NLP researcher quantifying LLM evaluation biases, from African-language sycophancy to judge-panel preference
2
papers here
2
reviewed by Evolve
2
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—
Publication-volume and productivity metrics are not available in this profile's data yet; only the two Evolve-reviewed 2026 preprints are on file. This axis would measure the scale, cadence, and completeness of the researcher's publication record, so it is left unscored.
Impact—
Citation counts, h-index, and downstream influence are absent from the current snapshot. This axis would measure how widely the researcher's work is cited and taken up by follow-up studies; rather than estimating from thin evidence, it is honestly left unscored.
Focus100/100
Focus scores 100 because the reviewed field distribution is fully concentrated: the fieldDist record lists natural language processing (LLM evaluation) as the sole field, and both 2026 first-author papers fall under it. One targets sycophancy in African-language QA, the other judge-panel preference, giving a tightly measurement-centered agenda rather than a spread across topics.
Collaboration41/100
Collaboration scores 41 with seven distinct co-authors recorded. Recurring names include Elvis Gyasi Owusu, Rose-Mary Owusuaa Mensah Gyening, Luke E. K. Achenie, Benjamin Tei Partey, and Nii-Nai Derrick Sowah. Both papers are first-authored without corresponding-author flags, consistent with small-team, co-led evaluation studies within a moderately connected but not yet extensive network.
Activity—
Recent publication velocity and ongoing research activity cannot be computed from the data provided, which covers only 2026-dated preprints. This axis would measure publishing cadence and engagement over time; no score is assigned at this stage.
Reproducibility78/100
Reproducibility scores 78, the mean across two scored papers (reproPapers: 2). AfriSyCo carries a high readiness rating, backed by its controlled factorial design over assertive framing, verification, and wording; the judge-panel study is rated medium, with Evolve noting its 6.7pp result is design-sensitive. Together they indicate careful measurement practice from the outset.
Research summary
David Ababio Awuni is an early-stage researcher in natural language processing whose reviewed work concentrates on a single theme: measuring how large language models behave as objects and agents of evaluation. In AfriSyCo (2026, arXiv, first author), he quantifies sycophancy around African-language content, showing in a controlled factorial design that assertive framing alone shifts LLM answers on African-language multiple-choice questions by +30.4 percentage points, while verification and wording effects were probed over only three prompt wordings. In Who Judges Matters (2026, first author), he measures family-conditioned preference in LLM-as-a-judge panels, applying a corrected estimator within a Bradley–Terry framework and finding a robust 6.7 percentage-point same-family lift across four judge families. Evolve rates the reproducibility of this portfolio at a mean of 78 across two papers, with AfriSyCo rated high readiness and the judge-panel study medium. The field distribution sits entirely within LLM evaluation (focus 100), and the collaboration score of 41 reflects seven distinct co-authors to date. Output, impact, and activity metrics are not yet available in the underlying data, so this profile should be read as an early snapshot of a measurement-focused NLP researcher.
Collaboration pattern
With seven distinct co-authors, David Ababio Awuni works in a small, stable team: Elvis Gyasi Owusu (listed twice in the source data), Rose-Mary Owusuaa Mensah Gyening, Luke E. K. Achenie, Benjamin Tei Partey, and Nii-Nai Derrick Sowah recur across the record. Both Evolve-reviewed papers are first-authored without corresponding-author designation, consistent with early-career, co-led evaluation studies. The collaboration score of 41 sits mid-range, matching a compact network rather than a broad one.
Research keywords
Highlights
Assertive framing shifts LLM answers on African-language MCQs by +30.4 points
A corrected estimator finds a 6.7pp same-family lift in LLM judge panels
High reproducibility standards across a young measurement portfolio
Timeline
- 2026First-author arXiv preprint AfriSyCo measures assertive framing, verification, and wording sensitivity around African-language content, finding a +30.4pp answer-switching effect.
- 2026First-author arXiv preprint Who Judges Matters quantifies family-conditioned preference in LLM-as-a-judge panels via a corrected Bradley–Terry estimator, reporting a 6.7pp same-family lift.
- 2026Both first-author preprints enter the Evolve catalog under natural language processing (LLM evaluation), establishing a measurement-focused research agenda with a 78 reproducibility mean.
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 come from an OpenAlex-based snapshot; output, impact, and activity axes lack the data to be scored.
- Institution, country, and ORCID are not recorded; the profile is name-resolved from public bibliographic data.
- Coverage is limited to the two Evolve-reviewed papers; other works by this author may exist outside the catalog.
- The co-author list contains the researcher's own name and a duplicate entry, indicating possible entity-resolution noise in the source data.