{"id":"W4412378051","doi":"10.1145/3726302.3730335","title":"Tip of the Tongue Query Elicitation for Simulated Evaluation","year":2025,"lang":"en","type":"article","venue":"","topic":"Speech and dialogue systems","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Microsoft (Canada)","funders":"National Institute of Standards and Technology","keywords":"Computer science; Tongue; Information retrieval; Query optimization; Artificial intelligence; Natural language processing; Linguistics","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":[0.007633768,0.001443637,0.000984112,0.0007710746,0.0004902462,0.00150214,0.002444443,0.001827201,0.007210396],"category_scores_gemma":[0.03623712,0.0005870694,0.0007541602,0.0007989967,0.0008245819,0.001350696,0.002372367,0.001460677,0.001935159],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001782335,"about_ca_system_score_gemma":0.001482393,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004687539,"about_ca_topic_score_gemma":0.006048978,"domain_scores_codex":[0.9918967,0.005846589,0.0004810564,0.0005775981,0.0009465974,0.0002514456],"domain_scores_gemma":[0.9636625,0.02887062,0.0008157663,0.003267895,0.002726525,0.0006567775],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.007263787,0.004730972,0.01018124,0.004033915,0.0003975955,0.0008263778,0.002663319,0.7118272,0.04169493,0.01820926,0.04101791,0.1571537],"study_design_scores_gemma":[0.0005990775,0.0009362252,0.0009617663,0.00007220674,0.00003128077,0.00007785772,0.0002042229,0.977235,0.008105407,0.005938186,0.005789616,0.00004913198],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4933088,0.0017384,0.4501408,0.001262301,0.0004191941,0.007076748,0.009626084,0.014533,0.02189465],"genre_scores_gemma":[0.7704482,0.000262701,0.2122123,0.0005032738,0.0000480368,0.006894292,0.006281441,0.0005229919,0.002826703],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007633768,"threshold_uncertainty_score":0.04037166,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01673934936747049,"score_gpt":0.3067126234972936,"score_spread":0.2899732741298231,"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."}}