Computational linguist at Linköping University — multilingual language models, interpretability, and Swedish NLP
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
22
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Publication-volume composites are not yet computable from this snapshot, so the axis is unscored. It would measure total and recent output — works per year, venue mix, and authorship roles across indexed sources.
Impact—
Citation-impact composites are unavailable in this snapshot, so the axis is left unscored. It would measure citation-based influence — citations per work, field-normalized rates, and landmark-paper signals — rather than raw counts alone.
Focus100/100
The focus axis scores 100/100: every classified work falls in a single field, computational linguistics (multilingual NLP interpretability). The topic profile agrees — natural language processing techniques (101 works) and syntax, semantics and linguistic variation (80) dominate, with semigroups and automata theory (54) as a formal undercurrent, and all three 2026 catalog papers target multilingual language models.
Collaboration39/100
Collaboration scores 39/100. The recent window shows only 6 distinct coauthors, and across the three reviewed 2026 papers he appears as last author twice and middle author once, never flagged corresponding. Recurring partners include Jenny Kunz, Marcel Bollmann, Lukas Borggren, Oskar Holmström, Romina Oji and Marc Braun — a small, stable cluster rather than a large distributed network.
Activity77/100
Activity scores 77/100. The OpenAlex counts-by-year record shows continuous publishing from 2001 to 2026, peaking at 19 works in 2010, with a steady recent rhythm of 4 works in 2024, 5 in 2025 and 4 in 2026 — 13 works over the last three years, a cadenced rather than sporadic record.
Reproducibility55/100
Reproducibility scores 55/100, a mean derived from one Evolve-reviewed paper (reproMean 55, reproPapers 1). Across the three reviewed preprints, readiness is medium for the interpretability comparison and the knowledge-transfer study, and low for the Swedish journalism continued pre-training, where proprietary data and models limit independent replication.
Research summary
Marco Kuhlmann is a computational linguistics researcher at Linköping University (Sweden) whose OpenAlex record spans 2001–2026 with an h-index of 22 and an i10-index of 36. His topic profile is anchored in natural language processing techniques (101 works) and syntax, semantics and linguistic variation (80), with a substantial formal strand in semigroups and automata theory (54), plus topic modeling (38) and early explainable-AI work (9). The three papers Evolve has reviewed so far — all 2026 arXiv preprints — define a coherent current agenda around multilingual language models: a last-author systematic comparison of four representation-sharing metrics across 21 models showing anisotropy-driven failures of interpretability methods; a study using training-data interventions to probe the limits of factual knowledge transfer from English into another language; and a continued pre-training effort adapting large language models to Swedish journalism on an 8.5-billion-token corpus with a new six-task benchmark. His focus dimension scores 100/100 on a single classified field, activity 77/100 on 13 works over the last three years with only 6 distinct recent coauthors, and reproducibility 55/100 from a single reviewed paper, while output and impact composites remain unscored for lack of data.
Collaboration pattern
The recent record shows a compact collaboration footprint: 6 distinct coauthors in the current window, with a stable recurring circle of Jenny Kunz, Marcel Bollmann, Lukas Borggren, Oskar Holmström, Romina Oji and Marc Braun. On the three 2026 catalog papers he appears as last author twice and middle author once, never flagged corresponding — a senior-supervisor pattern on small, Linköping-centered teams.
Research keywords
Highlights
Systematic comparison of multilingual interpretability metrics reveals anisotropy-driven failures
Training-data interventions probe the limits of cross-lingual factual transfer
Adapting large language models to Swedish journalism via continued pre-training
Timeline
- 2001Earliest year in the OpenAlex counts-by-year record, with 1 work — the start of a publication span running continuously through 2026.
- 2010Most productive year on record: 19 works, the highest single-year count in the 2001–2026 series.
- 2024Recent output settles into a steady cadence: 4 works in 2024, 5 in 2025 and 4 in 2026, totaling 13 works over the last three years.
- 2026Three arXiv preprints enter the Evolve catalog — a multilingual interpretability comparison, a factual-transfer intervention study, and Swedish journalism continued pre-training — with last authorship on two of the three.
Papers in the Evolve catalog
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.
- Bibliometrics are drawn from OpenAlex only; the citation-level data behind output and impact composites is unavailable, so those two radar axes are unscored.
- h-index (22) and i10-index (36) are quoted as provided and are not field- or career-stage-normalized.
- Only three Evolve-reviewed papers exist, all 2026 arXiv preprints; the reproducibility score is a mean over a single reviewed paper.
- Profile is name-resolved from public bibliographic data (OpenAlex); confirm identity against ORCID 0000-0002-2492-9872.