{"id":"W4413963629","doi":"10.1016/j.jss.2025.112604","title":"Syntactic multilingual probing of pre-trained language models of code","year":2025,"lang":"en","type":"article","venue":"Journal of Systems and Software","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"Federación Española de Enfermedades Raras; Ministerio de Ciencia, Innovación y Universidades","keywords":"Computer science; Natural language processing; Code (set theory); Linguistics; Artificial intelligence; Programming language; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001218607,0.001087524,0.0004249187,0.0005536988,0.0003255901,0.0009957635,0.0009090949,0.0007374655,0.001882566],"category_scores_gemma":[0.0109883,0.0004811924,0.0008408077,0.0004435228,0.001057874,0.003387552,0.002111131,0.002536385,0.0007132093],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008952991,"about_ca_system_score_gemma":0.001041957,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003000492,"about_ca_topic_score_gemma":0.00479357,"domain_scores_codex":[0.9992272,0.0002770775,0.00003411436,0.0002646826,0.0001152282,0.00008175941],"domain_scores_gemma":[0.9960568,0.002368535,0.0002957222,0.0007218934,0.0004435478,0.0001134957],"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":[0.0008241468,0.0002637681,0.02520739,0.000426007,0.0002665469,0.0005913406,0.002334288,0.4052631,0.08754183,0.02717815,0.004717793,0.4453856],"study_design_scores_gemma":[0.00001085851,0.0001509463,0.001871683,0.00001968413,0.00002412451,0.0000896555,0.0001797643,0.9621633,0.02188943,0.01230436,0.001265599,0.00003061823],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5160882,0.0002231448,0.4737957,0.000398267,0.00006891863,0.00007046907,0.0004531333,0.004688829,0.004213436],"genre_scores_gemma":[0.9378222,0.000082859,0.05898313,0.0001360106,0.00001301187,0.00007058746,0.000971043,0.0004441586,0.001477112],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003000492,"threshold_uncertainty_score":0.006495953,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01889991488062288,"score_gpt":0.2756434855286045,"score_spread":0.2567435706479816,"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."}}