Matej Martinc

Jožef Stefan Institute

Unclaimed · AI-structured profileORCID OpenAlex

NLP researcher benchmarking LLMs for readability assessment and political-bias evaluation at Jožef Stefan Institute

papers here

reviewed by Evolve

first-author

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 scored: the bibliometric snapshot provides no publication-volume totals for this researcher, so overall output cannot be computed. This axis would measure the number and cadence of papers across venues and years once a fuller publication record is available.

Impact—

Not scored: no citation counts or influence metrics are available in the input data, so impact cannot be estimated. This axis would measure citations, uptake and scholarly influence of the work once citation coverage is added.

Focus0/100

The focus score is 0/100, reflecting an exactly even 0.5/0.5 split in the field distribution between computational linguistics (NLP) and natural language processing / computational social science. Rather than scattered interests, the cataloged work is thematically coherent: both 2026 papers are systematic large-language-model evaluation studies, spanning readability assessment and political-bias measurement with a shared multilingual emphasis.

Collaboration36/100

The collaboration dimension scores 36/100, based on 5 distinct co-authors in the bibliometric snapshot — a compact, stable circle. Frequent collaborators include Raphaël Thieffry, Luka Debevc, Nishan Chatterjee, Antoine Doucet and Senja Pollak. On both reviewed 2026 papers Martinc appears in the last author position, consistent with a senior or supervising role within small research teams rather than large consortia.

Activity—

Not scored: the snapshot lacks year-by-year publication counts, so research tempo cannot be computed. This axis would measure recent publishing cadence and sustained engagement, though the two 2026 arXiv papers indicate current activity.

Reproducibility78/100

Reproducibility scores 78/100, computed as the mean across 2 Evolve-reviewed papers. The readability benchmark carries high reproducibility readiness — five open-weight LLMs and released protocols make replication straightforward — while the political-compass study is rated medium. Evolve's verdicts also flag fixable rigor gaps, such as missing significance testing, suggesting a generally transparent, artifact-friendly workflow.

Research summary

Matej Martinc is a researcher at the Jožef Stefan Institute working in natural language processing, with the bibliometric field distribution splitting evenly (0.5/0.5) between computational linguistics and the NLP/computational-social-science interface. Evolve's catalog currently contains two of his 2026 arXiv papers, on both of which he appears in the last author position, a placement consistent with a senior or supervising role. The first, 'Assessing Readability with LLMs', benchmarks five open-weight large language models on readability level prediction and finds that native reasoning and one-shot prompting help most; Evolve deems the findings credible while noting the lack of significance testing, and rates its reproducibility readiness high. The second, 'Navigating the digital spectrum', applies a rigorous multi-factor Political Compass Test across languages and designs, showing that LLM political coordinates are sensitive to prompting and persona choices; Evolve praises the methods but cautions that chat and downstream-fairness claims outrun the data. Across the two studies the mean reproducibility score is 78/100. The recorded collaboration footprint is compact — 5 distinct co-authors, including Raphaël Thieffry, Luka Debevc, Nishan Chatterjee, Antoine Doucet and Senja Pollak — yielding a collaboration score of 36/100. Output, impact and activity metrics are not yet available, so those radar axes remain unscored.

Collaboration pattern

The bibliometric snapshot records only 5 distinct co-authors, and the collaboration dimension scores 36/100 — a compact, stable circle of collaborators rather than large consortia. Frequent co-authors include Raphaël Thieffry, Luka Debevc, Nishan Chatterjee, Antoine Doucet and Senja Pollak. On both cataloged 2026 papers Martinc appears in the last author position, a placement consistent with a senior or supervising role within these small project teams.

Research keywords

Large language modelsLLM evaluation and benchmarkingReadability assessmentChain-of-thought promptingFew-shot in-context learningMultilingual evaluationLow-resource languagesPolitical bias measurementPolitical Compass TestPrompt sensitivityPersona steerabilityDownstream fairness

Highlights

Benchmarking readability assessment with five open-weight LLMs
The study evaluates five open large language models on readability level prediction, finding that native reasoning and one-shot prompting help most; Evolve deems the findings credible while noting the absence of significance testing, and rates reproducibility readiness as high.
Mapping LLM political coordinates with a multi-factor Political Compass design
A rigorous multi-factor Political Compass Test study shows that LLM political coordinates are sensitive to design choices such as prompts and personas. Evolve rates the methods strong, while cautioning that chat-based and downstream-fairness claims go beyond what the data supports.
A shared multilingual lens across the LLM evaluation agenda
Both cataloged 2026 papers list multilingual evaluation among their keywords, and the readability work explicitly targets low-resource languages, indicating a consistent focus on testing LLM behavior beyond English.

Timeline

2026Publishes 'Navigating the digital spectrum: Assessing political bias, stability, and downstream fairness in Large Language Models' on arXiv in the last author position; Evolve rates its multi-factor Political Compass methodology strong.
2026Publishes 'Assessing Readability with LLMs: The Role of Reasoning and Few-Shot Prompting' on arXiv as last author; the five-model benchmark receives a high reproducibility readiness rating from Evolve.
2026Both papers enter the Evolve catalog, anchoring a full-tier profile with a mean reproducibility score of 78/100 across the two studies.

Papers in the Evolve catalog

Frequent co-authors

Raphaël Thieffry×1 Luka Debevc×1 Nishan Chatterjee×1 Antoine Doucet×1 Senja Pollak×1

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.

Radar dimensions are computed from an OpenAlex-scope bibliometric snapshot; citation counts, h-index and the full publication list are outside this profile's data.
Only the two 2026 arXiv preprints reviewed by Evolve are covered; earlier and venue-published work is not assessed here.
Output, impact and activity axes are unscored because the underlying bibliometric data is insufficient; no values were estimated.
The profile is name-resolved from public bibliographic data (ORCID 0000-0002-7384-8112); country and career history beyond the institution name are not provided.