languages. In order to reach a very high, more than 98%, annotation accuracy, the presented system uses a trie-based suffix guesser, which enables the tool to handle words unseen in the training data effectively. The system yields state-of-the-art performance among language-independent tools for morphological annotation of Hungarian. For PoS tagging, it even outperforms the best hybrid tagger, which includes a language-specific morphological analyzer. © Springer-Verlag 2013.