{"id":"W4406499752","doi":"10.1109/cascon62161.2024.10837905","title":"Anaphora Resolution in Software Requirements Engineering: A Comparison of Generative NLP Pipelines and Encoder-Based Models","year":2024,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Computer science; Artificial intelligence; Resolution (logic); Pipeline transport; Natural language processing; Generative grammar; Software; Anaphora (linguistics); Programming language; Engineering","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":[0.006151619,0.000742309,0.0006902418,0.001467677,0.0005332322,0.001487021,0.001716455,0.001231316,0.001660309],"category_scores_gemma":[0.01416888,0.0007017944,0.001003789,0.001474776,0.0006932585,0.003844846,0.001196306,0.002004039,0.0008794373],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002449363,"about_ca_system_score_gemma":0.00239479,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01735689,"about_ca_topic_score_gemma":0.01886657,"domain_scores_codex":[0.9977396,0.001255215,0.0001268363,0.0003601447,0.000415852,0.0001025052],"domain_scores_gemma":[0.9864218,0.01108627,0.000368096,0.001123333,0.0008401608,0.0001603192],"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":[0.0007757875,0.0006268421,0.007200673,0.0005312523,0.0001727071,0.0001405875,0.0007634327,0.5595574,0.004779415,0.01942118,0.00328361,0.4027471],"study_design_scores_gemma":[0.00001605074,0.00005780987,0.0003654259,0.0000117096,0.00001833804,0.00002610647,0.00003247169,0.9950146,0.001378556,0.002576262,0.0004940036,0.000008694919],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2217785,0.001846706,0.75807,0.001753417,0.00007814283,0.0004300404,0.0008114134,0.007414929,0.007816817],"genre_scores_gemma":[0.7732146,0.0009837277,0.2211366,0.0002950466,0.00003655909,0.0002104761,0.001609469,0.0002769122,0.002236492],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01735689,"threshold_uncertainty_score":0.03451169,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04183350212277246,"score_gpt":0.3194555275864681,"score_spread":0.2776220254636956,"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."}}