---
title: "GATE: General Arabic Text Embedding for — Nacar 2025"
description: "Arabic has few strong sentence-embedding models and few good datasets for semantic textual similarity (STS)."
authors: ["Omer Nacar", "Anis Koubaa", "Serry Sibaee", "Yasser Al-Habashi", "Adel Ammar", "Wadii Boulila"]
year: 2025
venue: "arXiv:2505.24581"
source: "https://arxiv.org/abs/2505.24581"
doi: "10.48550/arxiv.2505.24581"
arxiv: "2505.24581"
licence: "CC BY-SA 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by-sa/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/nacar-2025-gate-general-arabic-text-embedding/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/nacar-2025-gate-general-arabic-text-embedding/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/2/) / GATE: General Arabic Text Embedding for Enhanced Semantic Textual Similarity with Matryoshka Representation Learning and Hybrid Loss Training

# GATE: General Arabic Text Embedding for Enhanced Semantic Textual Similarity with Matryoshka Representation Learning and Hybrid Loss Training

**Authors:** [Omer Nacar](https://md.research.elamur.ai/free/en/authors/omer-nacar/), [Anis Koubaa](https://md.research.elamur.ai/free/en/authors/anis-koubaa/), [Serry Sibaee](https://md.research.elamur.ai/free/en/authors/serry-sibaee/), [Yasser Al-Habashi](https://md.research.elamur.ai/free/en/authors/yasser-al-habashi/), [Adel Ammar](https://md.research.elamur.ai/free/en/authors/adel-ammar/), [Wadii Boulila](https://md.research.elamur.ai/free/en/authors/wadii-boulila/) · **Year:** 2025 · **Venue:** arXiv:2505.24581 · **Licence:** [CC BY-SA 4.0 (arXiv)](http://creativecommons.org/licenses/by-sa/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2505.24581) · [PDF](https://arxiv.org/pdf/2505.24581) · [DOI 10.48550/arxiv.2505.24581](https://doi.org/10.48550/arxiv.2505.24581) · [arXiv 2505.24581](https://arxiv.org/abs/2505.24581) · [OpenAlex](https://openalex.org/W4414857917)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14) · 7 in our library, of the 140884 papers whose bibliographies we read

## Summary
Arabic has few strong sentence-embedding models and few good datasets for semantic textual similarity (STS). Its rich root-and-pattern morphology, flexible word order and missing diacritics add to the difficulty. The authors address this with GATE, a family of Arabic embedding models.

To build training data, they machine-translated SNLI and MultiNLI into Arabic with an NMT pipeline (CTranslate2, SentencePiece) and checked the output by hand. The result is three subsets: graded STS pairs, NLI triplets and labelled entailment pairs. They then fine-tuned AraBERT, MARBERT, LaBSE, multilingual MPNet and multilingual-E5-small. Training used Matryoshka Representation Learning (MRL), which makes shorter prefixes of an embedding usable on their own, and a hybrid objective: a cosine-similarity loss for STS plus a softmax classification loss over NLI labels.

On the MTEB Arabic STS tasks, the best ~135M-parameter models beat e5-mistral-7b-instruct, udever-bloom-1b1 and OpenAI embeddings. Quality barely drops when vectors are cut to 64 dimensions. The work shows that small, language-specific models with targeted objectives can outperform much larger general systems.

## Key points
- **Training data:**
  - STS subset: 8.63K train / 1.68K test.
  - Triplet subset: 571K / 6.58K.
  - Pair-classification subset (entailment / neutral / contradiction): 981K / 19.7K.
  - All three are translated from SNLI and MultiNLI.
- **Best models:**
  - Arabic-Triplet-Matryoshka-V2, AraBERT-based and trained with Matryoshka plus triplet loss, has the top MTEB STS average of 69.99 (STS17 85.31).
  - GATE-AraBERT-V1, trained with the hybrid loss, averages 68.54.
- **Truncation results for Arabic-Triplet-Matryoshka-V2:** the STS average goes from 69.99 at 768 dimensions to 69.92 at 512, 69.86 at 256, 69.52 at 128 and 69.43 at 64.
- **Baselines:**
  - e5-mistral-7b-instruct: 68.00
  - udever-bloom-1b1: 68.07
  - text-embedding-3-large: 65.54
  - text-embedding-ada-002: 63.67
  - MARBERT-all-nli-triplet-Matryoshka: 67.19
  - Arabic-labse-Matryoshka: 66.76
- **Error analysis:**
  - On an unrelated pair with gold score 0.1 ("a man playing guitar" vs. "a man driving a car"), most models gave 0.32–0.48 because of shared words. GATE-AraBERT-V1 gave 0.04.
  - On a highly similar pair, GATE-AraBERT-V1 was more conservative (0.73), while the other models scored above 0.84.

## Interesting ideas
- Pairing a graded-similarity regression loss with a three-way entailment/neutral/contradiction classification loss lets one model learn both fine-grained closeness and discrete logical relations, which a single contrastive loss misses. — For Mova (frames): A prag scorer that compares two frames could be trained against both graded paraphrase scores and NLI labels, so it learns to separate "same event, different wording" from "contradicting event."
- Nested (Matryoshka) representations let a single trained vector be cut to a short prefix for cheap uses or kept full-length for precise ones, with little quality loss. — For Mova (tree): Hidden states in gparse could be trained so that a short prefix still supports arc scoring, which would allow a fast, low-memory parsing mode on weaker hardware.
- The error analysis uses lexically overlapping but semantically unrelated pairs as a probe, and it exposes a false-positive bias that average scores hide. — For Mova (learn): Evaluation suites could add negative-control pairs with high word overlap but different who-did-what-to-whom, to check that frame matching is not just counting shared lemmas.
- Translating established English NLI corpora into a low-resource language with manual spot checks is a cheap way to bootstrap a training set. Turned around, the original NLI triplets are a ready supply of premise/entailment/contradiction test items. — For Mova (answer): SNLI/MultiNLI triplets could serve as an explainable QA test, where the reader must justify each "follows / contradicts / unrelated" verdict from the world it built for the premise.
- Reporting the same metric at every truncation level (768 down to 64) is a clean ablation design that shows exactly how much capacity a task really needs. — For Mova (learn): The same per-size ablation table could be applied to the perceptron feature sets or gparse widths, to find the smallest configuration that keeps LAS close to the full model.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

Fields: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/)

## Cite

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{nacar2025gate,
  title = {{GATE: General Arabic Text Embedding for Enhanced Semantic Textual Similarity with Matryoshka Representation Learning and Hybrid Loss Training}},
  author = {Nacar, Omer and Koubaa, Anis and Sibaee, Serry and Al-Habashi, Yasser and Ammar, Adel and Boulila, Wadii},
  year = {2025},
  doi = {10.48550/arxiv.2505.24581},
  eprint = {2505.24581},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2505.24581}
}
```

- [← previous From Tradition to Transformation: Addressing Challenges and Trends in Arabic Syntax Mastery via Problem-Based Learning](https://md.research.elamur.ai/free/en/papers/masruroh-2025-tradition-transformation-addressing-challenges-trends/)
- [next → Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://md.research.elamur.ai/free/en/papers/comanici-2025-gemini-2-5-pushing-frontier/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/2/)

## References

37 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Muhammad Abdul-Mageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2020. Arbert & marbert: Deep bidirectional transformers for arabic. arXiv preprint arXiv:2101.01785. [↗](https://arxiv.org/abs/2101.01785) [our card](https://md.research.elamur.ai/free/en/papers/abdul-2021-mageed-arbert-marbert-deep-bidirectional-transformers/)
2. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. Arabert: Transformer-based model for arabic language understanding. arXiv preprint arXiv:2003.00104. [↗](https://arxiv.org/abs/2003.00104) [our card](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/)
3. Akari Asai, Sewon Min, Zexuan Zhong, and Danqi Chen. 2023. Retrieval-based language models and applications. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 6: Tutorial Abstracts), pages 41–46. [find](https://scholar.google.com/scholar?q=Akari+Asai%2C+Sewon+Min%2C+Zexuan+Zhong%2C+and+Danqi+Chen%2E+2023%2E+Retrieval%2Dbased+language+models+and+applications%2E+In+Proceedings+of+the+61st+Annual+Meeting+of+the+Association+for+Computational+Linguistics+)
4. Ghizlane Bourahouat, Manar Abourezq, and Najima Daoudi. 2024. Word embedding as a semantic feature extraction technique in arabic natural language processing: an overview. Int. Arab J. Inf. Technol., 21(2):313–325. [find](https://scholar.google.com/scholar?q=Ghizlane+Bourahouat%2C+Manar+Abourezq%2C+and+Najima+Daoudi%2E+2024%2E+Word+embedding+as+a+semantic+feature+extraction+technique+in+arabic+natural+language+processing%3A+an+overview%2E+Int%2E+Arab+J%2E+Inf%2E+Technol%2E%2C+) [our card](https://md.research.elamur.ai/free/en/papers/guellil-2019-arabic-natural-language-processing-overview/)
5. Samuel R Bowman, Gabor Angeli, Christopher Potts, and Christopher D Manning. 2015. A large annotated corpus for learning natural language inference. arXiv preprint arXiv:1508.05326. [↗](https://arxiv.org/abs/1508.05326)
6. Daniel Cer, Mona Diab, Eneko Agirre, Inigo Lopez-Gazpio, and Lucia Specia. 2017. Semeval-2017 task 1: Semantic textual similarity-multilingual and cross-lingual focused evaluation. arXiv preprint arXiv:1708.00055. [↗](https://arxiv.org/abs/1708.00055)
7. Jinhyuk Lee, Zhuyun Dai, Xiaoqi Ren, Blair Chen, Daniel Cer, Jeremy R Cole, Kai Hui, Michael Boratko, Rajvi Kapadia, Wen Ding, et al. 2024. Gecko: Versatile text embeddings distilled from large language models. arXiv preprint arXiv:2403.20327. [↗](https://arxiv.org/abs/2403.20327)
8. Fangxiaoyu Feng, Yinfei Yang, Daniel Cer, Naveen Arivazhagan, and Wei Wang. 2020. Language-agnostic bert sentence embedding. arXiv preprint arXiv:2007.01852. [↗](https://arxiv.org/abs/2007.01852) [our card](https://md.research.elamur.ai/free/en/papers/feng-2020-language-agnostic-bert-sentence-embedding/)
9. Yang Li and Tao Yang. 2018. Word embedding for understanding natural language: a survey. Guide to big data applications, pages 83–104. [find](https://scholar.google.com/scholar?q=Yang+Li+and+Tao+Yang%2E+2018%2E+Word+embedding+for+understanding+natural+language%3A+a+survey%2E+Guide+to+big+data+applications%2C+pages+83%E2%80%93104%2E)
10. Tianyu Gao, Xingcheng Yao, and Danqi Chen. 2021. Simcse: Simple contrastive learning of sentence embeddings. arXiv preprint arXiv:2104.08821. [↗](https://arxiv.org/abs/2104.08821) [our card](https://md.research.elamur.ai/free/en/papers/gao-2021-simcse-simple-contrastive-learning-sentence/)
11. Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, and Haofen Wang. 2023. Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997. [↗](https://arxiv.org/abs/2312.10997) [our card](https://md.research.elamur.ai/free/en/papers/gao-2023-retrieval-augmented-generation-large-language/)
12. Michael Gutmann and Aapo Hyvärinen. 2010. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. In Proceedings of the thirteenth international conference on artificial intelligence and statistics, pages 297–304. JMLR Workshop and Conference Proceedings. [find](https://scholar.google.com/scholar?q=Michael+Gutmann+and+Aapo+Hyv%C3%A4rinen%2E+2010%2E+Noise%2Dcontrastive+estimation%3A+A+new+estimation+principle+for+unnormalized+statistical+models%2E+In+Proceedings+of+the+thirteenth+international+conference+on+art)
13. Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. 2020. Momentum contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9729–9738. [find](https://scholar.google.com/scholar?q=Kaiming+He%2C+Haoqi+Fan%2C+Yuxin+Wu%2C+Saining+Xie%2C+and+Ross+Girshick%2E+2020%2E+Momentum+contrast+for+unsupervised+visual+representation+learning%2E+In+Proceedings+of+the+IEEE%2FCVF+conference+on+computer+vision+a) [our card](https://md.research.elamur.ai/free/en/papers/he-2019-momentum-contrast-unsupervised-visual-representation/)
14. Zehan Li, Xin Zhang, Yanzhao Zhang, Dingkun Long, Pengjun Xie, and Meishan Zhang. 2023. Towards general text embeddings with multi-stage contrastive learning. arXiv preprint arXiv:2308.03281. [↗](https://arxiv.org/abs/2308.03281) [our card](https://md.research.elamur.ai/free/en/papers/li-2023-towards-general-text-embeddings-multi/)
15. Shuaiqi Liu, Jiannong Cao, Ruosong Yang, and Zhiyuan Wen. 2022. Key phrase aware transformer for abstractive summarization. Information Processing & Management, 59(3):102913. [find](https://scholar.google.com/scholar?q=Shuaiqi+Liu%2C+Jiannong+Cao%2C+Ruosong+Yang%2C+and+Zhiyuan+Wen%2E+2022%2E+Key+phrase+aware+transformer+for+abstractive+summarization%2E+Information+Processing+%26+Management%2C+59%283%29%3A102913%2E)
16. Bill MacCartney and Christopher D Manning. 2008. Modeling semantic containment and exclusion in natural language inference. In Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 521–528. [find](https://scholar.google.com/scholar?q=Bill+MacCartney+and+Christopher+D+Manning%2E+2008%2E+Modeling+semantic+containment+and+exclusion+in+natural+language+inference%2E+In+Proceedings+of+the+22nd+International+Conference+on+Computational+Linguis)
17. Niklas Muennighoff, Nouamane Tazi, Loïc Magne, and Nils Reimers. 2022. Mteb: Massive text embedding benchmark. arXiv preprint arXiv:2210.07316. [↗](https://arxiv.org/abs/2210.07316) [our card](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-mteb-massive-text-embedding-benchmark/)
18. Omer Nacar and Anis Koubaa. 2024. Enhancing semantic similarity understanding in arabic nlp with nested embedding learning. arXiv preprint arXiv:2407.21139. [↗](https://arxiv.org/abs/2407.21139) [our card](https://md.research.elamur.ai/free/en/papers/nacar-2024-enhancing-semantic-similarity-understanding-arabic/)
19. Junqin Huang, Zhongjie Hu, Zihao Jing, Mengya Gao, and Yichao Wu. 2024. Piccolo2: General text embedding with multi-task hybrid loss training. arXiv preprint arXiv:2405.06932. [↗](https://arxiv.org/abs/2405.06932)
20. OpenAI. 2023. Openai embeddings documentation. [https://platform.openai.com/docs/](https://platform.openai.com/docs/) guides/embeddings. [↗](https://platform.openai.com/docs/)
21. Infgrad. 2024. Stella-mrl-large-zh-v3.5-1792d. Accessed: 2024-08-28. [find](https://scholar.google.com/scholar?q=Infgrad%2E+2024%2E+Stella%2Dmrl%2Dlarge%2Dzh%2Dv3%2E5%2D1792d%2E+Accessed%3A+2024%2D08%2D28%2E)
22. OpenAI. 2024. New embedding models and api updates. Accessed: 2024-08-28. [find](https://scholar.google.com/scholar?q=OpenAI%2E+2024%2E+New+embedding+models+and+api+updates%2E+Accessed%3A+2024%2D08%2D28%2E)
23. Seonhoon Kim, Inho Kang, and Nojun Kwak. 2019. Semantic sentence matching with densely-connected recurrent and co-attentive information. In Proceedings of the AAAI conference on artificial intelligence, volume 33, pages 6586–6593. [find](https://scholar.google.com/scholar?q=Seonhoon+Kim%2C+Inho+Kang%2C+and+Nojun+Kwak%2E+2019%2E+Semantic+sentence+matching+with+densely%2Dconnected+recurrent+and+co%2Dattentive+information%2E+In+Proceedings+of+the+AAAI+conference+on+artificial+intelligenc)
24. Amarnath Pathak, Partha Pakray, and Jereemi Bentham. 2019. English–mizo machine translation using neural and statistical approaches. Neural Computing and Applications, 31(11):7615–7631. [find](https://scholar.google.com/scholar?q=Amarnath+Pathak%2C+Partha+Pakray%2C+and+Jereemi+Bentham%2E+2019%2E+English%E2%80%93mizo+machine+translation+using+neural+and+statistical+approaches%2E+Neural+Computing+and+Applications%2C+31%2811%29%3A7615%E2%80%937631%2E)
25. Guillaume Klein, Yoon Kim, Yuntian Deng, Jean Senellart, and Alexander M Rush. 2017. Opennmt: Open-source toolkit for neural machine translation. arXiv preprint arXiv:1701.02810. [↗](https://arxiv.org/abs/1701.02810) [our card](https://md.research.elamur.ai/free/en/papers/klein-2017-opennmt-open-source-toolkit-neural/)
26. Darius Koenig, Sean Lee, and Aamir Shakir. 2024. Open source strikes bread - new fluffy embeddings model. Accessed: 2024-08-28. [find](https://scholar.google.com/scholar?q=Darius+Koenig%2C+Sean+Lee%2C+and+Aamir+Shakir%2E+2024%2E+Open+source+strikes+bread+%2D+new+fluffy+embeddings+model%2E+Accessed%3A+2024%2D08%2D28%2E)
27. Aditya Kusupati, Gantavya Bhatt, Aniket Rege, Matthew Wallingford, Aditya Sinha, Vivek Ramanujan, William Howard-Snyder, Kaifeng Chen, Sham Kakade, Prateek Jain, et al. 2022. Matryoshka representation learning. Advances in Neural Information Processing Systems, 35:30233–30249. [find](https://scholar.google.com/scholar?q=Aditya+Kusupati%2C+Gantavya+Bhatt%2C+Aniket+Rege%2C+Matthew+Wallingford%2C+Aditya+Sinha%2C+Vivek+Ramanujan%2C+William+Howard%2DSnyder%2C+Kaifeng+Chen%2C+Sham+Kakade%2C+Prateek+Jain%2C+et+al%2E+2022%2E+Matryoshka+representation)
28. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748–8763. PMLR. [find](https://scholar.google.com/scholar?q=Alec+Radford%2C+Jong+Wook+Kim%2C+Chris+Hallacy%2C+Aditya+Ramesh%2C+Gabriel+Goh%2C+Sandhini+Agarwal%2C+Girish+Sastry%2C+Amanda+Askell%2C+Pamela+Mishkin%2C+Jack+Clark%2C+et+al%2E+2021%2E+Learning+transferable+visual+models+fro) [our card](https://md.research.elamur.ai/free/en/papers/radford-2021-learning-transferable-visual-models-natural/)
29. N Reimers. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084. [↗](https://arxiv.org/abs/1908.10084)
30. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Nils+Reimers+and+Iryna+Gurevych%2E+2019%2E+Sentence%2Dbert%3A+Sentence+embeddings+using+siamese+bert%2Dnetworks%2E+In+Proceedings+of+the+2019+Conference+on+Empirical+Methods+in+Natural+Language+Processing%2E+Associ)
31. Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. 2022. Text embeddings by weakly-supervised contrastive pre-training. arXiv preprint arXiv:2212.03533. [↗](https://arxiv.org/abs/2212.03533) [our card](https://md.research.elamur.ai/free/en/papers/wang-2022-text-embeddings-weakly-supervised-contrastive/)
32. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2023. Improving text embeddings with large language models. arXiv preprint arXiv:2401.00368. [↗](https://arxiv.org/abs/2401.00368) [our card](https://md.research.elamur.ai/free/en/papers/wang-2024-improving-text-embeddings-large-language/)
33. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024. Multilingual e5 text embeddings: A technical report. arXiv preprint arXiv:2402.05672. [↗](https://arxiv.org/abs/2402.05672) [our card](https://md.research.elamur.ai/free/en/papers/wang-2024-multilingual-e5-text-embeddings-technical/)
34. Yongliang Wu, Shuliang Zhao, and Ruiqiang Guo. 2021. A novel community answer matching approach based on phrase fusion heterogeneous information network. Information Processing & Management, 58(1):102408. [find](https://scholar.google.com/scholar?q=Yongliang+Wu%2C+Shuliang+Zhao%2C+and+Ruiqiang+Guo%2E+2021%2E+A+novel+community+answer+matching+approach+based+on+phrase+fusion+heterogeneous+information+network%2E+Information+Processing+%26+Management%2C+58%281%29%3A102)
35. Shitao Xiao, Zheng Liu, Peitian Zhang, and Niklas Muennighof. 2023. C-pack: Packaged resources to advance general chinese embedding. arXiv preprint arXiv:2309.07597. [↗](https://arxiv.org/abs/2309.07597) [our card](https://md.research.elamur.ai/free/en/papers/xiao-2023-c-pack-packed-resources/)
36. Xin Zhang, Zehan Li, Yanzhao Zhang, Dingkun Long, Pengjun Xie, Meishan Zhang, and Min Zhang. 2023. Language models are universal embedders. arXiv preprint arXiv:2310.08232. [↗](https://arxiv.org/abs/2310.08232)
37. Ying Zhao, Tingyu Xia, Yunqi Jiang, and Yuan Tian. 2024. Enhancing inter-sentence attention for semantic textual similarity. Information Processing & Management, 61(1):103535. [find](https://scholar.google.com/scholar?q=Ying+Zhao%2C+Tingyu+Xia%2C+Yunqi+Jiang%2C+and+Yuan+Tian%2E+2024%2E+Enhancing+inter%2Dsentence+attention+for+semantic+textual+similarity%2E+Information+Processing+%26+Management%2C+61%281%29%3A103535%2E)