{"id":"W3185590891","doi":"10.22034/elt.2021.43932.2334","title":"A New Dilemma for Language Teachers and Students: Self-assessment or Teacher Assessment (Research Article)","year":2021,"lang":"en","type":"article","venue":"DOAJ (DOAJ: Directory of Open Access Journals)","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Dilemma; Self-assessment; Likert scale; Reading (process); Psychology; Reading comprehension; Mathematics education; Comprehension; Set (abstract data type); Test (biology); Medical education; Pedagogy; Computer science; Medicine","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":[0.1348198,0.0004796135,0.00185391,0.00295667,0.002121754,0.008887448,0.00182428,0.0022193,0.002070111],"category_scores_gemma":[0.1701538,0.0004036716,0.0005231415,0.001613064,0.01489927,0.01116504,0.006035287,0.004677589,0.001024936],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001820758,"about_ca_system_score_gemma":0.006407566,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009246653,"about_ca_topic_score_gemma":0.001537562,"domain_scores_codex":[0.8298737,0.1394121,0.008410388,0.005238059,0.01621288,0.000852661],"domain_scores_gemma":[0.8429657,0.1086942,0.008291474,0.01175014,0.0208816,0.007416994],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003865396,0.0006632176,0.03388281,0.001799253,0.0001529463,0.0003096246,0.0551976,0.0002811775,0.002135151,0.08002491,0.02825326,0.7969135],"study_design_scores_gemma":[0.0004801831,0.003769354,0.06126129,0.01055424,0.0002461934,0.007453871,0.1895541,0.008287851,0.003075231,0.4022994,0.3121205,0.0008977659],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1909465,0.07204178,0.3345343,0.3205077,0.01456183,0.001438951,0.0003622057,0.001324752,0.06428199],"genre_scores_gemma":[0.8163628,0.008516422,0.1465874,0.01883791,0.002620842,0.001680881,0.000117758,0.0002481797,0.005027789],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1348198,"threshold_uncertainty_score":0.7130036,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3253825672241737,"score_gpt":0.6670082438301432,"score_spread":0.3416256766059695,"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."}}