{"id":"W3001286655","doi":"10.17605/osf.io/n9pmd","title":"Inter-rater reliability for study selection of Spanish language articles using Google’s Neural Machine Translation (Google Translate)","year":2020,"lang":"en","type":"article","venue":"OSF Preprints (OSF Preprints)","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Machine translation; Selection (genetic algorithm); Natural language processing; Computer science; Reliability (semiconductor); Artificial intelligence; Inter-rater reliability; Information retrieval; Machine learning; Statistics; Mathematics","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":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001525212,0.0001712795,0.0002515983,0.00007908283,0.0001214527,0.00006232342,0.0007717626,0.0001020854,0.006437669],"category_scores_gemma":[0.0003038213,0.0001790278,0.0001204055,0.0003304519,0.00006568768,0.0004409268,0.0002333907,0.0002477363,0.002440042],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007267283,"about_ca_system_score_gemma":0.00009202785,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000166783,"about_ca_topic_score_gemma":0.00006684443,"domain_scores_codex":[0.9976561,0.0003174761,0.0004907872,0.001093826,0.0002166594,0.0002251733],"domain_scores_gemma":[0.9984249,0.0002182379,0.0001748905,0.0009692137,0.0001251814,0.0000875415],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005031814,0.003788822,0.3937389,0.0003772196,0.0003368714,0.000009926635,0.0509394,0.01758299,0.2607238,0.002674396,0.0004961464,0.2688284],"study_design_scores_gemma":[0.001545391,0.00003972627,0.04647287,0.00003638541,0.0001025042,0.00001585592,0.001205088,0.7901367,0.1545454,0.004476856,0.001008539,0.0004146767],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7193723,0.000002613911,0.2763054,0.001908963,0.000113614,0.0008830737,0.000006990684,0.0001420064,0.001265078],"genre_scores_gemma":[0.970227,0.000001663954,0.02902595,0.00009236174,0.00003279733,0.00008786684,0.000005462082,0.0000131378,0.000513738],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7725537,"threshold_uncertainty_score":0.9983367,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03353334232358544,"score_gpt":0.316249793916117,"score_spread":0.2827164515925315,"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."}}