Rate-Distortion-Perception Trade-off in Summarization

Arda, Enes; Yener, Aylin

Permalink

https://hdl.handle.net/2142/130289 Copy

Description

Title

Rate-Distortion-Perception Trade-off in Summarization

Author(s)

• Arda, Enes
• Yener, Aylin

Issue Date

2025-09-17

Keyword(s)

• Information theory
• Summarization
• Rate-distortion theory
• Rate-distortion-perception trade-off
• Language models

Abstract

This paper develops an information theoretic framework that formalizes the rate–distortion–perception trade-off for text summarizers. This trade-off, originally introduced in lossy compression, for example of images, quantifies the relationship between the length of a compressed representation (rate), the fidelity of the reconstruction (distortion), and its perceived naturalness (perception). Motivated by the structural similarities between lossy compression and text summarization, we formalize the rate-distortion-perception framework for text summaries. We define the summarizer rate–distortion–perception function (SRDPF) and show that it characterizes the fundamental performance limit of any summarizer. We then present an algorithm to compute the SRDPF for a given text source, and a practical method to estimate it from data. Experimental results demonstrate that the SRDPF provides a principled basis for evaluating summarizers and can guide the generation of higher-quality summaries. Additionally, we extend classical rate–distortion–perception results to the setting where the reference distribution differs from the source distribution.

Publisher

Allerton Conference on Communication, Control, and Computing

Series/Report Name or Number

2025 61st Allerton Conference on Communication, Control, and Computing Proceedings

ISSN

2836-4503

Type of Resource

Text

Genre of Resource

Conference Paper/Presentation

Language

eng

Handle URL

https://hdl.handle.net/2142/130289&&

Copyright and License Information

Copyright 2025 is held by Enes Arda and Aylin Yener.

Owning Collections