Jonne Sälevä j0ma

🌍 Information

Actively seeking AI/ML Engineer and research-oriented industry roles starting in 2026.
Passionate about projects in multilinguality and low-resource languages!

Multilingual AI & NLP
- Bridging the digital divide for all 7,000+ languages of the 🌎, of which only ~100 are supported by AI
- Focus on multilingual representation learning, leveraging external knowledge, and exploring data augmentation
- Developing high-quality resources and evaluation sets for as many languages as possible
Model Evaluation
- LLM evaluation by investigating compositional generalization capability of LLMs using morphological reasoning
- Uncertainty quantification in multilingual AI/NLP leaderboards using resampling and statistical design

reuben: REsampling Uncertainty Bounds for Evaluating NLP
- CLI tool for uncertainty-aware performance evaluation of NLP models
- Useful for leaderboards and multi-task/multi-lingual evaluation
- Links: repo and latest release

2025
- From tests to effect sizes: Quantifying uncertainty and statistical variability in multilingual and multitask NLP evaluation benchmarks (Sälevä et al, AACL 2025)
- OpenNER 1.0: Standardized Open-Access Named Entity Recognition Datasets in 50+ Languages (Palen-Michel et al @ EMNLP 2025)
- Evaluating Morphological Compositional Generalization in Large Language Models (Ismayilzada et al @ NAACL 2025)
2024
- Language Model Priors and Data Augmentation Strategies for Low-resource Machine Translation (Sälevä & Lignos @ ACL Findings 2024)
- ParaNames 1.0: A Multilingual Entity Name Corpus (Sälevä & Lignos @ LREC-COLING 2024)
- First place in the VarDial 2024 DSL-ML Shared Task 🥇 (Sälevä & Palen-Michel @ VarDial 2024)
2023
- What changes when you randomly choose BPE merge operations? Not much. (Sälevä & Lignos @ Insights Workshop @ EACL 2023) - Best Paper Award ✨

For more, see ACL Anthology