{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:7DKKB7LWVZMSOGQFOEVRJJHZUD","short_pith_number":"pith:7DKKB7LW","schema_version":"1.0","canonical_sha256":"f8d4a0fd76ae59271a05712b14a4f9a0f5a6bd228e6a38c8432ad409594e88fd","source":{"kind":"arxiv","id":"2104.06390","version":1},"attestation_state":"computed","paper":{"title":"Detoxifying Language Models Risks Marginalizing Minority Voices","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Albert Xu, Dan Klein, Eric Wallace, Eshaan Pathak, Maarten Sap, Suchin Gururangan","submitted_at":"2021-04-13T17:52:01Z","abstract_excerpt":"Language models (LMs) must be both safe and equitable to be responsibly deployed in practice. With safety in mind, numerous detoxification techniques (e.g., Dathathri et al. 2020; Krause et al. 2020) have been proposed to mitigate toxic LM generations. In this work, we show that current detoxification techniques hurt equity: they decrease the utility of LMs on language used by marginalized groups (e.g., African-American English and minority identity mentions). In particular, we perform automatic and human evaluations of text generation quality when LMs are conditioned on inputs with different "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2104.06390","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-13T17:52:01Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"7e7b8ed39d946a628e1d09245e6ced2bd4a1d60c3cbeefe54efeb98452d2ec19","abstract_canon_sha256":"b652c1f90fca2c8c472f096c586e54f3e35d3dcf480ef34afe181ff672d14d83"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:31:56.915610Z","signature_b64":"wTDTpMQxGw2+nRaf3kC5C90S03fths2lrb3FYWa62UnFWmDm8O76+v7W1+x7+i2KYJ2PLrYaPWp7ArxesHI1Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f8d4a0fd76ae59271a05712b14a4f9a0f5a6bd228e6a38c8432ad409594e88fd","last_reissued_at":"2026-07-05T02:31:56.915193Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:31:56.915193Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Detoxifying Language Models Risks Marginalizing Minority Voices","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Albert Xu, Dan Klein, Eric Wallace, Eshaan Pathak, Maarten Sap, Suchin Gururangan","submitted_at":"2021-04-13T17:52:01Z","abstract_excerpt":"Language models (LMs) must be both safe and equitable to be responsibly deployed in practice. With safety in mind, numerous detoxification techniques (e.g., Dathathri et al. 2020; Krause et al. 2020) have been proposed to mitigate toxic LM generations. In this work, we show that current detoxification techniques hurt equity: they decrease the utility of LMs on language used by marginalized groups (e.g., African-American English and minority identity mentions). In particular, we perform automatic and human evaluations of text generation quality when LMs are conditioned on inputs with different "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.06390","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2104.06390/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2104.06390","created_at":"2026-07-05T02:31:56.915248+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.06390v1","created_at":"2026-07-05T02:31:56.915248+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.06390","created_at":"2026-07-05T02:31:56.915248+00:00"},{"alias_kind":"pith_short_12","alias_value":"7DKKB7LWVZMS","created_at":"2026-07-05T02:31:56.915248+00:00"},{"alias_kind":"pith_short_16","alias_value":"7DKKB7LWVZMSOGQF","created_at":"2026-07-05T02:31:56.915248+00:00"},{"alias_kind":"pith_short_8","alias_value":"7DKKB7LW","created_at":"2026-07-05T02:31:56.915248+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12234","citing_title":"On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Study","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2305.10403","citing_title":"PaLM 2 Technical Report","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2202.03286","citing_title":"Red Teaming Language Models with Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2112.04359","citing_title":"Ethical and social risks of harm from Language Models","ref_index":295,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD","json":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD.json","graph_json":"https://pith.science/api/pith-number/7DKKB7LWVZMSOGQFOEVRJJHZUD/graph.json","events_json":"https://pith.science/api/pith-number/7DKKB7LWVZMSOGQFOEVRJJHZUD/events.json","paper":"https://pith.science/paper/7DKKB7LW"},"agent_actions":{"view_html":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD","download_json":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD.json","view_paper":"https://pith.science/paper/7DKKB7LW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.06390&json=true","fetch_graph":"https://pith.science/api/pith-number/7DKKB7LWVZMSOGQFOEVRJJHZUD/graph.json","fetch_events":"https://pith.science/api/pith-number/7DKKB7LWVZMSOGQFOEVRJJHZUD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD/action/storage_attestation","attest_author":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD/action/author_attestation","sign_citation":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD/action/citation_signature","submit_replication":"https://pith.science/pith/7DKKB7LWVZMSOGQFOEVRJJHZUD/action/replication_record"}},"created_at":"2026-07-05T02:31:56.915248+00:00","updated_at":"2026-07-05T02:31:56.915248+00:00"}