{"id":"W4415379075","doi":"10.3758/s13428-025-02806-z","title":"LexKO: A quick, reliable lexical test of Korean language proficiency","year":2025,"lang":"en","type":"article","venue":"Behavior Research Methods","topic":"Text Readability and Simplification","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba; University of Toronto","funders":"Kyung Hee University; University of Manitoba; City University of Hong Kong","keywords":"Language proficiency; Test (biology); Korean language; Multilingualism; Language assessment; Foreign language","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.007781638,0.0001097759,0.0002263277,0.000395591,0.0002060279,0.0001142363,0.001447749,0.0001206891,0.00003367712],"category_scores_gemma":[0.001894177,0.00009536908,0.00008784617,0.001996018,0.0003393605,0.0002475095,0.0005276887,0.000472116,0.00002046599],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008351919,"about_ca_system_score_gemma":0.0004422888,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000580952,"about_ca_topic_score_gemma":0.0000132718,"domain_scores_codex":[0.9969823,0.00100277,0.0003837032,0.0005358201,0.0006110146,0.0004843902],"domain_scores_gemma":[0.9969004,0.001292701,0.00005799018,0.001335381,0.0002966918,0.0001168541],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00001423325,0.001171865,0.02157569,0.0001116564,0.000006523277,0.00001066157,0.001694659,0.000004346873,0.2918729,0.07390288,0.0008403689,0.6087942],"study_design_scores_gemma":[0.0006057603,0.0008837106,0.2147085,0.0001270745,0.00003082859,0.0000160855,0.001453008,0.009156548,0.742803,0.009370979,0.02046911,0.000375335],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2994111,0.000938134,0.6863989,0.001556993,0.0002319125,0.001595004,0.00001214101,0.0002384691,0.00961724],"genre_scores_gemma":[0.6396102,0.00001381538,0.3557316,0.00002462951,0.00001840117,0.0003244023,0.000003505425,0.000007327032,0.004266039],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6084189,"threshold_uncertainty_score":0.3889039,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1770580664468663,"score_gpt":0.5632092123204916,"score_spread":0.3861511458736253,"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."}}