Named Entity Recognition in Estonian

Abstract

The task of Named Entity Recognition (NER) is to identify in text predefined units of information such as person names, organizations and locations. In this work, we address the problem of NER in Estonian using supervised learning approach. We explore common issues related to building a NER system such as the usage of language-agnostic and languagespecific features, the representation of named entity tags, the required corpus size and the need for linguistic tools. For system training and evaluation purposes, we create a gold standard NER corpus. On this corpus, our CRF-based system achieves an overall F1-score of 87%. 1