LQM

Linguistically Motivated Multidimensional Quality Metrics for Machine Translation

A linguistically grounded framework for diagnosing machine translation quality.

---

Overview

LQM is a linguistically motivated, multidimensional framework for analyzing machine translation quality.

This repository accompanies the paper:

LQM: Linguistically Motivated Multidimensional Quality Metrics for Machine Translation
Findings of ACL 2026

Paper: arXiv:2604.18490

---

Framework

LQM analyzes translation quality across six linguistic dimensions:

Level	Focus
Sociolinguistics	Dialect, register, variety, and social appropriateness
Pragmatics	Context, intent, implicature, and communicative function
Semantics	Meaning preservation, omission, addition, and mistranslation
Morphosyntax	Grammar, agreement, word order, and structure
Orthography	Spelling, punctuation, and writing conventions
Graphetics	Script-level and visual-form issues

---

Resources

File	Description
PROMPT.md	Translation prompt used in the experiments
LQM-Annotation-Guidelines.pdf	Annotation guidelines for applying LQM

---

Citation

If you use LQM or this repository, please cite:

@misc{magdy2026lqmlinguisticallymotivatedmultidimensional,
  title         = {LQM: Linguistically Motivated Multidimensional Quality Metrics for Machine Translation},
  author        = {Samar M. Magdy and Fakhraddin Alwajih and Abdellah El Mekki and Wesam El-Sayed and Muhammad Abdul-Mageed},
  year          = {2026},
  eprint        = {2604.18490},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2604.18490}
}