{"id":"W4313549837","doi":"10.1145/3551349.3556900","title":"AST-Probe: Recovering abstract syntax trees from hidden representations of pre-trained language models","year":2022,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Computer science; ENCODE; Natural language processing; Syntax; Artificial intelligence; Language model; Subspace topology; Representation (politics); Abstract syntax tree; Natural language","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":[0.0009387343,0.002636454,0.0008406129,0.001077198,0.0004511906,0.001033088,0.001540239,0.001318864,0.002912049],"category_scores_gemma":[0.006681967,0.0007159693,0.001809404,0.0008495739,0.000767818,0.003227451,0.001964107,0.003384152,0.001950221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006507149,"about_ca_system_score_gemma":0.001848701,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005419819,"about_ca_topic_score_gemma":0.00756921,"domain_scores_codex":[0.9993823,0.0001885146,0.00003472769,0.0001806399,0.0001162665,0.00009757549],"domain_scores_gemma":[0.9975764,0.001390454,0.0001523333,0.0004752589,0.0003099882,0.00009554937],"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":[0.0007732698,0.0002670149,0.004814052,0.0005835808,0.0002561339,0.0004689112,0.000901664,0.3123408,0.06711205,0.01293925,0.01649317,0.5830502],"study_design_scores_gemma":[0.00002257023,0.00009290837,0.0006597228,0.00002148198,0.00002602532,0.0000604485,0.0001062378,0.9729114,0.0125164,0.01194482,0.001611156,0.00002682885],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06485791,0.0002909461,0.9191964,0.0002500639,0.00008100281,0.0001017427,0.001591388,0.01264947,0.0009811404],"genre_scores_gemma":[0.534902,0.0003471939,0.4455696,0.0002723213,0.00005322741,0.0005003145,0.01304113,0.001549349,0.003764871],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005419819,"threshold_uncertainty_score":0.01077652,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02802503178024772,"score_gpt":0.2726645080868488,"score_spread":0.2446394763066011,"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."}}