Post-training for Efficient Communication via Convention Formation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Hua, Yilun, Wang, Evan, Artzi, Yoav
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912528125132800
author Hua, Yilun
Wang, Evan
Artzi, Yoav
author_facet Hua, Yilun
Wang, Evan
Artzi, Yoav
contents Humans communicate with increasing efficiency in multi-turn interactions, by adapting their language and forming ad-hoc conventions. In contrast, prior work shows that LLMs do not naturally show this behavior. We develop a post-training process to develop this ability through targeted fine-tuning on heuristically identified demonstrations of convention formation. We evaluate with two new benchmarks focused on this capability. First, we design a focused, cognitively-motivated interaction benchmark that consistently elicits strong convention formation trends in humans. Second, we create a new document-grounded reference completion task that reflects in-the-wild convention formation behavior. Our studies show significantly improved convention formation abilities in post-trained LLMs across the two evaluation methods.
format Preprint
id arxiv_https___arxiv_org_abs_2508_06482
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Post-training for Efficient Communication via Convention Formation
Hua, Yilun
Wang, Evan
Artzi, Yoav
Computation and Language
Artificial Intelligence
Machine Learning
Humans communicate with increasing efficiency in multi-turn interactions, by adapting their language and forming ad-hoc conventions. In contrast, prior work shows that LLMs do not naturally show this behavior. We develop a post-training process to develop this ability through targeted fine-tuning on heuristically identified demonstrations of convention formation. We evaluate with two new benchmarks focused on this capability. First, we design a focused, cognitively-motivated interaction benchmark that consistently elicits strong convention formation trends in humans. Second, we create a new document-grounded reference completion task that reflects in-the-wild convention formation behavior. Our studies show significantly improved convention formation abilities in post-trained LLMs across the two evaluation methods.
title Post-training for Efficient Communication via Convention Formation
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2508.06482