{"id":"W6910323100","doi":"10.48448/2254-my14","title":"Compositional Generalization in Dependency Parsing","year":2022,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mila - Quebec Artificial Intelligence Institute; McGill University; Minnow Environmental (Canada)","funders":"","keywords":"Dependency (UML); Parsing; Dependency grammar; Generalization; Set (abstract data type); Dependency theory (database theory); Focus (optics); Test set; Similarity (geometry)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0009489282,0.0002652191,0.0002529795,0.002224752,0.0003141315,0.0001237125,0.0008622761,0.0001164295,0.01952101],"category_scores_gemma":[0.00006955124,0.0002931128,0.00003870377,0.002610713,0.0006243911,0.0002531016,0.0003259291,0.0003642799,0.0009991891],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001197576,"about_ca_system_score_gemma":0.0009278876,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007025274,"about_ca_topic_score_gemma":0.002341006,"domain_scores_codex":[0.9966273,0.0001284863,0.0003372897,0.0007835645,0.001594422,0.0005289243],"domain_scores_gemma":[0.9990758,0.0000296606,0.0002772792,0.0004299588,0.0000689774,0.00011829],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004697387,0.001419172,0.02386096,0.0001265094,0.0000574169,0.0002738627,0.0007014938,0.02620374,0.05933097,0.1141,0.7640746,0.009804278],"study_design_scores_gemma":[0.002641725,0.0002294552,0.004420998,0.0005077977,0.0000841038,0.0002408711,0.0004788513,0.1495947,0.0008215142,0.01090001,0.8273554,0.00272458],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.002482015,0.0008128054,0.003605764,0.0002966305,0.001357216,0.0008692172,0.0005112936,0.000754999,0.9893101],"genre_scores_gemma":[0.3347357,0.000203724,0.12621,0.002837971,0.002770468,0.0003264753,0.0104656,0.004843582,0.5176066],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.4717035,"threshold_uncertainty_score":0.9999521,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02113026028888899,"score_gpt":0.302395428166994,"score_spread":0.281265167878105,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}