{"id":"W2805299437","doi":"10.3389/feduc.2018.00042","title":"Examining Score Report Language in Accountability Testing","year":2018,"lang":"en","type":"article","venue":"Frontiers in Education","topic":"Management and Organizational Studies","field":"Business, Management and Accounting","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Accountability; Context (archaeology); Conversation analysis; Conversation; Discourse analysis; Interpretation (philosophy); Sample (material); Situated; Empirical research; Language assessment; Psychology; Linguistics; Political science; Computer science; Pedagogy; Epistemology; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.02414196,0.0004477855,0.000384862,0.005209764,0.00184408,0.004340494,0.001081264,0.0005592454,0.001404211],"category_scores_gemma":[0.1073589,0.0003259789,0.0002383537,0.004213565,0.004407492,0.002470555,0.005067215,0.001169602,0.000305746],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002712559,"about_ca_system_score_gemma":0.001973722,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008822819,"about_ca_topic_score_gemma":0.01086621,"domain_scores_codex":[0.9581166,0.03033517,0.001979017,0.001119417,0.007412758,0.001037012],"domain_scores_gemma":[0.8629919,0.08516403,0.02657104,0.003943539,0.01956948,0.001759961],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001629778,0.00006512642,0.1836552,0.0003948116,0.00003725155,0.0009585721,0.7266723,0.0002827642,0.007993082,0.003127452,0.001274574,0.07537604],"study_design_scores_gemma":[0.00001035333,0.0002447032,0.2417835,0.0006889282,0.00004567019,0.001465076,0.7077864,0.002297797,0.006871834,0.002550571,0.03612213,0.0001330097],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9892659,0.0002556242,0.003714186,0.0007503817,0.00003853202,0.00004218078,0.0001573041,0.00005070966,0.005725158],"genre_scores_gemma":[0.9969138,0.000217473,0.001801236,0.0001182396,0.00003432867,0.00006335173,0.0001175223,0.00004005418,0.0006938752],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02414196,"threshold_uncertainty_score":0.1276764,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02881192560449489,"score_gpt":0.2610577279211443,"score_spread":0.2322458023166495,"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."}}