Parsing Morphologically Rich Languages with (Mostly) Off-The-Shelf Software and Word Vectors
Arne Köhn, Uyen Lao, Amir Zadeh, Kenji Sagae · 2014
As a contribution to the 2014 SPMRL shared task on parsing morphologically rich languages, we show that it is now possible to achieve high dependency accuracy using existing parsers without the need for intricate multi-parser schemes even if only small amounts of training data are available. We further show that the impact of using word vectors on parsing quality heavily depends on the amount of morphological information that is available. In addition, we discuss the use of parser scores for selection of morphological lattice paths, showing that there is much discriminative power in syntactic parsers for morphological disambiguation.