Nedjma Ousidhoum
LightNews LightNews
banner
nedjmaou-nlp.bsky.social
Nedjma Ousidhoum
@nedjmaou-nlp.bsky.social
120 followers 71 following 47 posts
#NLP #NLProc Lecturer (Assistant Professor) at Cardiff University.
http://nedjmaou.github.io
Posts Replies Media Videos
nedjmaou-nlp.bsky.social
Pleased to announce our #NLP SemEval Task 7: Everyday Knowledge Across Diverse Languages & Cultures.
We extend the BLEnD Benchmark to >30 language-culture pairs. [Our task is Junior-friendly, with live Q&A & tutorials.] 1/
September 22, 2025 at 11:14 AM
nedjmaou-nlp.bsky.social
And our SemEval task won the best task award! 🥳
Co-led with Shamsuddeen Muhammad and Saif M. Mohammad. Thanks to all the team members!
August 1, 2025 at 7:25 AM
nedjmaou-nlp.bsky.social
Our ✨BRIGHTER✨ paper won the a best resource paper award at @aclmeeting.bsky.social #ACL2025!
Big thanks to all the team members!
July 30, 2025 at 6:35 PM
nedjmaou-nlp.bsky.social
And ‪Naba Rizvi will be presenting:
📄"AUTALIC: A Dataset for Anti-AUTistic Ableist Language In Context"
🔗 aclanthology.org/2025.acl-lon...
📍 Session 3: IP-Orals – Resources and Evaluation 1
🗓️ Monday, July 28 🕑 14:00–15:30📍 Hall A
3/
July 27, 2025 at 9:44 AM
nedjmaou-nlp.bsky.social
Also presenting:
📄BRIGHTER: BRIdging the Gap in Human-Annotated Textual Emotion Recognition Datasets for 28 Languages
🔗 aclanthology.org/2025.acl-lon...
📍 Session 9: IP-Orals – Resources and Evaluation 2
🗓️ Tue, July 29 🕑 14:00–15:30📍 Hall B
2/
July 27, 2025 at 9:44 AM
nedjmaou-nlp.bsky.social
At #ACL2025 #ACL2025NLP In Vienna this week!
I’ll be presenting:
📄 Building Better: Avoiding Pitfalls in Developing Language Resources when Data is Scarce
🔗 aclanthology.org/2025.acl-lon...
📍 Panel 5: Multilinguality & Language Diversity
🗓️ Wed, July 30 🕘 09:00–10:30📍 Hall N.2
1/
July 27, 2025 at 9:44 AM
nedjmaou-nlp.bsky.social
Looking forward to seeing many of you at our next edition and do say hi if you’re attending #ACL2025 in Vienna! 3/3
July 16, 2025 at 12:18 PM
nedjmaou-nlp.bsky.social
We hope you also enjoyed the tutorial and the poster session! 2/3
July 16, 2025 at 12:18 PM
nedjmaou-nlp.bsky.social
Thank you to everyone who came along to our @cardiffnlp.bsky.social #NLP workshop, it was a pleasure to have you all!
A big thank you to our amazing speakers, and panelists!
1/3
July 16, 2025 at 12:18 PM
nedjmaou-nlp.bsky.social
The Cardiff #NLP Workshop kicks off this Monday (14 July)!
Full schedule www.cardiffnlpworkshop.org/schedule
✨ Talks, a tutorial, poster session and networking opportunities, an interdisciplinary session, and a panel discussion✨
Confirmations were sent. Looking forward to welcoming you to Cardiff!
July 10, 2025 at 11:59 AM
nedjmaou-nlp.bsky.social
❓ Do they receive appropriate recognition and credit?
Findings:
- Only 33% consistently received credit for their contributions.
- Problematic incentives persist, such as suggesting that serving a language/community is compensation enough. 3/
June 2, 2025 at 1:53 PM
nedjmaou-nlp.bsky.social
We discuss:
❓Why do researchers choose to work on these languages?
Findings:
- Many contributors are motivated by a personal or community connection to the language. 2/
June 2, 2025 at 1:53 PM
nedjmaou-nlp.bsky.social
🚨#ACL2025 position paper on key challenges in creating resources for mid- to low-resource lgges emphasising that data in #NLP is not merely a collection of tokens, but is rooted in cultural and social contexts.
Based on 81 responses from researchers working on >70lgges
📄 arxiv.org/abs/2410.12691 1/🧵
June 2, 2025 at 1:53 PM
nedjmaou-nlp.bsky.social
Join us in July for the Cardiff #NLP Workshop by completing this form: forms.gle/TvqnzAbKVYQn... (before 11 April):
5 confirmed speakers (& more to come!), a tutorial, short talks, and a poster session www.cardiffnlpworkshop.org
The registration is ✨free✨ with some affordable accommodation options!
April 7, 2025 at 9:43 AM
nedjmaou-nlp.bsky.social
Check out the accepted papers from our lab at #NAACL2025 and #ICLR2025!
(Preprints in the alt text. we're also on BlueSky cardiffnlp.bsky.social)
Accepted papers at #ICLR2025 AND #NAACL2025 from the CardiffNLP group WORLDCUISINES: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines https://arxiv.org/abs/2410.12705 AfriHate: A Multilingual Collection of Hate Speech and Abusive Language Datasets for African Languages https://arxiv.org/abs/2501.08284 Systematic Relational Reasoning With Epistemic Graph Neural Networks https://arxiv.org/abs/2407.17396 "Adapting Sentence-level Automatic Metrics for Document-LevelSimplification Evaluation" SANDWiCH: Semantical Analysis of Neighbours for Disambiguating Words in Context ad HocPerCul: A Story-Driven Cultural Evaluation of LLMs in Persian
February 11, 2025 at 4:10 PM
nedjmaou-nlp.bsky.social
The eval phase of our #nlp SemEval shared task on emotion will start soon! We have 32 datasets (in many low-resource languages) & 3 tracks A)multilabel, B)intensity and C)cross lingual. We'll have a Q&A session with the participants next week & a writing tutorial in Feb github.com/emotion-anal... 1/
SemEval 2025 Task 11: Bridging the Gap in Text-Based Emotion Detection. New emotion datasets in more than 30 languages (notably, many from Africa and Asia) Translations: - I'm speechless - my boss has offered me an unexpected promotion! - Floods now threaten southern Mozambique. - It's our fault. For being women... Shitty world. - Please pray for me, I am in trouble. - This is a kind of violence against the WOMAN MINISTER DAMARES. Action must be taken. - Cannot be, but why? - After all why didn't you try to correct this mistake!! - I would once again like to convey my Department's deepest condolences to the parents and family of Nadine. - This morning, after putting on one sock, I felt something big and hard inside. I reached in to pull it out, and oh my god, it was a huge wasp! It was still alive, flapping its wings. Since it didn’t sting me, I let it go, but it was terrifying! - Omar was thinking: "France is a painting with many colours." How could this faraway land be his [motherland]? And his mother? His mother is at home and her name is Aini. He doesn't have two mothers. Aini is not France. Omar realised that this was a lie
January 7, 2025 at 10:51 AM
nedjmaou-nlp.bsky.social
Only 33% of the respondents consistently received credit for their work and 67% did not. This sometimes happened for work that took people more than a month to finish and often due to problematic incentivisation. 3/
Histogram on the time that took the respondents to work on a project when not properly compensated and the incentives for that. 36 responded community membership, 26 said because they were acknowledged on the paper/website and 7 were somehow manipulated. >40% of the time the work took more than a day and up to more than a month. Stacked bar chart on credit attribution: 26 respondents have always received credit/financial compensation for their contributions, 16 often, 15 sometimes, 5 rarely, and 18 never did.
December 17, 2024 at 10:19 AM
nedjmaou-nlp.bsky.social
Based on 81 responses from researchers working on >70 languages: people often work on mid- to low-resource lgges because they are interested in languages/CL/NLP/ML/... But also because they'd like to work on *their* own languages & the data may be scarce/non-representative/... 👇🏻2/
Bar chart on why do people work on NLP for mid- to low-resource languages? - (Interest in NLP/CL/ML, 68%) - (Interest in languages, 68%) - (LT for their own language(s), 60%) - (Limitations in available resources, 60%) - (LLM Research, 59%) - (LT for many languages, 38%) Bar chart on observed limitations in available resources for a respondent's language(s) of interest: - (Data scarcity, 78%) - (Data (non)representativeness, 57%) - (Tools do not perform well, 54%) - (Misalignment with needs, 34%) - (Annotation quality, 25%) - (Data limited usefulness, 18%)
December 17, 2024 at 10:19 AM
nedjmaou-nlp.bsky.social
#nlp/#nlproc preprint on how to build better datasets/tools for mid- to low-resource languages while respecting the labor of the data workers: arxiv.org/abs/2410.12691
We examine 1) *why* people work on mid- to low-resource languages and 2) whether they get properly credited.🧵1/
Figure 1 from our paper. The main themes and targets of our survey. It is designed for NLP researchers and practitioners who have worked on non-high-resource languages (data curation, annotation, and/or model construction). Some of the questions focus on the perspectives of the subset highlighted in the figure, i.e., speakers who focus on their own languages.
December 17, 2024 at 10:19 AM