Priya Ronald D'Costa, Evan Rowbotham, Xinlan Emily Hu
{"title":"What you say or how you say it? Predicting Conflict Outcomes in Real and LLM-Generated Conversations","authors":"Priya Ronald D'Costa, Evan Rowbotham, Xinlan Emily Hu","doi":"arxiv-2409.09338","DOIUrl":null,"url":null,"abstract":"When conflicts escalate, is it due to what is said or how it is said? In the\nconflict literature, two theoretical approaches take opposing views: one\nfocuses on the content of the disagreement, while the other focuses on how it\nis expressed. This paper aims to integrate these two perspectives through a\ncomputational analysis of 191 communication features -- 128 related to\nexpression and 63 to content. We analyze 1,200 GPT-4 simulated conversations\nand 12,630 real-world discussions from Reddit. We find that expression features\nmore reliably predict destructive conflict outcomes across both settings,\nalthough the most important features differ. In the Reddit data, conversational\ndynamics such as turn-taking and conversational equality are highly predictive,\nbut they are not predictive in simulated conversations. These results may\nsuggest a possible limitation in simulating social interactions with language\nmodels, and we discuss the implications for our findings on building social\ncomputing systems.","PeriodicalId":501032,"journal":{"name":"arXiv - CS - Social and Information Networks","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Social and Information Networks","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.09338","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
When conflicts escalate, is it due to what is said or how it is said? In the
conflict literature, two theoretical approaches take opposing views: one
focuses on the content of the disagreement, while the other focuses on how it
is expressed. This paper aims to integrate these two perspectives through a
computational analysis of 191 communication features -- 128 related to
expression and 63 to content. We analyze 1,200 GPT-4 simulated conversations
and 12,630 real-world discussions from Reddit. We find that expression features
more reliably predict destructive conflict outcomes across both settings,
although the most important features differ. In the Reddit data, conversational
dynamics such as turn-taking and conversational equality are highly predictive,
but they are not predictive in simulated conversations. These results may
suggest a possible limitation in simulating social interactions with language
models, and we discuss the implications for our findings on building social
computing systems.