東北大学 助教 Assistant professor, Tohoku University. Also in charge of Dharmamitra in collaboration with BAIR, UC Berkeley. Research in ancient Asian languages.
Sendai, Japan
Joined November 2020
- Tweets2.2K
- Following1.8K
- Followers7.2K
- Likes5.7K
Pinned Tweet
Thrilled to share that I will be joining Tohoku University in Sendai, Japan, as tenure track assistant professor from autumn this year. Happy to work on AI for Sanskrit, Chinese, Tibetan, Pali, and Japanese, and to educate new generation of students who can work with these tools.
New preprint: Nehrdich, S., Harimoto, K., Sellmer, S., & Hellwig, O. “MITRA-lexicon: Lexica of Attested Translations from Sanskrit into English, Tibetan and Chinese”. A family of corpus-grounded Sanskrit-English, Sanskrit-Tibetan, and Sanskrit-Chinese lexica with 2.18 million headword entries and 10.4 million attestations, each linked to its sentence pair.
arxiv.org/abs/2610.25934 This is one of the resources that I found incredibly satisfying to create over the recent years. A web aplication is avialable at lexicon.dharmamitra.org !
Sebastian Nehrdich retweeted
📢 [🌎] 国際ワークショップ＆シンポジウム「AIとデジタル・ヒューマニティーズ」開催のお知らせ
日本国内の大学に所属する大学院生・ポスドク・助教が対象となります。関心のある方は積極的にご参加ください。
主催：ハーバード燕京研究所、Dharmamitra Project、東北大学統合日本学センター
🤖 Made with AI
Sebastian Nehrdich retweeted
カリフォルニア大学バークレー校で中国仏教枠の准教授または助教授の公募が一昨日から始まっています．East Asian Buddhism とありますが，求められているエリアは中国です．
aprecruit.berkeley.edu/JPF05…
Sebastian Nehrdich retweeted
Humanitextプロジェクトが、米国 Schmidt Sciences の国際研究プログラム「HAVI」に採択されました。42か国・501件の応募から19件の一つです。
AIが古典文献どうしの引用・言い換え・批判の候補を出し、専門家が一件ずつ確かめて、根拠をたどれる知識グラフにします。
humanitext.ai/ja/news/havi-2…
Sebastian Nehrdich retweeted
The updated Dharmamitra emacs package is now on MELPA: melpa.org/#/dharmamitra
Sebastian Nehrdich retweeted
New Release: We have the Dharmamitra Lexicon releases for Tibetan and Sanskrit now as standalone Stardict and Apple Dictionary files for free download! github.com/dharmamitra/dharm… 1/
Sebastian Nehrdich retweeted
🎊#受賞 #お知らせ
統合日本学センターのニアディッヒ・セバスチャン助教のプロジェクト「DharmaMitra」がAming Tu Prizeを受賞
#東北大学 #大学
⬇️詳細はこちら
tohoku.ac.jp/japanese/2026/0…
Sebastian Nehrdich retweeted
We just added to the Dharmamitra lexicon: Sanskrit ↔ English terms with 584k Sanskrit headwords (311k of them compounds & phrases), 1.13M sense pairs, 3.47M cited attestations from 402k aligned sentences across 54 texts!
The Dharmamitra team in 2037 when finding another edition of the Nīlamata Purāṇa that nobody digitized before
Sebastian Nehrdich retweeted
☺️ Glad to share my recently published paper on the “new translation” style of Xuanzang 玄奘 (ca. 602–664), a great Buddhist translator and cultural diplomat of the Tang dynasty.
To me, he is definitely a hard-rock star. ⭐️
Read the article here:
academia.edu/resource/work/1…
For a while there is this Fojin suite of tools which seems to repackage dharmamitra parallel aligned data and translation capabilities! I actually think this is a good thing, as it contributes to knowledge and the author is transparent with where the data comes from.
Sebastian Nehrdich retweeted
The MITRA team is very happy to announce that we got two research papers accepted at WMT2026:
1. Mitrasaṃgraha: the largest Sanskrit–English parallel corpus to date with 391k sentence pairs
2, MITRA: a large-scale parallel corpus and 9B translation model for Buddhist languages
AI brings a lot of good things, but it also leads to a massive devaluation of knowledge. This is a fundamental shift in what we value as humans in other humans.
Gemini 3.7 Flash is another Gemini release that doesn't improve in performance on Tibetan/Sanskrit/Chinese, so 3-flash-preview remains the best flash model in that lineage for now. They are getting more expensive with almost each release, and from 2027 onwards significantly so!
Dharmamitra is moving back to self-hosted models, and this makes me incredibly proud! Its not just better for the quality of the system, it is also better for privacy and means that we will be able to provide consistent high-quality service in the future as well. 1/
Google Gemini, as well as Claude Opus, which used to be the best-performing commercial models for classical Asian language translation, have been rather inconsistent over the last year, and recent updates brought performance stagnation or even decrease with rising API costs. 2/