{"id":"W4383069305","doi":"10.1177/02655322231179134","title":"Fairness of using different English accents: The effect of shared L1s in listening tasks of the Duolingo English test","year":2023,"lang":"en","type":"article","venue":"Language Testing","topic":"Phonetics and Phonology Research","field":"Psychology","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"Brock University","funders":"","keywords":"Active listening; Psychology; Stress (linguistics); Test (biology); Interlanguage; Vocabulary; Dictation; Linguistics; Task (project management); Hindi; Communication","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05847613,0.0005535468,0.0006600661,0.0006298637,0.001085292,0.001914751,0.0007575943,0.0007678882,0.001182708],"category_scores_gemma":[0.1640635,0.000420373,0.0006202831,0.0003171373,0.002398172,0.001862976,0.003694753,0.001266797,0.000337652],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006339923,"about_ca_system_score_gemma":0.0006180952,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001090895,"about_ca_topic_score_gemma":0.002264256,"domain_scores_codex":[0.95171,0.03043661,0.003753136,0.004363387,0.008815587,0.0009212358],"domain_scores_gemma":[0.6981454,0.240359,0.02795635,0.02117159,0.008810516,0.003557123],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01761842,0.002189803,0.7263381,0.0003865261,0.0008199843,0.000557832,0.02989757,0.001740327,0.0844661,0.00154769,0.0004234676,0.1340142],"study_design_scores_gemma":[0.0002234212,0.005909177,0.9531065,0.0001205352,0.0002940425,0.0007106285,0.004002206,0.002249994,0.03041326,0.001486908,0.001336811,0.0001463486],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9928139,0.0001444125,0.004023078,0.0001274018,0.00003864456,0.00008621126,0.0000220578,0.00002164722,0.002722617],"genre_scores_gemma":[0.9971232,0.00003298772,0.002153491,0.0001489828,0.00002796179,0.00004952927,0.00002416686,0.00002279969,0.0004169266],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05847613,"threshold_uncertainty_score":0.3092551,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03822191359303156,"score_gpt":0.3483421313789877,"score_spread":0.3101202177859562,"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."}}