{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:NUW4Z3TJDBFXVGXS2T3MYOYCZZ","short_pith_number":"pith:NUW4Z3TJ","schema_version":"1.0","canonical_sha256":"6d2dccee69184b7a9af2d4f6cc3b02ce447b77037d5b6dd1ff3107274ac64a26","source":{"kind":"arxiv","id":"2101.00288","version":2},"attestation_state":"computed","paper":{"title":"Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Daniel S. Weld, Jeffrey Heer, Marco Tulio Ribeiro, Tongshuang Wu","submitted_at":"2021-01-01T18:34:22Z","abstract_excerpt":"While counterfactual examples are useful for analysis and training of NLP models, current generation methods either rely on manual labor to create very few counterfactuals, or only instantiate limited types of perturbations such as paraphrases or word substitutions. We present Polyjuice, a general-purpose counterfactual generator that allows for control over perturbation types and locations, trained by finetuning GPT-2 on multiple datasets of paired sentences. We show that Polyjuice produces diverse sets of realistic counterfactuals, which in turn are useful in various distinct applications: i"},"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":"2101.00288","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2021-01-01T18:34:22Z","cross_cats_sorted":[],"title_canon_sha256":"37c8f635ba604517fb0cc05ded68253a35fc5f41bf9d3257cf80559efce57c50","abstract_canon_sha256":"c933015334c45c7a3fc24f9b7f3863864b60fed4f21a969ca7193078c6530980"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:44:53.305362Z","signature_b64":"vLvxEocG9kqoB9XDhVYgLk8lbrJasLsUU510I52LKFWjlFrWzXKwtVByOiryWabOX8droE8Rtk5XuGAD/FRyCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d2dccee69184b7a9af2d4f6cc3b02ce447b77037d5b6dd1ff3107274ac64a26","last_reissued_at":"2026-07-05T02:44:53.304741Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:44:53.304741Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Daniel S. Weld, Jeffrey Heer, Marco Tulio Ribeiro, Tongshuang Wu","submitted_at":"2021-01-01T18:34:22Z","abstract_excerpt":"While counterfactual examples are useful for analysis and training of NLP models, current generation methods either rely on manual labor to create very few counterfactuals, or only instantiate limited types of perturbations such as paraphrases or word substitutions. We present Polyjuice, a general-purpose counterfactual generator that allows for control over perturbation types and locations, trained by finetuning GPT-2 on multiple datasets of paired sentences. We show that Polyjuice produces diverse sets of realistic counterfactuals, which in turn are useful in various distinct applications: i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2101.00288","kind":"arxiv","version":2},"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/2101.00288/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":"2101.00288","created_at":"2026-07-05T02:44:53.304800+00:00"},{"alias_kind":"arxiv_version","alias_value":"2101.00288v2","created_at":"2026-07-05T02:44:53.304800+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2101.00288","created_at":"2026-07-05T02:44:53.304800+00:00"},{"alias_kind":"pith_short_12","alias_value":"NUW4Z3TJDBFX","created_at":"2026-07-05T02:44:53.304800+00:00"},{"alias_kind":"pith_short_16","alias_value":"NUW4Z3TJDBFXVGXS","created_at":"2026-07-05T02:44:53.304800+00:00"},{"alias_kind":"pith_short_8","alias_value":"NUW4Z3TJ","created_at":"2026-07-05T02:44:53.304800+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2507.06419","citing_title":"Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2509.11206","citing_title":"Evalet: Evaluating Large Language Models through Functional Fragmentation","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09548","citing_title":"Retrieval-Augmented Large Language Models for Evidence-Informed Guidance on Cannabidiol Use in Older Adults","ref_index":70,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ","json":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ.json","graph_json":"https://pith.science/api/pith-number/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/graph.json","events_json":"https://pith.science/api/pith-number/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/events.json","paper":"https://pith.science/paper/NUW4Z3TJ"},"agent_actions":{"view_html":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ","download_json":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ.json","view_paper":"https://pith.science/paper/NUW4Z3TJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2101.00288&json=true","fetch_graph":"https://pith.science/api/pith-number/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/graph.json","fetch_events":"https://pith.science/api/pith-number/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/action/storage_attestation","attest_author":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/action/author_attestation","sign_citation":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/action/citation_signature","submit_replication":"https://pith.science/pith/NUW4Z3TJDBFXVGXS2T3MYOYCZZ/action/replication_record"}},"created_at":"2026-07-05T02:44:53.304800+00:00","updated_at":"2026-07-05T02:44:53.304800+00:00"}