{"id":"W3188759137","doi":"10.18653/v1/2022.wassa-1.14","title":"Improving Social Meaning Detection with Pragmatic Masking and Surrogate Fine-Tuning","year":2022,"lang":"en","type":"preprint","venue":"","topic":"Hate Speech and Cyberbullying Detection","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; Natural Sciences and Engineering Research Council of Canada; Compute Canada","keywords":"Computer science; Masking (illustration); Exploit; Set (abstract data type); Artificial intelligence; Shot (pellet); Meaning (existential); Natural language processing; Class (philosophy); Language model; Baseline (sea); Machine learning; Training set; Contrast (vision); Domain (mathematical analysis); Test set; Speech recognition; Psychology; Mathematics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002247988,0.001778704,0.001052606,0.0009929521,0.0006820403,0.001548424,0.001446477,0.002071614,0.002652184],"category_scores_gemma":[0.009672486,0.0004914381,0.001095592,0.0005445679,0.001361541,0.003390421,0.002666724,0.00304892,0.002219523],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006461355,"about_ca_system_score_gemma":0.001256435,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002549108,"about_ca_topic_score_gemma":0.005578375,"domain_scores_codex":[0.9985215,0.0005989163,0.00006996551,0.0004344367,0.0002480112,0.0001270534],"domain_scores_gemma":[0.997748,0.00132456,0.0001350653,0.0003743042,0.0003099455,0.0001081278],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008903693,0.0006167507,0.005974106,0.0005376188,0.000257775,0.0004681384,0.00107279,0.09863501,0.08290617,0.0178108,0.02429581,0.7665347],"study_design_scores_gemma":[0.00005666855,0.0002159323,0.001408035,0.00005030874,0.00004962243,0.0001850998,0.0002669675,0.9535649,0.01445143,0.02577421,0.00392236,0.00005454035],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1311757,0.001359431,0.8480722,0.00177226,0.0004114995,0.0001950695,0.0006517228,0.00968725,0.006674848],"genre_scores_gemma":[0.7496848,0.000275744,0.238276,0.00132393,0.0002335373,0.0002363197,0.002107208,0.0006327893,0.007229779],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.002652184,"threshold_uncertainty_score":0.01188862,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01062833523970794,"score_gpt":0.2189080723559419,"score_spread":0.2082797371162339,"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."}}