Research notebook

Notes

A public notebook for paper-reading memos, research thoughts, and things I want to think through a little further. Mostly around NLP, AI, multilinguality, and evaluation.

How these notes are made

Most notes start as GenAI-assisted drafts from roughly 10–20 turns of my own paper discussions. I review, revise, and make the final publication decision.

Process details

For a typical paper, I use GenAI as a discussion partner to question interpretations, connect related work, and explore possible follow-up ideas from my own research interests and perspective.

The discussion log is then used to generate an initial draft. These notes are not raw model outputs or neutral paper summaries; the published version reflects my own interpretation, interests, and final editorial decisions.

Ideas are open invitations

Ideas at the end of notes are open invitations. Want to test one together? Contact me. Prefer to run with it yourself? Please do — and let me know what you find.

2026-09-01

Does Romanization Really Help Multilingual Transfer?

A short note on romanized multilingual pretraining, why it improves cross-script transfer, and why the mechanism may be more complicated than lexical overlap alone.

#paper-reading #multilingual #llm #cross-lingual-transfer #romanization #tokenization

2026-09-01

What Does Multilingual LLM Output Resemble?

A short note on translationese in multilingual LLM generation, and a question that may be more useful than asking whether a model "thinks in English."

#paper-reading #multilingual #llm #translationese

2026-09-01

Starting Notes

A small kickoff for this new space — and a first test of how I want to keep and share research notes.

#meta #research-notes