---
title: "VLCAP at ImageEval 2025 Shared Task: Multimodal Arabic"
description: "The authors present VLCAP, a two-stage pipeline that avoids direct end-to-end captioning. 21 references, cited by 1 paper here."
authors: ["Passant Elchafei", "Amany Fashwan"]
year: 2025
venue: "Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks"
source: "https://aclanthology.org/2025.arabicnlp-sharedtasks.56/"
doi: "10.18653/v1/2025.arabicnlp-sharedtasks.56"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/elchafei-2025-vlcap-imageeval-2025-shared-task/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/elchafei-2025-vlcap-imageeval-2025-shared-task/)

[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/13/) / VLCAP at ImageEval 2025 Shared Task: Multimodal Arabic Captioning with Interpretable Visual Concept Integration

# VLCAP at ImageEval 2025 Shared Task: Multimodal Arabic Captioning with Interpretable Visual Concept Integration

**Authors:** [Passant Elchafei](https://md.research.elamur.ai/free/en/authors/passant-elchafei/) and [Amany Fashwan](https://md.research.elamur.ai/free/en/authors/amany-fashwan/) · **Year:** 2025 · **Venue:** [Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2025](https://md.research.elamur.ai/free/en/venues/arabicnlp/2025/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2025.arabicnlp-sharedtasks.56/) · [PDF](https://aclanthology.org/2025.arabicnlp-sharedtasks.56.pdf) · [DOI 10.18653/v1/2025.arabicnlp-sharedtasks.56](https://doi.org/10.18653/v1/2025.arabicnlp-sharedtasks.56) · [OpenAlex](https://openalex.org/W4416037776)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/elchafei-2025-vlcap-imageeval-2025-shared-task/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper tackles Arabic image captioning for Subtask 2 of the ImageEval 2025 shared task. In this subtask a system receives one image and must return a short, culturally appropriate Arabic description. The authors present VLCAP, a two-stage pipeline that avoids direct end-to-end captioning.

- **Stage 1:** a multilingual CLIP-style encoder (mCLIP, AraCLIP or Jina V4) ranks entries of an Arabic label vocabulary by image–text cosine similarity. The vocabulary combines frequent content words from the 2,718 training captions with about 21K Visual Genome concepts translated into Arabic.
- **Stage 2:** an Arabic prompt is built around roughly 25–30 of the highest-ranked labels and passed, together with the image, to Qwen-VL or Gemini Pro Vision.

This yields six encoder–decoder combinations. The submitted system ranked first among teams on the official cosine-similarity metric (60.01) and second on the GPT-4o judge (33.05). BLEU-1 stayed far below the organisers' baseline.

The work matters because it shows that explicit, inspectable concept labels can steer a large vision–language model. It also shows that the choice of label extractor can change the judged quality of captions noticeably.

## Key points
- **Data:** the shared-task dataset has 2,718 training, 76 validation and 753 test images. The captions are human-written, mostly in MSA with occasional dialect, and the gold captions for validation and test are hidden.
- **Best configurations:**
  - Gemini + mCLIP had the best BLEU-1 (5.34) and cosine similarity (60.01).
  - The best LLM-judge score was 36.33. The abstract assigns it to AraCLIP + Qwen-VL, but Table 1 lists it under Gemini + AraCLIP; Qwen + AraCLIP has 31.40 there.
- **Baseline comparison:** the baseline reached BLEU-1 16.98, cosine 58.46 and LLM judge 30.82. VLCAP beats it on cosine and judge scores in most configurations, but it is much lower on BLEU-1.
- **Official evaluation:**
  - Cosine similarity was computed over TF–IDF n-grams after Arabic normalisation, so it measures lexical overlap.
  - The LLM judge was GPT-4o.
  - Manual rating covered 5% of the test set.
- **Manual scores for VLCAP:**
  - Completeness 2.67 and Accuracy 2.97 are the top values among teams.
  - Cultural Relevance 2.57 and Conciseness 3.17 trail several other teams in Table 4, although the text calls them second.

## Interesting ideas
- **Separate recognition from wording.** First extract a short list of explicit, inspectable concepts, then let the generator verbalise them; failures can then be traced to either step.
- **Build the label inventory from two sources.** Combine frequent content words from the task's own corpus with a large general-domain concept list, so the inventory covers both in-domain wording and broad coverage.
- **Cross every front-end with every back-end.** Running all extractor × generator combinations revealed that the front-end choice shifted judged quality more than the generator did.
- **Report metrics that disagree.** The lexical, overlap-based and LLM-judge metrics ranked the configurations differently; the paper reports all of them instead of picking one.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (back): Generation from the meaning graph could feed the realiser an explicit list of graph concepts as a checklist, so omissions or hallucinations are attributable to a specific node.
- **I2** (world): The ontology lexicon could merge frequent lemmas from the target texts (fables, school tests) with WordNet/Wikidata concepts, flagging which source each entry came from.
- **I3** (learn): Evaluating each tokenizer/parser variant crossed with each downstream reader would show which stage actually drives answer accuracy before optimising the wrong component.
- **I4** (learn): Evaluations of answers or generated text should always pair a surface-overlap metric with a meaning-level check, treating disagreement between them as a diagnostic signal.

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

## Institutions

- [Alexandria University](https://md.research.elamur.ai/free/en/universities/alexandria-university/) — University, Alexandria, Egypt · 39 works · 287 citations
- [Universität Ulm](https://md.research.elamur.ai/free/en/universities/universitat-ulm/) — University, Ulm, Germany · 4 works · 89 citations

📚 Shelves (2): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works

🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [🖼️ Vision & language](https://md.research.elamur.ai/free/en/topics/multimodal/) · 657 works

## Cite

Cite as: Passant Elchafei and Amany Fashwan. 2025. VLCAP at ImageEval 2025 Shared Task: Multimodal Arabic Captioning with Interpretable Visual Concept Integration. Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks.

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

```bibtex
@inproceedings{elchafei2025vlcap,
  title = {{VLCAP at ImageEval 2025 Shared Task: Multimodal Arabic Captioning with Interpretable Visual Concept Integration}},
  author = {Passant Elchafei and Amany Fashwan},
  booktitle = {Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks},
  year = {2025},
  doi = {10.18653/v1/2025.arabicnlp-sharedtasks.56},
  url = {https://doi.org/10.18653/v1/2025.arabicnlp-sharedtasks.56}
}
```

- [← previous Using NLI to Identify Potential Collocation Transfer in L2 English](https://md.research.elamur.ai/free/en/papers/yang-2025-nli-identify-potential-collocation-transfer/)
- [next → Where Frameworks (Dis)agree: A Study of Discourse Segmentation](https://md.research.elamur.ai/free/en/papers/ogrodniczuk-2025-where-frameworks-dis-agree-study/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/13/)

## References

21 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Muhammad Al-Barham, Imad Afyouni, Khalid Almubarak, Ashraf Elnagar, Ayad Turky, and Ibrahim Hashem. 2024. [Araclip: Cross-lingual learning for effective arabic image retrieval](https://md.research.elamur.ai/free/en/papers/barham-2024-araclip-cross-lingual-learning-effective/). In Proceedings of The Second Arabic Natural Language Processing Conference, pages 102–110. Association for Computational Linguistics.
2. Rasha Saleh Al-Malki and Arwa Yousuf Al-Aama. 2023. Arabic captioning for images of clothing using deep learning. Sensors, 23(8). [↗](https://doi.org/10.3390/s23083783)
3. Huda Al-muzaini, Tasniem N., and Hafida Benhidour. 2018. Automatic arabic image captioning using rnn-lstm-based language model and cnn. International Journal of Advanced Computer Science and Applications, 9. [find](https://scholar.google.com/scholar?q=Huda+Al%2Dmuzaini%2C+Tasniem+N%2E%2C+and+Hafida+Benhidour%2E+2018%2E+Automatic+arabic+image+captioning+using+rnn%2Dlstm%2Dbased+language+model+and+cnn%2E+International+Journal+of+Advanced+Computer+Science+and+Applicati)
4. Rabee Al-Qasem and Mohannad Hendi. 2025. [Impactai at imageeval 2025 shared task: Region-aware transformers for arabic image captioning: A case study on the palestinian narrative](https://md.research.elamur.ai/free/en/papers/qasem-2025-impactai-imageeval-2025-shared-task/). In Proceedings of the Third Arabic NLP Conference (ArabicNLP 2025). Co-located with EMNLP 2025, Nov 5–9.
5. Ashwaq Alsayed, Thamir M. Qadah, and Muhammad Arif. 2023. A performance analysis of transformer-based deep learning models for arabic image captioning. J. King Saud Univ. Comput. Inf. Sci., 35(9). [↗](https://doi.org/10.1016/j.jksuci.2023.101750)
6. Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew Dai, Anja Hauth, Katie Millican, David Silver, Melvin Johnson, Ioannis Antonoglou, Julian Schrittwieser, Amelia Glaese, Jilin Chen, Emily Pitler, Timothy Lillicrap, Angeliki Lazaridou, and Orhan Firat. 2023. [Gemini: A family of highly capable multimodal models](https://md.research.elamur.ai/free/en/papers/gemini-team-2023-gemini-family-highly-capable-multimodal/).
7. Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023. [Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond](https://md.research.elamur.ai/free/en/papers/bai-2023-qwen-vl-versatile-vision-language/). Ahlam Bashiti, Alaa Aljabari, Hadi Hamoud, Md. [↗](https://doi.org/10.48550/arxiv.2308.12966)
8. Rafiul Biswas, Bilal Shalash, Mustafa Jarrar, Fadi Zaraket, George Mikros, Ehsaneddin Asgari, and Wajdi Zaghouani. 2025. [ImageEval 2025: The First Arabic Image Captioning Shared Task](https://md.research.elamur.ai/free/en/papers/bashiti-2025-imageeval-2025-first-arabic-image/). In Proceedings of the Third Arabic Natural Language Processing Conference (ArabicNLP 2025). Co-located with EMNLP 2025, November 5–9.
9. Abdulkadir Shehu Bichi, Ismail Dauda Abubakar, Fatima Muhammad Adam, Aminu Musa, Auwal Umar Ahmed, Abubakar Ibrahim, Khadija Salihu Aua, Aisha Mustapha Ahmed, and Mahmud Said Ahmed. 2025. Codezone research group at imageeval 2025 shared task: Arabic image captioning using blip and m2m100: A two-stage translation approach for imageeval 2025. In Proceedings of the Third Arabic NLP Conference (ArabicNLP 2025). Co-located with EMNLP 2025, Nov 5–9. [find](https://scholar.google.com/scholar?q=Abdulkadir+Shehu+Bichi%2C+Ismail+Dauda+Abubakar%2C+Fatima+Muhammad+Adam%2C+Aminu+Musa%2C+Auwal+Umar+Ahmed%2C+Abubakar+Ibrahim%2C+Khadija+Salihu+Aua%2C+Aisha+Mustapha+Ahmed%2C+and+Mahmud+Said+Ahmed%2E+2025%2E+Codezone+res)
10. Guanhua Chen, Lu Hou, Yun Chen, Wenliang Dai, Lifeng Shang, Xin Jiang, Qun Liu, Jia Pan, and Wenping Wang. 2023. mCLIP: Multilingual CLIP via cross-lingual transfer. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13028– 13043, Toronto, Canada. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Guanhua+Chen%2C+Lu+Hou%2C+Yun+Chen%2C+Wenliang+Dai%2C+Lifeng+Shang%2C+Xin+Jiang%2C+Qun+Liu%2C+Jia+Pan%2C+and+Wenping+Wang%2E+2023%2E+mCLIP%3A+Multilingual+CLIP+via+cross%2Dlingual+transfer%2E+In+Proceedings+of+the+61st+Annual+)
11. Samar Elbedwehy and T. Medhat. 2023. Improved arabic image captioning model using feature concatenation with pre-trained word embedding. Neural Comput. Appl., 35(26):19051–19067. [↗](https://doi.org/10.1007/s00521-023-08744-1)
12. Ranjay Krishna, Yuke Zhu, Oliver Groth, Justin Johnson, Kenji Hata, Joshua Kravitz, Stephanie Chen, Yannis Kalantidis, Li-Jia Li, David Shamma, Michael Bernstein, and Fei-Fei Li. 2017. [Visual genome: Connecting language and vision using crowdsourced dense image annotations](https://md.research.elamur.ai/free/en/papers/krishna-2016-visual-genome-connecting-language-vision/). International Journal of Computer Vision, 123. [↗](https://doi.org/10.1007/s11263-016-0981-7)
13. Abdelrahman Mohamed, Fakhraddin Alwajih, El Moatez Billah Nagoudi, Alcides Alcoba, and Muhammad Abdul-Mageed. 2023. [Violet: A vision-language model for arabic image captioning with gemini decoder](https://md.research.elamur.ai/free/en/papers/mohamed-2023-violet-vision-language-model-arabic/). pages 1–11.
14. Mariam Saeed, Sarah Elshabrawy, Abdelrahman Ha-grass, Mazen Yasser, and Ayman Khalafallah. 2025. [Averroes at imageeval 2025 shared task: Advancing arabic image captioning with augmentation and two-stage generation](https://md.research.elamur.ai/free/en/papers/saeed-2025-averroes-imageeval-2025-shared-task/). In Proceedings of the Third Arabic NLP Conference (ArabicNLP 2025). Co-located with EMNLP 2025, Nov 5–9.
15. Muhy Eddin Za’ter and Bashar Talafha. 2022. Benchmarking and improving arabic automatic image captioning through the use of multi-task learning paradigm. [find](https://scholar.google.com/scholar?q=Muhy+Eddin+Za%E2%80%99ter+and+Bashar+Talafha%2E+2022%2E+Benchmarking+and+improving+arabic+automatic+image+captioning+through+the+use+of+multi%2Dtask+learning+paradigm%2E)
16. Jingyi Zhang, Jiaxing Huang, Sheng Jin, and Shijian Lu. 2024. Vision-language models for vision tasks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(8):5625–5644. [↗](https://doi.org/10.1109/tpami.2024.3369699)
17. Obeida Eljundi, Mohamad Dhaybi, Kotaiba Mokadam, Hazem Hajj, and Daniel Asmar. 2020. Resources and end-to-end neural network models for arabic image captioning. pages 233–241. [find](https://scholar.google.com/scholar?q=Obeida+Eljundi%2C+Mohamad+Dhaybi%2C+Kotaiba+Mokadam%2C+Hazem+Hajj%2C+and+Daniel+Asmar%2E+2020%2E+Resources+and+end%2Dto%2Dend+neural+network+models+for+arabic+image+captioning%2E+pages+233%E2%80%93241%2E)
18. Jonathan Emami, Pierre Nugues, Ashraf Elnagar, and Imad Afyouni. 2022. [Arabic image captioning using pre-training of deep bidirectional transformers](https://md.research.elamur.ai/free/en/papers/emami-2022-arabic-image-captioning-pre-training/). In Proceedings of the 15th International Conference on Natural Language Generation, pages 40–51, Waterville, Maine, USA and virtual meeting. Association for Computational Linguistics.
19. Wen Gendy and Dularia Patel. 2024. Advancements in computer vision: A comprehensive survey of image processing and interdisciplinary applications. Academic Journal of Science and Technology, 13(2):28–34. [↗](https://doi.org/10.54097/5e1cqw59)
20. Michael Günther, Saba Sturua, Mohammad Akram, Isabelle Mohr, Andrei Ungureanu, Sedigheh Eslami, Scott Martens, Bo Wang, Nan Wang, and Han Xiao. 2025. jina-embeddings-v4: Universal embeddings for multimodal multilingual retrieval. Muhammad Abu Horaira, Farhan Amin, Sakibul Hasan, Md. [↗](https://doi.org/10.48550/arxiv.2506.18902)
21. Tanvir Ahammed Shawon, and Muhammad Ibrahim Khan. 2025. Phantomtroupe at imageeval shared task: Multimodal arabic image captioning through translation-based fine-tuning of llm models. In Proceedings of the Third Arabic NLP Conference (ArabicNLP 2025). Co-located with EMNLP 2025, Nov 5–9. [find](https://scholar.google.com/scholar?q=Tanvir+Ahammed+Shawon%2C+and+Muhammad+Ibrahim+Khan%2E+2025%2E+Phantomtroupe+at+imageeval+shared+task%3A+Multimodal+arabic+image+captioning+through+translation%2Dbased+fine%2Dtuning+of+llm+models%2E+In+Proceedings+o)

*Updated 1 November 2026*