Named Entity Recognition System for Urdu

Proceedings of COLING 2012: Technical Papers, pages 2507–2518, COLING 2012, Mumbai, December 2012. Named Entity Recognition System for Urdu UmrinderPal Singh1,1, Vishal Goyal2,1 Gurpreet Singh Lehal3,1 (1) Department of Computer Science, Punjabi University Patiala umrinderpal@gmail.com1, vishal.pup@gmail.com2, gslehal@gmail.com3 ABSTRACT Named Entity Recognition (NER) is a task which helps in finding out Persons name, Location names, Brand names, Abbreviations, Date, Time etc and classifies them into predefined different categories. NER plays a major role in various Natural Language Processing (NLP) fields like Information Extraction, Machine Translations and Question Answering. This paper describes the problems of NER in the context of Urdu Language and provides relevant solutions. The system is developed to tag thirteen different Named Entities (NE), twelve NE proposed by IJCNLP-08 and Izaafats. We have used the Rule Based approach and developed the various rules to extract the Named Entities in the given Urdu text. Keywords: NER, Urdu NER, Urdu, Rule Based 2507 1. Introduction Named Entity Recognition (NER) is a subtask of Information Extraction (IE). NER extracts and classifies the true Named Entities in text. NER system is widely used in different tasks of Natural Language Processing (NLP) and in many commercial applications on internet like Search Engine. Accuracy of NER system is directly reflected in NLP applications. So, accurate working of NER system is very important. NER system can be used for one's personal interest like company manager wants to know all the names involved in specific text document. S.No: Tag Name Example 1 Person Name دشرا(Arshad) 2 Location ا�یٹپ (Patiala) 3 Organizations س�یار (Reliance) 4 Terms سٹاڈ�پس (Spondylitis) 5 Designation �ظعا �ی�و (President) 6 Title Person ���ج (Mr.) 7 Title Object ��ئ�ٹ ��تسود�ہ (Hindustan Time) 8 Brand گ�س�یس (Samsung) 9 Measure ل�س 10 (10 Years) 10 Number ود ،کیا (One, Two) 11 Date/Time ��وتکا 12 (12 October) 12 Abbreviation یس ی� ی� (BBC) TABLE 1- Different Named Entity Tags Named Entities mentioned above were proposed at IJCNLP-08 workshop [17]. Named Entities can be domain specifics like NER system to identify entities in scientific data. The NER system can be developed using three approaches, 'Rule-Based', 'Machine Learning' (HMM, SVM, CRF, Decision Tree) and 'Hybrid' approach. The Rule-Based system is difficult to develop as one should know the language and grammar rules. These kinds of systems are domain specific. Machine learning approach provides different Statistical NLP tools to train NER system. Statistical tools provide fast way to develop NER system but the accuracy of the system is dependent on annotated training data. For greater accuracy we need to train the NER system with large amount of annotated data. Hybrid approach is a combination of both Rule Base and Statistical based. 2. Related Work NER system came in focus during the sixth Message Understanding Conference (MUC-6) [6]. After that many NER systems were developed. Most of these systems were developed for European languages and all systems were highly accurate. For south Asian languages, NER systems yet in developing phase. IJCNLP-08 workshop played a major role in development of NER Systems for …

Content truncated to 500 words