{"id":"https://openalex.org/W7140143472","doi":"https://doi.org/10.18653/v1/2026.eacl-industry.37","title":"SAGE: An Agentic Explainer Framework for Interpreting SAE Features in Language Models","display_name":"SAGE: An Agentic Explainer Framework for Interpreting SAE Features in Language Models","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140143472","doi":"https://doi.org/10.18653/v1/2026.eacl-industry.37"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.eacl-industry.37","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.eacl-industry.37","pdf_url":"https://aclanthology.org/2026.eacl-industry.37.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":"Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.eacl-industry.37.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Jiaojiao Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiaojiao Han","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Wujiang Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wujiang Xu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Mingyu Jin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mingyu Jin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Mengnan Du","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mengnan Du","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":0.26077242,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"483","last_page":"495"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.5353000164031982,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.5353000164031982,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.07660000026226044,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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":0.040300000458955765,"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/feature","display_name":"Feature (linguistics)","score":0.3463999927043915},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.2937000095844269},{"id":"https://openalex.org/keywords/interpretation","display_name":"Interpretation (philosophy)","score":0.2808000147342682},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.27810001373291016},{"id":"https://openalex.org/keywords/language-understanding","display_name":"Language understanding","score":0.2685000002384186}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5641000270843506},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.44020000100135803},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.36890000104904175},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36800000071525574},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3463999927043915},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.32739999890327454},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2937000095844269},{"id":"https://openalex.org/C527412718","wikidata":"https://www.wikidata.org/wiki/Q855395","display_name":"Interpretation (philosophy)","level":2,"score":0.2808000147342682},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.27810001373291016},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.2685000002384186},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2605000138282776},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.2551000118255615},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2531999945640564}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.eacl-industry.37","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.eacl-industry.37","pdf_url":"https://aclanthology.org/2026.eacl-industry.37.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":"Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.eacl-industry.37","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.eacl-industry.37","pdf_url":"https://aclanthology.org/2026.eacl-industry.37.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":"Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140143472.pdf","grobid_xml":"https://content.openalex.org/works/W7140143472.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1,111],"models":[2,112],"(LLMs)":[3],"have":[4,27],"achieved":[5],"remarkable":[6],"progress,":[7],"yet":[8],"their":[9,20],"internal":[10],"mechanisms":[11],"remain":[12],"largely":[13],"opaque,":[14],"posing":[15],"a":[16,30,49,68,79],"significant":[17],"challenge":[18],"to":[19,93,126],"safe":[21],"and":[22,96,122],"reliable":[23],"deployment.Sparse":[24],"autoencoders":[25],"(SAEs)":[26],"emerged":[28],"as":[29],"promising":[31],"tool":[32],"for":[33,87],"decomposing":[34],"LLM":[35],"representations":[36],"into":[37,73],"more":[38],"interpretable":[39],"features,":[40],"but":[41],"explaining":[42],"the":[43],"features":[44,106],"captured":[45],"by":[46,82],"SAEs":[47,108],"remains":[48],"challenging":[50],"task.In":[51],"this":[52],"work,":[53],"we":[54],"propose":[55],"SAGE":[56,115],"(SAE":[57],"AGentic":[58],"Explainer),":[59],"an":[60,74],"agent-based":[61],"framework":[62],"that":[63,114],"recasts":[64],"feature":[65],"interpretation":[66],"from":[67,107],"passive,":[69],"single-pass":[70],"generation":[71],"task":[72],"active,":[75],"explanationdriven":[76],"process.SAGE":[77],"implements":[78],"rigorous":[80],"methodology":[81],"systematically":[83],"formulating":[84],"multiple":[85],"explanations":[86,99,117],"each":[88],"feature,":[89],"designing":[90],"targeted":[91],"experiments":[92],"test":[94],"them,":[95],"iteratively":[97],"refining":[98],"based":[100],"on":[101,105],"empirical":[102],"activation":[103],"feedback.Experiments":[104],"of":[109],"diverse":[110],"demonstrate":[113],"produces":[116],"with":[118],"significantly":[119],"higher":[120],"generative":[121],"predictive":[123],"accuracy":[124],"compared":[125],"state-of-theart":[127],"baselines.The":[128],"code":[129],"is":[130],"available":[131],"at":[132],"https:":[133],"//github.com/jiujiubuhejiu/SAGE.":[134]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-02-13T00:00:00"}