Alexander Mehler

—

Unclaimed · AI-structured profileOpenAlex

Senior (last-author) contributor to Arabic NLP shared-task systems in stance detection and MT error-span tagging.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-27) 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 scored: the snapshot contains only the two reviewed 2026 arXiv papers, far too little to quantify output. This axis would measure publication volume, venue spread, and productivity trends over a multi-year record.

Impact—

Not yet scored: no citation counts, h-index, or influence signals are available. This axis would measure citation-based impact, including total citations, citation velocity, and uptake of the shared-task systems by other groups.

Focus—

Not yet scored: with only two 2026 papers and no venue or affiliation history, thematic concentration cannot be estimated. This axis would measure how consistently an author's work clusters around specific topics over time.

Collaboration28/100

The only scored axis: 28/100, with just 3 distinct co-authors recorded across the bibliometric snapshot. Both 2026 catalog papers list Mehler in the last author slot with the same compact TTLab team, and Ali Abusaleh and Bhuvanesh Verma recur as frequent partners. This is a small, stable, lab-centric collaboration footprint rather than a broad or international network — typical of an evaluation-driven shared-task group.

Activity—

Not yet scored: the record contains only 2026-dated items, so publishing cadence and recent-productivity trends cannot be computed. This axis would measure output frequency and research continuity across recent years.

Reproducibility—

Not yet scored: no quantitative reproducibility metrics exist in the snapshot. Evolve's qualitative review does flag both 2026 papers as medium repro-readiness, but a numeric axis score would require code release, artifact, and benchmark verification data.

Research summary

Alexander Mehler is a natural language processing researcher whose catalogued work centers on Arabic-language NLP, published through the TTLab team on arXiv in 2026. In the reviewed record he appears consistently in the last (senior) author position on two shared-task system papers, though on neither as corresponding author. The first, at StanceEval-2026, presents CLASP-Ar, a cloze-style prompting approach for Arabic stance detection built on masked language modeling, pattern-exploiting training, and verbalizer-constrained decoding; Evolve's verdict notes 71.36/74.14 Favg2 with useful ablations, while flagging unexplained encoder conflicts and a dev-test gap. The second, at AlexandriaX-2026, applies a fine-tuned surface tagger with focal loss to Arabic machine-translation error-span detection and classification, ranking 3rd with 40.91/42.00, with honest ablations but rare-class F1 near zero. Both papers carry a medium repro-readiness flag. The bibliometric snapshot is otherwise thin: only three distinct co-authors are recorded, and collaboration is the sole scored dimension (28/100), reflecting a compact, lab-centric network around recurring partners such as Ali Abusaleh and Bhuvanesh Verma. Output, impact, focus, activity, and reproducibility lack sufficient underlying data to score, so this should be read as an early-stage, evaluation-driven profile rather than a complete career record.

Collaboration pattern

Collaboration data shows a deliberately compact footprint: 3 distinct co-authors and a collaboration score of 28/100, the only scored dimension. Ali Abusaleh and Bhuvanesh Verma are the recurring partners on both 2026 TTLab shared-task papers, where Mehler sits in the last (senior) author position without being corresponding author. The pattern suggests a small, stable, lab-internal team writing evaluation-driven system papers rather than a wide external network.

Research keywords

Arabic NLPstance detectionmasked language modelingpattern-exploiting trainingverbalizer-constrained decodingmachine translation quality estimationerror span detectionsequence taggingfocal lossclass imbalanceArabic dialectsshared-task evaluation

Highlights

CLASP-Ar: cloze-style prompting for Arabic stance detection at StanceEval-2026
Last-authored TTLab system paper applies verbalizer-constrained masked-language-model prompting with pattern-exploiting training to Arabic stance detection, reaching 71.36/74.14 Favg2 with useful ablations — though encoder conflicts and a dev-test gap remain unexplained (medium repro-readiness).
3rd place in Arabic MT error-span detection at AlexandriaX-2026
A fine-tuned surface tagger trained with focal loss ranked 3rd with 40.91/42.00 in the AlexandriaX-2026 Arabic machine-translation error-span shared task, with honest ablations — rare-class F1 stays near zero under severe class imbalance (medium repro-readiness).
A senior-author pattern across 2026 shared-task entries
On both catalogued 2026 papers Mehler holds the last (senior) author position — not corresponding author — working with a compact TTLab team and recurring partners Ali Abusaleh and Bhuvanesh Verma, a signature of a supervisory role in an evaluation-driven Arabic NLP group.

Timeline

2026TTLab enters StanceEval-2026 with CLASP-Ar, a cloze-style prompting system for Arabic stance detection, reporting 71.36/74.14 Favg2 on arXiv.
2026TTLab's fine-tuned surface tagger with focal loss ranks 3rd (40.91/42.00) in the AlexandriaX-2026 Arabic MT error-span shared task.
2026Both shared-task papers appear on arXiv with Mehler as last author and medium repro-readiness flags, within a three-person distinct co-author pool.
2026The 2026 record consolidates a focused Arabic NLP evaluation agenda with recurring co-authors Ali Abusaleh and Bhuvanesh Verma.

Papers in the Evolve catalog

Frequent co-authors

Ali Abusaleh×2 Bhuvanesh Verma×1 Bhuvanesh Verma×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.

Name resolution: this profile is resolved from public bibliographic data and the two 2026 TTLab arXiv papers; no ORCID or institutional affiliation was provided, and 'Alexander Mehler' may refer to more than one researcher.
Bibliometrics come from a single public bibliometric snapshot; only the collaboration dimension (28/100, 3 distinct co-authors) had sufficient data to score.
Paper outcomes (71.36/74.14 Favg2; rank 3 with 40.91/42.00) are quoted from Evolve's review verdicts on arXiv shared-task papers, not independently verified leaderboard results.
Output, impact, focus, activity, and reproducibility axes are unscored because the snapshot contains only two 2026 papers; read this as an early-stage profile.