{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T17:58:40Z","timestamp":1781805520955,"version":"3.54.5"},"reference-count":75,"publisher":"Public Library of Science (PLoS)","issue":"6","license":[{"start":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T00:00:00Z","timestamp":1781740800000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"Ministry of Education of the Republic of Korea and the National Research Foundation of Korea","award":["NRF-2020S1A5B5A16082578"],"award-info":[{"award-number":["NRF-2020S1A5B5A16082578"]}]}],"content-domain":{"domain":["www.plosone.org"],"crossmark-restriction":false},"short-container-title":["PLoS One"],"abstract":"<jats:p>\n                    Transformer-based language models have demonstrated sensitivity to a range of linguistic dependencies, yet it remains unclear how they represent information-structural focus and integrate discourse and lexical focus cues during ellipsis resolution. We investigated GPT-style models\u2019 interpretation of elliptical remnant continuations in double-object constructions by manipulating contextual focus via preceding interrogatives (\n                    <jats:italic>who<\/jats:italic>\n                    vs.\n                    <jats:italic>what<\/jats:italic>\n                    ) and lexical focus via the particle\n                    <jats:italic>only<\/jats:italic>\n                    , whose surface position was varied. Using word-by-word surprisal as an index of processing difficulty, we conducted three experiments with GPT-2 models (Small\u2013XL) and GPT-Neo. In Experiment 1 (no\n                    <jats:italic>only<\/jats:italic>\n                    ), models robustly tracked the\n                    <jats:italic>wh<\/jats:italic>\n                    -induced discourse focus, assigning higher surprisal to remnants that mismatched the contextually focused constituent. In Experiment 2 (\n                    <jats:italic>only<\/jats:italic>\n                    preceding the indirect object), contextual focus continued to dominate, indicating that discourse cues were maintained despite the presence of a competing lexical marker. In Experiment 3 (\n                    <jats:italic>only<\/jats:italic>\n                    preceding the direct object), lexical focus effects became stronger: models favored remnants aligned with the lexically biased direct object, consistent with locality-based cue weighting when\n                    <jats:italic>only<\/jats:italic>\n                    is adjacent to that object. Comparisons with human reaction-time data revealed broad convergence in contextual-focus sensitivity but divergence when the remnant was compatible with one cue but not the other, with GPT-style models exhibiting a stronger bias toward alignment with\n                    <jats:italic>only<\/jats:italic>\n                    than humans. Together, these findings suggest that the tested models maintain discourse-level focus representations while integrating multiple focus cues in a proximity-sensitive manner, revealing both overlap and limits in their alignment with human processing.\n                  <\/jats:p>","DOI":"10.1371\/journal.pone.0351924","type":"journal-article","created":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T17:49:54Z","timestamp":1781804994000},"page":"e0351924","update-policy":"https:\/\/doi.org\/10.1371\/journal.pone.corrections_policy","source":"Crossref","is-referenced-by-count":0,"title":["Focus shifts in contextual and lexical cue interactions in GPT models"],"prefix":"10.1371","volume":"21","author":[{"given":"Wonil","family":"Chung","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0003-1076-0310","authenticated-orcid":true,"given":"Keonwoo","family":"Koo","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"340","published-online":{"date-parts":[[2026,6,18]]},"reference":[{"key":"pone.0351924.ref001","article-title":"Using computational models to test syntactic learnability.","author":"E Wilcox","year":"2021","journal-title":"Proc Annu Meet Cogn Sci Soc"},{"issue":"6","key":"pone.0351924.ref002","doi-asserted-by":"crossref","first-page":"875","DOI":"10.1080\/01690965.2012.668197","article-title":"Processing contextual and lexical cues to focus: Evidence from eye movements in reading","volume":"28","author":"A Sauermann","year":"2012","journal-title":"Language and Cognitive Processes"},{"key":"pone.0351924.ref003","first-page":"147","article-title":"Contrastive Focus.","volume-title":"The Notions of Information Structure","author":"M. Zimmermann","year":"2007"},{"issue":"10","key":"pone.0351924.ref004","doi-asserted-by":"crossref","first-page":"1423","DOI":"10.1080\/17470210601100563","article-title":"Focus identification during sentence comprehension: evidence from eye movements","volume":"60","author":"KB Paterson","year":"2007","journal-title":"Q J Exp Psychol (Hove)"},{"issue":"2","key":"pone.0351924.ref005","doi-asserted-by":"crossref","first-page":"199","DOI":"10.1017\/S0022226700016613","article-title":"Notes on transitivity and theme in English: Part 2","volume":"3","author":"MAK Halliday","year":"1967","journal-title":"J Ling"},{"key":"pone.0351924.ref006","author":"Y Hao","year":"2020"},{"issue":"3","key":"pone.0351924.ref007","doi-asserted-by":"crossref","first-page":"375","DOI":"10.1207\/s15516709cog0000_25","article-title":"An activation-based model of sentence processing as skilled memory retrieval","volume":"29","author":"RL Lewis","year":"2005","journal-title":"Cogn Sci"},{"key":"pone.0351924.ref008","doi-asserted-by":"crossref","first-page":"393","DOI":"10.1017\/S0022226707004628","article-title":"Machine learning theory and practice as a source of insight into universal grammar","volume":"43","author":"S Lappin","year":"2007","journal-title":"J Linguist"},{"issue":"3","key":"pone.0351924.ref009","doi-asserted-by":"crossref","first-page":"1126","DOI":"10.1016\/j.cognition.2007.05.006","article-title":"Expectation-based syntactic comprehension","volume":"106","author":"R Levy","year":"2008","journal-title":"Cognition"},{"key":"pone.0351924.ref010","doi-asserted-by":"crossref","first-page":"777963","DOI":"10.3389\/frai.2022.777963","article-title":"Comparison of Structural Parsers and Neural Language Models as Surprisal Estimators","volume":"5","author":"B-D Oh","year":"2022","journal-title":"Front Artif Intell"},{"key":"pone.0351924.ref011","doi-asserted-by":"crossref","first-page":"1451","DOI":"10.1162\/tacl_a_00612","article-title":"Testing the predictions of surprisal theory in 11 languages","volume":"11","author":"EG Wilcox","year":"2023","journal-title":"Trans Assoc Comput Linguist"},{"key":"pone.0351924.ref012","article-title":"On the predictive power of neural language models for human real-time comprehension behavior","author":"E Wilcox","year":"2020","journal-title":"arXiv preprint"},{"key":"pone.0351924.ref013","first-page":"1","article-title":"A probabilistic earley parser as a psycholinguistic model.","author":"J Hale","year":"2001","journal-title":"Second meeting of the North American Chapter of the Association for Computational Linguistics on Language technologies 2001 - NAACL \u201901"},{"key":"pone.0351924.ref014","author":"N Chomsky","year":"1971","journal-title":"Deep structure, surface structure, and semantics: An interdisciplinary reader in philosophy, linguistics and psychology"},{"issue":"1","key":"pone.0351924.ref015","first-page":"23","article-title":"Extraposition and the complement principle","volume":"21","author":"PW Culicover","year":"1990","journal-title":"Linguistic Inquiry"},{"issue":"1","key":"pone.0351924.ref016","first-page":"41","article-title":"Questions in Montague English","volume":"10","author":"CL Hamblin","year":"1973","journal-title":"Foundations of Language"},{"key":"pone.0351924.ref017","doi-asserted-by":"crossref","first-page":"1320","DOI":"10.3389\/fpsyg.2019.01320","article-title":"Proximity and Same Case Marking Do Not Increase Attraction Effect in Comprehension: Evidence From Eye-Tracking Experiments in Korean","volume":"10","author":"N Kwon","year":"2019","journal-title":"Front Psychol"},{"key":"pone.0351924.ref018","volume-title":"Sense and sensitivity: How focus determines meaning","author":"DI Beaver","year":"2009"},{"issue":"3","key":"pone.0351924.ref019","doi-asserted-by":"crossref","first-page":"228","DOI":"10.1016\/j.bandl.2007.04.004","article-title":"Violations of information structure: an electrophysiological study of answers to wh-questions","volume":"102","author":"HW Cowles","year":"2007","journal-title":"Brain Lang"},{"key":"pone.0351924.ref020","article-title":"ERP correlates of focus accentuation in Dutch.","author":"DV Dimitrova","journal-title":"Proceedings of the Speech and Prosody 2010a"},{"key":"pone.0351924.ref021","doi-asserted-by":"crossref","DOI":"10.21437\/SpeechProsody.2010-80","article-title":"Focus particles and prosody processing in Dutch: Evidence from ERPs","author":"DV Dimitrova","year":"2010","journal-title":"Speech Prosody"},{"issue":"1","key":"pone.0351924.ref022","doi-asserted-by":"crossref","first-page":"49","DOI":"10.1016\/0010-0277(79)90010-6","article-title":"Semantic focus and sentence comprehension","volume":"7","author":"A Cutler","year":"1979","journal-title":"Cognition"},{"key":"pone.0351924.ref023","doi-asserted-by":"crossref","first-page":"202","DOI":"10.3765\/salt.v6i0.2767","article-title":"On the interface principles for intonational focus","author":"M Rooth","year":"1996","journal-title":"Semantics and Linguistic Theory"},{"issue":"2","key":"pone.0351924.ref024","doi-asserted-by":"crossref","first-page":"245","DOI":"10.2307\/417867","article-title":"Identificational Focus Versus Information Focus","volume":"74","author":"K\u00c9 Kiss","year":"1998","journal-title":"Language"},{"key":"pone.0351924.ref025","doi-asserted-by":"crossref","DOI":"10.1075\/sigla.4","author":"M Rochemont","year":"1986","journal-title":"Focus in generative grammar"},{"issue":"4","key":"pone.0351924.ref026","doi-asserted-by":"crossref","first-page":"365","DOI":"10.1177\/002383099503800403","article-title":"Focus, accent, and argument structure: Effects on language comprehension","volume":"38","author":"S Birch","year":"1995","journal-title":"Language and Speech"},{"issue":"4","key":"pone.0351924.ref027","doi-asserted-by":"crossref","first-page":"571","DOI":"10.1016\/S0749-596X(02)00018-9","article-title":"Effects of varying focus and accenting of adjuncts on the comprehension of utterances","volume":"47","author":"S Birch","year":"2002","journal-title":"Journal of Memory and Language"},{"issue":"1","key":"pone.0351924.ref028","doi-asserted-by":"crossref","first-page":"64","DOI":"10.3758\/BF03197663","article-title":"Intonational marking of given and new information: some consequences for comprehension","volume":"11","author":"JK Bock","year":"1983","journal-title":"Mem Cognit"},{"issue":"5","key":"pone.0351924.ref029","doi-asserted-by":"crossref","first-page":"1512","DOI":"10.1121\/1.395195","article-title":"Accents, focus distribution, and the perceived distribution of given and new information: An experiment","volume":"82","author":"SG Nooteboom","year":"1987","journal-title":"J Acoust Soc Am"},{"issue":"5","key":"pone.0351924.ref030","doi-asserted-by":"crossref","first-page":"530","DOI":"10.1016\/S0022-5371(74)80005-8","article-title":"Surface structure and presupposition","volume":"13","author":"PA Hornby","year":"1974","journal-title":"Journal of Verbal Learning and Verbal Behavior"},{"issue":"3","key":"pone.0351924.ref031","doi-asserted-by":"crossref","first-page":"271","DOI":"10.1016\/0749-596X(85)90028-2","article-title":"The role of discourse structure in understanding anaphora","volume":"24","author":"BC Malt","year":"1985","journal-title":"Journal of Memory and Language"},{"issue":"11","key":"pone.0351924.ref032","doi-asserted-by":"crossref","DOI":"10.1371\/journal.pone.0206616","article-title":"Cue integration during sentence comprehension: Electrophysiological evidence from ellipsis","volume":"13","author":"AE Martin","year":"2018","journal-title":"PLoS One"},{"issue":"1","key":"pone.0351924.ref033","doi-asserted-by":"crossref","first-page":"75","DOI":"10.1177\/00238309000430010301","article-title":"Focus and the Interpretation of Pitch Accent: Disambiguating Embedded Questions","volume":"43","author":"A Schafer","year":"2000","journal-title":"Lang Speech"},{"issue":"5","key":"pone.0351924.ref034","doi-asserted-by":"crossref","first-page":"549","DOI":"10.1016\/0022-5371(76)90049-9","article-title":"Thematic structure and the integration of linguistic information","volume":"15","author":"M Singer","year":"1976","journal-title":"Journal of Verbal Learning and Verbal Behavior"},{"key":"pone.0351924.ref035","author":"N Kadmon","year":"2001","journal-title":"Formal pragmatics: semantics, pragmatics, preposition, and focus"},{"key":"pone.0351924.ref036","doi-asserted-by":"crossref","first-page":"14691","DOI":"10.18653\/v1\/2024.acl-long.787","article-title":"Mission: Impossible Language Models.","author":"J Kallini","year":"2024","journal-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"},{"issue":"5","key":"pone.0351924.ref037","doi-asserted-by":"crossref","first-page":"653","DOI":"10.3758\/BF03211306","article-title":"Linguistic focus affects eye movements during reading","volume":"25","author":"S Birch","year":"1997","journal-title":"Mem Cognit"},{"key":"pone.0351924.ref038","author":"R Jackendoff","year":"1972","journal-title":"Semantic interpretation in generative grammar"},{"issue":"3","key":"pone.0351924.ref039","first-page":"369","article-title":"The status of thematic relations in linguistic theory","volume":"18","author":"R Jackendoff","year":"1987","journal-title":"Linguistic inquiry"},{"key":"pone.0351924.ref040","first-page":"430","article-title":"Language and Knowledge","author":"J Kaminsky","year":"1959","journal-title":"Revue Internationale de Philosophie"},{"key":"pone.0351924.ref041","first-page":"190","article-title":"Concessive relations as the dual of causal relations","volume":"12","author":"E K\u00f6nig","year":"1991","journal-title":"Semantic universals and universal semantics"},{"key":"pone.0351924.ref042","doi-asserted-by":"crossref","first-page":"243","DOI":"10.1556\/ALing.55.2008.3-4.2","article-title":"Basic notions of information structure","volume":"55","author":"M Krifka","year":"2008","journal-title":"Acta Linguistica Hungarica"},{"issue":"1","key":"pone.0351924.ref043","doi-asserted-by":"crossref","first-page":"75","DOI":"10.1007\/BF02342617","article-title":"A theory of focus interpretation","volume":"1","author":"M Rooth","year":"1992","journal-title":"Nat Lang Seman"},{"key":"pone.0351924.ref044","author":"M Rooth","year":"1985","journal-title":"Association with focus"},{"key":"pone.0351924.ref045","volume-title":"The processing cost of reference set computation: Guess patterns in acquisition","author":"T Reinhart","year":"1999"},{"key":"pone.0351924.ref046","article-title":"Attention is all you need","volume":"30","author":"A Vaswani","year":"2017","journal-title":"Advances in neural information processing systems"},{"key":"pone.0351924.ref047","doi-asserted-by":"crossref","first-page":"891","DOI":"10.1162\/tacl_a_00403","article-title":"Sensitivity as a complexity measure for sequence classification tasks","volume":"9","author":"M Hahn","year":"2021","journal-title":"Trans Assoc Comput Linguist"},{"key":"pone.0351924.ref048","article-title":"Modern language models refute Chomsky\u2019s approach to language.","author":"ST. Piantadosi","journal-title":"From fieldwork to linguistic theory: A tribute to Dan Everett. Berlin: Language Science Press"},{"key":"pone.0351924.ref049","article-title":"Language models are unsupervised multitask learners","author":"A Radford","year":"2019","journal-title":"OpenAI blog"},{"key":"pone.0351924.ref050","article-title":"RNNs as psycholinguistic subjects: Syntactic state and grammatical dependency","author":"R Futrell","year":"2018","journal-title":"arXiv preprint"},{"key":"pone.0351924.ref051","doi-asserted-by":"crossref","DOI":"10.1007\/978-3-031-02165-7","volume-title":"Neural network methods for natural language processing","author":"Y Goldberg","year":"2017"},{"key":"pone.0351924.ref052","doi-asserted-by":"crossref","first-page":"104510","DOI":"10.1016\/j.jml.2024.104510","article-title":"Large-scale benchmark yields no evidence that language model surprisal explains syntactic disambiguation difficulty","volume":"137","author":"KJ Huang","year":"2024","journal-title":"J Mem Lang"},{"issue":"1","key":"pone.0351924.ref053","doi-asserted-by":"crossref","DOI":"10.1353\/lan.2019.0015","article-title":"What Can Linguistics and Deep Learning Contribute to Each Other? Response to Pater","volume":"95","author":"T Linzen","year":"2019","journal-title":"Language"},{"issue":"1","key":"pone.0351924.ref054","doi-asserted-by":"crossref","first-page":"195","DOI":"10.1146\/annurev-linguistics-032020-051035","article-title":"Syntactic Structure from Deep Learning","volume":"7","author":"T Linzen","year":"2021","journal-title":"Annu Rev Linguist"},{"key":"pone.0351924.ref055","article-title":"Targeted syntactic evaluation of language models","author":"R Marvin","year":"2018","journal-title":"arXiv preprint"},{"key":"pone.0351924.ref056","article-title":"Modeling garden path effects without explicit hierarchical syntax.","author":"M Van Schijndel","year":"2018","journal-title":"Proceedings of the Annual Meeting of the Cognitive Science Society"},{"key":"pone.0351924.ref057","doi-asserted-by":"crossref","first-page":"377","DOI":"10.1162\/tacl_a_00321","article-title":"BLiMP: The benchmark of linguistic minimal pairs for English","volume":"8","author":"A Warstadt","year":"2020","journal-title":"Transactions of the Association for Computational Linguistics"},{"key":"pone.0351924.ref058","article-title":"Comparing gated and simple recurrent neural network architectures as models of human sentence processing.","author":"C Aurnhammer","year":"2019","journal-title":"Proceedings of the Annual Meeting of the Cognitive Science Society"},{"key":"pone.0351924.ref059","author":"ER Futrell","year":"2019"},{"key":"pone.0351924.ref060","doi-asserted-by":"crossref","first-page":"10","DOI":"10.18653\/v1\/W18-0102","article-title":"Predictive power of word surprisal for reading times is a linear function of language model quality.","author":"A Goodkind","year":"2018","journal-title":"Proceedings of the 8th Workshop on Cognitive Modeling and Computational Linguistics (CMCL 2018)"},{"key":"pone.0351924.ref061","author":"L He","year":"2024"},{"key":"pone.0351924.ref062","doi-asserted-by":"crossref","first-page":"1725","DOI":"10.18653\/v1\/2020.acl-main.158","article-title":"A systematic assessment of syntactic generalization in neural language models.","author":"J Hu","year":"2020","journal-title":"Proceedings of the 58th annual meeting of the association for computational linguistics"},{"key":"pone.0351924.ref063","article-title":"Accounting for agreement phenomena in sentence comprehension with transformer language models: Effects of similarity-based interference on surprisal and attention","author":"SH R","year":"2021","journal-title":"arXiv preprint"},{"key":"pone.0351924.ref064","article-title":"What do RNN language models learn about filler-gap dependencies?","author":"E Wilcox","year":"2018","journal-title":"arXiv preprint"},{"key":"pone.0351924.ref065","first-page":"8402","article-title":"Injecting structural hints: Using language models to study inductive biases in language learning.","author":"I Papadimitriou","year":"2023","journal-title":"Findings of the Association for Computational Linguistics: EMNLP 2023"},{"key":"pone.0351924.ref066","first-page":"1877","article-title":"Language models are few-shot learners","volume":"2020","author":"B Brown","year":"2020","journal-title":"Adv Neural Inf Process Syst"},{"key":"pone.0351924.ref067","first-page":"2","author":"D Bates","year":"2015","journal-title":"Package \u2018lme4\u2019. Convergence"},{"issue":"5","key":"pone.0351924.ref068","doi-asserted-by":"crossref","first-page":"171","DOI":"10.1111\/lnc3.12075","article-title":"Syntactic Memory in the Comprehension of Reflexive Dependencies: an Overview","volume":"8","author":"B Dillon","year":"2014","journal-title":"Language and Linguist Compass"},{"issue":"1","key":"pone.0351924.ref069","doi-asserted-by":"crossref","first-page":"116","DOI":"10.1037\/h0030186","article-title":"Retrieval processes in recognition memory: Effects of associative context","volume":"87","author":"E Tulving","year":"1971","journal-title":"J Exp Psychol"},{"key":"pone.0351924.ref070","first-page":"405","article-title":"Relation between encoding specificity and levels of processing","author":"E Tulving","year":"1979","journal-title":"Levels of Processing in Human Memory"},{"issue":"2","key":"pone.0351924.ref071","doi-asserted-by":"crossref","first-page":"206","DOI":"10.1016\/j.jml.2009.04.002","article-title":"Agreement attraction in comprehension: Representations and processes","volume":"61","author":"MW Wagers","year":"2009","journal-title":"Journal of Memory and Language"},{"issue":"1","key":"pone.0351924.ref072","doi-asserted-by":"crossref","first-page":"60","DOI":"10.1016\/j.bandl.2013.01.010","article-title":"Reconciling time, space and function: a new dorsal-ventral stream model of sentence comprehension","volume":"125","author":"I Bornkessel-Schlesewsky","year":"2013","journal-title":"Brain Lang"},{"issue":"3","key":"pone.0351924.ref073","doi-asserted-by":"crossref","first-page":"457","DOI":"10.1007\/s42113-024-00206-8","article-title":"Memory Retrieval in Online Sentence Parsing: Empirical Evidence, Computational Modelling, and Simulations","volume":"7","author":"H Fujita","year":"2024","journal-title":"Comput Brain Behav"},{"key":"pone.0351924.ref074","doi-asserted-by":"crossref","first-page":"30","DOI":"10.18653\/v1\/2024.cmcl-1.3","article-title":"Locally biased transformers better align with human reading times.","author":"AG de Varda","year":"2024","journal-title":"Proceedings of the Workshop on Cognitive Modeling and Computational Linguistics"},{"key":"pone.0351924.ref075","article-title":"On the importance of local information in transformer based models","author":"M Pande","year":"2020","journal-title":"arXiv preprint"}],"container-title":["PLOS One"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pone.0351924","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T17:50:04Z","timestamp":1781805004000},"score":1,"resource":{"primary":{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pone.0351924"}},"subtitle":[],"editor":[{"given":"Issa","family":"Atoum","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"editor"}]}],"short-title":[],"issued":{"date-parts":[[2026,6,18]]},"references-count":75,"journal-issue":{"issue":"6","published-online":{"date-parts":[[2026,6,18]]}},"URL":"https:\/\/doi.org\/10.1371\/journal.pone.0351924","relation":{},"ISSN":["1932-6203"],"issn-type":[{"value":"1932-6203","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,6,18]]}}}