{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:J5WZ34F2CGX6CI37E5JZBHBOHY","short_pith_number":"pith:J5WZ34F2","schema_version":"1.0","canonical_sha256":"4f6d9df0ba11afe1237f2753909c2e3e3d1f234df77ea34cf89215ef18a0e6f6","source":{"kind":"arxiv","id":"1907.01752","version":4},"attestation_state":"computed","paper":{"title":"On the Weaknesses of Reinforcement Learning for Neural Machine Translation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Leshem Choshen, Lior Fox, Omri Abend, Zohar Aizenbud","submitted_at":"2019-07-03T06:15:14Z","abstract_excerpt":"Reinforcement learning (RL) is frequently used to increase performance in text generation tasks, including machine translation (MT), notably through the use of Minimum Risk Training (MRT) and Generative Adversarial Networks (GAN). However, little is known about what and how these methods learn in the context of MT. We prove that one of the most common RL methods for MT does not optimize the expected reward, as well as show that other methods take an infeasibly long time to converge. In fact, our results suggest that RL practices in MT are likely to improve performance only where the pre-traine"},"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":"1907.01752","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-07-03T06:15:14Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3676114c68dacf0ecc1448470c6fa70fc77bc196b914184620780999f662b17e","abstract_canon_sha256":"053b01b3caacdfea9be2fca7e9bebb216e5107c84edac4ed4769e381ef0dc31e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:33:34.650432Z","signature_b64":"VLcpG5cNUtm4edFf1lIMOaq/fyIKsPsZ48hR0IuQFJ8fxrTEWzClEknlACgCGqmOR2ahIWePlrXrThBNIJXvAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4f6d9df0ba11afe1237f2753909c2e3e3d1f234df77ea34cf89215ef18a0e6f6","last_reissued_at":"2026-07-05T00:33:34.649961Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:33:34.649961Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Weaknesses of Reinforcement Learning for Neural Machine Translation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Leshem Choshen, Lior Fox, Omri Abend, Zohar Aizenbud","submitted_at":"2019-07-03T06:15:14Z","abstract_excerpt":"Reinforcement learning (RL) is frequently used to increase performance in text generation tasks, including machine translation (MT), notably through the use of Minimum Risk Training (MRT) and Generative Adversarial Networks (GAN). However, little is known about what and how these methods learn in the context of MT. We prove that one of the most common RL methods for MT does not optimize the expected reward, as well as show that other methods take an infeasibly long time to converge. In fact, our results suggest that RL practices in MT are likely to improve performance only where the pre-traine"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1907.01752","kind":"arxiv","version":4},"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/1907.01752/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":"1907.01752","created_at":"2026-07-05T00:33:34.650018+00:00"},{"alias_kind":"arxiv_version","alias_value":"1907.01752v4","created_at":"2026-07-05T00:33:34.650018+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1907.01752","created_at":"2026-07-05T00:33:34.650018+00:00"},{"alias_kind":"pith_short_12","alias_value":"J5WZ34F2CGX6","created_at":"2026-07-05T00:33:34.650018+00:00"},{"alias_kind":"pith_short_16","alias_value":"J5WZ34F2CGX6CI37","created_at":"2026-07-05T00:33:34.650018+00:00"},{"alias_kind":"pith_short_8","alias_value":"J5WZ34F2","created_at":"2026-07-05T00:33:34.650018+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2304.06767","citing_title":"RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment","ref_index":96,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY","json":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY.json","graph_json":"https://pith.science/api/pith-number/J5WZ34F2CGX6CI37E5JZBHBOHY/graph.json","events_json":"https://pith.science/api/pith-number/J5WZ34F2CGX6CI37E5JZBHBOHY/events.json","paper":"https://pith.science/paper/J5WZ34F2"},"agent_actions":{"view_html":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY","download_json":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY.json","view_paper":"https://pith.science/paper/J5WZ34F2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1907.01752&json=true","fetch_graph":"https://pith.science/api/pith-number/J5WZ34F2CGX6CI37E5JZBHBOHY/graph.json","fetch_events":"https://pith.science/api/pith-number/J5WZ34F2CGX6CI37E5JZBHBOHY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY/action/storage_attestation","attest_author":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY/action/author_attestation","sign_citation":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY/action/citation_signature","submit_replication":"https://pith.science/pith/J5WZ34F2CGX6CI37E5JZBHBOHY/action/replication_record"}},"created_at":"2026-07-05T00:33:34.650018+00:00","updated_at":"2026-07-05T00:33:34.650018+00:00"}