{"id":"https://openalex.org/W4416034516","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1210","title":"Local Normalization Distortion and the Thermodynamic Formalism of Decoding Strategies for Large Language Models","display_name":"Local Normalization Distortion and the Thermodynamic Formalism of Decoding Strategies for Large Language Models","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034516","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1210"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1210","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1210","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1210.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.1210.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5083538214","display_name":"Tom Kempton","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tom Kempton","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5082052901","display_name":"Stuart A. Burrell","orcid":"https://orcid.org/0000-0002-6333-1750"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Stuart Burrell","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.17027535,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"22216","last_page":"22231"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12072","display_name":"Machine Learning and Algorithms","score":0.1014999970793724,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12072","display_name":"Machine Learning and Algorithms","score":0.1014999970793724,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.10010000318288803,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.09589999914169312,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/formalism","display_name":"Formalism (music)","score":0.6832000017166138},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5145000219345093},{"id":"https://openalex.org/keywords/normalization","display_name":"Normalization (sociology)","score":0.49230000376701355},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.2759999930858612},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.2759000062942505}],"concepts":[{"id":"https://openalex.org/C73301696","wikidata":"https://www.wikidata.org/wiki/Q5469984","display_name":"Formalism (music)","level":3,"score":0.6832000017166138},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5396999716758728},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5163000226020813},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5145000219345093},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.49230000376701355},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3926999866962433},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.367000013589859},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.36419999599456787},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.2759999930858612},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C2781170535","wikidata":"https://www.wikidata.org/wiki/Q30587856","display_name":"Noisy data","level":2,"score":0.2567000091075897}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.1210","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1210","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1210.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},{"id":"pmh:oai:pure.atira.dk:openaire_cris_publications/802ba8f6-badd-49c9-8773-28d1b5839771","is_oa":true,"landing_page_url":"https://research.manchester.ac.uk/en/publications/802ba8f6-badd-49c9-8773-28d1b5839771","pdf_url":null,"source":{"id":"https://openalex.org/S4306400662","display_name":"Research Explorer (The University of Manchester)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I28407311","host_organization_name":"University of Manchester","host_organization_lineage":["https://openalex.org/I28407311"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Kempton, T & Burrell, S 2025, Local Normalization Distortion and the Thermodynamic Formalism of Decoding Strategies for Large Language Models. in Findings of the Association for Computational Linguistics: EMNLP 2025.","raw_type":"info:eu-repo/semantics/conferenceObject"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1210","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1210","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1210.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034516.pdf","grobid_xml":"https://content.openalex.org/works/W4416034516.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Advances":[0],"in":[1,11,48,57,78,108],"hardware":[2],"and":[3,23,84,111,126,133,142,155],"language":[4,13,68,80],"model":[5],"architecture":[6],"have":[7],"spurred":[8],"a":[9,58,120],"revolution":[10],"natural":[12],"generation.However,":[14],"autoregressive":[15],"models":[16,69],"compute":[17],"probability":[18],"distributions":[19],"over":[20],"next-token":[21],"choices,":[22],"sampling":[24],"from":[25],"these":[26],"distributions,":[27],"known":[28],"as":[29,75],"decoding,":[30],"has":[31],"received":[32],"significantly":[33],"less":[34],"attention":[35],"than":[36],"other":[37],"design":[38,151],"choices.Existing":[39],"decoding":[40,65,73,124,153],"strategies":[41,66,125],"are":[42,51],"largely":[43],"based":[44],"on":[45,136],"heuristics,":[46],"resulting":[47],"methods":[49],"that":[50,115],"difficult":[52],"to":[53,102,106],"apply":[54],"or":[55],"improve":[56],"principled":[59],"manner.We":[60],"develop":[61],"the":[62,79,86,94,97,128,140,150,156],"theory":[63,83],"of":[64,81,96,123,130,144,152,158],"for":[67,139,149],"by":[70],"expressing":[71],"popular":[72],"algorithms":[74,154],"equilibrium":[76],"states":[77],"ergodic":[82],"stating":[85],"objective":[87],"functions":[88],"they":[89],"optimize.Using":[90],"this,":[91],"we":[92],"analyze":[93],"effect":[95,135],"local":[98,116],"normalization":[99,117],"step":[100],"required":[101],"make":[103],"probabilities":[104],"sum":[105],"one":[107],"top-k,":[109],"nucleus,":[110],"temperature":[112],"sampling.We":[113],"argue":[114],"distortion":[118,132],"is":[119],"fundamental":[121],"defect":[122],"quantify":[127],"size":[129],"this":[131],"its":[134],"mathematical":[137],"proxies":[138],"quality":[141],"diversity":[143],"generated":[145],"text.This":[146],"yields":[147],"conclusions":[148],"detection":[157],"machine-generated":[159],"text.":[160]},"counts_by_year":[],"updated_date":"2026-07-02T09:51:11.867554","created_date":"2025-11-08T00:00:00"}