SpanBERT: Improving Pre-training by Representing and Predicting Spans
Open accessThe DOI matches its Crossref record; 53 of the 53 references it cites were found in a registry or an index.
Abstract
We present SpanBERT, a pre-training method that is designed to better represent and predict spans of text. Our approach extends BERT by (1) masking contiguous random spans, rather than random tokens, and (2) training the span boundary representations to predict the entire content of the masked span, without relying on the individual token representations within it. SpanBERT consistently outperforms BERT and our better-tuned baselines, with substantial gains on span selection tasks such as question answering and coreference resolution. In particular, with the same training data and model size as BERT large , our single model obtains 94.6% and 88.7% F1 on SQuAD 1.1 and 2.0 respectively. We also achieve a new state of the art on the OntoNotes coreference resolution task (79.6% F1), strong performance on the TACRED relation extraction benchmark, and even gains on GLUE. 1
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“XLM [46] performs MLM on a concatenation of parallel bilingual sentence pairs, called Translation Language Modeling (TLM).”
From Pre-trained models for natural language processing: A survey (2020) · cites this paper as [46]
Experimental — accuracy not yet measuredIn the citing paper
“Conneau and Lample [46] tried to initialize the entire encoder and decoder by a multilingual pre-trained BERT model and showed a significant improvement could be achieved on unsupervised MT and English-Romanian supervised MT.”
From Pre-trained models for natural language processing: A survey (2020) · cites this paper as [46]
In the citing paper
“In the experiment, this approach can surpass the BERT-style pretraining proposed by Conneau and Lample [46] both on unsupervised MT and English-Romanian supervised MT.”
From Pre-trained models for natural language processing: A survey (2020) · cites this paper as [46]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019) · arXiv · doi:10.48550/arxiv.1907.11692 · Cited by 16,515
Self-citationDecoupled Weight Decay Regularization
Loshchilov, I., & Hutter, F. (2017) · arXiv · doi:10.48550/arxiv.1711.05101 · Cited by 12,522
Deep Contextualized Word Representations
Peters, M., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., & Zettlemoyer, L. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1202 · Cited by 11,983
Self-citationRecursive Deep Models for Semantic Compositionality Over a Sentiment Treebank
Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C. D., Ng, A., & Potts, C. (2013) · Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d13-1170 · Cited by 6,878 · cited 4×
“…Understanding Evaluation (GLUE) benchmark (Wang et al., 2019) consists of 9 sentence-level classification tasks: • Two sentence-level classification tasks including CoLA (Warstadt et al., 2018) for evaluating linguistic acceptability and SST-2 (Socher et al., 2013) for sentiment classification.”
BackgroundInfluentialXLNet: Generalized Autoregressive Pretraining for Language Understanding
Yang, Z., Dai, Z., Yang, Y., Carbonell, J., Salakhutdinov, R., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1906.08237 · Cited by 6,687 · cited 4×
“While others show the benefits of adding more data (Yang et al., 2019) and increasing model size (Lample and Conneau, 2019), this work demonstrates the importance of designing good pretraining tasks and objectives, which can also have a significant impact.”
BackgroundMethodInfluentialSQuAD: 100,000+ Questions for Machine Comprehension of Text
Rajpurkar, P., Zhang, J., Lopyrev, K., & Liang, P. (2016) · Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d16-1264 · Cited by 6,208
Universal Language Model Fine-tuning for Text Classification
Howard, J., & Ruder, S. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1031 · Cited by 4,100 · cited 2×
“, 2018) have had immense impact on NLP lately, particularly as methods for initializing a large model before fine-tuning it for a specific task (Howard and Ruder, 2018; Radford et al., 2018; Devlin et al., 2019).”
Method
Research works printed but not linked: 7
- Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. 2016. Layer normalization. arXiv preprint arXiv:1607.06450 . · arXiv
- Andrew M Dai and Quoc V Le. 2015. Semisupervised sequence learning. In Advances in Neural Information Processing Systems (NIPS) , pages 3079-3087.
- Adam Fisch, Alon Talmor, Robin Jia, Minjoon Seo, Eunsol Choi, and Danqi Chen. 2019. MRQA 2019 shared task: Evaluating generalization in reading comprehension. In Proceedings of 2nd Machine Reading for Reading Comprehension (MRQA) Workshop at EMNLP .
- Dan Hendrycks and Kevin Gimpel. 2016. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415 . · arXiv
- Mandar Joshi, Eunsol Choi, Omer Levy, Daniel Weld, and Luke Zettlemoyer. 2019a. pair2vec: Compositional word-pair embeddings for crosssentence inference. In North American Association for Computational Linguistics (NAACL) , pages 3597-3608.
- Liu Yinhan, 2019, arxiv preprint arXiv:1907.11692 · arXiv
- Mitchell Stern, Noam Shazeer, and Jakob Uszkoreit. 2018. Blockwise parallel decoding for deep autoregressive models. In Advances in Neural Information Processing Systems (NIPS) .
Cited by
Items 901–1,000 of 1,012
PALT: Parameter-Lite Transfer of Language Models for Knowledge Graph Completion
Shen, J., Wang, C., Yuan, Y., Han, J., Ji, H., Sen, K., Zhang, M., & Song, D. X. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.281 · Cited by 8
Can Anaphora Resolution Improve Extractive Query-Focused Multi-Document Summarization?
Lamsiyah, S., Mahdaouy, A. E., & Schommer, C. (2023) · IEEE Access · doi:10.1109/access.2023.3314524 · Cited by 6
Domain Knowledge-Driven Relation Extraction Methods
Chen, B., & Yuan, G. (2024) · Journal of Data Science and Intelligent Systems · doi:10.47852/bonviewjdsis42022524 · Cited by 7
"The Data Says Otherwise" — Towards Automated Fact-checking and Communication of Data Claims
Fu, Y., Guo, S., Hoffswell, J., Bursztyn, V. S., Rossi, R. A., & Stasko, J. (2024) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3654777.3676359 · Cited by 16
“Third, although we utilize GPT to perform the pre-processing steps, the downstream tasks (e.g., coreference, ellipsis resolution) have been extensively explored in NLP research, with pre-trained statistical models showing increasing capabilities [8, 44].”
CMAAN: Cross-Modal Aggregation Attention Network for Next POI Recommendation
Zhuang, Z., Liu, L., Qi, H., Shen, Y., & Yin, B. (2024) · IEEE Transactions on Computational Social Systems · doi:10.1109/tcss.2024.3513947 · Cited by 6
“Given a trajectory, we also use n-gram masking [39] to generate masked records.”
A discrete convolutional network for entity relation extraction
Yang, W., Qin, Y., Wang, K., Hu, Y., Huang, R., & Chen, Y. (2025) · Neural Networks · doi:10.1016/j.neunet.2024.107117 · Cited by 8
Bhardwaj, S., Alfonso-Hermelo, D., Langlais, P., Bernier-Colborne, G., Goutte, C., & Simard, M. (2020) · International Conference on Computational Linguistics · doi:10.18653/v1/2020.coling-main.576 · Cited by 12
“(Kudo and Richardson, 2018) and performs whole word masking, which has been shown to be preferable (Joshi et al., 2019).”
BackgroundOntoGUM: Evaluating Contextualized SOTA Coreference Resolution on 12 More Genres
Zhu, Y., Pradhan, S., & Zeldes, A. (2021) · doi:10.18653/v1/2021.acl-short.59 · Cited by 28
Towards Semantics-Enhanced Pre-Training: Can Lexicon Definitions Help Learning Sentence Meanings?
Ren, X., Sun, X., Wang, H., & Liu, Q. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i15.17619 · Cited by 6
“…use a masked language model objective and considerable efforts have been devoted to discovering better sequence corrupting scheme (Sun et al. 2019; Joshi et al. 2020; Wang et al. 2020; Clark et al. 2020), applying auto-regressive objectives (Song et al. 2019; Dong et al. 2019; Lewis et al. 2020;…”
MethodSAS: Self-Augmentation Strategy for Language Model Pre-training
Xu, Y., Zhang, J., Ru, H., Ge, L., Yang, C., Yang, C. H., & Wu, Y. N. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i10.21412 · Cited by 8
“…tokens boosts BERT’s performance, while pairing sentences and undertaking a corresponding next sentence prediction (NSP) task does not; the SpanBERT (Joshi et al. 2019) masks spans of tokens in each input sequence and then uses two boundary tokens of each masked span to predict each token in the…”
BackgroundLiu, T., Jiang, Y., Cotterell, R., & Sachan, M. (2022) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2022.naacl-main.189 · Cited by 10
“We closely follow the hyperparameter settings of Joshi et al. (2020) and build our models upon the codebase of Xu and Choi (2020) 9 under Apache License 2.0.”
BackgroundMethodInfluentialMaterials genome engineering-based hydrogen storage materialsdatabase and its applications
Huang, P., Cai, D., Lin, H., Liu, J., Li, Z., Li, B., Zou, Y., Chu, H., Sun, L., & Xu, F. (2022) · Scientia Sinica Chimica · doi:10.1360/ssc-2022-0157 · Cited by 6
LogiTorch: A PyTorch-based library for logical reasoning on natural language
Helwe, C., Clavel, C., & Suchanek, F. M. (2022) · Conference on Empirical Methods in Natural Language Processing: System Demonstrations (EMNLP) · doi:10.18653/v1/2022.emnlp-demos.25 · Cited by 16
“In LogiTorch, we implemented a wrapper over a finetuned SpanBERT (Joshi et al., 2020) for coreference resolution.”
MethodA Joint Domain-Specific Pre-Training Method Based on Data Enhancement
Gan, Y., Lu, G., Su, Z., Lei, W., Zhou, J., Jiang, J., & Chen, D. (2023) · Applied Sciences · doi:10.3390/app13074115 · Cited by 7
Self-citationText Representation for Sentiment Analysis: From Static to Dynamic
Gavali, P. M., & Shiragave, S. K. (2023) · International Conference on Smart Data Intelligence (ICSMDI) · doi:10.1109/icsmdi57622.2023.00025 · Cited by 6
“SpanBERT [15] enhances the span prediction outcomes by masking multiple continuous words instead of single word masking.”
BackgroundResearch on Relation Classification Tasks Based on Cybersecurity Text
Shi, Z., Li, H., Zhao, D., & Pan, C. (2023) · Mathematics · doi:10.3390/math11122598 · Cited by 6
Luo, Z., Yan, S., & Luo, S. (2023) · Mathematics · doi:10.3390/math11122733 · Cited by 7
Multitask Pretraining with Structured Knowledge for Text-to-SQL Generation
Giaquinto, R. A., Zhang, D., Kleiner, B., Li, Y., Tan, M., Bhatia, P., Nallapati, R. M., & Ma, X. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.620 · Cited by 10
Knowledgeable Parameter Efficient Tuning Network for Commonsense Question Answering
Zhao, Z., Hu, L., Zhao, H., Shao, Y., & Wang, Y. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.503 · Cited by 6
“For the query-related knowledge piece k Q i , we mask 15% tokens in total at the span level and predict them in the same way as SpanBERT (Joshi et al., 2020).”
MethodAsk-and-Verify: Span Candidate Generation and Verification for Attribute Value Extraction
Ding, Y., Liang, Y., Zalmout, N., Li, X., Grant, C., & Weninger, T. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-industry.9 · Cited by 6
“Recently, NLP researchers have developed span-based models (Shen et al., 2021; Joshi et al., 2020; Yamada et al., 2020; Li et al., 2020) to obtain state-of-the-art performance on span-based or entity-centered tasks like named entity recognition (Ding et al., 2021; Huang et al., 2015), entity…”
BackgroundXDoc: Unified Pre-training for Cross-Format Document Understanding
Chen, J., Lv, T., Cui, L., Zhang, C., & Wei, F. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.71 · Cited by 14
AI, Human–Robot Interaction, and Natural Language Processing
McLoughlin, I. V., & Indurkhya, N. (2023) · Cambridge University Press eBooks · doi:10.1017/9781108989275.021 · Cited by 8
Kang, T., Li, S., Tang, J., Li, D., Wang, P., & Wang, T. (2024) · Data Mining and Knowledge Discovery · doi:10.1007/s10618-023-00998-6 · Cited by 7
“…computed through matrix multiplication to directly obtain final cosine similarity score, which is calculated as, Finally, we predict corresponding 𝐂 𝐭 with maximum probability for each ( 𝐡 , 𝐫 ) , For contextualized loss, we apply the InfoNCE loss (Chen et al. 2020) with an additive margin.”
Singh, G. V., Firdaus, M., Chauhan, D. S., Ekbal, A., & Bhattacharyya, P. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i17.29864 · Cited by 13
Korenčić, D., Chulvi, B., Casals, X. B., Toselli, A. H., Taulé, M., & Rosso, P. (2024) · Expert Systems · doi:10.1111/exsy.13671 · Cited by 17
“Additionally, we experimented with the SpanBERT model (Joshi et al., 2020) as the transformer backbone.”
PatentTransformer: A Framework for Personalized Patent Claim Generation.
Lee, J.-S. (2019) · Cited by 6
A SURVEY ON SARCASM DETECTION APPROACHES
aboobaker, J., & Dr., E. I. (2020) · Indian Journal of Computer Science and Engineering · doi:10.21817/indjcse/2020/v11i6/201106048 · Cited by 6
SCRIPT: Self-Critic PreTraining of Transformers
Nijkamp, E., Pang, B., Wu, Y. N., & Xiong, C. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.409 · Cited by 7
“…is dominated by masked language modeling (MLM) for bi-directional encoders, such as BERT (Devlin et al., 2018; Yang et al., 2019; Liu et al., 2019; Joshi et al., 2020; Lan et al., 2019; Lewis et al., 2020; Jiao et al., 2019), and causal masking for uni-directional auto-regressive decoders…”
BackgroundFrom What to Why: Improving Relation Extraction with Rationale Graph
Zhang, Z., Yu, B., Shu, X., Xue, M., Liu, T., & Guo, L. (2021) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2021.findings-acl.8 · Cited by 13
“Typically, existing BERT-based RE solutions first concatenate target entities with the text or mark them in the input sequence with special tokens, and then directly take h cls as the input of final classification module (Joshi et al., 2020; Yu et al., 2020).”
MethodWVOQ at SemEval-2021 Task 6: BART for Span Detection and Classification
Roele, C. (2021) · International Workshop on Semantic Evaluation (SemEval) · doi:10.18653/v1/2021.semeval-1.32 · Cited by 9
How May I Help You? Using Neural Text Simplification to Improve Downstream NLP Tasks
Van, H., Tang, Z. S., & Surdeanu, M. (2021) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2021.findings-emnlp.343 · Cited by 17
Deep learning for detecting logic-flaw-exploiting network attacks: An end-to-end approach
Zou, Q., Singhal, A., Sun, X., & Liu, P. (2021) · Journal of Computer Security · doi:10.3233/jcs-210101 · Cited by 8
“Recently, some pre-trained LLMs (e.g., BERT [11], RoBERTa [27], SpanBERT [22], etc.) have been released and proven to be effective not only in natural language processing [5] (the proposed usage), but also in other tasks like object detection [8].”
Understanding Mention Detector-Linker Interaction in Neural Coreference Resolution
Wu, Z., & Gardner, M. (2021) · Workshop on Computational Models of Reference, Anaphora and Coreference (CRAC) · doi:10.18653/v1/2021.crac-1.16 · Cited by 10
Incorporating Constituent Syntax for Coreference Resolution
Jiang, F., & Cohn, T. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i10.21329 · Cited by 6
Enhancing Targeted Minority Class Prediction in Sentence-Level Relation Extraction
Baek, H.-R., & Choi, Y. S. (2022) · Sensors · doi:10.3390/s22134911 · Cited by 10
“We compared our model with the following models: (1) C-GCN [3], (2) LUKE [5], (3) SpanBERT [42], (4) KnowBERT [4], (5) RoBERTa-large [43], and (6) RE-marker [32].”
ResultMulti-Faceted Knowledge-Driven Pre-Training for Product Representation Learning
Zhang, D., Liu, Y., Yuan, Z., Fu, Y., Chen, H., & Xiong, H. (2022) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2022.3200921 · Cited by 6
“Following PLMs, many endeavors (e.g., SpanBERT [10], ERNIE [27], etc.) have been made for further optimization.”
MethodSelf-citationMeasuring design compliance using neural language models: an automotive case study
Parthasarathy, D., Ekelin, C., Karri, A. P., Sun, J., & Moraitis, P. (2022) · doi:10.1145/3558489.3559067 · Cited by 7
“One option is to modify the masking recipe like [29], which masks selected phrases and [16], which masks larger spans of tokens.”
BackgroundScaboro, S., Portelli, B., Chersoni, E., Santus, E., & Serra, G. (2022) · Experimental Biology and Medicine · doi:10.1177/15353702221128577 · Cited by 12
Research on Chinese News Text Classification Based on ERNIE Model
Zhang, W. (2023) · Smart innovation, systems and technologies · doi:10.1007/978-981-19-7184-6_8 · Cited by 6
Revisiting text decomposition methods for NLI-based factuality scoring of summaries
Glover, J., Fancellu, F., Jagannathan, V., Gormley, M. R., & Schaaf, T. (2022) · Workshop on Generation, Evaluation and Metrics (GEM) · doi:10.18653/v1/2022.gem-1.7 · Cited by 17
Is Attention Always Needed? A Case Study on Language Identification from Speech
Mandal, A., Pal, S., Dutta, I., Bhattacharya, M., & Naskar, S. K. (2022) · Natural language processing. · doi:10.1017/nlp.2024.22 · Cited by 7
“, 2019), SpanBERT (Joshi et al., 2020), XLNet (Yang et al.”
Transfer learning and subword sampling for asymmetric-resource one-to-many neural translation
Grönroos, S.-A., Virpioja, S., & Kurimo, M. (2020) · Machine Translation · doi:10.1007/s10590-020-09253-x · Cited by 10
Joint Coreference Resolution and Character Linking for Multiparty Conversation
Bai, J., Zhang, H., Song, Y., & Xu, K. (2021) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2021.eacl-main.43 · Cited by 7
Khosla, S., Fiacco, J., & Rosé, C. P. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.130 · Cited by 10
CDˆ2CR: Co-reference resolution across documents and domains
Ravenscroft, J., Clare, A. J., Cattan, A., Dagan, I., & Liakata, M. (2021) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2021.eacl-main.21 · Cited by 16
LICHEE: Improving Language Model Pre-training with Multi-grained Tokenization
Guo, W., Zhao, M. M., Zhang, L., Niu, D., Luo, J., Liu, Z., Li, Z., & Tang, J. (2021) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2021.findings-acl.119 · Cited by 6
“To obtain a more accurate model, prior studies attempt to incorporate coarse-grained information into models trained with fine-grained tokenization by masking sequences of consecutive tokens in the pre-training stage (Joshi et al., 2020; Cui et al., 2019).”
InfluentialNEREL: A Russian Dataset with Nested Named Entities, Relations and Events
Loukachevitch, N., Artemova, E., Batura, T. V., Braslavski, P., Denisov, I., Ivanov, V. V., Manandhar, S., Pugachev, A., & Tutubalina, E. V. (2021) · Electronic scientific archive of UrFU (Ural Federal University) · doi:10.48550/arxiv.2108.13112 · Cited by 7
Reusing Monolingual Pre-Trained Models by Cross-Connecting Seq2seq Models for Machine Translation
Oh, J., & Choi, Y. S. (2021) · Applied Sciences · doi:10.3390/app11188737 · Cited by 9
Zhang, H., Zhao, X., & Song, Y. (2021) · Workshop on Computational Models of Reference, Anaphora and Coreference (CRAC) · doi:10.18653/v1/2021.crac-1.1 · Cited by 12
Open-domain conversational search assistants: the Transformer is all you need
Ferreira, R., Leite, M., Semedo, D., & Magalhães, J. (2022) · Information Retrieval · doi:10.1007/s10791-022-09403-0 · Cited by 9
“In Joshi et al. (2020), a BERT model (Devlin et al. 2019) is used for the same task, with an optimization objective based on corrupting spans of text, during training and fine-tuning, greatly increasing performance over (Lee et al. 2017).”
MethodGraphCache: Message Passing as Caching for Sentence-Level Relation Extraction
Wang, Y., Chen, M., Zhou, W., Cai, Y., Liang, Y., & Hooi, B. (2022) · Findings of the Association for Computational Linguistics: NAACL · doi:10.18653/v1/2022.findings-naacl.128 · Cited by 11
Chanaa, H., Nfaoui, E. H., & Loqman, C. (2023) · Lecture notes in networks and systems · doi:10.1007/978-3-031-29857-8_95 · Cited by 7
Experimental Standards for Deep Learning in Natural Language Processing Research
Ulmer, D., Bassignana, E., Müller-Eberstein, M., Varab, D., Zhang, M., van der Goot, R., Hardmeier, C., & Plank, B. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.196 · Cited by 25
“Other ways of distributing models is setting hyperlinks in the repository (e.g., Joshi et al., 2020), to load the models from the checkpoints they have been saved to.”
MethodImproving BERT Fine-tuning with Embedding NormalizationPreprint
Zhou, W., Du, J., & Ren, X. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.03918 · Cited by 9
“Transformer-based sentence encoders including BERT (Devlin et al., 2018) and its variants (Joshi et al., 2019; Liu et al., 2019) have achieved tremendous success on a wide range of natural language processing tasks.”
BackgroundLiu, L., Song, Z., Zheng, X., & He, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.04639 · Cited by 9
“Joshi et al. (2019a) also propose SpanBERT to better represent and predict spans of text and achieve a new state-of-the-art result (Avg.”
BackgroundMethodHybrid Generative-Contrastive Representation LearningPreprint
Kim, S., Kim, S., & Lee, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.06162 · Cited by 13
“Learning representations without human annotation has recently achieved remarkable progress, especially in natural language processing and computer vision [11, 29, 37, 25, 27, 9, 44, 43, 2, 34, 48, 22, 41, 19, 7, 6, 3, 17, 47, 4, 12].”
Linguistically Informed Masking for Representation Learning in the Patent DomainPreprint
Althammer, S., Buckley, M., Hofstätter, S., & Hanbury, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.05768 · Cited by 10
“The extensive and varied further research results based on the BERTmodel [15, 19, 25, 40] show the generality and flexibility of the representations.”
BackgroundSelf-supervised Contrastive Cross-Modality Representation Learning for Spoken Question AnsweringPreprint
You, C., Chen, N., & Zou, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.03381 · Cited by 17
“Although, recent work (Liu et al., 2019; Lan et al., 2020; Devlin et al., 2018; Joshi et al., 2020) found that utilizing this auxiliary task can improve model performance, it only focuses on learning syntactic and semantic representations of the word in token-level.”
BackgroundExploring The Landscape of Distributional Robustness for Question Answering Models
Awadalla, A., Wortsman, M., Ilharco, G., Min, S., Magnusson, I., Hajishirzi, H., & Schmidt, L. (2022) · arXiv · doi:10.48550/arxiv.2210.12517 · Cited by 6
Hu, Y., Wang, H., Chen, J., Zhang, C., Ren, M., Dong, M., Zhang, H., Hu, Z., & Dong, X. T. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.129291 · Cited by 8
Generative Artificial Intelligence Models: A Survey
Almuammar, M., Alsultan, K., & Altukhaim, Y. (2026) · Artificial Intelligence Review · doi:10.1007/s10462-026-11546-1 · Cited by 9
“There are wide range of applications of GANs including image editing; image translations; text translations; audio enhancement; improving cyberse-curity by detecting security anomalies, assisting in password cracking, and producing more (Jiao et al. 2020 ) Compressed, fast inference Depends on teacher model, weaker in…”
ContextAD: Context‐Aware Acronym Disambiguation with Siamese BERT Network
Ou, L., Yao, Y., Luo, X., Li, X., & Chen, K. (2023) · International Journal of Intelligent Systems · doi:10.1155/2023/5014355 · Cited by 6
“Singh and Kumar used the SpanBERT [13] model to transform the acronym disambiguation problem into a span prediction problem [14].”
MethodCharacter Coreference Resolution in Movie Screenplays
Baruah, S., & Narayanan, S. S. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.654 · Cited by 15
TACR: A Table Alignment-based Cell Selection Method for HybridQA
Wu, J., Xu, Y., Gao, Y., Lou, J., Karlsson, B. F., & Okumura, M. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.409 · Cited by 8
Yu, J., Paun, S., Camilleri, M., Garcia, P., Chamberlain, J., Kruschwitz, U., & Poesio, M. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.54 · Cited by 10
EIBC: a deep learning framework for Chinese toponym recognition with multiple layers
Zhao, Y., Zhang, D., Jiang, L., Liu, Q., Liu, Y., & Liao, Z. (2024) · Journal of Geographical Systems · doi:10.1007/s10109-024-00441-4 · Cited by 5
“Some researches (Cui et al. 2021; Joshi et al. 2020) focus on extending the objective of BERT’s MLM from masking individual tokens to contiguous sequences of n tokens(n-grams), but the effectiveness and reliability of the continuously masking strategies are lower because of the prediction of tokens…”
Self-citationXiao, Q., Ren, L., Yang, J., Chen, Y., Jiang, S., & Wang, D. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.124475 · Cited by 5
Evaluating the Effectiveness of Fine-Tuning Large Language Model for Domain-Specific Task
Dabhi, S., Martínez, J., & Poursardar, F. (2024) · IEEE International Conference on Information Reuse and Integration (IRI) · doi:10.1109/iri62200.2024.00045 · Cited by 5
“Similarly, Joshi [3] introduced SpanBERT, which enhances the modeling of span-level representations during pre-training, leading to improved performance on span-based tasks like SQuAD.”
MethodPeng, Q., Luo, X., Yuan, Y., Gu, F., Shen, H., & Huang, Z. (2024) · International Journal of Web Information Systems · doi:10.1108/ijwis-09-2024-0277 · Cited by 7
A prompt tuning method based on relation graphs for few-shot relation extraction
Zhang, Z., Yang, Y., & Chen, B. (2025) · Neural Networks · doi:10.1016/j.neunet.2025.107214 · Cited by 8
Li, Y., Zhang, M., Zhang, Z., & Wang, Y. (2024) · ACM/IEEE Joint Conference on Digital Libraries · doi:10.1145/3677389.3702584 · Cited by 6
Wang, J., Jatowt, A., & Cai, Y. (2025) · ACM Transactions on the Web · doi:10.1145/3723352 · Cited by 12
“Further innovations include XLM [44], which incorporates translation language modeling (TLM) for cross-lingual training, XLNet [33] with its permuted language modeling (PLM), and SpanBERT [33], which extends BERT by masking contiguous spans and employing a span boundary objective.”
Adaptive Attention and Feature Embedding for Enhanced Entity Extraction Using an Improved Bert Model
Wu, L., Gao, J., Liao, X., Zheng, H., Hu, J., & Bao, R. (2024) · International Conference on Communication Technology and Information Technology (ICCTIT) · doi:10.1109/icctit64404.2024.10928668 · Cited by 7
Bringing Two Worlds Together: The Convergence of Large Language Models and Knowledge Graphs
Cai, L., Kang, Y., Yu, C., Fu, Y., Zhang, H., & Zhao, Y. (2024) · International Conference on Automation, Robotics and Computer Engineering (ICARCE) · doi:10.1109/icarce63054.2024.00046 · Cited by 9
“SpanBERT [8] improves relation extraction with spanbased pretraining.”
ForestryBERT: A pre-trained language model with continual learning adapted to changing forestry text
Tan, J., Zhang, H., Yang, J., Liu, Y., Zheng, D., & Liu, X. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.113706 · Cited by 7
Self-citationFernando, A., & Ranathunga, S. (2025) · Knowledge and Information Systems · doi:10.1007/s10115-025-02520-4 · Cited by 6
“Thefollow-upworkmasked consecutivesub-wordsintextspans[12]andcorrelatedtextspanswithPoint-wiseMutual Information(PMI)masking[13].”
Patel, R., Patel, S. A., James, V., Singh, Y. R., Shah, V., Thakar, V., & Prajapati, B. G. (2025) · Bioprinting · doi:10.1016/j.bprint.2025.e00437 · Cited by 5
A Study on Entity Resolution for Email Conversations
Dakle, P. P., Desai, T., & MOLDOVAN, D. I. (2020) · Language Resources and Evaluation · Cited by 5
Reasoning Requirements for Indirect Speech Act Interpretation
Sarathy, V., Tsuetaki, A., Roque, A., & Scheutz, M. J. (2020) · International Conference on Computational Linguistics · doi:10.18653/v1/2020.coling-main.433 · Cited by 5
“The first insight is that word meanings (in the case of BERT) and span meanings (in the case of SpanBERT (Joshi et al., 2020)) cannot be understood separate from their context, which in these cases is primarily linguistic context.”
BackgroundConstructing a cross-document event coreference corpus for Dutch
Langhe, L. D., Clercq, O. D., & Hoste, V. (2022) · Language Resources and Evaluation · doi:10.1007/s10579-022-09597-1 · Cited by 5
“…in coreference research has shifted from traditional machine learning approaches such as maximum entropy models (Ahn, 2006) and support vector machines (Chen and Ng, 2014) to deep learning architectures (Nguyen et al., 2016) and span-based embedding models (Lee et al., (2018; Joshi et al., 2020) .”
CustRE: a rule based system for family relations extraction from english text
Mumtaz, R., & Qadir, M. A. (2022) · Knowledge and Information Systems · doi:10.1007/s10115-022-01687-4 · Cited by 5
“Joshi et al. present SpanBERT [14], a modification of BERT that represents and predicts spans of tokens, instead of individual tokens.”
BackgroundMethodResultInfluentialA Multi-tasking and Multi-stage Chinese Minority Pre-trained Language Model
Li, B., Weng, Y., Sun, B., & Li, S. (2022) · Communications in computer and information science · doi:10.1007/978-981-19-7960-6_10 · Cited by 9
“We used a 15% mask rate and a maximum of 3-grams for span masking [42].”
MethodEnhancing Semantic Relation Classification With Shortest Dependency Path Reasoning
Li, J., Shuang, K., Guo, J., Shi, Z., & Wang, H. (2023) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2023.3265205 · Cited by 15
“[28] replace the token masking with span masking Fig.”
BackgroundChen, Y., Li, X., Li, A., Li, Y., Yang, X., Lin, Z., Yu, S., & Tang, X. (2023) · JMIR Formative Research · doi:10.2196/47434 · Cited by 8
“For example, Mandar et al [31] proposed a pretrained model, SpanBERT, which was experimentally demonstrated to generally outperform the BERT model in terms of relation extraction and coreference resolution.”
MethodZhou, X., Fu, Q., Chen, J., Liu, L., Wang, Y., Lu, Y., & Wu, H. (2023) · Journal of Biomedical Informatics · doi:10.1016/j.jbi.2023.104445 · Cited by 7
From task to evaluation: an automatic text summarization review
Lu, L., Liu, Y., Xu, W., Li, H., & Sun, G. (2023) · Artificial Intelligence Review · doi:10.1007/s10462-023-10582-5 · Cited by 10
“BART Lewis et al. (2019) tested five different document corruption methods then filter out the most effective one which inspired by Span-BERT (Joshi et al. 2020).”
MethodSelf-citationResearch on Chinese short text classification based on ERNIE-TEXTCNN model
Wang, M., & Xu, J. (2023) · International Conference on Advanced Algorithms and Neural Networks (AANN) · doi:10.1117/12.3005031 · Cited by 5
A Survey of Cross-Lingual Text Classification and Its Applications on Fake News Detection
Lan, L., Huang, T., Li, Y., & Song, Y. (2023) · World Scientific Annual Review of Artificial Intelligence · doi:10.1142/s2811032323500030 · Cited by 8
Sun, S., Jun, D., & Qin, X. (2023) · Aslib Journal of Information Management · doi:10.1108/ajim-05-2023-0180 · Cited by 5
Multi-level Contrastive Learning for Script-based Character Understanding
Li, D., Zhang, H., Li, Y., & Yang, S. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.366 · Cited by 11
Automatic Analysis of Substantiation in Scientific Peer Reviews
Guo, Y., Shang, G., Rennard, V., Vazirgiannis, M., & Clavel, C. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.684 · Cited by 34
Hop, Union, Generate: Explainable Multi-hop Reasoning without Rationale Supervision
Zhao, W., Chiu, J., Cardie, C., & Rush, A. M. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.1001 · Cited by 13
FormaT5: Abstention and Examples for Conditional Table Formatting with Natural Language
Singh, M. P., Cambronero, J., Gulwani, S., Le, V., Negreanu, C., Nouri, E., Raza, M., & Verbruggen, G. (2023) · Proceedings of the VLDB Endowment · doi:10.14778/3632093.3632111 · Cited by 13
KIMedQA: towards building knowledge-enhanced medical QA models
Zafar, A., Sahoo, S. K., Varshney, D., Das, A., & Ekbal, A. (2024) · Journal of Intelligent Information Systems · doi:10.1007/s10844-024-00844-1 · Cited by 11
“SpanBERT Joshi et al. (2020) improved on BERT Devlin et al. (2018) by masking contiguous token spans and introducing a span boundary target.”
MiLMo:Minority Multilingual Pre-Trained Language Model
Deng, J., Shi, H., Yu, X., Bao, W., Sun, Y., & Zhao, X. (2023) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc53992.2023.10393961 · Cited by 5
“After BERT, various BERT variants have emerged, such as ALBERT [9], SpanBERT [10] and RoBERta [11].”
BackgroundMethodBigText-QA: Question Answering over a Large-Scale Hybrid Knowledge Graph
Xu, J., Biryukov, M., Theobald, M., & Venugopal, V. E. (2024) · Lecture notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering · doi:10.1007/978-3-031-52265-9_3 · Cited by 6
“NED AIDA-Light [37], REL [23], ELQ [31] CR SpanBERT:2018 [30], SpanBERT:2020 [24] Background KB YAGO , WikiData Projects which involve the stage of text (pre-)processing typically apply either an entire end-to-end suite of annotation tools, such as NLTK [5], StanfordNLP [35], SpaCy 7 , or a specific component from it…”
DSTEA: Improving Dialogue State Tracking via Entity Adaptive pre-training
Lee, Y., Kim, T., Yoon, H., Kang, P., Bang, J., & Kim, M. (2024) · Knowledge-Based Systems · doi:10.1016/j.knosys.2024.111542 · Cited by 5
“However, this masking strategy trains information about a single segment (Joshi et al., 2020) and has a limitation that models cannot learn related sub-word tokens together.”
BackgroundZhu, Y., Peng, S., Pradhan, S., & Zeldes, A. (2023) · International Joint Conference on Natural Language Processing and Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL) · doi:10.18653/v1/2023.ijcnlp-short.14 · Cited by 9
A syntax-enhanced parameter generation network for multi-source cross-lingual event extraction
Huang, W., Zhang, J., Tian, T., & Ji, D. (2024) · Knowledge-Based Systems · doi:10.1016/j.knosys.2024.111585 · Cited by 5
Tan, Y., Liu, Q., Liu, T., Liu, H., Wang, S.-M., & Chen, Z. (2024) · Applied Sciences · doi:10.3390/app14104259 · Cited by 5
Metadata
- Authors
- Joshi, M., Chen, D., Liu, Y., Weld, D. S., Zettlemoyer, L., & Levy, O.
- Affiliations
- University of Washington (US)
- Princeton University (US)
- Meta (United States) (US)
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2020-12
- Volume
- 8
- Pages
- 64-77
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W2962753370
- Cited by
- 1,827
- References
- 51
- Citations by year
- 2026: 84 · 2025: 169 · 2024: 214 · 2023: 373 · 2022: 342 · 2021: 393 · 2020: 224
- FWCI
- 130.82
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- cc-by
- Locations
- Transactions of the Association for Computational Linguistics (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University)
- Funders
- University of Washington
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Keywords
- SpanBERT, BERT pre-training, span prediction, question answering, coreference resolution, relation extraction, text representation learning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.