{"id":"https://openalex.org/W7166846637","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1748","title":"GSM-Noise: Exploring and Enhancing Large Language Models\u2019 Reasoning under Noisy Inputs","display_name":"GSM-Noise: Exploring and Enhancing Large Language Models\u2019 Reasoning under Noisy Inputs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166846637","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1748"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1748","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1748","pdf_url":"https://aclanthology.org/2026.findings-acl.1748.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: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1748.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139780580","display_name":"Zhengxin Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhengxin Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139799739","display_name":"Chengyu Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chengyu Huang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139775828","display_name":"Xufu Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xufu Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139831532","display_name":"Dan Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dan Zhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074353443","display_name":"Jinyan Su","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jinyan Su","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139791683","display_name":"Claire Cardie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Claire Cardie","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":0.83623258,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"35020","last_page":"35045"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.24779999256134033,"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/T10028","display_name":"Topic Modeling","score":0.24779999256134033,"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/T12031","display_name":"Speech and dialogue systems","score":0.08420000225305557,"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":0.05939999967813492,"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/noise","display_name":"Noise (video)","score":0.29660001397132874},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.2793999910354614},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.265500009059906},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.26100000739097595},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.24869999289512634}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6157000064849854},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5335999727249146},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.35409998893737793},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.29660001397132874},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2793999910354614},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.265500009059906},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.24869999289512634},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.24289999902248383},{"id":"https://openalex.org/C9652623","wikidata":"https://www.wikidata.org/wiki/Q190109","display_name":"Field (mathematics)","level":2,"score":0.2409999966621399}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1748","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1748","pdf_url":"https://aclanthology.org/2026.findings-acl.1748.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: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.1748","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1748","pdf_url":"https://aclanthology.org/2026.findings-acl.1748.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: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166846637.pdf","grobid_xml":"https://content.openalex.org/works/W7166846637.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,101,114],"(LLMs)":[3],"have":[4],"demonstrated":[5],"impressive":[6],"reasoning":[7,44,97,196],"capabilities,":[8],"yet":[9],"they":[10],"often":[11],"struggle":[12],"when":[13],"dealing":[14],"with":[15,26,169,172,176],"complex,":[16],"ill-formed,":[17],"or":[18],"noisy":[19,122],"inputs":[20,130],"that":[21,95,126,160],"frequently":[22],"occur":[23],"in":[24,51,137,197],"interactions":[25],"real":[27],"users.LLMs":[28],"typically":[29],"lack":[30],"crucial":[31],"refining":[32],"capabilities":[33],"needed":[34],"to":[35,65,89,143,164,188],"filter":[36],"out":[37],"irrelevant":[38],"details,":[39],"restructure":[40],"key":[41],"points":[42],"before":[43],"over":[45],"the":[46,96,180,190],"text":[47],"and":[48,54,104,154,174,192],"responding,":[49],"resulting":[50],"suboptimal":[52],"performance":[53,166],"incorrect":[55],"answers.From":[56],"an":[57,184],"information":[58],"theory":[59],"perspective,":[60],"this":[61,75,145],"behavior":[62],"is":[63],"akin":[64],"decoding":[66],"a":[67,81],"high-entropy":[68],"problem":[69],"without":[70],"first":[71,78,128],"reducing":[72,132],"its":[73],"entropy.In":[74],"work,":[76],"we":[77,124],"introduce":[79],"GSM-Noise,":[80],"benchmark":[82],"featuring":[83],"gradeschool":[84],"math":[85],"problems":[86],"systematically":[87],"perturbed":[88],"reflect":[90],"real-world":[91,198],"input":[92,161],"variability.We":[93],"show":[94,159],"ability":[98],"of":[99,182,194],"open-source":[100],"(e.g.,":[102],"LLaMA":[103],"Qwen":[105],"series)":[106],"can":[107],"be":[108],"compromised":[109],"by":[110],"noise,":[111],"while":[112],"closed-source":[113],"are":[115],"more":[116],"robust.To":[117],"improve":[118],"LLM":[119,195],"robustness":[120,191],"under":[121],"conditions,":[123],"propose":[125],"LLMs":[127],"refine":[129],"-thereby":[131],"their":[133],"entropy":[134],"-before":[135],"engaging":[136],"in-depth":[138],"analysis.We":[139],"investigate":[140],"three":[141],"approaches":[142],"instill":[144],"refinement":[146,162,186],"capability:":[147],"prompt":[148],"engineering":[149],"(PE),":[150],"supervised":[151],"finetuning":[152],"(SFT),":[153],"reinforcement":[155],"learning":[156],"(RL).Experimental":[157],"results":[158,178],"leads":[163],"consistent":[165],"gains:":[167],"2-12%":[168],"PE,":[170],"4-13%":[171],"SFT,":[173],"3-25%":[175],"RL.These":[177],"highlight":[179],"importance":[181],"incorporating":[183],"explicit":[185],"phase":[187],"enhance":[189],"reliability":[193],"scenarios.":[199]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}