Post-edits Are Preferences Too

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Berger, Nathaniel, Exel, Miriam, Huck, Matthias, Riezler, Stefan
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915163990392832
author Berger, Nathaniel
Exel, Miriam
Huck, Matthias
Riezler, Stefan
author_facet Berger, Nathaniel
Exel, Miriam
Huck, Matthias
Riezler, Stefan
contents Preference Optimization (PO) techniques are currently one of the state of the art techniques for fine-tuning large language models (LLMs) on pairwise preference feedback from human annotators. However, in machine translation, this sort of feedback can be difficult to solicit. Additionally, Kreutzer et al. (2018) have shown that, for machine translation, pairwise preferences are less reliable than other forms of human feedback, such as 5-point ratings. We examine post-edits to see if they can be a source of reliable human preferences by construction. In PO, a human annotator is shown sequences $s_1$ and $s_2$ and asked for a preference judgment, %$s_1 > s_2$; while for post-editing, editors create $s_1$ and know that it should be better than $s_2$. We attempt to use these implicit preferences for PO and show that it helps the model move towards post-edit-like hypotheses and away from machine translation-like hypotheses. Furthermore, we show that best results are obtained by pre-training the model with supervised fine-tuning (SFT) on post-edits in order to promote post-edit-like hypotheses to the top output ranks.
format Preprint
id arxiv_https___arxiv_org_abs_2410_02320
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Post-edits Are Preferences Too
Berger, Nathaniel
Exel, Miriam
Huck, Matthias
Riezler, Stefan
Computation and Language
Artificial Intelligence
Machine Learning
Preference Optimization (PO) techniques are currently one of the state of the art techniques for fine-tuning large language models (LLMs) on pairwise preference feedback from human annotators. However, in machine translation, this sort of feedback can be difficult to solicit. Additionally, Kreutzer et al. (2018) have shown that, for machine translation, pairwise preferences are less reliable than other forms of human feedback, such as 5-point ratings. We examine post-edits to see if they can be a source of reliable human preferences by construction. In PO, a human annotator is shown sequences $s_1$ and $s_2$ and asked for a preference judgment, %$s_1 > s_2$; while for post-editing, editors create $s_1$ and know that it should be better than $s_2$. We attempt to use these implicit preferences for PO and show that it helps the model move towards post-edit-like hypotheses and away from machine translation-like hypotheses. Furthermore, we show that best results are obtained by pre-training the model with supervised fine-tuning (SFT) on post-edits in order to promote post-edit-like hypotheses to the top output ranks.
title Post-edits Are Preferences Too
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2410.02320