Israel Abebe | እስራኤል አበበ

@SIC_Saar | Multimodal Learning (Vision +NLP) | Software Developer | #MachineLearning | 🇪🇹

[AfricaNLP @Africa_NLP]

🚨 AfricaNLP 2026 First Call for Paper

The 7th AfricaNLP Workshop at EACL 2026 (Rabat, Morocco) invites papers on Multilingual Multimodal LLMs for African languages.

👉 Submission deadline: December 19 2025 (AoE)
Details: https://t.co/aBDcSRcJxJ

#AfricaNLP #EACL2026 #NLP

[David Ifeoluwa Adelani 🇳🇬 @davlanade]

Join my lab! I’m currently recruiting 1-2 PhD students for admission in the fall of 2026 at
@Mila_Quebec

https://t.co/kOGpVOSTGB

Are you interested in multilingual NLP + VLMs / AI Safety? I would encourage you to apply.
Deadline: December 1

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

CaMMT: Benchmarking Culturally Aware Multimodal Machine Translation

Can images provide cultural grounding for translation? Does adding visual context boost multimodal MT performance?

paper - https://t.co/mfXvaD5jLm

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

Check out the following works we have accepted in #EMNLP2025

🧵

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

AfroXLMR-Social: Adapting Pre-trained Language Models for African Languages Social Media Text

Exploring which pre-training strategies matter most for improving downstream tasks on African social media data.

paper - https://t.co/S4zM9Ser0s

[David Ifeoluwa Adelani 🇳🇬 @davlanade]

5 papers accepted at #EMNLP2025: pp below

- SSA-COMET https://t.co/jdAuAUQHqL @SenyuLiReal

- AfriDOC-MT https://t.co/mPF1J2b3P9 &
African NLP survey https://t.co/4kejiqCHeB @alabi_jesujoba

- TransWebLLM https://t.co/8ESodpEZ2Q

- AfroXLMR-Social https://t.co/qh08yh4Cx0

[Natural Language Processing Papers @HEI]

Bridging the Culture Gap: A Framework for LLM-Driven Socio-Cultural Localization of Math Word Problems in Low-Resource Languages. https://t.co/genZvu7DGv

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

Working on LLMs is exciting—until you're the one stuck relying on one to handle your customer service issue.

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

Our work has been accepted to the #EMNLP2024 Findings! 🎉

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

This week I will be presenting our work "Walia-LLM: Enhancing Amharic-LLaMA by Integrating Task-Specific and Generative Datasets" at 5th #AfricanNLP Workshop at @iclr_conf .

[EquianoAI @EquianoAI]

You are invited to this workshop we are organising on "How Good are AI Language Models in Global Languages?" on 2 July at @KingsCollegeLon for @londondataweek. The aim of the workshop is to work with AI researchers to define global AI benchmarks and evaluation methods.

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

I watched this being built 😁

[Ömercan Yazici @pearcoding]

The Computer Graphics chair at @Saar_Uni just finished building our own whiteboard plotter. We used seven @STAEMars pens. Wonder what more colors could do?
#cg #saarland

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

Gpt-4o vs Gemini 1.5
https://t.co/Tj04k00ntk

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

wake up 🌍 "human-translated benchmark dataset for 16 typologically-diverse low-resource African languages covering three tasks" just dropped 🥳

[David Ifeoluwa Adelani 🇳🇬 @davlanade]

I'm very happy to share our new benchmark, IrokoBench--a human translated benchmark dataset for 16 African languages covering:
- NLI (AfriXNLI)
- Maths reasoning (AfriMGSM)
- Multi-choice QA (AfriMMLU)

https://t.co/wbk4L5cGzs

Project funded by @LacunaFund with @MasakhaneNLP

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

👀

[Yann LeCun @ylecun]

If you are a student interested in building the next generation of AI systems, don't work on LLMs

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

Checkout our work https://t.co/Fjb7w8Pk5S @y_mitiku @AtnafuLambebo @tadesseit @amaniopia Eyasu Shiferaw Yonas Chanie @WalelignS @seyyaw

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

We showed that we can improve performances by only integrating these datasets. We also evaluated the performance of GPT- 4 for this tasks.

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

Additionally we explored the MT performance of these models comparing them to SOTA translation models.

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

To avoid instruction template overfitting we collected instruction templates for each task and worked on a pipeline to generate alpaca format dataset .

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

This week I will be presenting our work "Walia-LLM: Enhancing Amharic-LLaMA by Integrating Task-Specific and Generative Datasets" at 5th #AfricanNLP Workshop at @iclr_conf .

[Israel Abebe | እስራኤል አበበ @Israel_Abebe]

In this work we created new instruction datasets from task specific datasets and also worked on collecting new instruction dataset to improve creative writing.