Interpretable textual entailment and knowledge-augmented NLI for LLMs — Universidad Autónoma del Estado de Morelos
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-08) 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—
The output axis is not scored: publication-volume data such as total paper and preprint counts is unavailable in the current snapshot. Only the two Evolve-reviewed 2026 arXiv preprints are known, so career-wide productivity cannot yet be measured.
Impact—
The impact axis lacks sufficient data: citation counts, citation-based influence metrics and venue reception are not yet in the snapshot. It would measure how widely the NLI and interpretability work is taken up by other studies.
Focus0/100
Radar's focus axis reads 0. The raw field distribution shows essentially all reviewed output in one line — natural language processing / computational linguistics (0.5 and 0.5 across two label variants of the same field). This is extreme topical concentration: both 2026 papers target textual entailment / natural language inference, with no recorded dispersion into other fields.
Collaboration32/100
Collaboration scores 32, based on a small network of 4 distinct coauthors. Recurring partners include David Torres-Moreno and Asela Reig-Alamillo; on the 2026 semantic-abstraction paper the researcher takes the last author position. Both reviewed papers are multi-author arXiv preprints, suggesting a compact, stable team rather than a large distributed network.
Activity—
The activity axis is unscored because longitudinal publication data is missing. It would measure output cadence over time — e.g., papers per recent year — but the snapshot covers only two 2026 preprints, so recency trends cannot be assessed.
Reproducibility55/100
Reproducibility scores 55, the mean over 2 papers, both rated medium readiness. Evolve's verdicts temper this: the SLITE feature model's 96% SICK-CE / 83% SICK results near RoBERTa lack significance tests and non-SICK evaluation, and the +11.6 p.p. consensus gains for small LLMs need scrutiny of consensus training and baselines.
Research summary
Jorge Hermosillo-Valadez is a natural language processing researcher at the Universidad Autónoma del Estado de Morelos (ORCID 0000-0001-9040-767X) whose Evolve-visible work centers on natural language inference (NLI) and textual entailment. Two 2026 arXiv preprints define the current profile. The first, Linguistic Features for Interpretable Textual Entailment, introduces SLITE, a hybrid symbolic-distributional logistic-regression model built on 17 interpretable features that reaches 96% on SICK-CE and 83% on SICK — near RoBERTa — though Evolve notes the evaluation is SICK-only and lacks significance testing. The second, Semantic Abstraction for NLI, proposes a methodological framework that uses ConceptNet relation groups and decision-tree consensus to discover and compensate semantic knowledge and reasoning gaps in large language models, lifting small-LLM NLI by a reported +11.6 percentage points; Evolve judges the gains plausibly real but calls for scrutiny of consensus training and baselines. Both papers carry medium reproducibility readiness, and the reproducibility dimension scores 55 across 2 papers. The radar shows a highly concentrated research focus — effectively one field, NLP/computational linguistics — and a compact collaboration network of 4 distinct coauthors (collaboration score 32), with David Torres-Moreno and Asela Reig-Alamillo as recurring partners. Output, impact and activity axes remain unscored for lack of data, making this an early-stage, preprint-centric profile.
Collaboration pattern
The radar collaboration score is 32, drawn from only 4 distinct coauthors. The recurring pattern is a compact in-house team: David Torres-Moreno appears repeatedly among frequent coauthors, alongside Asela Reig-Alamillo. On the 2026 semantic-abstraction paper Jorge Hermosillo-Valadez takes the last author slot (not marked corresponding), consistent with a senior-but-shared authorship culture within a small UAEM-centered NLP group rather than large multi-institution consortia.
Research keywords
Highlights
SLITE: 17 interpretable linguistic features approach transformer-level entailment accuracy
Semantic abstraction lifts small-LLM natural language inference by +11.6 percentage points
A ConceptNet-anchored research line linking interpretability and LLM gap analysis
Timeline
- 2026Publishes SLITE, a 17-feature interpretable hybrid model for textual entailment, reaching 96% SICK-CE and 83% SICK — near RoBERTa.
- 2026Posts the semantic abstraction framework for NLI to arXiv in the last author position, reporting +11.6 p.p. gains for small LLMs via ConceptNet relation groups and decision-tree consensus.
- 2026Both preprints enter the Evolve catalog with medium reproducibility readiness; the reproducibility dimension records a mean of 55 over 2 papers.
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.
- Radar bibliometrics are OpenAlex-derived only; publication counts, citation totals and the output/impact/activity scores were not available.
- The profile rests on 2 Evolve-reviewed arXiv preprints, both dated 2026; earlier or non-arXiv work is not represented.
- Country metadata is missing; identity is resolved via ORCID 0000-0001-9040-767X from public bibliographic data.
- Both papers carry medium reproducibility readiness; Evolve flags SICK-only evaluation, missing significance tests, and scrutiny of consensus training and baselines.