Coreference resolution, as a fundamental task in natural language processing, plays a crucial role in various downstream applications, including machine translation, question answering, and text summarization. However, current mainstream research is predominantly focused on high-resource languages, while studies on coreference resolution for low-resource languages such as Tibetan remain limited. To address the shortage of relevant datasets and applied research in the field of Tibetan natural language processing, we constructed a Tibetan coreference resolution corpus, TiCoref-2025, covering both news and literary genres. Building upon this corpus, we proposed a neural network model, TiCoref, that integrates multiple modeling strategies. This model utilises BERT-base-Tibetan as its base encoder, incorporates BiLSTM to enhance long-distance dependency modeling, and introduces a Conditional Random Field (CRF) to improve the accuracy of mention boundary recognition. Additionally, to address the characteristics of literary texts, a Word-Knowledge Module (WKM) is specifically introduced to provide external prior knowledge, thereby improving the model's ability to capture complex coreference relations, and the effectiveness of this strategy in genre-specific scenarios has been validated. Experimental results show that the proposed method achieved F1 scores of 78.70%, 71.08%, and 68.22% on the official MUC, , metrics, respectively, validating the model's applicability and robustness on low-resource Tibetan texts.
共指消解(Coreference Resolution,CR)任务作为自然语言处理(Natural Language Processing,NLP)中的关键任务,在机器翻译、文本摘要、问答系统等多个下游任务中发挥着重要作用[1-4],其目标是识别文本中指向同一个实体的不同指称表述。例如,在句子“近年来,人工智能技术得到了广泛应用,许多企业将它视为提升竞争力的关键。”一句中,“人工智能技术”和“它”均为同一实体的不同表述,而共指消解任务即在于明确这些实体提及之间的指代关系[5,6]。
(2)模型方法:采用端到端神经共指消解模型架构,结合藏文预训练语言模型构建跨度表示。为增强长距离依存关系的建模能力,引入双向长短时记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)。同时集成条件随机场(Conditional Random Field, CRF)模块对提及边界进行序列标注优化,通过细化边界的判定,提升提及识别的整体准确率。
LiuY M, PengX Y, CaoJ N, et al. Bridging context gaps: leveraging coreference resolution for long contextual understanding[C]. Proceedings of the International Conference on Learning Representations, 2025.
[4]
WongB. A survey on the development and applications of coreference resolution[R]. University of California, Berkeley, 2023.
HobbsJ R. Resolving pronoun references[J]. Lingua, 1978, 44:311-338.
[8]
GroszB J, WeinsteinS, JoshiA K. Centering: A framework for modeling the local coherence of discourse[J]. Computational Linguistics, 1995, 21(2):203-225.
[9]
GeN, HaleJ, CharniakE. A statistical approach to anaphora resolution[C]. Proc Sixth Workshop on Very Large Corpora, 1998:161-170.
[10]
NgV, CardieC. Improving machine learning approaches to coreference resolution[C]. Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, 2002:104-111.
WisemanS, RushA M, ShieberS M. Learning global features for coreference resolution[C]. Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies(NAACL-HLT). San Diego, California: ACL, 2016:994-1004.
[13]
ParkC, ChoiK H, LeeC, et al. Korean coreference resolution with guided mention pair model using deep learning[J]. ETRI Journal, 2016, 38(6):1207-1217.
[14]
JoshiM, LevyO, WeldD S, et al. BERT for coreference resolution: Baselines and analysis[C]. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Hong Kong, China: Association for Computational Linguistics, 2019:5803-5808.
[15]
LeeK, HeL, ZettlemoyerL. Higher-order coreference resolution with coarse-to-fine inference[C]. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. New Orleans, USA: Association for Computational Linguistics, 2018:687-692.
[16]
KirstainY, RamO, LevyO. Coreference resolution without span representations[C]. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021:14-19.
[17]
HsuB, HorwoodG. Contrastive representation learning for cross-document coreference resolution of events and entities[C]. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2022:3644-3655.
[18]
WangY, JinH. Neural coreference resolution based on reinforcement learning[J]. arXiv preprint, 2022:2212.09028.
[19]
YaoY, LiZ, ZhaoH. Learning event-aware measures for event coreference resolution[C]. Findings of the Association for Computational Linguistics, 2023:13542-13556.
RahmanA, NgV. Coreference resolution with world knowledge[C]. Proceedings of the 49 th Annual Meeting of the Association for Computational Linguistics. Portland, Oregon, 2011:814-824.
HovyE, MarcusM, PalmerM, et al. OntoNotes: The 90% solution[C]. Proceedings of the Human Language Technology Conference of the ACL. New York, USA: Association for Computational Linguistics, 2006:57-60.
[29]
LinR. A survey on the state of the art in coreference resolution[R]. University of California, Berkeley, 2023.
LiuR, MaoR, LuuA T, et al. A brief survey on recent advances in coreference resolution[J]. Artificial Intelligence Review, 2023, 56:14439-14481.
[32]
DevlinJ, ChangM W, LeeK, 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.
[33]
LanZ, ChenM, GoodmanS, et al. ALBERT: A lite BERT for self-supervised learning of language representations[C]. Proceedings of the International Conference on Learning Representations (ICLR), 2020.
[34]
LiuS, DengJ, SunY, et al. Tibert: Tibetan pre-trained language model[C]. 2022 IEEE International Conference on Systems, Man, and Cybernetics, 2022:2956-2961.