CINO: A Chinese Minority Pre-trained Language Model

Abstract

Multilingual pre-trained language models have shown impressive performance on cross-lingual tasks. It greatly facilitates the applications of natural language processing on low-resource languages. However, there are still some languages that the existing multilingual model does not perform well on. In this paper, we propose CINO (Chinese Minority Pre-trained Language Model), a multilingual pre-trained language model for Chinese minority languages. It covers Standard Chinese, Cantonese, and six other Chinese minority languages. To evaluate the cross-lingual ability of the multilingual models on the minority languages, we collect documents from Wikipedia and build a text classification dataset WCM (Wiki-Chinese-Minority). We test CINO on WCM and two other text classification tasks. Experiments show that CINO outperforms the baselines notably. The CINO model and the WCM dataset will be made publicly available.

Tasks

Reproductions