Giulia Rambelli, Emmanuele Chersoni, Davide Testa, Philippe Blache, Alessandro Lenci
{"title":"Neural Generative Models and the Parallel Architecture of Language: A Critical Review and Outlook","authors":"Giulia Rambelli, Emmanuele Chersoni, Davide Testa, Philippe Blache, Alessandro Lenci","doi":"10.1111/tops.12733","DOIUrl":null,"url":null,"abstract":"According to the parallel architecture, syntactic and semantic information processing are two separate streams that interact selectively during language comprehension. While considerable effort is put into psycho‐ and neurolinguistics to understand the interchange of processing mechanisms in human comprehension, the nature of this interaction in recent neural Large Language Models remains elusive. In this article, we revisit influential linguistic and behavioral experiments and evaluate the ability of a large language model, GPT‐3, to perform these tasks. The model can solve semantic tasks autonomously from syntactic realization in a manner that resembles human behavior. However, the outcomes present a complex and variegated picture, leaving open the question of how Language Models could learn structured conceptual representations.","PeriodicalId":47822,"journal":{"name":"Topics in Cognitive Science","volume":"11 1","pages":""},"PeriodicalIF":2.9000,"publicationDate":"2024-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Topics in Cognitive Science","FirstCategoryId":"102","ListUrlMain":"https://doi.org/10.1111/tops.12733","RegionNum":2,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"PSYCHOLOGY, EXPERIMENTAL","Score":null,"Total":0}
引用次数: 0
Abstract
According to the parallel architecture, syntactic and semantic information processing are two separate streams that interact selectively during language comprehension. While considerable effort is put into psycho‐ and neurolinguistics to understand the interchange of processing mechanisms in human comprehension, the nature of this interaction in recent neural Large Language Models remains elusive. In this article, we revisit influential linguistic and behavioral experiments and evaluate the ability of a large language model, GPT‐3, to perform these tasks. The model can solve semantic tasks autonomously from syntactic realization in a manner that resembles human behavior. However, the outcomes present a complex and variegated picture, leaving open the question of how Language Models could learn structured conceptual representations.
期刊介绍:
Topics in Cognitive Science (topiCS) is an innovative new journal that covers all areas of cognitive science including cognitive modeling, cognitive neuroscience, cognitive anthropology, and cognitive science and philosophy. topiCS aims to provide a forum for: -New communities of researchers- New controversies in established areas- Debates and commentaries- Reflections and integration The publication features multiple scholarly papers dedicated to a single topic. Some of these topics will appear together in one issue, but others may appear across several issues or develop into a regular feature. Controversies or debates started in one issue may be followed up by commentaries in a later issue, etc. However, the format and origin of the topics will vary greatly.