Deep Neural Learning Approaches for Latvian Morphological Tagging
Paikens P emacr teris · Frontiers in artificial intelligence and applications · 2016
This paper describes ongoing research on improvements of morphological analysis, disambiguation and POS tagging for the Latvian language. Authors apply recent advances in sequential tagging with neural networks and word embeddings calculated from unlabeled corpus to improve morphological tagging accuracy. These approaches allow to reduce the fine-grained morphological tag word error rate from 7.9% of earlier best systems to 6.2%, and coarse-grained POS tag error rate from 3.6% to 2.2%.