CNRS computational linguist: MT evaluation, multilingual tokenization, code-switch LID, preference inference
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: publication-volume metrics (paper counts, venues, years of coverage) are absent from the current bibliographic snapshot, so this axis cannot be computed. It would measure total and annual research output across indexed venues.
Impact—
Not yet scored: citation counts, h-index, and venue-level impact data are missing from this snapshot. The axis would measure citations to and uptake of the researcher's papers across the indexed literature.
Focus0/100
The focus score is 0/100 because the observed field distribution is perfectly uniform: computational linguistics, machine translation evaluation, tokenization for multilingual LLMs, and computational social science/NLP each hold 0.25. Within the four catalogued 2026 papers this reads as deliberate breadth — an evaluation-methods toolkit applied to four problem areas — rather than scattering, though a four-paper sample is too small to settle the question.
Collaboration56/100
Collaboration scores 56/100, with 16 distinct coauthors across the four catalogued papers and 12 recurring partners, including Joanna Radoła, Josep Maria Crego, Rachel Bawden, Ziqian Peng, Franck Signe, and Hippolyte Pilchen. Authorship positions — two middle, two last, none corresponding — suggest a senior-contributor pattern: joining project teams and anchoring papers at the end of the author list rather than leading them as corresponding author.
Activity—
Not yet scored: the current snapshot lacks year-by-year publication counts, so temporal productivity cannot be computed. This axis would measure publishing cadence and continuity — outputs per year and gaps — across the researcher's full indexed career.
Reproducibility78/100
Reproducibility is the strongest measured axis: a 78/100 mean across all four scored papers. Evolve rates the code-switched LID paper (ILP over MaskLID, well ablated) and the preference-inference benchmark paper (new metrics plus a 1M-vote multilingual dataset) as high reproducibility readiness; the tokenizer and terminology-evaluation papers are rated medium, the latter limited by small scale and absent significance tests.
Research summary
François Yvon is a researcher at the French National Centre for Scientific Research (CNRS) whose catalogued work sits at the intersection of computational linguistics, multilingual NLP, and computational social science. The Evolve catalog holds four 2026 arXiv papers spanning four distinct problem areas in equal measure: modular tokenization for multilingual LLMs, variation-aware terminology evaluation for machine translation, code-switched language identification reformulated as integer linear programming, and collective-centric evaluation of preference inference for participatory democracy. This perfectly even spread — each field holding 0.25 of the observed distribution — yields a focus score of 0/100 and frames Yvon as an evaluation-methods contributor rather than a single-track specialist. Authorship positions are consistently non-leading in the sample: two middle-author and two last-author slots, with no corresponding-author paper. The collaboration footprint is moderate at 56/100, built on 16 distinct coauthors and repeated work with Joanna Radoła, Josep Maria Crego, Rachel Bawden, and Ziqian Peng. Reproducibility is the strongest signal: a 78/100 mean across four scored papers, with the ILP-based language-identification paper (well ablated, 125 Latin-script languages) and the preference-inference benchmark (new metrics plus a million-vote multilingual dataset) rated high readiness, while the tokenizer and terminology papers sit at medium.
Collaboration pattern
Yvon's catalogued work is team-based: 16 distinct coauthors across four 2026 papers, with 12 recurring partners such as Joanna Radoła, Josep Maria Crego, Rachel Bawden, Ziqian Peng, Franck Signe, Hippolyte Pilchen, and Édouard Grave. No paper lists him as corresponding author; he appears twice as middle author and twice as last author, a pattern consistent with a senior researcher contributing methods and oversight across several parallel collaborations.
Research keywords
Highlights
Modular tokenizers give each language its own subword inventory
Variation-aware terminology evaluation for machine translation
Code-switched language identification as integer linear programming
Collective-centric evaluation of preference inference
Timeline
- 2026Published modular per-language tokenizers for efficient multilingual LLMs (arXiv, middle authorship).
- 2026Proposed variation-aware terminology evaluation (EM/TCR/CTV) for machine translation on EN–FR NLP corpora (arXiv).
- 2026Last-authored an ILP reformulation of code-switched language identification with large exact-match gains over 125 Latin-script languages.
- 2026Last-authored a collective-centric evaluation benchmark with new metrics and a 1M-vote multilingual dataset for preference inference.
Papers in the Evolve catalog
To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs
Improving Term Evaluation in Machine Translation: Variation Matters
Improving Language Identification for Code-Switched Utterances with Integer Linear Programming
Toward Collective-Centric Evaluation of Preference Inference for Participatory Democracy
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric dimensions are computed from public bibliographic data (OpenAlex-derived) only; citation counts, h-index, and venue-level impact metrics were not available.
- Only the four 2026 arXiv papers reviewed by Evolve are in scope; earlier career output is not represented in this profile.
- The output, impact, and activity radar axes are unscored because the underlying data is insufficient.
- Profile is name-resolved from public bibliographic records via ORCID 0000-0002-7972-7442; verify identity before citing career facts.