{"id":"W2786266018","doi":"10.3138/cmlr.3670","title":"First Language Test Bias? Comparing French-Speaking and Polish-Speaking Participants’ Performance on the Peabody Picture Vocabulary Test","year":2018,"lang":"en","type":"article","venue":"Canadian Modern Language Review/ La Revue canadienne des langues vivantes","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Cognate; Vocabulary; Psychology; Peabody Picture Vocabulary Test; Test (biology); Linguistics; Word (group theory); Meaning (existential); Contrast (vision); Logistic regression; Descriptive statistics; Vocabulary development; Statistics; Computer science; Artificial intelligence; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0009929573,0.0005342491,0.0006694154,0.0003254399,0.0008022634,0.0001993686,0.0006080891,0.0001923743,0.005973747],"category_scores_gemma":[0.002070397,0.0004225187,0.0001495705,0.0005053761,0.000475084,0.0001948026,0.00008881455,0.0006978044,0.0002680292],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003179522,"about_ca_system_score_gemma":0.00009658241,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.09619557,"about_ca_topic_score_gemma":0.4303747,"domain_scores_codex":[0.9967254,0.000307088,0.0005449367,0.0007601517,0.0001500578,0.001512306],"domain_scores_gemma":[0.9964129,0.001241911,0.0002741909,0.001057451,0.0001410858,0.0008724924],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00004070663,0.000201593,0.1157951,0.003726468,0.000468689,0.007703491,0.7520414,0.00003742185,0.004646556,0.001726366,0.009164447,0.1044478],"study_design_scores_gemma":[0.001484502,0.00073299,0.78067,0.01975323,0.0005675994,0.003403382,0.06647564,0.00922812,0.0003086826,0.00006790656,0.1147351,0.00257282],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.903596,0.07816702,0.00002895848,0.0008498728,0.000268574,0.0005300035,0.0001782451,0.0001425943,0.01623877],"genre_scores_gemma":[0.9857031,0.0007070893,0.0001124751,0.01095216,0.001057472,0.00008263889,0.0001262167,0.0001127633,0.00114609],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6855658,"threshold_uncertainty_score":0.9998227,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03728438863522641,"score_gpt":0.2738895015692373,"score_spread":0.2366051129340109,"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."}}