{"id":"https://openalex.org/W7204900746","doi":"https://doi.org/10.18653/2026.emnlp-main.751","title":"Toward Latent Language Model Skills Steering and Optimization: An Empirical Study","display_name":"Toward Latent Language Model Skills Steering and Optimization: An Empirical Study","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7204900746","doi":"https://doi.org/10.18653/2026.emnlp-main.751"},"language":"en","primary_location":{"id":"doi:10.18653/2026.emnlp-main.751","doi":"https://doi.org/10.18653/2026.emnlp-main.751","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.751","pdf_url":"https://aclanthology.org/2026.emnlp-main.751.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://aclanthology.org/2026.emnlp-main.751.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5146784277","display_name":"Xunyi Jiang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I57206974","display_name":"New York University","ror":"https://ror.org/0190ak572","country_code":"US","type":"education","lineage":["https://openalex.org/I57206974"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xunyi Jiang","raw_affiliation_strings":["New York University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"New York University","institution_ids":["https://openalex.org/I57206974"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101641321","display_name":"Junda Wu","orcid":"https://orcid.org/0009-0006-5146-4843","observed_orcids":["https://orcid.org/0009-0006-5146-4843"]},"institutions":[{"id":"https://openalex.org/I1306409833","display_name":"Adobe Systems (United States)","ror":"https://ror.org/059tvcg64","country_code":"US","type":"company","lineage":["https://openalex.org/I1306409833"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Junda Wu","raw_affiliation_strings":["Adobe Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Adobe Research","institution_ids":["https://openalex.org/I1306409833"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5143432528","display_name":"Yuxin Xiong","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yuxin Xiong","raw_affiliation_strings":["UC San Diego"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC San Diego","institution_ids":["https://openalex.org/I36258959"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111520079","display_name":"Shuying Yu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sheldon Yu","raw_affiliation_strings":["UC San Diego"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC San Diego","institution_ids":["https://openalex.org/I36258959"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122093612","display_name":"Tong Yu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1306409833","display_name":"Adobe Systems (United States)","ror":"https://ror.org/059tvcg64","country_code":"US","type":"company","lineage":["https://openalex.org/I1306409833"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tong Yu","raw_affiliation_strings":["Adobe Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Adobe Research","institution_ids":["https://openalex.org/I1306409833"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063845534","display_name":"David T. Arbour","orcid":"https://orcid.org/0000-0002-9932-7657","observed_orcids":["https://orcid.org/0000-0002-9932-7657"]},"institutions":[{"id":"https://openalex.org/I1306409833","display_name":"Adobe Systems (United States)","ror":"https://ror.org/059tvcg64","country_code":"US","type":"company","lineage":["https://openalex.org/I1306409833"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"David Arbour","raw_affiliation_strings":["Adobe Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Adobe Research","institution_ids":["https://openalex.org/I1306409833"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056964468","display_name":"Ritwik Sinha","orcid":"https://orcid.org/0000-0002-0344-1368","observed_orcids":["https://orcid.org/0000-0002-0344-1368"]},"institutions":[{"id":"https://openalex.org/I1306409833","display_name":"Adobe Systems (United States)","ror":"https://ror.org/059tvcg64","country_code":"US","type":"company","lineage":["https://openalex.org/I1306409833"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ritwik Sinha","raw_affiliation_strings":["Adobe Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Adobe Research","institution_ids":["https://openalex.org/I1306409833"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021827617","display_name":"Julian McAuley","orcid":"https://orcid.org/0000-0003-0955-7588","observed_orcids":["https://orcid.org/0000-0003-0955-7588"]},"institutions":[{"id":"https://openalex.org/I36258959","display_name":"University of California San Diego","ror":"https://ror.org/0168r3w48","country_code":"US","type":"education","lineage":["https://openalex.org/I36258959"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Julian McAuley","raw_affiliation_strings":["UC San Diego"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC San Diego","institution_ids":["https://openalex.org/I36258959"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5137782294","display_name":"Hongyi Wen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I57206974","display_name":"New York University","ror":"https://ror.org/0190ak572","country_code":"US","type":"education","lineage":["https://openalex.org/I57206974"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hongyi Wen","raw_affiliation_strings":["New York University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"New York University","institution_ids":["https://openalex.org/I57206974"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"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":0.69606735,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"17373","last_page":"17392"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5835000276565552,"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":0.5835000276565552,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.2401999980211258,"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":0.021299999207258224,"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/large-language-models","display_name":"large language models","score":0.8379999995231628},{"id":"https://openalex.org/keywords/procedural-skills","display_name":"procedural skills","score":0.800000011920929},{"id":"https://openalex.org/keywords/latent-representations","display_name":"latent representations","score":0.5460000038146973},{"id":"https://openalex.org/keywords/activation-space","display_name":"activation space","score":0.6389999985694885},{"id":"https://openalex.org/keywords/skill-composition","display_name":"skill composition","score":0.6600000262260437},{"id":"https://openalex.org/keywords/algorithmic-personalization","display_name":"algorithmic personalization","score":0.9589999914169312},{"id":"https://openalex.org/keywords/multi-step-reasoning","display_name":"multi-step reasoning","score":0.7910000085830688},{"id":"https://openalex.org/keywords/program-execution","display_name":"program execution","score":0.9380000233650208}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7186999917030334},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.6317999958992004},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.5873000025749207},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5404999852180481},{"id":"https://openalex.org/C183003079","wikidata":"https://www.wikidata.org/wiki/Q1000371","display_name":"Personalization","level":2,"score":0.527999997138977},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.4375999867916107},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.4327000081539154},{"id":"https://openalex.org/C113230428","wikidata":"https://www.wikidata.org/wiki/Q7247149","display_name":"Procedural modeling","level":2,"score":0.42489999532699585},{"id":"https://openalex.org/C124469403","wikidata":"https://www.wikidata.org/wiki/Q1813993","display_name":"Procedural knowledge","level":3,"score":0.42340001463890076},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.37310001254081726},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.36550000309944153},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.35499998927116394},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.35030001401901245},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3271999955177307},{"id":"https://openalex.org/C164749973","wikidata":"https://www.wikidata.org/wiki/Q18606","display_name":"Procedural memory","level":3,"score":0.3215000033378601},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.3012999892234802},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.29269999265670776},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2840000092983246},{"id":"https://openalex.org/C120558287","wikidata":"https://www.wikidata.org/wiki/Q1418502","display_name":"Procedural programming","level":4,"score":0.2824000120162964},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.27489998936653137},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.27390000224113464},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.26989999413490295},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.26010000705718994}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/2026.emnlp-main.751","doi":"https://doi.org/10.18653/2026.emnlp-main.751","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.751","pdf_url":"https://aclanthology.org/2026.emnlp-main.751.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2608.29459","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29459","pdf_url":"https://arxiv.org/pdf/2608.29459","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.29459","doi":"https://doi.org/10.48550/arxiv.2608.29459","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.29459","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/2026.emnlp-main.751","doi":"https://doi.org/10.18653/2026.emnlp-main.751","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.751","pdf_url":"https://aclanthology.org/2026.emnlp-main.751.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204900746.pdf","grobid_xml":"https://content.openalex.org/works/W7204900746.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7216179301","https://openalex.org/W7164899787","https://openalex.org/W7129059528","https://openalex.org/W7163699798","https://openalex.org/W7124467363","https://openalex.org/W7141388408","https://openalex.org/W7164372372","https://openalex.org/W7162468227","https://openalex.org/W7215129359","https://openalex.org/W7157156342"],"abstract_inverted_index":{"Skills,":[0],"as":[1,28,57,75],"a":[2,96,141,150],"useful":[3],"abstraction":[4],"for":[5],"the":[6,39,49],"procedural":[7,44,69,93,145],"capabilities":[8,45],"of":[9,41,144],"large":[10],"language":[11],"models":[12,16],"(LLMs),":[13],"capture":[14],"how":[15,42],"perform":[17],"structured,":[18],"multi-step":[19],"reasoning":[20],"and":[21,51,80,123],"program":[22],"execution.Existing":[23],"approaches":[24],"typically":[25],"treat":[26],"skills":[27,71,94],"explicit,":[29],"surface-level":[30],"constructs":[31],"specified":[32],"through":[33,158],"prompts":[34],"or":[35],"programs,":[36],"leaving":[37],"open":[38],"question":[40],"such":[43],"are":[46],"represented":[47,74],"inside":[48],"model":[50,107],"whether":[52,68,81],"they":[53,148],"can":[54,72,87,102,112],"be":[55,73,103],"manipulated":[56],"structured":[58],"objects":[59],"in":[60,77],"latent":[61,151],"space.In":[62],"this":[63],"empirical":[64],"study,":[65],"we":[66],"investigate":[67],"LLM":[70,146],"directions":[76,86,101,111,118,128],"activation":[78],"space":[79],"vector-space":[82,97,152],"operations":[83],"over":[84,126],"these":[85],"express":[88],"skill-level":[89],"behaviors.We":[90],"find":[91],"that":[92,154],"admit":[95,149],"representation:":[98],"individual":[99],"skill":[100,127],"activated":[104],"to":[105,114],"shift":[106],"behavior;":[108],"independently":[109],"extracted":[110],"compose":[113],"form":[115],"higher-level":[116],"skills.Contrastive":[117],"yield":[119],"context-conditioned":[120],"algorithmic":[121],"personalization":[122],"optimization":[124],"trajectories":[125],"evolve":[129],"non-monotonically,":[130],"with":[131],"intermediate":[132],"states":[133],"often":[134],"surpassing":[135],"fully":[136],"optimized":[137],"solutions.These":[138],"results":[139],"support":[140],"representation-level":[142],"view":[143],"skills:":[147],"organization":[153],"allows":[155],"direct":[156],"manipulation":[157],"internal":[159],"interventions.":[160]},"counts_by_year":[],"updated_date":"2026-11-01T11:40:01.658920","created_date":"2026-09-02T00:00:00"}