{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:6EQ3SFQQONAGEZJANJ5OJSSKYN","short_pith_number":"pith:6EQ3SFQQ","schema_version":"1.0","canonical_sha256":"f121b9161073406265206a7ae4ca4ac368a05a1d1cf42659621153463748c6e3","source":{"kind":"arxiv","id":"2205.10964","version":2},"attestation_state":"computed","paper":{"title":"The Geometry of Multilingual Language Model Representations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Benjamin K. Bergen, Tyler A. Chang, Zhuowen Tu","submitted_at":"2022-05-22T23:58:24Z","abstract_excerpt":"We assess how multilingual language models maintain a shared multilingual representation space while still encoding language-sensitive information in each language. Using XLM-R as a case study, we show that languages occupy similar linear subspaces after mean-centering, evaluated based on causal effects on language modeling performance and direct comparisons between subspaces for 88 languages. The subspace means differ along language-sensitive axes that are relatively stable throughout middle layers, and these axes encode information such as token vocabularies. Shifting representations by lang"},"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":"2205.10964","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-05-22T23:58:24Z","cross_cats_sorted":[],"title_canon_sha256":"be83bf7ccda1f39c801fc333bfc32bafa0af6c874f4e8530bcaabf25d1578b97","abstract_canon_sha256":"b997132c0461c00a4aa744540313875745fc5f647698b7a611636108a4c76ad2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:09:18.257248Z","signature_b64":"eXq+Hhq7E9A5/jLYMBD8mT1TRS8qoI4OFMwcChJD1cTkNk68Y7l9UzrPobzHj8ExliM6QBYyUT08vzvtGb4wBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f121b9161073406265206a7ae4ca4ac368a05a1d1cf42659621153463748c6e3","last_reissued_at":"2026-07-05T05:09:18.256911Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:09:18.256911Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Geometry of Multilingual Language Model Representations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Benjamin K. Bergen, Tyler A. Chang, Zhuowen Tu","submitted_at":"2022-05-22T23:58:24Z","abstract_excerpt":"We assess how multilingual language models maintain a shared multilingual representation space while still encoding language-sensitive information in each language. Using XLM-R as a case study, we show that languages occupy similar linear subspaces after mean-centering, evaluated based on causal effects on language modeling performance and direct comparisons between subspaces for 88 languages. The subspace means differ along language-sensitive axes that are relatively stable throughout middle layers, and these axes encode information such as token vocabularies. Shifting representations by lang"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.10964","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/2205.10964/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":"2205.10964","created_at":"2026-07-05T05:09:18.256970+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.10964v2","created_at":"2026-07-05T05:09:18.256970+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.10964","created_at":"2026-07-05T05:09:18.256970+00:00"},{"alias_kind":"pith_short_12","alias_value":"6EQ3SFQQONAG","created_at":"2026-07-05T05:09:18.256970+00:00"},{"alias_kind":"pith_short_16","alias_value":"6EQ3SFQQONAGEZJA","created_at":"2026-07-05T05:09:18.256970+00:00"},{"alias_kind":"pith_short_8","alias_value":"6EQ3SFQQ","created_at":"2026-07-05T05:09:18.256970+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.24077","citing_title":"LWM-CDE: A Representation Space for Wireless Data Reasoning and Transferability","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25384","citing_title":"GeoMathCode: Understanding Interleaved Math-Code Reasoning for Geometry Problem Solving","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2412.04497","citing_title":"Opportunities and Challenges of Large Language Models for Low-Resource Languages in Humanities Research","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2505.14990","citing_title":"Language Specific Knowledge: Do Models Know Better in X than in English?","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN","json":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN.json","graph_json":"https://pith.science/api/pith-number/6EQ3SFQQONAGEZJANJ5OJSSKYN/graph.json","events_json":"https://pith.science/api/pith-number/6EQ3SFQQONAGEZJANJ5OJSSKYN/events.json","paper":"https://pith.science/paper/6EQ3SFQQ"},"agent_actions":{"view_html":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN","download_json":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN.json","view_paper":"https://pith.science/paper/6EQ3SFQQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.10964&json=true","fetch_graph":"https://pith.science/api/pith-number/6EQ3SFQQONAGEZJANJ5OJSSKYN/graph.json","fetch_events":"https://pith.science/api/pith-number/6EQ3SFQQONAGEZJANJ5OJSSKYN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN/action/storage_attestation","attest_author":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN/action/author_attestation","sign_citation":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN/action/citation_signature","submit_replication":"https://pith.science/pith/6EQ3SFQQONAGEZJANJ5OJSSKYN/action/replication_record"}},"created_at":"2026-07-05T05:09:18.256970+00:00","updated_at":"2026-07-05T05:09:18.256970+00:00"}