{"id":"W3181314628","doi":"10.1017/s0272263121000449","title":"A DIFFERENT PERSPECTIVE ON THE LIMITATIONS OF SIZE AND LEVELS TESTS OF WRITTEN RECEPTIVE VOCABULARY KNOWLEDGE","year":2021,"lang":"en","type":"article","venue":"Studies in Second Language Acquisition","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Perspective (graphical); Vocabulary; Linguistics; Psychology; Natural language processing; Mathematics education; Artificial intelligence; Computer science; Philosophy","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0003015825,0.0001672433,0.0003532768,0.0001245127,0.00009096757,0.000009102176,0.00009626705,0.00008252764,0.01238054],"category_scores_gemma":[0.001198245,0.0001262533,0.00008538822,0.0003098641,0.0002846499,0.00005929535,0.00009302614,0.0002332754,0.00001972345],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008628646,"about_ca_system_score_gemma":0.00002721112,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002342438,"about_ca_topic_score_gemma":0.0002969261,"domain_scores_codex":[0.9985075,0.0005268469,0.0003344727,0.0003212681,0.0001203215,0.0001895852],"domain_scores_gemma":[0.9952574,0.003932862,0.0001741894,0.0003159427,0.0002891245,0.00003045874],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002045414,0.0007210684,0.0008625395,0.0001411134,0.0007840942,0.00023182,0.910854,0.000005232873,0.03371237,0.04032113,0.001858597,0.01030353],"study_design_scores_gemma":[0.001014282,0.0002568578,0.2361047,0.000235025,0.00004905113,0.00003451839,0.7546895,0.000003869649,0.006278066,0.001050208,0.0001288153,0.0001550654],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9158685,0.05114514,0.000009980488,0.0004795265,0.0002242659,0.0002072811,0.00006790755,0.00001825895,0.03197919],"genre_scores_gemma":[0.9963118,0.0001629431,0.00008506932,0.001816887,0.0001041737,0.00005824917,0.00001544588,0.00001673665,0.001428629],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2352422,"threshold_uncertainty_score":0.9885223,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08114140722840063,"score_gpt":0.3803840345125217,"score_spread":0.2992426272841211,"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."}}