{"id":"W4285148079","doi":"10.18653/v1/2022.findings-acl.293","title":"Local Structure Matters Most: Perturbation Study in NLU","year":2022,"lang":"en","type":"article","venue":"Findings of the Association for Computational Linguistics: ACL 2022","topic":"Topic Modeling","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research; McGill University; Polytechnique Montréal; Mila - Quebec Artificial Intelligence Institute","funders":"Canada First Research Excellence Fund; Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Computer science; Perturbation (astronomy); Word order; Phenomenon; Artificial neural network; Invariant (physics); Artificial intelligence; Natural language processing; Mathematics; Physics","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":[0.002466983,0.0003408302,0.0005736956,0.0004403256,0.0004108349,0.0009341705,0.000609073,0.0007648727,0.001004198],"category_scores_gemma":[0.03353014,0.0002995041,0.0003069224,0.0004278127,0.001296472,0.002709759,0.0009393271,0.00161894,0.0001893349],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000642793,"about_ca_system_score_gemma":0.0002884997,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001541064,"about_ca_topic_score_gemma":0.001158738,"domain_scores_codex":[0.9986019,0.0009196848,0.00004789995,0.0002370875,0.000142276,0.00005131652],"domain_scores_gemma":[0.9796521,0.01679408,0.0008062698,0.001939623,0.000496909,0.0003110039],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002732866,0.001064435,0.0444022,0.0005336584,0.0004686507,0.0006758727,0.004042729,0.6395189,0.145335,0.03391996,0.002252222,0.1250535],"study_design_scores_gemma":[0.00004116764,0.0002901046,0.01446959,0.0000219124,0.00004894717,0.000108375,0.000371936,0.9240426,0.02087468,0.03900556,0.0006899219,0.00003526151],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9346654,0.0004768881,0.06195209,0.0005671018,0.00002773028,0.00004684801,0.00007889848,0.0003223946,0.001862655],"genre_scores_gemma":[0.9949645,0.00006364839,0.004639853,0.0000497527,0.00001291066,0.00001963836,0.00005190942,0.00004494576,0.0001527405],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.002466983,"threshold_uncertainty_score":0.0130468,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01015079981417298,"score_gpt":0.2441797445615391,"score_spread":0.2340289447473661,"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."}}