Improving Named Entity Recognition by Jointly Learning to Disambiguate Morphological Tags

Proceedings of the 27th International Conference on Computational Linguistics, pages 2082–2092 Santa Fe, New Mexico, USA, August 20-26, 2018. 2082 Improving Named Entity Recognition by Jointly Learning to Disambiguate Morphological Tags Onur G¨ung¨or Bo˘gazic¸i University, Istanbul, Turkey onurgu@boun.edu.tr Suzan ¨Usk¨udarlı Bo˘gazic¸i University, Istanbul, Turkey suzan.uskudarli@boun.edu.tr Tunga G¨ung¨or Bo˘gazic¸i University, Istanbul, Turkey gungort@boun.edu.tr Abstract Previous studies have shown that linguistic features of a word such as possession, genitive or other grammatical cases can be employed in word representations of a named entity recognition (NER) tagger to improve the performance for morphologically rich languages. However, these taggers require external morphological disambiguation (MD) tools to function which are hard to obtain or non-existent for many languages. In this work, we propose a model which alle- viates the need for such disambiguators by jointly learning NER and MD taggers in languages for which one can provide a list of candidate morphological analyses. We show that this can be done independent of the morphological annotation schemes, which differ among languages. Our experiments employing three different model architectures that join these two tasks show that joint learning improves NER performance. Furthermore, the morphological disambiguator’s performance is shown to be competitive. Title and Abstract in Turkish Bic¸imbilimsel Etiketleri Ayrıs¸tırmayı Birlikte ¨O˘grenerek Varlık ˙Ismi Tanıma Bas¸arısını Artırmak Daha ¨onceki c¸alıs¸malar, bic¸imbilimsel olarak zengin dillerdeki varlık ismi tanıma (VAT) bas¸arısını artırmak ic¸in s¨ozc¨uklerin iyelik, genitif ve benzeri hˆallerinin kullanılabilece˘gini g¨ostermis¸tir. Ancak, bu t¨urden varlık ismi tanıma is¸aretleyicilerinin c¸alıs¸abilmesi ic¸in elde edilmesi zor veya bazı diller ic¸in imkansız olan dıs¸sal bic¸imbilimsel ayrıs¸tırıcılara (BA) ihtiyac¸ vardır. Bu c¸alıs¸mada, bu t¨ur ayrıs¸tırıcılara olan ihtiyacı ortadan kaldırmak ic¸in VAT ve BA g¨orevlerini aynı anda c¸¨ozen ve aday bic¸imbilimsel c¸¨oz¨umlemelerin sunulabildi˘gi dillere uygulanabilen bir model ¨onerilmektedir. Bunun dillere g¨ore de˘gis¸en bic¸imbilimsel is¸aretleme s¸emalarından ba˘gımsız olarak yapılabildi˘gi g¨osterilmis¸tir. Bu iki g¨orevi aynı anda gerc¸ekles¸tiren ¨uc¸ farklı model mimarisi kullanarak yaptı˘gımız deneyler birlikte ¨o˘grenmenin VAT bas¸arısını artırdı˘gını g¨ostermis¸tir. Buna ek olarak, bic¸imbilimsel ayrıs¸tırıcının bas¸arısının ¨onceki c¸alıs¸malarla kars¸ılas¸tırılabilir oldu˘gu g¨or¨ulm¨us¸t¨ur. 1 Introduction Named entity recognition (NER) is the task of selecting the portions of text which refer to an entity that designate a person, location or organization. This makes it a basic natural language processing (NLP) task closely related to relation extraction, knowledge base population, and entity linking. Works that represent the current state of the art in NER generally start by representing words with pretrained word embeddings, embeddings which rely on surface form characters (Lample et al., 2016; Ma and Hovy, 2016). These architectures feed the word representations to a bidirectional long short- term memory layer (Bi-LSTM) to represent the context where the disambiguation between the possible entities is undertaken by decoding on trellis provided by a conditional random field (CRF) model. This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 2083 When these models are trained and evaluated for morphologically rich languages (MRLs), it has been shown that using embeddings based on characters or linguistic properties of the word such as morpholog- ical features indicating a grammatical case …

Content truncated to 500 words