{"id":"W3209039755","doi":"10.18653/v1/2021.emnlp-main.821","title":"IndoNLI: A Natural Language Inference Dataset for Indonesian","year":2021,"lang":"en","type":"article","venue":"Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing","topic":"Topic Modeling","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Universitas Indonesia; York University; Samsung; National Science Foundation","keywords":"Annotation; Indonesian; Sentence; Test set; Set (abstract data type); Inference; Data set","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00136268,0.001464061,0.0008390729,0.002736604,0.001273521,0.001234693,0.001847341,0.001658905,0.01081394],"category_scores_gemma":[0.005044593,0.0004802059,0.0007408656,0.002232438,0.0006360148,0.002140397,0.00179399,0.002141649,0.01137615],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001744611,"about_ca_system_score_gemma":0.002205404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01364511,"about_ca_topic_score_gemma":0.03218947,"domain_scores_codex":[0.9981651,0.0004857442,0.0002714191,0.0006153373,0.0003463194,0.0001161526],"domain_scores_gemma":[0.9973979,0.001006218,0.0002762184,0.0006135142,0.0005093798,0.0001968104],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0007707928,0.0009670305,0.01749924,0.005246401,0.000241115,0.002156754,0.002256936,0.006330294,0.01786698,0.006519645,0.795597,0.1445479],"study_design_scores_gemma":[0.0003497424,0.0002511561,0.08519321,0.0006117635,0.0001963447,0.0026465,0.002662491,0.05062395,0.02261049,0.009092948,0.8254596,0.0003017908],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"empirical","genre_scores_codex":[0.05711932,0.002061597,0.02139697,0.00120106,0.0004040064,0.0008615474,0.8796455,0.0140655,0.02324462],"genre_scores_gemma":[0.03894811,0.0002585581,0.0232452,0.0003263016,0.0000501041,0.0008208487,0.9327601,0.0003824104,0.003208442],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01364511,"threshold_uncertainty_score":0.03617626,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06648941278896932,"score_gpt":0.43358885394361,"score_spread":0.3670994411546407,"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."}}