HMM-Based Dari Named Entity Recognition for Information Extraction

Named Entity Recognition (NER) is the fundamental subtask of information extraction systems that labels elements into categories such as persons, organizations or locations. The task of NER is to detect and classify words that are parts of sentences. This paper describes a statistical approach to modeling NER in Dari language. Dari and Pashto are low resources languages, spoken as official languages in Afghanistan. Unlike other languages, named entity detection approaches differ in Dari. Since in Dari language there is no capitalization for identifying named entities. We seek to bridge the gap between Dari linguistic structure and supervised learning model that predict the sequences of words paired with a sequence of tags as outputs. Dari corpus was developed from the collection of news, reports and articles based on the original orthographic structure of the Dari language. The experimental result of named entity recognition performance presents 94% accuracy.