UIUC NLP researcher: executable verification of LLM table claims and minimal-pair tests of Arabic dialect bias
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 yet computable from the available data, which covers only the two Evolve-reviewed papers. The output axis measures total publication volume, venue record, and year-over-year productivity, none of which can be derived from the current snapshot.
Impact—
No citation counts, h-index, or downstream-uptake records exist in this snapshot, so impact cannot be scored. This axis would measure how often the researcher's evaluation frameworks and dialectal bias benchmarks are cited, reused, and adopted by other work.
Focus0/100
The focus dimension scores 0/100, computed from a field distribution that splits evenly at 0.5 between sociolinguistic bias in LLMs and table-to-text generation with LLM evaluation. With only two cataloged papers, one per subfield, this even split reflects breadth across two adjacent evaluation topics rather than scattered drift. The score should be read as an artifact of a very small corpus; it can be expected to stabilize as more of the researcher's output enters the record.
Collaboration39/100
The collaboration dimension scores 39/100, based on 6 distinct coauthors in the record: Ryan Dolan, Paul de Nijs, Jonathan Dunn, Gunjan Anand, Ayush Singh, and Aleksandre Maskharashvili. Both cataloged papers are first-authored without corresponding-author flags, suggesting a contributor within small research teams rather than a large-network hub. The moderate score reflects a compact, early-stage collaboration footprint; the six-name roster indicates steady teamwork across the two evaluation projects.
Activity—
Insufficient data: only two 2026 papers are on record, with no full publication timeline or submission cadence. The activity axis measures recent output rate and sustained research engagement over consecutive years, which requires a longer publication history.
Reproducibility55/100
The reproducibility dimension scores a mean of 55 across both scored papers (reproPapers = 2), each rated medium reproducibility readiness. Evolve's reviews support this middle rating: the executable-verification paper includes honest human auditing that aids replication, but relies on synthetic data and confounded models; the dialect-bias paper provides convincing minimal-pair methodology while its downstream effects remain patchier. Together they suggest reusable artifacts with real but manageable reproduction barriers.
Research summary
Mai Mohamed Eida is a natural language processing researcher at the University of Illinois Urbana-Champaign whose work, as captured in the Evolve catalog, centers on evaluation methodology for large language models. The profile rests on two first-authored arXiv papers from 2026, and both are evaluation papers: one introduces executable Python checkers that verify whether LLM inference generation over tables yields correct quantified statements; the other uses minimal pairs and targeted syntactic evaluation to argue that LLMs favor Cairene Egyptian Arabic over the less prestigious Sa'idi dialect. This dual orientation shows up directly in the field distribution, which splits exactly 0.5/0.5 between table-to-text generation and LLM evaluation, and sociolinguistic bias in LLMs. Evolve's verdicts are encouraging on method design — the executable-verification idea is judged sound and the dialect minimal-pair evidence convincing — while flagging shared weaknesses: synthetic data, confounded model comparisons, and absent significance tests in the former; smaller, patchier downstream degradation in the latter. Both papers carry medium reproducibility readiness, giving a reproducibility mean of 55 across two scored papers. Collaboration is currently compact: 6 distinct coauthors and a collaboration score of 39. Output, impact, and activity metrics are not yet computable from the available record.
Collaboration pattern
Collaboration is currently compact: 6 distinct coauthors — Ryan Dolan, Paul de Nijs, Jonathan Dunn, Gunjan Anand, Ayush Singh, and Aleksandre Maskharashvili — supporting a collaboration score of 39/100. Both cataloged papers are first-authored with no corresponding-author role recorded, consistent with an active contributor inside small teams spanning the two evaluation projects, rather than a senior-led or large-consortium pattern.
Research keywords
Highlights
Executable verification for quantified table claims
Evidence that LLMs rank Arabic dialects by prestige
A throughline of targeted, checkable evaluation
Across both papers, evaluation is designed around precise, checkable targets — generalized quantifiers verified by code, dialect contrasts isolated by minimal pairs — a methodological signature linking the two subfields that each hold 0.5 of the field distribution.
Timeline
- 2026First-authored arXiv paper on non-prestigious Arabic dialectal bias in LLMs appears and enters the Evolve catalog with medium reproducibility readiness.
- 2026First-authored arXiv paper on evaluating LLM inference generation through executable verification appears and enters the Evolve catalog with medium reproducibility readiness.
- 2026Both papers are reviewed by Evolve, establishing a two-paper record split evenly across LLM evaluation and sociolinguistic bias, with a reproducibility mean of 55.
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.
- Dimension scores, coauthor counts, and field distribution come from public bibliometric data only; citation and venue coverage may lag, and arXiv preprints may not yet be fully indexed.
- The profile rests on only 2 reviewed papers (both 2026 arXiv preprints); output, impact, and activity dimensions are missing because no broader publication record is available.
- No ORCID or country is recorded, and no peer-reviewed venue appears in the data; name resolution relies on public bibliographic matching and may need confirmation.
- The reproducibility mean of 55 is based on 2 papers only and should not be generalized.