Publication details

KernelTagger – a PoS Tagger for Very Small Amount of Training Data



Year of publication 2017
Type Article in Proceedings
Conference Proceedings of the Eleventh Workshop on Recent Advances in Slavonic Natural Language Processing, RASLAN 2017
MU Faculty or unit

Faculty of Informatics

Field Informatics
Keywords PoS tagging; morphological tagging; language model; Czech
Description The paper describes a new Part of speech (PoS) tagger which can learn a PoS tagging language model from very short annotated text with the use of much bigger non-annotated text. Only several sentences could be used for training to achieve much better accuracy than a baseline. The results cannot be compared to the results of state-of-the-art taggers but it could be used during the annotation process for a pre-annotation.
Related projects: