---
title: "LIM-LIG at SemEval-2017 Task1 — paper by Nagoudi et al. (2017)"
description: "The LIM-LIG system builds a sentence vector by summing word embeddings from a pretrained Arabic… 17 references, cited by 4 papers here."
authors: ["El Moatez Billah Nagoudi", "Jérémy Ferrero", "Didier Schwab"]
year: 2017
venue: "Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)"
source: "https://aclanthology.org/S17-2017/"
doi: "10.18653/v1/s17-2017"
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/nagoudi-2017-lim-lig-semeval-2017-task1/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/nagoudi-2017-lim-lig-semeval-2017-task1/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/4/) / LIM-LIG at SemEval-2017 Task1: Enhancing the Semantic Similarity for Arabic Sentences with Vectors Weighting

# LIM-LIG at SemEval-2017 Task1: Enhancing the Semantic Similarity for Arabic Sentences with Vectors Weighting

**Authors:** [El Moatez Billah Nagoudi](https://md.research.elamur.ai/free/en/authors/el-moatez-billah-nagoudi/), Jérémy Ferrero and [Didier Schwab](https://md.research.elamur.ai/free/en/authors/didier-schwab/) · **Year:** 2017 · **Venue:** [Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2017](https://md.research.elamur.ai/free/en/venues/semeval/2017/)) · **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/S17-2017/) · [PDF](https://aclanthology.org/S17-2017.pdf) · [DOI 10.18653/v1/s17-2017](https://doi.org/10.18653/v1/s17-2017) · [OpenAlex](https://openalex.org/W2626047032)

**Citations:** 17 citations (OpenAlex, 2026-10-14) · [4](https://md.research.elamur.ai/free/en/papers/nagoudi-2017-lim-lig-semeval-2017-task1/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper tackles the first Arabic monolingual track of the SemEval-2017 Semantic Textual Similarity task. Systems must give each pair of Arabic sentences a graded score from 0 to 5 that agrees with human judgements. The LIM-LIG system builds a sentence vector by summing word embeddings from a pretrained Arabic CBOW model and compares two sentences with cosine similarity. The authors tested three ways of weighting the summed vectors: no weights, IDF weights estimated on a background news corpus, and weights tied to each word's part of speech, tuned on the task's training data. Before scoring, the text is normalized: stop words, punctuation and diacritics are removed, letter variants are unified, and digits are mapped to a placeholder. POS weighting was their best official run and placed 2nd in the Arabic track. A version built after the competition multiplies IDF and POS weights together and does better still. The work shows that cheap, interpretable salience weights can substantially improve plain embedding averaging for a lower-resource language.

## Key points
- The word vectors come from the Arabic CBOW model of Zahran et al. (2015), trained on over 5.8 billion words. Sources include Arabic Wikipedia, the BBC and CNN Arabic corpus, MultiUN, KSU corpus, Meedan, Osac and Arabase.
- IDF is computed per sentence rather than per document, as log(S / W_S), using the BBC and CNN Arabic corpus as background.
- The POS tagger is that of G. Braham et al. (2012). An illustrative weight set is proper noun 0.7, noun 0.5, verb 0.4, adjective 0.3 and preposition 0.1, with the actual weights set empirically on the SemEval-2017 Task 1 training data.
- Pearson correlation on the 250 pairs of the STS 2017 Arabic evaluation set:
  - no weighting: 0.5957
  - IDF: 0.7309
  - POS: 0.7463 (ranked 2nd)
  - mixed IDF×POS, post-competition: 0.7667, higher than every submitted Arabic run.
- Future work proposed by the authors: combining these methods with n-gram, fingerprint and linguistic-resource techniques.

## Interesting ideas
- Weighting each word by its grammatical category, with proper nouns and nouns highest and prepositions lowest, is a simple and transparent way to say which words carry a sentence's content.
- Two independent salience signals, a corpus rarity score (IDF) and a grammatical-role score (POS), are combined by multiplying them, and the product beats either signal alone.
- Category weights are tuned on the task's training data, while the rarity statistics come from a separate, genre-matched background corpus counted at sentence level.
- The unweighted embedding sum is reported next to the weighted variants, which shows exactly how much each added component contributes.
- Before comparison, digits are mapped to a single placeholder token and spelling variants of letters are unified. Surface variation that carries no meaning then cannot lower the similarity score.

## For Mova

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

- **I1** (world): Category-based salience weights could rank which entities and predicates matter most when matching a question's world against a text's world, and the ranking stays explainable.
- **I2** (answer): When a reader scores candidate answer sentences, a word's rarity and its role in the frame could be multiplied to give one auditable relevance weight.
- **I3** (learn): Mova could likewise estimate lexical rarity from an in-genre corpus such as fables or school texts instead of generic web data, which keeps the statistics matched to the domain being evaluated.
- **I4** (learn): A weighted vector-sum similarity is a ready-made bag-of-words baseline or negative control, useful for showing that Mova's tree- and frame-based comparisons actually gain over pure lexical overlap.
- **I5** (tokens): A NUM placeholder at the lemma level, with the original value kept for later use in the world model, could keep numeric variation from affecting lexical matching.

*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

- [Université Grenoble Alpes](https://md.research.elamur.ai/free/en/universities/universite-grenoble-alpes/) — University, Saint-Martin-d'Hères, France · 20 works · 8,059 citations
- [Laboratoire d'Informatique de Grenoble](https://md.research.elamur.ai/free/en/labs/laboratoire-d-informatique-de-grenoble/) — Research institute, Grenoble, France · 21 works · 688 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: El Moatez Billah Nagoudi, Jérémy Ferrero and Didier Schwab. 2017. LIM-LIG at SemEval-2017 Task1: Enhancing the Semantic Similarity for Arabic Sentences with Vectors Weighting. Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017).

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

```bibtex
@inproceedings{nagoudi2017lim,
  title = {{LIM-LIG at SemEval-2017 Task1: Enhancing the Semantic Similarity for Arabic Sentences with Vectors Weighting}},
  author = {El Moatez Billah Nagoudi, Jérémy Ferrero and Didier Schwab},
  booktitle = {Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)},
  year = {2017},
  doi = {10.18653/v1/s17-2017},
  url = {https://doi.org/10.18653/v1/s17-2017}
}
```

- [← previous Exploring English Teachers' attitudes towards Pronunciation issues and varieties of English in three circles of world Englishes](https://md.research.elamur.ai/free/en/papers/khatib-2017-exploring-english-teachers-attitudes-towards/)
- [next → Revisiting the ISO standard for dialogue act annotation](https://md.research.elamur.ai/free/en/papers/bunt-2017-revisiting-iso-standard-dialogue-act/)
- [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/4/)

## References

17 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Hazem M Raafat, Mohamed A Zahran, and Mohsen Rashwan. 2013. Arabase-a database combining different arabic resources with lexical and semantic information. In KDIR/KMIS, pages 233–240. [find](https://scholar.google.com/scholar?q=Hazem+M+Raafat%2C+Mohamed+A+Zahran%2C+and+Mohsen+Rashwan%2E+2013%2E+Arabase%2Da+database+combining+different+arabic+resources+with+lexical+and+semantic+information%2E+In+KDIR%2FKMIS%2C+pages+233%E2%80%93240%2E)
2. Eneko Agirre, Carmen Baneab, Claire Cardiec, Daniel Cerd, Mona Diabe, Aitor Gonzalez-Agirrea, Weiwei Guof, Inigo Lopez-Gazpioa, Montse Maritx-alara, Rada Mihalceab, et al. 2015. [Semeval-2015 task 2: Semantic textual similarity, english, spanish and pilot on interpretability](https://md.research.elamur.ai/free/en/papers/agirre-2015-semeval-2015-task-2-semantic/). In Proceedings of the 9th international workshop on semantic evaluation (SemEval 2015), pages 252–263.
3. Motaz K Saad and Wesam Ashour. 2010. Osac: Open source arabic corpora. In 6th ArchEng Int. Symposiums, EEECS, volume 10. [↗](https://doi.org/10.13140/2.1.4664.9288)
4. Gerard Salton and Christopher Buckley. 1988. Term-weighting approaches in automatic text retrieval. Information processing & management, 24(5):513– 523. [↗](https://doi.org/10.1016/0306-4573(88)90021-0)
5. Yu Chen and Andreas Eisele. 2012. Multiun v2: Un documents with multilingual alignments. In LREC, pages 2500–2504. [find](https://scholar.google.com/scholar?q=Yu+Chen+and+Andreas+Eisele%2E+2012%2E+Multiun+v2%3A+Un+documents+with+multilingual+alignments%2E+In+LREC%2C+pages+2500%E2%80%932504%2E)
6. Didier Schwab. 2005. Approche hybride-lexicale et thématique-pour la modélisation, la détection et lex-ploitation des fonctions lexicales en vue de lanalyse sémantique de texte. Ph.D. thesis, Université Montpellier II. [find](https://scholar.google.com/scholar?q=Didier+Schwab%2E+2005%2E+Approche+hybride%2Dlexicale+et+the%CC%81matique%2Dpour+la+mode%CC%81lisation%2C+la+de%CC%81tection+et+lex%2Dploitation+des+fonctions+lexicales+en+vue+de+lanalyse+se%CC%81mantique+de+texte%2E+Ph%2ED%2E+thesis%2C+Univ)
7. Ronan Collobert and Jason Weston. 2008. A unified architecture for natural language processing: Deep neural networks with multitask learning. In Proceedings of the 25th international conference on Machine learning, pages 160–167. ACM. [find](https://scholar.google.com/scholar?q=Ronan+Collobert+and+Jason+Weston%2E+2008%2E+A+unified+architecture+for+natural+language+processing%3A+Deep+neural+networks+with+multitask+learning%2E+In+Proceedings+of+the+25th+international+conference+on+Mac)
8. Jörg Tiedemann. 2012. Parallel data, tools and interfaces in opus. In LREC, volume 2012, pages 2214– 2218. [find](https://scholar.google.com/scholar?q=Jo%CC%88rg+Tiedemann%2E+2012%2E+Parallel+data%2C+tools+and+interfaces+in+opus%2E+In+LREC%2C+volume+2012%2C+pages+2214%E2%80%93+2218%2E)
9. Jérémy Ferrero, Frédéric Agnès, Laurent Besacier, and Didier Schwab. 2017. Using Word Embedding for Cross-Language Plagiarism Detection. In European Association for Computational Linguistics (EACL), Volume ”short papers” EACL 2017, Valence, Spain, April. [find](https://scholar.google.com/scholar?q=Je%CC%81re%CC%81my+Ferrero%2C+Fre%CC%81de%CC%81ric+Agne%CC%80s%2C+Laurent+Besacier%2C+and+Didier+Schwab%2E+2017%2E+Using+Word+Embedding+for+Cross%2DLanguage+Plagiarism+Detection%2E+In+European+Association+for+Computational+Linguistics+%28EAC)
10. Joseph Turian, Lev Ratinov, and Yoshua Bengio. 2010. [Word representations: a simple and general method for semi-supervised learning](https://md.research.elamur.ai/free/en/papers/turian-2010-word-representations-simple-general-method/). In Proceedings of the 48th annual meeting of the association for computational linguistics, pages 384–394. Association for Computational Linguistics. ksucorpus. 2012. King saud university corpus, [http://ksucorpus.ksu.edu.sa/ar/](http://ksucorpus.ksu.edu.sa/ar/) (accessed january 20,2017). [↗](http://ksucorpus.ksu.edu.sa/ar/)
11. WikiAr. 2006. Arabic wikipedia corpus, [http://linguatools.org/tools/corpora/wikipedia-monolingual-corpora/](http://linguatools.org/tools/corpora/wikipedia-monolingual-corpora/), (accessed january 21,2017). [↗](http://linguatools.org/tools/corpora/wikipedia-monolingual-corpora/)
12. Christina Lioma and Roi Blanco. 2009. Part of speech based term weighting for information retrieval. In European Conference on Information Retrieval, pages 412–423. Springer. [find](https://scholar.google.com/scholar?q=Christina+Lioma+and+Roi+Blanco%2E+2009%2E+Part+of+speech+based+term+weighting+for+information+retrieval%2E+In+European+Conference+on+Information+Retrieval%2C+pages+412%E2%80%93423%2E+Springer%2E)
13. Meedan. 2012. Meedan’s open source arabic english, [https://github.com/anastaw/meedan-memory](https://github.com/anastaw/meedan-memory), (accessed january 20,2017). [↗](https://github.com/anastaw/meedan-memory)
14. Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013a. [Efficient estimation of word representations in vector space](https://md.research.elamur.ai/free/en/papers/mikolov-2013-efficient-estimation-word-representations-vector/). In In: ICLR: Proceeding of the International Conference on Learning Representations Workshop Track, pages 1301–3781. [↗](https://doi.org/10.48550/arxiv.1301.3781)
15. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013b. [Distributed representations of words and phrases and their compositionality](https://md.research.elamur.ai/free/en/papers/mikolov-2013-distributed-representations-words-phrases-their/). In Advances in neural information processing systems, pages 3111–3119. [↗](https://doi.org/10.48550/arxiv.1310.4546)
16. Tomas Mikolov, Wen-tau Yih, and Geoffrey Zweig. 2013c. [Linguistic regularities in continuous space word representations](https://md.research.elamur.ai/free/en/papers/mikolov-2013-linguistic-regularities-continuous-space-word/). In Hlt-naacl, volume 13, pages 746–751.
17. Andriy Mnih and Geoffrey E Hinton. 2009. A scalable hierarchical distributed language model. In D. Koller, D. Schuurmans, Y. Bengio, and L. Bottou, editors, Advances in Neural Information Processing Systems 21, pages 1081–1088. Curran Associates, Inc. [find](https://scholar.google.com/scholar?q=Andriy+Mnih+and+Geoffrey+E+Hinton%2E+2009%2E+A+scalable+hierarchical+distributed+language+model%2E+In+D%2E+Koller%2C+D%2E+Schuurmans%2C+Y%2E+Bengio%2C+and+L%2E+Bottou%2C+editors%2C+Advances+in+Neural+Information+Processing+S)

*Updated 1 November 2026*