NLP researcher at VTT Finland working on low-resource NER annotation and vision-language hallucination detection
4
papers here
4
reviewed by Evolve
3
first-author
2
corresponding
—
works · OpenAlex
5
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 output score is available yet: total publication volume and yearly paper counts have not been computed for this profile. Only two 2026 first-author arXiv papers are in scope, too few to characterize research output.
Impact—
No impact score is available yet: citation counts, h-index and venue-level influence have not been measured. Both cataloged papers are recent 2026 arXiv preprints, so citation-based impact will need time and further tracking before it can be scored.
Focus0/100
The focus axis currently scores 0/100, reflecting a field distribution split exactly 50/50 between two NLP subfields — computational linguistics and vision-language modeling — with neither yet dominant. With only two cataloged papers, this even split reads as early-career breadth: both strands are methodologically linked through automatic label quality, but the record lacks the concentrated output that focus scoring rewards.
Collaboration53/100
Collaboration scores 36/100 with 5 distinct coauthors detected across the bibliometric footprint. Both cataloged papers list Toqeer Ehsan as first author and not corresponding, consistent with contributing inside established research teams. The frequent-coauthor set comprises Thamar Solorio, Nico Penttilä, Richard Schmidt, Arash Hajikhani and Victoria Palacin, though the catalog does not map individuals to specific papers. The small measured network reads as early-stage rather than sparse, and shared-task work typically implies multi-author conventions.
Activity75/100
No activity score is available yet: recent publication cadence and year-over-year throughput cannot be computed from current data. Only two papers, both dated 2026, are in scope — insufficient to trace a sustainable activity trend.
Reproducibility49/100
This is the strongest measured axis: a mean reproducibility score of 78 across 2 evaluated papers. The SHROOM-Visions committee paper rates high readiness, helped by transparent aggregation over five vision-language judges, while the NER annotation-correction framework rates medium, with Evolve citing missing significance testing and ablations. The absence of statistical significance tests recurs in both reviews, making it the clearest single lever for lifting an already solid score.
Research summary
Toqeer Ehsan is a researcher at VTT Technical Research Centre of Finland whose Evolve-cataloged work sits squarely in natural language processing, split exactly 50/50 between computational linguistics and vision-language modeling. The profile currently rests on two first-author 2026 arXiv papers. The first, "A Scalable Framework for Automated NER Annotation Correction in Low-Resource Languages", applies self-training and pseudo-labeling to repair missing named-entity labels in Urdu, Shahmukhi and Sindhi datasets, reporting gains of +1.4–4.0 F1; Evolve's review flags the absence of significance testing and ablations. The second, the "Vroom-Vroom" entry at SHROOM-Visions, deploys a five-model vision-language committee with majority voting to detect hallucinated spans, winning 3 of 4 shared-task languages; the noted weaknesses are a vote threshold tuned on the test split and, again, no significance tests. A common methodological thread runs through both: automatic or ensemble-driven label quality for NLP systems, from low-resource annotation to multimodal hallucination detection. Measured bibliometrics are early-stage. Reproducibility is the strongest axis, with a mean score of 78 across 2 evaluated papers (high readiness for the shared-task system, medium for the NER framework). Collaboration scores 36 with 5 distinct coauthors, and focus scores 0 because the record is evenly split across two subfields. Output, impact and activity axes remain unscored pending citation and volume data.
Collaboration pattern
The measured collaboration footprint is compact: 5 distinct coauthors and a collaboration score of 36/100. Both cataloged papers are first-authored and non-corresponding, consistent with contributing within established teams rather than leading correspondence. The frequent-coauthor list comprises Thamar Solorio, Nico Penttilä, Richard Schmidt, Arash Hajikhani and Victoria Palacin, though the catalog does not map individual coauthors to specific papers; shared-task participation also reflects team-based authorship conventions.
Research keywords
Highlights
Automated NER annotation repair for Urdu, Shahmukhi and Sindhi
Five-judge VLM committee wins 3 of 4 SHROOM-Visions languages
Timeline
- 2026First-author arXiv paper introduces a scalable framework for automated NER annotation correction in low-resource languages (Urdu, Shahmukhi, Sindhi), reporting +1.4–4.0 F1 from self-training and pseudo-labeling.
- 2026First-author "Vroom-Vroom" system for the SHROOM-Visions shared task detects hallucinated spans in vision-language outputs using a five-model committee, winning 3 of 4 languages.
- 2026Both first-author papers enter the Evolve catalog; a mean reproducibility readiness of 78 is recorded across the two evaluations, while output, impact and activity metrics remain unscored.
Papers in the Evolve catalog
- first authorcorresponding
Multi-Task Learning by using Contextualized Word Representations for Syntactic Parsing of a Morphologically Rich Language
- corresponding
From Policy Documents to Structured Survey Responses: Evaluating Large Language Models for Policy Monitoring
- first author
A Scalable Framework for Automated NER Annotation Correction in Low-Resource Languages
- first author
Vroom-Vroom at SHROOM-Visions: A Multi-Judge Committee for Detecting Hallucinated Spans in Vision-Language Outputs
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 derive from OpenAlex-based radar computation only; citation counts, h-index and total publication volume are not available.
- Profile is name-resolved from public bibliographic data (ORCID 0000-0002-6724-6705); assessment covers only the two papers reviewed in the Evolve catalog.
- Reproducibility mean (78) is based on only 2 evaluated papers, so the score is sensitive to any single new review.
- Output, impact and activity axes are unscored (dimMissing) and no estimates are provided.