{"id":"W4212921657","doi":"10.1044/2021_jslhr-21-00369","title":"Evaluating the Modified-Shortened Token Test as a Working Memory and Language Assessment Tool","year":2022,"lang":"en","type":"article","venue":"Journal of Speech Language and Hearing Research","topic":"Language Development and Disorders","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Working memory; Test (biology); Language assessment; Short-term memory; Security token; Language acquisition; Memory test","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":[0.007297889,0.0001295092,0.0002365571,0.0003297841,0.0007406319,0.000197475,0.0003318861,0.00005458363,0.001521734],"category_scores_gemma":[0.0004463412,0.00009210058,0.00006191353,0.0004095862,0.0001040296,0.0001047103,0.0004933208,0.001303554,0.000007781198],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001094622,"about_ca_system_score_gemma":0.0001987914,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007929115,"about_ca_topic_score_gemma":0.00004345378,"domain_scores_codex":[0.9973022,0.0004886436,0.0003878407,0.0002430455,0.00110929,0.0004689716],"domain_scores_gemma":[0.998385,0.000980036,0.0001474213,0.0002623765,0.0001140012,0.0001112137],"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":[0.0004313976,0.0002620438,0.03129533,0.00005709702,0.0002290071,0.003389661,0.12242,0.00005015399,0.02655829,0.0007906655,0.001733429,0.8127829],"study_design_scores_gemma":[0.005761754,0.003200518,0.1989538,0.0002292523,0.0001314637,0.004278271,0.7808282,0.001317164,0.001228507,0.00129301,0.002122478,0.000655523],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9753981,0.006531599,0.00001384854,0.001380769,0.0001761039,0.0002892818,0.000002556969,0.00001554653,0.01619218],"genre_scores_gemma":[0.9932019,0.00008008509,0.0009348079,0.0003076529,0.0003424047,0.00003203964,0.000003429433,0.00002426867,0.00507348],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8121274,"threshold_uncertainty_score":0.999391,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.139177161784714,"score_gpt":0.4823641336463894,"score_spread":0.3431869718616753,"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."}}