TiKEM:基于知识增强的藏文预训练语言模型(TiKEM: Knowledge Enhanced Tibetan Pre-trained Language Model)

Junjie Deng (邓俊杰), Long Chen (陈龙), Yan Zhang (张廷), YUan Sun (孙媛), Xiaobin Zhao (赵小兵)


Abstract
“预训练语言模型在中英文领域有着优异的表现,而低资源语言数据获取难度大,预训练语言模型在低资源语言如藏文上的研究刚取得初步进展。现有的藏文预训练语言模型,使用大规模无结构的文本语料库进行自监督学习,缺少外部知识指导,知识记忆能力和知识推理能力受限。为了解决以上问题,本文构建含有50万个三元组知识的藏文知识增强预训练数据集,联合结构化的知识表示和无结构化的文本表示,训练基于知识增强的藏文预训练语言模型TiKEM,以提高模型的知识记忆和推理能力。最后,本文在文本分类、实体关系分类和机器阅读理解三个下游任务中验证了模型的有效性。”
Anthology ID:
2023.ccl-1.12
Volume:
Proceedings of the 22nd Chinese National Conference on Computational Linguistics
Month:
August
Year:
2023
Address:
Harbin, China
Editors:
Maosong Sun, Bing Qin, Xipeng Qiu, Jing Jiang, Xianpei Han
Venue:
CCL
SIG:
Publisher:
Chinese Information Processing Society of China
Note:
Pages:
135–144
Language:
Chinese
URL:
https://aclanthology.org/2023.ccl-1.12
DOI:
Bibkey:
Cite (ACL):
Junjie Deng, Long Chen, Yan Zhang, YUan Sun, and Xiaobin Zhao. 2023. TiKEM:基于知识增强的藏文预训练语言模型(TiKEM: Knowledge Enhanced Tibetan Pre-trained Language Model). In Proceedings of the 22nd Chinese National Conference on Computational Linguistics, pages 135–144, Harbin, China. Chinese Information Processing Society of China.
Cite (Informal):
TiKEM:基于知识增强的藏文预训练语言模型(TiKEM: Knowledge Enhanced Tibetan Pre-trained Language Model) (Deng et al., CCL 2023)
Copy Citation:
PDF:
https://aclanthology.org/2023.ccl-1.12.pdf