[4Xin2] Poster session 2

[4Xin2-110] Multi-Lingual Prompt: Few-shot inference using multilingual examples in low-resource languages

〇Yasuhisa Kato¹, Masahiro Kaneko^1,2, Naoaki Okazaki¹ (1.Tokyo Institute of Technology, 2.Mohamed bin Zayed University of Artificial Intelligence)

Keywords:GPT, Few-shot inference, multi language, low resource language

While large language models demonstrate high performance on English tasks, it is known that their performance
relatively decreases when solving tasks in low-resource languages. To address this issue, this paper proposes few-
shot inference using multilingual examples. In this study, we employed the natural language inference task as a
metric to measure the inference performance of large language models under low-resource languages. As a result,
our proposed method showed performance improvements on two tasks, FEVER and ANLI.