{"id":"W7008088209","doi":"","title":"Beyond maximum-likelihood training: analysis and methods for building robust language generation models","year":2025,"lang":"en","type":"dissertation","venue":"eScholarship@McGill (McGill)","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Natural language generation; Natural language; Key (lock); Matching (statistics); Abstraction","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":[0.00847161,0.001341851,0.001795425,0.00169646,0.001119053,0.002662478,0.003124031,0.002742548,0.006066276],"category_scores_gemma":[0.04657643,0.001669912,0.00176237,0.001757148,0.001442992,0.005610664,0.002619214,0.004808231,0.002472033],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001020633,"about_ca_system_score_gemma":0.00142445,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005811876,"about_ca_topic_score_gemma":0.005257664,"domain_scores_codex":[0.9959111,0.002351723,0.0001908246,0.0007755419,0.000609197,0.0001617161],"domain_scores_gemma":[0.9537665,0.04159915,0.0007478338,0.001963207,0.001628962,0.0002944365],"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":[0.0003686239,0.0001667102,0.002573613,0.0003334322,0.0002359121,0.0001572356,0.0003436307,0.4036215,0.003900748,0.02483384,0.01384169,0.5496231],"study_design_scores_gemma":[0.0000135639,0.00001840924,0.0002496274,0.00001830082,0.00002185784,0.00002626248,0.00001401642,0.974019,0.001585508,0.02305326,0.0009684216,0.00001176649],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.003576829,0.0004014469,0.993888,0.0002771385,0.00003046209,0.00003137004,0.0001179506,0.001203221,0.0004735626],"genre_scores_gemma":[0.1809835,0.0007831333,0.807059,0.0005904882,0.0002559987,0.0004445193,0.001989557,0.002299129,0.005594521],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.00847161,"threshold_uncertainty_score":0.04480267,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03253397473445307,"score_gpt":0.3230812356243357,"score_spread":0.2905472608898826,"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."}}