Beyond Binary Rewrite Decisions: Separating Rewrite-Intensity Control from Level Prediction in a Frozen-LLM Generation-and-Selection Pipeline for Korean Easy-Read Drafting
- 1Department of Artificial Intelligence Convergence, Graduate School, Pukyong National University, Busan 48513, Republic of Korea
- 2Department of Computer Engineering and Artificial Intelligence, Pukyong National University, Busan 48513, Republic of Korea
Abstract
Editors of published Korean Easy-Read documents changed 50 aligned sentences by widely varying amounts (edit distance 0.39–0.89), motivating rewrite intensity as a control variable for large language models (LLMs). We ask whether a requested intensity is realized per sentence and whether the level to request can be predicted from the source, using a generation-and-selection pipeline over a frozen LLM (EXAONE-3.5-7.8B-Instruct) in two sequential protocols with criteria hash-frozen before generation. Realized intensity is normalized character edit distance, which measures change, not readability, meaning preservation or Easy-Read quality. A four-level protocol (v1) ordered outputs on average but failed the sentence-level criterion (14/50 under its zero-tolerance rule). A three-level follow-up (v2), designed after that failure with several simultaneous changes, met its criterion for the whole pipeline, including its intensity-aware selector: 41/50 sources were non-decreasingly ordered at the declared tolerance (24/50 with strict increases; post hoc, 30/50 without the selector’s intensity-match term at that tolerance). Three runs of one LLM proxy perceived the ordering (consistency, not validity); no human evaluated the outputs. Stronger levels reduced source similarity without monotonically reducing lexical difficulty. Six source features did not support prediction of the LLM-proxy-preferred level under a pre-specified rule (quadratic-weighted κ −0.098 vs. 0.012), so no adaptive allocation was built. Rewrite-intensity control and level allocation therefore require separate validation.
Related
- Selection rules for the harmonic spectroscopy of animal decisions
- Separating benchmark discrimination from experimental retrieval in synthetic-lethality prediction
- When blood is not thicker than water: Social information landscapes offset the food-fear trade-off in habitat selection decisions
- Pocket-Conditioned Generation and Down-Selection of Non-Oxime Reactivators
- Ultrafast and Accurate Selection of High-Quality Protein Complex Models from Large-Scale Prediction
BibTeX
@article{lee2026beyond,
title = {{Beyond Binary Rewrite Decisions: Separating Rewrite-Intensity Control from Level Prediction in a Frozen-LLM Generation-and-Selection Pipeline for Korean Easy-Read Drafting}},
author = {Jeong-Eun Lee and Sang Ho Oh},
journal = {Applied Sciences},
year = {2026},
month = oct,
doi = {10.3390/app162010253},
url = {https://doi.org/10.3390/app162010253},
note = {CC BY},
}
RIS
TY - JOUR
TI - Beyond Binary Rewrite Decisions: Separating Rewrite-Intensity Control from Level Prediction in a Frozen-LLM Generation-and-Selection Pipeline for Korean Easy-Read Drafting
AU - Jeong-Eun Lee
AU - Sang Ho Oh
PY - 2026
DA - 2026/10/26
JO - Applied Sciences
DO - 10.3390/app162010253
UR - https://doi.org/10.3390/app162010253
AB - Editors of published Korean Easy-Read documents changed 50 aligned sentences by widely varying amounts (edit distance 0.39–0.89), motivating rewrite intensity as a control variable for large language models (LLMs). We ask whether a requested intensity is realized per sentence and whether the level to request can be predicted from the source, using a generation-and-selection pipeline over a frozen LLM (EXAONE-3.5-7.8B-Instruct) in two sequential protocols with criteria hash-frozen before generation. Realized intensity is normalized character edit distance, which measures change, not readability, meaning preservation or Easy-Read quality. A four-level protocol (v1) ordered outputs on average but failed the sentence-level criterion (14/50 under its zero-tolerance rule). A three-level follow-up (v2), designed after that failure with several simultaneous changes, met its criterion for the whole pipeline, including its intensity-aware selector: 41/50 sources were non-decreasingly ordered at the declared tolerance (24/50 with strict increases; post hoc, 30/50 without the selector’s intensity-match term at that tolerance). Three runs of one LLM proxy perceived the ordering (consistency, not validity); no human evaluated the outputs. Stronger levels reduced source similarity without monotonically reducing lexical difficulty. Six source features did not support prediction of the LLM-proxy-preferred level under a pre-specified rule (quadratic-weighted κ −0.098 vs. 0.012), so no adaptive allocation was built. Rewrite-intensity control and level allocation therefore require separate validation.
SN - 2076-3417
ER -