{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0007664965,0.0002967149,0.0002890958,0.0002530292,0.0008969242,0.0007088443,0.000478163,0.0001502304,0.00003986202],"category_scores_gemma":[0.00004857253,0.0002802756,0.00007051315,0.0003269717,0.00003382619,0.0003671734,0.001737334,0.000966022,0.000004124984],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001729046,"about_ca_system_score_gemma":0.0001291249,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005694799,"about_ca_topic_score_gemma":0.0004206431,"domain_scores_codex":[0.9979238,0.0001850787,0.0002874735,0.0007769579,0.000449241,0.0003774631],"domain_scores_gemma":[0.9990438,0.00007727746,0.0003442538,0.0003844524,0.00007219268,0.00007801913],"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":[0.00004625795,0.00004438446,0.0002394697,0.0007031624,0.0001674198,0.0001298112,0.007341098,0.006696688,0.01177513,0.005142499,0.0000280844,0.967686],"study_design_scores_gemma":[0.0009230074,0.0003188616,0.0008324638,0.0002327527,0.0001032419,0.0002625546,0.001079101,0.9789264,0.009620633,0.005500236,0.0008971454,0.001303588],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08529862,0.00005384071,0.9090101,0.0001597306,0.0006532878,0.0003469715,0.000001615759,0.0007091969,0.003766702],"genre_scores_gemma":[0.9488408,0.000006043445,0.05041961,0.0000533742,0.0001589589,0.0000893339,0.000007352777,0.00003387301,0.000390658],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9722297,"threshold_uncertainty_score":0.999965,"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."}}