{"id":"W2896111887","doi":"10.5539/elt.v11n11p98","title":"Item Types: Their Effect on the Sensitivity of Multiple-Choice Cloze Tests","year":2018,"lang":"en","type":"article","venue":"English Language Teaching","topic":"Reading and Literacy Development","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Tsing Hua University","keywords":"Reading comprehension; Psychology; Reading (process); Comprehension; Multiple choice; Mathematics education; Linguistics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1553874,0.002066474,0.002390227,0.005046101,0.00100771,0.003085295,0.001834359,0.004181718,0.005547791],"category_scores_gemma":[0.5124674,0.001152219,0.002646575,0.002714279,0.002339594,0.004297265,0.003329864,0.003319442,0.001295369],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009729076,"about_ca_system_score_gemma":0.000575286,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009814823,"about_ca_topic_score_gemma":0.00116948,"domain_scores_codex":[0.6954923,0.1941839,0.04090499,0.02044454,0.04705817,0.001916102],"domain_scores_gemma":[0.1450066,0.7913617,0.02584459,0.0190052,0.01653122,0.002250621],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.05736277,0.003942419,0.6772495,0.002169052,0.006698723,0.001911135,0.00886589,0.005443544,0.05526017,0.003058022,0.005051367,0.1729874],"study_design_scores_gemma":[0.001980987,0.0171132,0.845263,0.00109638,0.003836233,0.006087506,0.001560364,0.0194182,0.07792608,0.004922411,0.02000156,0.0007941286],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9189073,0.005107065,0.05363245,0.0007715735,0.001541767,0.003560349,0.001140757,0.00104354,0.01429513],"genre_scores_gemma":[0.9494316,0.0005741048,0.04193195,0.001314603,0.0002839733,0.00157882,0.0008225645,0.0008404011,0.003222091],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1553874,"threshold_uncertainty_score":0.821777,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01330263660055979,"score_gpt":0.2972927804988638,"score_spread":0.283990143898304,"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."}}