Saved in:
Bibliographic Details
Main Authors: Zhang, Luke, Vasselli, Justin, Khan, Aditya, Ng, York Hay, Lee, En-Shiun Annie
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.09098
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916071751024640
author Zhang, Luke
Vasselli, Justin
Khan, Aditya
Ng, York Hay
Lee, En-Shiun Annie
author_facet Zhang, Luke
Vasselli, Justin
Khan, Aditya
Ng, York Hay
Lee, En-Shiun Annie
contents We propose Dynamic Meta-Metrics (DMM), a framework for machine translation evaluation that learns source-sentence conditioned combinations of existing metrics. Rather than relying on a single static ensemble or language-specific weighting, DMM adapts the metric combination based on properties of the source segment. We study hard conditioning, which fits an interpretable combiner per cluster, and an exploratory soft-conditioned extension whose weights vary continuously with source-cluster responsibilities. We evaluate DMM on the WMT Metrics Shared Task data across multiple language pairs using pairwise agreement measures at the system and segment levels. Across settings, MLP-based combinations outperform linear and Gaussian process-based ensembles, and introducing soft conditioning yields gains over linear models.
format Preprint
id arxiv_https___arxiv_org_abs_2605_09098
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Dynamic Meta-Metrics: Source-Sentence Conditioned Weighting for MT Evaluation
Zhang, Luke
Vasselli, Justin
Khan, Aditya
Ng, York Hay
Lee, En-Shiun Annie
Computation and Language
We propose Dynamic Meta-Metrics (DMM), a framework for machine translation evaluation that learns source-sentence conditioned combinations of existing metrics. Rather than relying on a single static ensemble or language-specific weighting, DMM adapts the metric combination based on properties of the source segment. We study hard conditioning, which fits an interpretable combiner per cluster, and an exploratory soft-conditioned extension whose weights vary continuously with source-cluster responsibilities. We evaluate DMM on the WMT Metrics Shared Task data across multiple language pairs using pairwise agreement measures at the system and segment levels. Across settings, MLP-based combinations outperform linear and Gaussian process-based ensembles, and introducing soft conditioning yields gains over linear models.
title Dynamic Meta-Metrics: Source-Sentence Conditioned Weighting for MT Evaluation
topic Computation and Language
url https://arxiv.org/abs/2605.09098