--- title: "Generating Answering Patterns from Factoid Arabic Questions" description: "The paper addresses factoid question answering for Arabic, a language with few QA systems and limited freely available tools. 2 references." authors: ["Essia Bessaies", "Slim Mesfar", "Henda Ben Ghezala"] year: 2017 venue: "Proceedings of the Linguistic Resources for Automatic Natural Language Generation - LiRA@NLG" source: "https://aclanthology.org/W17-3803/" doi: "10.18653/v1/w17-3803" 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/bessaies-2017-generating-answering-patterns-factoid-arabic/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/bessaies-2017-generating-answering-patterns-factoid-arabic/) [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/5/) / Generating Answering Patterns from Factoid Arabic Questions # Generating Answering Patterns from Factoid Arabic Questions **Authors:** Essia Bessaies, [Slim Mesfar](https://md.research.elamur.ai/free/en/authors/slim-mesfar/) and Henda Ben Ghezala · **Year:** 2017 · **Venue:** Proceedings of the Linguistic Resources for Automatic Natural Language Generation - LiRA@NLG · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/W17-3803/) · [PDF](https://aclanthology.org/W17-3803.pdf) · [DOI 10.18653/v1/w17-3803](https://doi.org/10.18653/v1/w17-3803) · [OpenAlex](https://openalex.org/W2789348537) **Citations:** 3 citations (OpenAlex, 2026-10-14) ## Summary The paper addresses factoid question answering for Arabic, a language with few QA systems and limited freely available tools. The authors split QA into three stages (question analysis, answer pattern generation, answer extraction) and cover only the first two, in the medical domain. Using the NooJ platform and its Arabic module, they extend dictionary entries with three properties: the nominal predicate, synonyms, and support verbs. A cascade of morpho-syntactic local grammars, together with a rule-based medical NER, labels each question with its expected answer Type, its Topic, and its Focus. A paraphrase and transformation component then uses these labels to produce many candidate answer sentences in nominal, verbal, prepositional, adverbial, active, and passive forms, combined with slots such as time expressions. On their data, NER reaches an F-measure of 0.88 and question annotation reaches 0.73. A single example question yields hundreds of patterns. The work matters because it shows how a hand-built lexical and grammar resource can anticipate the many surface forms an answer may take, which is useful when answer sentences are long and stylistically varied. ## Key points - The system is built in NooJ on top of the Arabic module (Mesfar, 2010). Lexical entries gain extra fields: NPred (nominal predicate), Syn (synonyms), and Sup (support verbs). For example, the verb "to discover" is linked to the noun "discovery", the synonym "to find", and support verbs such as "occur". - The medical NER (ENAMEX+MEDIC grammar) recognizes six kinds of entities: persons, organizations, locations, diseases, viruses, and treatments. Measured against manual extraction, it scores P 0.90, R 0.82, F 0.88. - Question annotation with the grammar cascade scores roughly P 0.75, R 0.72, F 0.73. Each question receives three labels: Type (time, quantity, person…), Topic (what the question is about), and Focus (the property of the topic being sought). - For the question "When did the discovery of AIDS occur?", the generator produces hundreds of answer-pattern combinations. Some valid forms are still not covered. - Errors come mainly from long, complex questions and from structures the grammars lack. Answer extraction and "why"/"how" questions are left for future work. ## Interesting ideas - Storing the nominal predicate, synonyms, and support verbs directly in each lexical entry lets one lexicon drive both analysis and paraphrase, and the variant forms come from explicit lexical links rather than from embeddings. - Rather than asking how to read the question, the system expands it into the set of declarative sentences that would answer it, across voices and phrase types, and then searches for those. - Splitting a question into an expected answer Type, a Topic, and a Focus gives a small, inspectable structure that guides later steps. - The generator combines a declarative core with a typed slot such as a Timex expression, so the answer slot has a known semantic type. - Each module is scored separately against manual extraction (NER on its own, then question annotation), and the authors note that downstream generation quality depends on upstream errors. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (world): Verb entries in the compiled knowledge could carry links to their nominalization and light-verb forms ("discover" ↔ "the discovery occurred"), so that "X discovered Y" and "the discovery of Y happened" map to the same event. - **I2** (answer): Fable and school-test readers could turn a question frame into expected answer frames (active, passive, nominalized) and match them against the world of the text, with each match explainable. - **I3** (frames): A question speech act could be represented as a frame with an open slot (the Type), a known entity (the Topic), and the queried property (the Focus), which tells the answer stage exactly which role to fill. - **I4** (back): When generating from the meaning graph, an unknown argument could be realized as a typed placeholder, giving templates that are useful both for answer verification and for testing the generator. - **I5** (learn): Evaluation could report per-stage scores and trace how often an answer failure starts in NER or tagging, so that repair rules target the actual source of the error. *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 - [Manouba University](https://md.research.elamur.ai/free/en/universities/manouba-university/) — University, Manouba, Tunisia · 5 works · 266 citations - [National School of Computer Science](https://md.research.elamur.ai/free/en/universities/national-school-of-computer-science/) — University, Manouba, Tunisia · 3 works · 259 citations 📚 Shelves (5): [🖥️ 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 · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works 🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works ## Cite Cite as: Essia Bessaies, Slim Mesfar and Henda Ben Ghezala. 2017. Generating Answering Patterns from Factoid Arabic Questions. Proceedings of the Linguistic Resources for Automatic Natural Language Generation - LiRA@NLG. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{bessaies2017generating, title = {{Generating Answering Patterns from Factoid Arabic Questions}}, author = {Essia Bessaies, Slim Mesfar and Henda Ben Ghezala}, booktitle = {Proceedings of the Linguistic Resources for Automatic Natural Language Generation - LiRA@NLG}, year = {2017}, doi = {10.18653/v1/w17-3803}, url = {https://doi.org/10.18653/v1/w17-3803} } ``` - [← previous CEVO: Comprehensive EVent Ontology Enhancing Cognitive Annotation](https://md.research.elamur.ai/free/en/papers/shekarpour-2017-cevo-comprehensive-event-ontology-enhancing/) - [next → Parallel Hardware for Faster Morphological Analysis](https://md.research.elamur.ai/free/en/papers/damaj-2019-parallel-hardware-faster-morphological-analysis/) - [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/) ## References 2 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Sman Bekhti and Maryam Alharbi. 2013. Aquasys: A question answering system for Arabic. In Proceedings of WSeas International Conference. Recent Advances in Computer Engineering Series, no. 12. WSEAS, pp 130-139 Yassine Benajiba, Paolo Rosso and Abdelouahid Lyhyaoui. 2007. Implementation of the ArabiQA Question Answering System’s components. In Proceedings of the Workshop on Arabic Natural Language Processing, 2nd Information Communication Technologies Int. Symposium, ICTIS-2007, Fez, Morroco, pp 3-5. [find](https://scholar.google.com/scholar?q=Sman+Bekhti+and+Maryam+Alharbi%2E+2013%2E+Aquasys%3A+A+question+answering+system+for+Arabic%2E+In+Proceedings+of+WSeas+International+Conference%2E+Recent+Advances+in+Computer+Engineering+Series%2C+no%2E+12%2E+WSEAS%2C+) 2. Wissal Brini, Mariem Ellouze, Slim Mesfar and Lamia Belguith. 2009. An Arabic question-answering system for factoid questions. In Natural Language Processing [find](https://scholar.google.com/scholar?q=Wissal+Brini%2C+Mariem+Ellouze%2C+Slim+Mesfar+and+Lamia+Belguith%2E+2009%2E+An+Arabic+question%2Danswering+system+for+factoid+questions%2E+In+Natural+Language+Processing) *Updated 1 November 2026*