计算机科学

计算机科学 ›› 2026, Vol. 53 ›› Issue (8): 266-275.doi: 10.11896/jsjkx.260300033

• 人工智能 • 上一篇    下一篇

从字节到语义:藏文命名实体识别的新范式

李庆凯1,2, 群诺1,2,3, 倪胜巧1,4, 杨进1,5   

  1. 1 西藏大学信息科学技术学院 拉萨 850000
    2 西藏信息化省部共建协同创新中心 拉萨 850000
    3 西藏大学藏语智能全国重点实验室 拉萨 850000
    4 四川大学机械工程学院 成都 610207
    5 四川大学网络空间安全学院 成都 610207
  • 收稿日期:2026-03-09 修回日期:2029-06-06 出版日期:2026-08-15 发布日期:2026-08-17
  • 通讯作者: 倪胜巧(nishengqiao@scu.edu.cn)
  • 作者简介:(leeqk@stu.utibet.edu.cn)
  • 基金资助:
    国家自然科学基金(62162057);西藏自治区自然科学基金重点项目(XZ202401ZR0040)

From Bytes to Semantics:New Paradigm for Tibetan Named Entity Recognition

LI Qingkai1,2, QUN Nuo1,2,3, NI Shengqiao1,4, YANG Jin1,5   

  1. 1 School of Information Science and Technology, Xizang University, Lhasa 850000, China
    2 Collaborative Innovation Center for Tibet informatization by MOE and Tibet Autonomous Region, Lhasa 850000, China
    3 State Key Laboratory of Tibetan Intelligence, Xizang University, Lhasa 850000, China
    4 School of Mechanical Engineering, Sichuan University, Chengdu 610207, China
    5 School of Cyber Science and Engineering, Sichuan University, Chengdu 610207, China
  • Received:2026-03-09 Revised:2029-06-06 Published:2026-08-15 Online:2026-08-17
  • About author:LI Qingkai,born in 2000,postgraduate.His main research interests include na-tural language processing and machine learning.
    NI Shengqiao,born in 1982,associate professor,Ph.D.His main research interests include machine learning and computer science professional education.
  • Supported by:
    National Natural Science Foundation of China(62162057) and Key Project of Natural Science Foundation of Tibet Autonomous Region(XZ202401ZR0040).

PDF (PC)

114

摘要： 针对藏文命名实体识别任务中标注数据稀缺、分词困难及长实体识别效果不佳等挑战,提出一种基于字节级全局建模与门控特征融合机制的命名实体识别框架。该框架利用Byte Latent Transformer模型直接对原始UTF-8字节序列建模,利用其基于下一字节熵的动态分块机制获得不依赖固定词表的全局上下文表示,以规避传统分词器破坏实体边界并增强对未登录词的鲁棒性;同时设计门控线性条件融合模块,将对齐后的全局上下文特征以可控强度、按序列位置自适应地动态注入到保留精确边界信息的局部特征中,从而形成互补表示并提升跨度建模能力。在此基础上,模型结合BiLSTM-CRF完成序列建模与标签解码,兼顾全局语义建模与局部边界判别能力。在TibetanAI_NER和TibNER两个数据集上的实验结果表明,模型F1值相比各自最优模型分别提升11.5个百分点和4.53个百分点;消融实验进一步验证了全局特征与门控融合机制的协同贡献。结果表明,该框架在分词受限及低资源场景下能稳定提升实体边界判别能力,尤其对长实体识别更为有效,为低资源语言命名实体识别提供了一种可行的建模思路。

关键词: 藏文命名实体识别, 字节级建模, 门控特征融合, 混合粒度策略, 边界感知

Abstract: Tibetan named entity recognition faces several practical challenges,including scarce annotated data,unreliable tokenization,and degraded performance on long entities.To address these issues,this paper proposes a NER framework based on byte-level global modeling and a gated feature fusion mechanism.Specifically,the framework employs the Byte Latent Transformer to directly model raw UTF-8 byte sequences,and leverages its entropy-based dynamic patching mechanism to obtain global contextual representations without relying on a fixed vocabulary.This effectively avoids entity boundary corruption introduced by conventional tokenizers and improves robustness to out-of-vocabulary patterns.Meanwhile,a gated linear conditional fusion module is designed to inject aligned global contextual features into local boundary-preserving representations with controllable strength in an adaptive,position-wise manner,thereby forming complementary features and enhancing span modeling capability.On this basis,the model combines BiLSTM and CRF to perform sequence modeling and label decoding,jointly preserving global semantic modeling and local boundary discrimination.Experimental results on the TibetanAI_NER and TibNER datasets show that the proposed model improves the F1 score by 11.5 and 4.53 percentage points,respectively,over the corresponding state-of-the-art baselines.Ablation studies further verify the synergistic contributions of the global features and the gated fusion mechanism.Overall,the proposed framework consistently strengthens entity boundary discrimination under segmentation-constrained and low-resource scenarios,with particularly clear gains on long entity recognition,providing a feasible modeling approach for NER in low-resource languages.

Key words: Tibetan named entity recognition, Byte-level modeling, Gated feature fusion, Hybrid granularity strategy, Boundary-aware

中图分类号: 

  • TP391
[1] MAYHEW S,BLEVINS T,LIU S,et al.Universal NER:Agold-standard multilingual named entity recognition benchmark[C]//Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies.2024:4322-4337.
[2] GAO F,HUANG C,LIU Y,et al.Tlue:A tibetan language understanding evaluation benchmark[C]//Proceedings of the 2025 Conference on Empirical Methods in Natural Language Proces-sing.2025:35059-35085.
[3] PATTNAYAK P,PATEL H,AGARWAL A.Tokenizationmatters:Improving zero-shotner for indic languages[C]//2025 IEEE International Conference on Electro Information Technology(eIT).IEEE,2025:456-462.
[4] LI D,DU S,LI P,et al.LE-NER:A Chinese NER Model Based on Lexical Enhancement[M]//Advanced Data Mining and Applications.Singapore:Springer,2025:344-359.
[5] PAGNONI A,PASUNURU R,RODRIGUEZ P,et al.Byte latent transformer:Patches scale better than tokens[C]//Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics.2025:9238-9258.
[6] JIAYANG J,LI Y C,ZONG C Q,et al.Tibetan Person Name Recognition Based on Fusion of Maximum Entropy and Conditional Random Field Models[J].Journal of Chinese Information Processing,2014,28(1):107-112.
[7] YUE J.Part-of-speech tagging of corpus based onBiLSTM-CRF model[C]//Proceedings of the 2025 4th International Confe-rence on Distributed Computing and Electrical Circuits and Electronics(ICDCECE).IEEE,2025:1-5.
[8] ZHANG J,ZHANG Z,YESHI L,et al.Tibetan medical named entity recognition based on syllable-word-sentence embedding transformer[J].CAAI Transactions on Intelligence Technology,2025,10(4):1148-1158.
[9] HU S,GUAN J,LI W.Chinese named entity recognition basedon multi-level information extraction[C]//Proceedings of the 2022 3rd International Conference on Electronic Communication and Artificial Intelligence(IWECAI).IEEE,2022:554-557.
[10] DEVLIN J,CHANG M W,LEE K,et al.Bert:Pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies.2019:4171-4186.
[11] CONNEAU A,KHANDELWAL K,GOYAL N,et al.Unsupervised cross-lingual representation learning at scale[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics.2020:8440-8451.
[12] POLITOV A,SHKALIKOV O,JÄKEL R,et al.Revisiting Projection-based Data Transfer for Cross-Lingual Named Entity Recognition in Low-Resource Languages[C]//Proceedings of the Joint 25th Nordic Conference on Computational Linguistics and 11th Baltic Conference on HumanLanguage Technologies(NoDaLiDa/Baltic-HLT 2025).2025:499-507.
[13] ZHU G,XIAO R,WANG H,et al.Large margin representation learning for robust cross-lingual named entity recognition[C]//Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics.2025:4270-4291.
[14] AFRICA D D,SALHAN S,WEISS Y,et al.Meta-Pretrainingfor Zero-Shot Cross-Lingual Named Entity Recognition in Low-Resource Philippine Languages[C]//Proceedings of the 5th Workshop on Multilingual Representation Learning(MRL 2025).2025:106-127.
[15] RADCHENKO V,DRUSHCHAK N.Improving Named Entity Recognition for Low-Resource Languages Using Large Language Models:A Ukrainian Case Study[C]//Proceedings of the Fourth Ukrainian Natural Language Processing Workshop(UNLP 2025).2025:27-35.
[16] ZHANG Z,LEE S Y M,ZHANG D,et al.Zero-shot Cross-lingual NER via Mitigating Language Difference:An Entity-aligned Translation Perspective[C]//Findings of the Association for Computational Linguistics:EMNLP 2025.2025:4541-4557.
[17] CLARK J H,GARRETTE D,TURC I,et al.Canine:Pre-training an efficient tokenization-free encoder for language representation[J].Transactions of the Association for Computational Linguistics,2022,10:73-91.
[18] XUE L,BARUA A,CONSTANT N,et al.ByT5:Towards a token-free future with pre-trained byte-to-byte models[J].Trans-actions of the Association for Computational Linguistics,2022,10:291-306.
[19] LI S,XU J,ZHANG M.EMBYTE:Decomposition and Com-pression Learning for Small yet Private NLP[C]//Findings of the Association for Computational Linguistics:EMNLP 2025.2025:7182-7201.
[20] ERONEN J,PTASZYNSKI M,MASUI F.Zero-shot cross-lingual transfer language selection using linguistic similarity[J].Information Processing & Management,2023,60(3):103250.
[21] WANG P,CHEN Y,LIU C,et al.An attention-enhanced LSTM model for Chinese educational named entity recognition[C]//Proceedings of the 2024 5th International Conference on Information Science and Education(ICISE-IE).IEEE,2024:272-277.
[22] RATCHATORN T,TANAKA M.Adaptive Adversarial Cross-Entropy Loss for Sharpness-Aware Minimization[C]//2024 IEEE International Conference on Image Processing(ICIP).IEEE,2024:479-485.
[23] LIU J,SUN M,ZHANG W,et al.DAE-NER:dual-channel attention enhancement for Chinese named entity recognition[J].Computer Speech & Language,2024,85:101581.
[24] VARSHNEY N,CHATTERJEE A,PARMAR M,et al.Investigating acceleration ofLLaMA inference by enabling interme-diate layer decoding via instruction tuning with ‘LITE’[C]//Findings of the Association for Computational Linguistics:NAACL 2024.2024:3656-3677.
[25] LI J,FANG A,SMYRNIS G,et al.Datacomp-lm:In search ofthe next generation of training sets for language models[J].Advances in Neural Information Processing Systems,2024,37:14200-14282.
[26] YU T,ZHANG Y,YONG C.Tibetan named entity recognition based on few-shot learning[J].Computer and Modernization,2023,39(5):13-19.
[27] ZHOUM K,OUJIAN C R,DAOJI C D,et al.TibNER:a Tibetan named entity recognition dataset[J].China Scientific Data(Chinese & English Online Edition),2024,9(4):17-27.
[28] LAMAOJIE,WANMAC D,YONGCUO,et al.Tibetan medical named entity recognition via adversarial training and iterative dilated convolution[J].Plateau Science Research,2025,9(1):105-118.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市两江新区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发