NLP researcher at Johns Hopkins working on document authoring, legal NLP, and efficient long-form generation
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Output metrics are not yet available in this snapshot: total publication counts and venue-normalized productivity could not be computed, so this axis — which would measure the volume and consistency of research output over time — is left unscored here.
Impact—
Impact cannot be assessed: the snapshot reports h-index and i10-index as 0, which indicates missing citation coverage rather than measured zero impact. This axis would measure citation-based influence of the work once citation data becomes available.
Focus0/100
The focus score is 0, and the raw field distribution explains why: the snapshot splits evenly across three fields — natural language processing, legal NLP / computational law, and human-computer interaction and NLP (efficient long-form generation) — at 33.3% each. Topic counts reinforce the spread, led by topic modeling (14) and information retrieval and search behavior (7), followed by AI in law (4), NLP techniques (3), and software engineering research (2).
Collaboration55/100
Collaboration scores 55, based on 15 distinct coauthors recorded in the snapshot. The recurring coauthor list spans the NLP and legal-domain sides of the agenda, including Alexander Martin, Luis Brena, William Jurayj, Gregory Deyesu, Zaid Al-Huneidi, Andrew Blair-Stanek, Zhengping Jiang, Gonzalo Ramos, Jina Suh, Shiqian Rachel Ng, and Elias Stengel-Eskin. All three reviewed 2026 papers are team-authored, with Van Durme in last author position twice and middle author once.
Activity84/100
Activity is the strongest axis at 84: the snapshot credits 17 works in the most recent three-year window, with countsByYear attributing all 17 to 2026. Every Evolve-reviewed paper is a 2026 arXiv preprint, consistent with a concentrated current burst of output rather than a long accumulation; the score should be read against that narrow snapshot window.
Reproducibility55/100
Reproducibility scores 55, a mean computed over 3 reviewed papers, each rated medium repro-readiness. Evolve's verdicts flag concrete risks: LLM-judge circularity and absent human or significance evidence in the document-authoring work; reliable labels but weak support edges and poor cross-case retrieval generalization in the legal corpus; and simulated, judge-based headline gains in the early-exit study. The mid-range score reflects solid artifacts tempered by evaluation dependencies.
Research summary
Benjamin Van Durme is a researcher at Johns Hopkins University whose recent work, as name-resolved from an OpenAlex snapshot, splits evenly across three fields — natural language processing, legal NLP / computational law, and human-computer interaction and NLP (efficient long-form generation) — each at 33.3% of the field distribution. Activity is the standout axis: the snapshot records 17 works in the most recent three-year window, with all 17 attributed to 2026, yielding an activity score of 84 alongside 15 distinct coauthors and a collaboration score of 55. Three Evolve-reviewed 2026 arXiv preprints define the current agenda. Knowledge Pull Requests for Continual Document Authoring makes updates to living documents reviewable at the claim level. Mining Legal Arguments in U.S. Corporate Case Law contributes the first expert-annotated, tree-structured argument corpus for I.R.C. §368 tax cases. FrankenReport formulates early exiting in deep-research report generation as an expected value of computation. Evolve's verdicts are broadly encouraging but share a methodological thread — LLM-as-a-judge evaluation with limited human or significance evidence — and all three papers carry a medium repro-readiness rating, averaging 55 across 3 papers. Topic counts (topic modeling 14; information retrieval and search behavior 7; AI in law 4; NLP techniques 3; software engineering research 2) confirm a deliberately multi-line portfolio.
Collaboration pattern
The snapshot lists 15 distinct coauthors, and the recurring list spans his NLP and legal-NLP lines: Alexander Martin, Luis Brena, William Jurayj, Gregory Deyesu, Zaid Al-Huneidi, Zhengping Jiang, Elias Stengel-Eskin, Gonzalo Ramos, Jina Suh, and Shiqian Rachel Ng, with Andrew Blair-Stanek among the frequent collaborators. All three reviewed papers are multi-authored, with Van Durme last author twice and middle author once; the list also contains his own name, likely a linkage artifact.
Research keywords
Highlights
Claim-level review flow for living documents
First expert-annotated argument tree corpus for U.S. §368 tax cases
Expected value of computation for early exit in report generation
Timeline
- 2026Published Knowledge Pull Requests for Continual Document Authoring on arXiv, introducing claim-level, reviewable updates to continually maintained documents (last author).
- 2026Published Mining Legal Arguments in U.S. Corporate Case Law on arXiv, the first expert-annotated tree-structured argument corpus for I.R.C. §368 tax cases (last author).
- 2026Co-authored FrankenReport on arXiv, formulating early exiting in long-form deep-research report generation as an expected value of computation (middle author).
- 2026OpenAlex snapshot attributes all 17 works in the most recent three-year window to 2026, the sole year with recorded output in countsByYear.
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.
- Bibliometrics are name-resolved from an OpenAlex snapshot (identity: openalex); output and impact axes are unscored because citation coverage is missing — h-index and i10-index are recorded as 0.
- The activity window is narrow: countsByYear contains only 2026 (17 works), so multi-year output trends cannot be assessed.
- The frequent-coauthor list includes the researcher's own name, suggesting record-linkage noise in the underlying bibliographic data.
- Reproducibility rests on only 3 Evolve-reviewed arXiv preprints (reproMean 55, all medium); peer-reviewed venue status and the fuller publication history are not captured.