Adaptive simplification of English sentences using unsupervised semantic representations: integrating audience needs with improved pre-training strategies
Digital Scholarship in the Humanities · 13.11.2026
DOI: 10.1093/llc/fqag147 ↗Kurz gefasst
Fachartikel in der Zeitschrift Digital Scholarship in the Humanities, online erschienen am 13.11.2026. Verfasst von Zhang Wenlong und 1 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben.
Abstract (Originalsprache)
Text simplification can improve access to information when source sentences exceed a reader’s language proficiency. This study developed an adaptive English sentence simplification framework that combines unsupervised semantic representations with reader-level controls and does not require a parallel complex-simple corpus. Bidirectional Encoder Representations from Transformers embeddings were corrected with singular value decomposition (SVD) by removing the five dominant directions. Mean pairwise cosine similarity decreased from μ = 0.75 (σ = 0.12) to μ = 0.62 (σ = 0.08), indicating reduced anisotropy. The corpus contained 10,000 English sentences from academic, professional, and general sources. Reader adaptation used seven proficiency tiers anchored to the Common European Framework of Reference for Languages, with an additional expert tier. System quality was evaluated with System Output Against References and Against the Input Sentence (SARI), Flesch-Kincaid Grade Level (FKGL), cosine similarity, clustering measures, and blinded expert ratings. SARI increased from 0.40 to 0.57, a 42.5 per cent relative increase, while FKGL decreased from 9.0 to 7.0. After SVD, cosine-similarity skewness decreased from 0.8 to 0.1, and the silhouette score increased from 0.45 to 0.68. Ten experts rated 200 randomized sentence pairs; the weighted mean score was 4.32/5.00 (ICC(2, k) = 0.89). Average processing time was 150 ms per sentence. The results indicate that geometric correction and reader-specific decision rules can improve simplification quality while preserving meaning.
Zitieren
APA
Wenlong, Z., & Jianxin, Z. (2026). Adaptive simplification of English sentences using unsupervised semantic representations: integrating audience needs with improved pre-training strategies. Digital Scholarship in the Humanities. https://doi.org/10.1093/llc/fqag147
Vancouver
Wenlong Z, Jianxin Z. Adaptive simplification of English sentences using unsupervised semantic representations: integrating audience needs with improved pre-training strategies. Digital Scholarship in the Humanities. 2026. doi:10.1093/llc/fqag147
BibTeX
@article{Wenlong2026,
author = {Wenlong, Zhang and Jianxin, Zhou},
title = {Adaptive simplification of English sentences using unsupervised semantic representations: integrating audience needs with improved pre-training strategies},
journal = {Digital Scholarship in the Humanities},
year = {2026},
doi = {10.1093/llc/fqag147}
}
EUVIMED