LLM evaluation researcher measuring sycophancy in African-language NLP and bias in LLM-as-judge panels
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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—
No publication-count metrics are available yet: the snapshot covers only the two Evolve-reviewed papers. This axis would measure cumulative output — total publications, venue types, and per-year counts — once a fuller publication record is indexed.
Impact—
Citation-based impact cannot be assessed: no citation counts, h-index, or venue influence data are present in the snapshot. This axis would measure how often the work is cited and taken up by other researchers.
Focus100/100
The focus score is 100, the maximum, because the field distribution places every reviewed paper in a single category — natural language processing (LLM evaluation). Both 2026 papers sit squarely in measurement-oriented LLM evaluation: sycophancy and wording sensitivity around African-language multiple-choice content, and family-conditioned preference in LLM-as-a-judge panels. This is a deliberately narrow, coherent line rather than a spread across topics, though the sample is only two papers.
Collaboration41/100
The collaboration score is 41, on the lower end, with 7 distinct co-authors recorded in the snapshot. Both reviewed papers are multi-authored: Owusu is the last author on AfriSyCo and a middle author on Who Judges Matters, and is listed as corresponding author on neither. Recurring co-authors include David Ababio Awuni, Rose-Mary Owusuaa Mensah Gyening, Luke E. K. Achenie, Benjamin Tei Partey, and Nii-Nai Derrick Sowah, suggesting a stable small team rather than a wide network.
Activity—
Recent activity is not yet measurable: with both papers dated 2026 and no prior-year records provided, publication tempo cannot be computed. This axis would measure output frequency and research continuity across the most recent years.
Reproducibility78/100
The reproducibility score is 78, the mean across 2 assessed papers. Evolve rates AfriSyCo's reproducibility readiness as high — a controlled factorial design with quantified effects such as the +30.4 percentage-point framing shift — and Who Judges Matters as medium, reflecting a corrected estimator (6.7 pp same-family lift) whose magnitude is design-sensitive. Together they indicate a measurement-first habit of reporting effects that others can re-run.
Research summary
Elvis Gyasi Owusu is a natural language processing researcher whose reviewed work centers on the evaluation of large language models, with two 2026 arXiv papers both classified under LLM evaluation. The first, AfriSyCo, examines sycophancy around African-language content: a controlled factorial experiment found assertive framing shifts LLM answers on African-language multiple-choice questions by +30.4 percentage points, while also probing verification behaviour and sensitivity to prompt wording — though only three wordings were tested. The second, Who Judges Matters, measures family-conditioned preference in LLM-as-a-judge panels, using a corrected estimator to report a robust 6.7 percentage-point same-family lift across four judge families built on Bradley–Terry modelling. Evolve rates the reproducibility of these papers high and medium respectively, consistent with a reproducibility score of 78 across 2 papers. The bibliometric snapshot shows 7 distinct co-authors (collaboration score 41) and a field distribution concentrated entirely in one area (focus score 100). Affiliation, country, and ORCID are not recorded, so this profile is name-resolved from public bibliographic data and covers only Evolve-reviewed output.
Collaboration pattern
The snapshot records 7 distinct co-authors and a collaboration score of 41, indicating a moderate, team-based pattern rather than a large network. Recurring names include David Ababio Awuni, Rose-Mary Owusuaa Mensah Gyening, Luke E. K. Achenie, Benjamin Tei Partey, and Nii-Nai Derrick Sowah. Both reviewed papers are multi-authored: Owusu is last author on AfriSyCo and a middle author on Who Judges Matters, without corresponding-author role on either.
Research keywords
Highlights
Assertive framing drives +30.4 pp answer switching on African-language MCQs
Robust 6.7 pp same-family lift across four LLM judge families
A measurement-first evaluation toolkit
Across both 2026 papers the common thread is controlled measurement: factorial designs isolating framing effects and corrected estimators removing bias in judge comparisons, yielding a mean reproducibility score of 78 across the 2 papers and a fully concentrated focus score of 100.
Timeline
- 2026AfriSyCo appears on arXiv: a controlled factorial experiment links assertive framing to a +30.4 pp increase in LLM answer switching on African-language MCQs, with reproducibility readiness rated high.
- 2026Who Judges Matters appears on arXiv: a corrected estimator reports a robust 6.7 pp same-family lift across four LLM judge families, with reproducibility readiness rated medium.
- 2026Both reviewed papers fall in a single field category (LLM evaluation), producing a focus score of 100, a collaboration score of 41 with 7 distinct co-authors, and a mean reproducibility score of 78.
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.
- Radar scores and co-author counts come from the provided bibliometric snapshot only; no institution, country, or ORCID is recorded for this name.
- Coverage is limited to two Evolve-reviewed 2026 arXiv preprints; the output, impact, and activity dimensions lack sufficient underlying data.
- 'Elvis Gyasi Owusu' may not be a unique name; this profile is name-resolved from public bibliographic data and could merge or omit namesakes.