---
title: "Qamosy at Arabic Reverse Dictionary shared task: Semi Decoder"
description: "This paper presents the Qamosy team's system for the KSAA-RD Arabic Reverse Dictionary shared task. 12 references, cited by 4 papers here."
authors: ["Serry Sibaee", "Samar Ahmad", "Ibrahim Khurfan", "Vian Sabeeh", "Ahmed Bahaaulddin", "Hanan Belhaj", "Abdullah Alharbi"]
year: 2023
venue: "Proceedings of ArabicNLP 2023"
source: "https://aclanthology.org/2023.arabicnlp-1.41/"
doi: "10.18653/v1/2023.arabicnlp-1.41"
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/sibaee-2023-qamosy-arabic-reverse-dictionary-shared/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sibaee-2023-qamosy-arabic-reverse-dictionary-shared/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/12/) / Qamosy at Arabic Reverse Dictionary shared task: Semi Decoder Architecture for Reverse Dictionary with SBERT Encoder

# Qamosy at Arabic Reverse Dictionary shared task: Semi Decoder Architecture for Reverse Dictionary with SBERT Encoder

**Authors:** [Serry Sibaee](https://md.research.elamur.ai/free/en/authors/serry-sibaee/), [Samar Ahmad](https://md.research.elamur.ai/free/en/authors/samar-ahmad/), [Ibrahim Khurfan](https://md.research.elamur.ai/free/en/authors/ibrahim-khurfan/), [Vian Sabeeh](https://md.research.elamur.ai/free/en/authors/vian-sabeeh/), [Ahmed Bahaaulddin](https://md.research.elamur.ai/free/en/authors/ahmed-bahaaulddin/), [Hanan Belhaj](https://md.research.elamur.ai/free/en/authors/hanan-belhaj/) and [Abdullah Alharbi](https://md.research.elamur.ai/free/en/authors/abdullah-alharbi/) · **Year:** 2023 · **Venue:** [Proceedings of ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/2023/)) · **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/2023.arabicnlp-1.41/) · [PDF](https://aclanthology.org/2023.arabicnlp-1.41.pdf) · [DOI 10.18653/v1/2023.arabicnlp-1.41](https://doi.org/10.18653/v1/2023.arabicnlp-1.41) · [OpenAlex](https://openalex.org/W4389518328)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · [4](https://md.research.elamur.ai/free/en/papers/sibaee-2023-qamosy-arabic-reverse-dictionary-shared/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
This paper presents the Qamosy team's system for the KSAA-RD Arabic Reverse Dictionary shared task. A reverse dictionary takes a description of a concept and returns the matching word. The authors note that Arabic has few such resources, partly because suitable definition datasets are scarce.

The organizers supplied entries drawn from the LMF Contemporary Arabic dictionary. Each entry pairs a gloss with two target vectors for the headword: skip-gram (sgns) and Electra embeddings. The task is to predict the target vector from the gloss.

The system has two stages:
1. A frozen multilingual SBERT model encodes each gloss into a fixed 512-dimensional vector.
2. A "semi-decoder", a widening-then-narrowing MLP trained with MSE loss, regresses that vector onto the target embedding space.

The authors tested seven MLP configurations on Electra and reused the best ones for sgns. The system placed 2nd by the Rank metric on both tracks. The work shows that an off-the-shelf sentence encoder plus a simple regression head is a competitive, low-cost baseline for a low-resource reverse dictionary.

## Key points
- **Data:** 58,010 JSON entries (Train 45,200 / Dev 6,400 / Test 6,410). Each entry has the fields id, word, gloss, enId, sgns and electra.
- **Encoder:** 'distiluse-base-multilingual-cased' from Sentence Transformers (SBERT), giving 512-dimensional gloss vectors.
- **Semi-decoder:** hidden layers sized 8×, 4×, 2× and 1× the input width. It projects to 256 dimensions (Electra) or 300 dimensions (sgns) and is trained with an MSE objective.
- **Results:** On Electra, Model 3 (GELU, 1000 epochs) achieved the best Rank of 28.05%. On sgns, the best Rank was 30.78%. The ReLU baseline stayed near a Rank of 50%, roughly chance level.
- **Training dynamics:** 100–300 epochs underfit the model, and more than 2000 epochs led to overfitting, so the useful range was about 1000–2000 epochs. GELU worked best among the activations, while dropout had only a moderate effect.

## Interesting ideas
- Keep a strong pretrained sentence encoder frozen and train only a small MLP that maps its output into a target word-vector space; this gives a cheap, fast baseline for definition-to-word lookup.
- One baseline had acceptable MSE and cosine scores but a Rank near 50%, meaning its ranking was effectively random; a rank-based metric therefore exposes models that look fine on distance scores.
- The task used tiered metrics: Rank decides first, then MSE breaks ties, then cosine similarity, giving one unambiguous model ordering.
- To save compute, the authors searched architectures on one target space (Electra) and reused only the winners on the second (sgns).
- A reverse dictionary turns a description into a word, which also makes it a probe of how well a model represents meaning.

## For Mova

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

- **I1** (world): definitions of unknown words could be mapped to suggest candidate WordNet/Wikidata entries, with each suggestion then accepted or rejected by explicit ontology checks.
- **I2** (learn): every evaluation could include a chance-level rank reference as a negative control, catching components whose average-similarity scores hide useless orderings.
- **I3** (learn): a fixed primary/secondary/tertiary metric order would make choices between parser or teacher variants reproducible and explainable.
- **I4** (tree): gparse hyperparameters could be tuned on one treebank and only the top configurations re-run on other treebanks or UD dialects.
- **I5** (back): during generation, a meaning-graph description of a concept could be matched to the best single lexical item, and missed matches would point to gaps in the graph.

*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

- [King Abdulaziz University](https://md.research.elamur.ai/free/en/universities/king-abdulaziz-university/) — University, Jeddah, Saudi Arabia · 31 works · 508 citations
- [Libyan Academy](https://md.research.elamur.ai/free/en/universities/libyan-academy/) — University, Tripoli, Libya · 2 works · 7 citations
- [Middle Technical University](https://md.research.elamur.ai/free/en/universities/middle-technical-university/) — University, Baghdad, Iraq · 2 works · 7 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Serry Sibaee, Samar Ahmad, Ibrahim Khurfan, Vian Sabeeh, Ahmed Bahaaulddin, Hanan Belhaj and Abdullah Alharbi. 2023. Qamosy at Arabic Reverse Dictionary shared task: Semi Decoder Architecture for Reverse Dictionary with SBERT Encoder. Proceedings of ArabicNLP 2023.

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

```bibtex
@inproceedings{sibaee2023qamosy,
  title = {{Qamosy at Arabic Reverse Dictionary shared task: Semi Decoder Architecture for Reverse Dictionary with SBERT Encoder}},
  author = {Serry Sibaee, Samar Ahmad, Ibrahim Khurfan, Vian Sabeeh, Ahmed Bahaaulddin, Hanan Belhaj and Abdullah Alharbi},
  booktitle = {Proceedings of ArabicNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.arabicnlp-1.41},
  url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.41}
}
```

- [← previous NLPeople at NADI 2023 Shared Task: Arabic Dialect Identification with Augmented Context and Multi-Stage Tuning](https://md.research.elamur.ai/free/en/papers/elkaref-2023-nlpeople-nadi-2023-shared-task/)
- [next → Raphael at ArAIEval Shared Task: Understanding Persuasive Language and Tone, an LLM Approach](https://md.research.elamur.ai/free/en/papers/shukla-2023-raphael-araieval-shared-task-understanding/)
- [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/12/)

## References

12 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Mohammad Taher Pilehvar. 2019. On the importance of distinguishing word meaning representations: A case study on reverse dictionary mapping. In 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), pages 2151–2156. [find](https://scholar.google.com/scholar?q=Mohammad+Taher+Pilehvar%2E+2019%2E+On+the+importance+of+distinguishing+word+meaning+representations%3A+A+case+study+on+reverse+dictionary+mapping%2E+In+Proceedings+of+the+2019+Conference+of+the+North+American)
2. Fanchao Qi, Lei Zhang, Yanhui Yang, Zhiyuan Liu, and Maosong Sun. 2020. Wantwords: An open-source online reverse dictionary system. EMNLP 2020, page 175. [find](https://scholar.google.com/scholar?q=Fanchao+Qi%2C+Lei+Zhang%2C+Yanhui+Yang%2C+Zhiyuan+Liu%2C+and+Maosong+Sun%2E+2020%2E+Wantwords%3A+An+open%2Dsource+online+reverse+dictionary+system%2E+EMNLP+2020%2C+page+175%2E)
3. Nils Reimers and Iryna Gurevych. 2019. [Sentence-bert: Sentence embeddings using siamese bert-networks](https://md.research.elamur.ai/free/en/papers/reimers-2019-sentence-bert-sentence-embeddings-siamese/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
4. Nils Reimers and Iryna Gurevych. 2020. [Making monolingual sentence embeddings multilingual using knowledge distillation](https://md.research.elamur.ai/free/en/papers/reimers-2020-making-monolingual-sentence-embeddings-multilingual/). In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
5. Sushrut Thorat and Varad Choudhari. 2016. Implementing a reverse dictionary, based on word definitions, using a node-graph architecture. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2797–2806. [↗](https://doi.org/10.48550/arxiv.1606.00025)
6. Aarchi Agrawal, KS Ashin Shanly, Kavita Vaishnaw, and Mayank Singh. 2021. Reverse dictionary using an improved cbow model. In Proceedings of the 3rd ACM India Joint International Conference on Data Science & Management of Data (8th ACM IKDD CODS & 26th COMAD), pages 420–420. [find](https://scholar.google.com/scholar?q=Aarchi+Agrawal%2C+KS+Ashin+Shanly%2C+Kavita+Vaishnaw%2C+and+Mayank+Singh%2E+2021%2E+Reverse+dictionary+using+an+improved+cbow+model%2E+In+Proceedings+of+the+3rd+ACM+India+Joint+International+Conference+on+Data+Sc)
7. Nihed Bendahman, Julien Breton, Lina Nicolaieff, Mokhtar Boumedyen Billami, Christophe Bortolaso, and Youssef Miloudi. 2022. Bl. research at semeval-2022 task 1: Deep networks for reverse dictionary using embeddings and lstm autoencoders. In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 94–100. [find](https://scholar.google.com/scholar?q=Nihed+Bendahman%2C+Julien+Breton%2C+Lina+Nicolaieff%2C+Mokhtar+Boumedyen+Billami%2C+Christophe+Bortolaso%2C+and+Youssef+Miloudi%2E+2022%2E+Bl%2E+research+at+semeval%2D2022+task+1%3A+Deep+networks+for+reverse+dictionary+u)
8. Lei Zhang, Fanchao Qi, Zhiyuan Liu, Yasheng Wang, Qun Liu, and Maosong Sun. 2020. Multi-channel reverse dictionary model. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 312–319. [find](https://scholar.google.com/scholar?q=Lei+Zhang%2C+Fanchao+Qi%2C+Zhiyuan+Liu%2C+Yasheng+Wang%2C+Qun+Liu%2C+and+Maosong+Sun%2E+2020%2E+Multi%2Dchannel+reverse+dictionary+model%2E+In+Proceedings+of+the+AAAI+conference+on+artificial+intelligence%2C+volume+34%2C+p)
9. Pinzhen Chen and Zheng Zhao. 2022. A unified model for reverse dictionary and definition modelling. AACL-IJCNLP 2022, page 8. [↗](https://doi.org/10.48448/g4ew-nf04)
10. Michael A Hedderich, Andrew Yates, Dietrich Klakow, and Gerard de Melo. 2019. Using multi-sense vector embeddings for reverse dictionaries. In Proceedings of the 13th International Conference on Computational Semantics-Long Papers, pages 247–258. [↗](https://doi.org/10.48550/arxiv.1904.01451)
11. Arman Malekzadeh, Amin Gheibi, and Ali Mohades. 2021. Predict: persian reverse dictionary. arXiv preprint arXiv:2105.00309. [↗](https://arxiv.org/abs/2105.00309)
12. Oscar Méndez, Hiram Calvo, and Marco A Moreno-Armendáriz. 2013. A reverse dictionary based on [find](https://scholar.google.com/scholar?q=Oscar+M%C3%A9ndez%2C+Hiram+Calvo%2C+and+Marco+A+Moreno%2DArmend%C3%A1riz%2E+2013%2E+A+reverse+dictionary+based+on)

*Updated 1 November 2026*