A Cross-Lingual Framework for Named Entity Dataset Construction Using Knowledge Transfer
A Cross-Lingual Framework for Named Entity Dataset Construction Using Knowledge Transfer
Reyhaneh Boroomand1 Marziea Rahimi2
1) Faculty of Computer Engineering, Shahrood University of Technology, Shahrood, Iran.
2) Faculty of Computer Engineering, Shahrood University of Technology, Shahrood, Iran.
IOI Code
:
AAHD-ABKDGK
Publication :
National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
In this study, a novel approach for creating a top-performing Persian named entity recognition dataset using cross-lingual transfer learning has been proposed. The method proposed here can overcome the main limitations that exist in the existing projection-based approaches through translation-aware alignment, dual matching, and noise filtering methods. The created dataset works well in many different named entity recognition models. This paper proposes a new approach and a valuable linguistic resource for low-resource language processing tasks. In future work, we intend to focus on extending the proposed method into new languages, employing active learning methods, and improving the alignment step with the help of multilingual representations.
Keywords : named entity recognition (ner) cross lingual transfer learning low resource languages natural language processing annotation projection