KoBigBird-large: Transformation of Transformer for Korean Language Understanding

• DOI: 10.48550/arxiv.2309.10339
• Publication Date: 2023-01-01
• Authors: Yang, Kisu; Jang, Yoonna; Lee, Taewoo; Seong, Jinwoo; Lee, Hyungjin; Jang, Hwanseok; Lim, Heuiseok
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Computation and Language (cs.CL)

Accepted at IJCNLP-AACL 2023<br/>This work presents KoBigBird-large, a large size of Korean BigBird that achieves state-of-the-art performance and allows long sequence processing for Korean language understanding. Without further pretraining, we only transform the architecture and extend the positional encoding with our proposed Tapered Absolute Positional Encoding Representations (TAPER). In experiments, KoBigBird-large shows state-of-the-art overall performance on Korean language understanding benchmarks and the best performance on document classification and question answering tasks for longer sequences against the competitive baseline models. We publicly release our model here.<br/>