Enhancing Low-Resource NER via Knowledge Transfer from LLM

This paper presents a study for low-resource language NER via knowledge transfer using large pre-trained language models. The goals of the study are to enhance the performance of the proposed model for low-resource language NER through knowledge obtained from PLMs with high-resource languages and to analyze the effectiveness of how PLMs contribute to NER in terms of knowledge transfer. To achieve these objectives, experiments related to knowledge transferring from monolingual and multilingual PLMs, different training settings, and few-shot learning cases are conducted. Experimental results show that the knowledge transfer learning (KTL) approach improves performance by approximately 7% over the previous best in the case of Kazakh, a typical low-resource language.