I'm a researcher, developer, and PhD candidate at the Faculty of Translation and Interpreting (FTI), University of Geneva 🇨🇭. I specialize in multilingual machine translation for medical communication, with a focus on low-resource languages, human evaluation, and the use of large language models.
- 🏥 Machine Translation for Medical Interactions
- 🌍 Low-Resource and Multilingual NLP
- 🤖 Large Language Models (LLMs), In-Context Learning
- 🧾 Human Evaluation & Semantic Metrics
- 🔁 Instruction Tuning, Domain Adaptation, RAG
My dissertation is titled:
"Evaluating Large Language Models for Low-Resource Multilingual Machine Translation in Medical Interactions"
As part of the PictoDr and BabelDr projects, I evaluate and build translation systems for over 140 language combinations — including translation into pictographs — to support communication between healthcare providers and patients.
- Frameworks: Hugging Face Transformers, MarianNMT, NLLB, Mistral, OpenNMT, Pytorch, Moses
- Techniques: Multilingual MT, domain adaptation, in-context learning, instruction tuning
- Evaluation: BLEU, ChrF, COMET, concept-level F1 (UMLS), human evaluation (adequacy, fluency, usability)
- Retrieval & Representation: BM25, LASER, UMLS-based concept mapping, gloss alignment, giza++
- Cumulative Link Mixed Models (CLMM)
- Fleiss’ Kappa (inter-annotator agreement)
- Regression, ANOVA, correlation analysis
- Likert scale design and analysis
- Languages: Python, PHP, JavaScript, Bash
- Core Libraries: PyTorch, NumPy, Pandas, Scikit-learn, Matplotlib, Seaborn
- NLP & ML: SentencePiece, Tokenizers, Accelerate, datasets
- Security: bcrypt, Argon2, PHPMailer
- GDPR compliance: consent, encryption, data retention
- Web & backend dev (PHP, MySQL, custom forms)
- Git, GitHub Actions, Docker (basic usage)
- Windows sever R12 and Ubuntu server
- LaTeX (custom
Thesis.cls, TikZ, BibTeX) - Markdown, Overleaf
- Reviewing: CLIN32, Languages and Resources in Springer Nature 2024, COLING 2024, AT4SSL, MT Summit 2025
- Built a multilingual MT training pipeline with multitask validation and biomedical vocabulary adaptation
- Conducted large-scale human evaluations of medical translation (Arabic, Spanish, Farsi, Albanian, Tigrinya…)
- Released models and data via Hugging Face: Models)
- 🌐 jonathanmutal.github.io
- 📧 jonathan.mutal [at] unige.ch
- 🐙 Explore my projects below ↓
Thanks for visiting! 👋