{"id":"W4410907667","doi":"10.5430/wjel.v15n6p209","title":"Evaluating AI Applications for Enhancing Listening Comprehension among EFL students in English language in Real-World Scenarios","year":2025,"lang":"en","type":"article","venue":"World Journal of English Language","topic":"Subtitles and Audiovisual Media","field":"Arts and Humanities","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Listening comprehension; Active listening; Computer science; Comprehension; Mathematics education; Natural language processing; Linguistics; Psychology; Programming language; Communication","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001173086,0.0001903752,0.0004587827,0.0009657982,0.0001428248,0.0002317604,0.0003474197,0.00004348021,0.0002350911],"category_scores_gemma":[0.0008536808,0.0001714337,0.000158089,0.0003537782,0.00006790573,0.0003189758,0.00009482337,0.0005283165,0.000002451741],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001857192,"about_ca_system_score_gemma":0.00007843072,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005709672,"about_ca_topic_score_gemma":0.09255359,"domain_scores_codex":[0.9981574,0.000112157,0.0008218969,0.0002187232,0.0003610733,0.0003287845],"domain_scores_gemma":[0.9983088,0.000580471,0.000348674,0.0001994726,0.0004881152,0.00007446045],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002599703,0.0006852102,0.06670678,0.0005479063,0.0001658532,0.0001402794,0.8833758,0.001233657,0.00305276,0.009220411,0.002914987,0.03169639],"study_design_scores_gemma":[0.01365642,0.0004389137,0.04141648,0.01056099,0.0004285188,0.00000272174,0.7752235,0.002147123,0.004269808,0.0006835809,0.1497968,0.001375174],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9785022,0.001265228,0.00006031583,0.00005050849,0.001375722,0.0006173672,0.0000126472,0.00004762073,0.01806842],"genre_scores_gemma":[0.9912387,0.00002347751,0.0004449764,0.0002740464,0.002185075,0.00006466587,0.00001704627,0.00002835477,0.005723638],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1468818,"threshold_uncertainty_score":0.9240049,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02784656890959413,"score_gpt":0.3522484213975462,"score_spread":0.3244018524879521,"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."}}