{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:HIGE7GOH32K5UEJBZFWD5HNJ34","short_pith_number":"pith:HIGE7GOH","schema_version":"1.0","canonical_sha256":"3a0c4f99c7de95da1121c96c3e9da9df2d2a7c13668e034f9ee63362e3f84519","source":{"kind":"arxiv","id":"1910.14659","version":3},"attestation_state":"computed","paper":{"title":"Masked Language Model Scoring","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","eess.AS","stat.ML"],"primary_cat":"cs.CL","authors_text":"Davis Liang, Julian Salazar, Katrin Kirchhoff, Toan Q. Nguyen","submitted_at":"2019-10-31T17:51:21Z","abstract_excerpt":"Pretrained masked language models (MLMs) require finetuning for most NLP tasks. Instead, we evaluate MLMs out of the box via their pseudo-log-likelihood scores (PLLs), which are computed by masking tokens one by one. We show that PLLs outperform scores from autoregressive language models like GPT-2 in a variety of tasks. By rescoring ASR and NMT hypotheses, RoBERTa reduces an end-to-end LibriSpeech model's WER by 30% relative and adds up to +1.7 BLEU on state-of-the-art baselines for low-resource translation pairs, with further gains from domain adaptation. We attribute this success to PLL's u"},"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":"1910.14659","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-10-31T17:51:21Z","cross_cats_sorted":["cs.LG","eess.AS","stat.ML"],"title_canon_sha256":"a96dd03b27e0f4d65a67935d17121ef7ab955503be06b02463c84d2f0c97f009","abstract_canon_sha256":"421881c0ada7895d55056af941c5cbb2b004371c2ac781264ca479d0b7d5d779"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:04:04.500266Z","signature_b64":"FTY5KnXOHqhYDh1KDrTIFgsyRBaHoOHJf7KA56OVD9tKMbKAp1XMoRe5bpD8ZtDKYSS7FPoMJmgnlLoQXrOjDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a0c4f99c7de95da1121c96c3e9da9df2d2a7c13668e034f9ee63362e3f84519","last_reissued_at":"2026-07-05T02:04:04.499860Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:04:04.499860Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Masked Language Model Scoring","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","eess.AS","stat.ML"],"primary_cat":"cs.CL","authors_text":"Davis Liang, Julian Salazar, Katrin Kirchhoff, Toan Q. Nguyen","submitted_at":"2019-10-31T17:51:21Z","abstract_excerpt":"Pretrained masked language models (MLMs) require finetuning for most NLP tasks. Instead, we evaluate MLMs out of the box via their pseudo-log-likelihood scores (PLLs), which are computed by masking tokens one by one. We show that PLLs outperform scores from autoregressive language models like GPT-2 in a variety of tasks. By rescoring ASR and NMT hypotheses, RoBERTa reduces an end-to-end LibriSpeech model's WER by 30% relative and adds up to +1.7 BLEU on state-of-the-art baselines for low-resource translation pairs, with further gains from domain adaptation. We attribute this success to PLL's u"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1910.14659","kind":"arxiv","version":3},"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/1910.14659/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":"1910.14659","created_at":"2026-07-05T02:04:04.499913+00:00"},{"alias_kind":"arxiv_version","alias_value":"1910.14659v3","created_at":"2026-07-05T02:04:04.499913+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1910.14659","created_at":"2026-07-05T02:04:04.499913+00:00"},{"alias_kind":"pith_short_12","alias_value":"HIGE7GOH32K5","created_at":"2026-07-05T02:04:04.499913+00:00"},{"alias_kind":"pith_short_16","alias_value":"HIGE7GOH32K5UEJB","created_at":"2026-07-05T02:04:04.499913+00:00"},{"alias_kind":"pith_short_8","alias_value":"HIGE7GOH","created_at":"2026-07-05T02:04:04.499913+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17489","citing_title":"Online LLM Selection via Constrained Bandits with Time-Varying Demand","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2304.01373","citing_title":"Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling","ref_index":247,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12220","citing_title":"Learning Project-wise Subsequent Code Edits via Interleaving Neural-based Induction and Tool-based Deduction","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05483","citing_title":"Can We Trust a Black-box LLM? LLM Untrustworthy Boundary Detection via Bias-Diffusion and Multi-Agent Reinforcement Learning","ref_index":30,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34","json":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34.json","graph_json":"https://pith.science/api/pith-number/HIGE7GOH32K5UEJBZFWD5HNJ34/graph.json","events_json":"https://pith.science/api/pith-number/HIGE7GOH32K5UEJBZFWD5HNJ34/events.json","paper":"https://pith.science/paper/HIGE7GOH"},"agent_actions":{"view_html":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34","download_json":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34.json","view_paper":"https://pith.science/paper/HIGE7GOH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1910.14659&json=true","fetch_graph":"https://pith.science/api/pith-number/HIGE7GOH32K5UEJBZFWD5HNJ34/graph.json","fetch_events":"https://pith.science/api/pith-number/HIGE7GOH32K5UEJBZFWD5HNJ34/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34/action/storage_attestation","attest_author":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34/action/author_attestation","sign_citation":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34/action/citation_signature","submit_replication":"https://pith.science/pith/HIGE7GOH32K5UEJBZFWD5HNJ34/action/replication_record"}},"created_at":"2026-07-05T02:04:04.499913+00:00","updated_at":"2026-07-05T02:04:04.499913+00:00"}