NLP researcher at NTUA evaluating gender bias in translation metrics and risk awareness in LLM agents
2
papers here
2
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—
Publication-volume data — papers per year and venue — is not yet available for this profile, so research output cannot be scored; the two Evolve-reviewed 2026 arXiv papers are currently the only confirmed outputs on record.
Impact—
Citation counts, h-index and downstream influence are not yet available, so this axis stays unscored; it would measure how widely the gender-bias and agent-safety benchmarks are adopted, reused and cited by other evaluation research.
Focus0/100
The field distribution splits exactly 50/50 between natural language processing (machine translation evaluation) and AI safety (LLM agent evaluation), which keeps the focus score at the low end rather than concentrated in one line of work. Both 2026 papers are, however, evaluation-centric studies — one auditing MT metrics for occupational gender bias, the other benchmarking risk awareness in agent trajectories — so the low score reflects two adjacent evaluation problems rather than unrelated research directions.
Collaboration53/100
The coauthor network counts 13 distinct collaborators with a mid-range collaboration score of 53. Orfeas Menis Mastromichalakis and Giorgos Filandrianos each appear twice in the frequency list, marking the stable core, while Wafaa Mohammed, Giuseppe Attanasio, Ruoxi Shang, Christina-Maria Androna, Yu Feng, Aniruddhan Ramesh, Rico Angell and Shang Hong Sim appear once, indicating per-project international partnerships alongside the recurring core.
Activity—
Temporal activity — publishing cadence, year-by-year distribution and recent gaps — cannot be computed from data covering only the two 2026 catalog entries; this axis would track sustained productivity over time and remains unscored.
Reproducibility55/100
Reproducibility readiness averages 55 across the 2 papers scored, both rated medium. The gender-bias study rests on a documented occupation-balanced GAMBIT+ subset and WMT26 evaluation runs, while the AURA-Eval verdict flags reliance on an LLM judge and a single seed source. Artifacts are thus partially reusable, but neither paper reaches turnkey replication readiness.
Research summary
Chrysoula Zerva is a researcher affiliated with the National Technical University of Athens (ORCID 0000-0002-4031-9492) whose cataloged work concentrates on evaluation methodology across two adjacent areas of natural language processing and AI safety. Her 2026 Evolve-reviewed output splits evenly between machine translation evaluation and LLM agent evaluation. As last author she appears on a benchmarking study of gender bias in machine translation evaluation metrics across occupations, built on an occupation-balanced GAMBIT+ subset of WMT26 evaluators, which found a masculine preference in 126 of 131 statistically significant runs with effect sizes varying widely. As a middle author she contributed to AURA-Eval, a framework for evaluating acting under risk awareness in LLM agent trajectories, which augments tool-use trajectories to show unsafe acts rise when no safe path exists; its verdict notes solid validation but flags reliance on an LLM judge and a single seed source. Her coauthor network spans 13 distinct collaborators, with Orfeas Menis Mastromichalakis and Giorgos Filandrianos recurring and international partners including Wafaa Mohammed, Giuseppe Attanasio and Yu Feng. Reproducibility readiness averages 55 across her two scored papers, both rated medium. Output, impact and activity metrics are not yet available, so this is an early-stage, evaluation-focused profile built from public bibliographic data.
Collaboration pattern
With 13 distinct coauthors on record, the pattern combines a stable recurring core — Orfeas Menis Mastromichalakis and Giorgos Filandrianos each appear twice in the frequency list — with per-project international partners such as Wafaa Mohammed, Giuseppe Attanasio, Ruoxi Shang, Yu Feng, Aniruddhan Ramesh, Rico Angell and Shang Hong Sim. Cataloged authorship spans last and middle positions, with no corresponding-author role in either paper.
Research keywords
Highlights
Gender bias benchmark across occupations for WMT26 MT evaluators
AURA-Eval: risk awareness in LLM agent trajectories
Timeline
- 2026Last-author arXiv paper benchmarks gender bias in machine translation evaluation metrics across occupations using an occupation-balanced GAMBIT+ subset against WMT26 evaluators.
- 2026Middle-author arXiv paper introduces AURA-Eval, a framework for evaluating acting under risk awareness in LLM agent tool-use trajectories.
- 2026Both Evolve catalog entries land in the same year, splitting evenly between machine translation evaluation and LLM agent safety.
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.
- Bibliometric scores (collaboration, focus, reproducibility) come from Evolve's radar over public bibliometric data; the output, impact and activity axes lack sufficient data and are not scored.
- The profile covers only the two Evolve-reviewed 2026 arXiv papers; other publications by this author are not reflected here.
- Country-level affiliation data is missing; institutional attribution relies on the provided NTUA affiliation and ORCID 0000-0002-4031-9492.
- Reproducibility readiness (mean 55) is based on only 2 papers and reflects Evolve's assessment, not independent replication.