{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":22,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":22,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"8eedc4ff3541","filters":{"venue":"Language Assessment Quarterly"}},"results":[{"id":"W2096160735","doi":"10.1080/15434300701375832","title":"Three Generations of DIF Analyses: Considering Where It Has Been, Where It Is Now, and Where It Is Going","year":2007,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":386,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Differential item functioning; Contrast (vision); Psychology; Cognitive psychology; Item response theory; Econometrics; Computer science; Psychometrics; Artificial intelligence; Mathematics; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.08492024923436474,"gpt":0.4363221750131228,"spread":0.351401925778758,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005792451,0.0004242767,0.0005519639,0.0001943643,0.0002984112,0.000223461,0.0003612596,0.0002868678,0.01493608],"category_scores_gemma":[0.00001015566,0.0003789917,0.0002008102,0.0003921827,0.0001821099,0.0003079022,0.00004759144,0.000456175,0.0002403061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008695065,"about_ca_system_score_gemma":0.0001166745,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001096836,"about_ca_topic_score_gemma":0.01301482,"domain_scores_codex":[0.9969435,0.0001482274,0.000929441,0.0008176497,0.0005243055,0.0006369313],"domain_scores_gemma":[0.9979244,0.000459609,0.0004142513,0.0007597716,0.0001867823,0.0002552242],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0001081645,0.001775553,0.2446399,0.0003096328,0.001423623,0.000288283,0.06773006,0.000005953319,0.02098111,0.004693085,0.6274801,0.03056452],"study_design_scores_gemma":[0.006072184,0.003364352,0.5359789,0.0009524424,0.0008627624,0.0002267555,0.2781354,0.0004938076,0.0007855789,0.007092046,0.1636461,0.002389612],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8150021,0.005624855,0.02692896,0.0475564,0.001170833,0.000859707,0.0002762704,0.0001407408,0.1024401],"genre_scores_gemma":[0.9720086,0.0001037086,0.01214098,0.006666999,0.0004935285,0.000086279,0.00008568882,0.00005445157,0.008359737],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.463834,"threshold_uncertainty_score":0.9998662,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2112485328","doi":"10.1080/15434303.2015.1010726","title":"Teachers’ Grading Decision Making: Multiple Influencing Factors and Methods","year":2015,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":138,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Grading (engineering); Psychology; Mathematics education; English language; Multivariate analysis of variance; Statistics; Engineering; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.04392888403571169,"gpt":0.4426289617588582,"spread":0.3987000777231465,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002215324,0.0002165094,0.0002843175,0.0001521183,0.0004279438,0.000414697,0.0003085519,0.0001270034,0.00009018436],"category_scores_gemma":[0.0002537668,0.0001872555,0.00007811721,0.0003026884,0.0001150815,0.0006288868,0.00006152047,0.0002338022,0.00001224679],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002860255,"about_ca_system_score_gemma":0.0001520136,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002273368,"about_ca_topic_score_gemma":0.001829075,"domain_scores_codex":[0.9976478,0.0005109212,0.000293084,0.0003663804,0.0006636434,0.0005181679],"domain_scores_gemma":[0.9984816,0.0007907837,0.0001499985,0.0002398501,0.00007483536,0.0002629701],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.000008713797,0.000048826,0.5265206,0.000004065987,0.0000347703,0.0000151809,0.376552,0.000001497073,0.0009257618,0.001428913,0.0002979,0.09416176],"study_design_scores_gemma":[0.0009254496,0.0002467533,0.2873632,0.00006240925,0.00005070329,0.00000158546,0.7060346,0.000321111,0.0000403012,0.00169127,0.002886578,0.0003760849],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9721803,0.0002567901,0.01473934,0.0001306893,0.0005319277,0.0003126729,0.000002428814,0.0001793628,0.01166647],"genre_scores_gemma":[0.9285433,0.000006457129,0.07078302,0.0001119689,0.0002082865,0.00001600317,0.000006128478,0.00002115638,0.0003036593],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3294826,"threshold_uncertainty_score":0.7636056,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2975930148","doi":"10.1080/15434303.2019.1671392","title":"Incorporating Translanguaging in Language Assessment: The Case of a Test for University Professors","year":2019,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Second Language Learning and Teaching","field":"Arts and Humanities","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa","funders":"","keywords":"Translanguaging; Operationalization; Active listening; Test (biology); Psychology; Task (project management); Competence (human resources); Mathematics education; Pedagogy; Computer science; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.01280746938865259,"gpt":0.2800899580955242,"spread":0.2672824887068716,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008636062,0.000211378,0.0003256055,0.0002000446,0.000262463,0.0001266804,0.0002857111,0.00005959388,0.0007758964],"category_scores_gemma":[0.00002438496,0.0001658575,0.0001408452,0.0001170893,0.00009220163,0.0003249333,0.00002163214,0.0004686462,0.000008716105],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008230501,"about_ca_system_score_gemma":0.00009434831,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.008453423,"about_ca_topic_score_gemma":0.01855964,"domain_scores_codex":[0.9985888,0.0002565996,0.0003422966,0.0003160389,0.0001576463,0.0003386077],"domain_scores_gemma":[0.9984386,0.0007919899,0.0002770411,0.0003954507,0.00005241733,0.00004452506],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00003858461,0.0003515043,0.01317507,0.0004519245,0.0001105925,0.001585262,0.8307458,0.00004418907,0.004901102,0.06497666,0.00007673892,0.08354262],"study_design_scores_gemma":[0.001865448,0.0005292157,0.001471548,0.0001575561,0.00006228724,0.00006321251,0.9895734,0.005173319,0.00006696027,0.0001011924,0.0006397872,0.0002960652],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9737643,0.0001600327,0.0003975346,0.0002729594,0.0002256582,0.0007648789,0.0001023287,0.00009111827,0.02422117],"genre_scores_gemma":[0.9949095,4.5575e-7,0.001116537,0.0001370979,0.0001730404,0.00001987522,0.00007423347,0.00003425888,0.003535028],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1588277,"threshold_uncertainty_score":0.9993491,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4389294437","doi":"10.1080/15434303.2023.2288253","title":"Validity Arguments for Automated Essay Scoring of Young Students’ Writing Traits","year":2023,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Text Readability and Simplification","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Writing assessment; Consistency (knowledge bases); Context (archaeology); Vocabulary; Psychology; Inference; Argument (complex analysis); Formative assessment; Trait; Artificial intelligence; Natural language processing; Computer science; Mathematics education; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.0296680655231284,"gpt":0.359881211422216,"spread":0.3302131458990876,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001113247,0.0001393121,0.0002202543,0.0001334027,0.0001303798,0.0002162125,0.0007928991,0.00005846357,0.000006747511],"category_scores_gemma":[0.00002646708,0.0001383288,0.0000959745,0.0004695973,0.00002138302,0.0005182773,0.00006507311,0.0000880709,0.00001430536],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009068496,"about_ca_system_score_gemma":0.00007510196,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006737553,"about_ca_topic_score_gemma":0.00003418868,"domain_scores_codex":[0.9982348,0.0001219398,0.0004217942,0.0003978873,0.0004916967,0.0003318607],"domain_scores_gemma":[0.9989762,0.0002295976,0.0001845904,0.0004533818,0.0000901913,0.00006602624],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00003311682,0.00147708,0.153676,0.001031753,0.0002530589,0.00007972993,0.09369092,0.0002957399,0.1997186,0.03360007,0.001006978,0.5151369],"study_design_scores_gemma":[0.001115786,0.0006074037,0.8524105,0.0001003498,0.00002777803,0.000006214538,0.005072051,0.1351477,0.003642167,0.00149702,0.00004980215,0.0003232574],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7880567,0.0000082329,0.2101454,0.0001180656,0.0002157385,0.0004129795,0.00001994741,0.0008196266,0.0002032719],"genre_scores_gemma":[0.9802725,9.708205e-7,0.01933962,0.00002606102,0.00005208327,0.0001117667,0.00004693149,0.00001196375,0.0001381106],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6987345,"threshold_uncertainty_score":0.5640885,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2015223621","doi":"10.1080/15434303.2014.981334","title":"Interpreting the Impact of the Ontario Secondary School Literacy Test on Second Language Students Within an Argument-Based Validation Framework","year":2015,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":24,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University","funders":"","keywords":"Argument (complex analysis); Literacy; Context (archaeology); Mathematics education; Test (biology); Curriculum; Reading (process); Ell; Language proficiency; Empirical research; Psychology; Pedagogy; Language assessment; Linguistics; Teaching method; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.01646036534847547,"gpt":0.3890332495887422,"spread":0.3725728842402667,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002156588,0.0002811441,0.0002830522,0.00009695641,0.0004360736,0.0007812304,0.001259779,0.0001380839,0.0022243],"category_scores_gemma":[0.0002022937,0.0001712414,0.0002250081,0.0003708061,0.0001524563,0.0006747794,0.00007754562,0.000809064,0.00003803763],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009341819,"about_ca_system_score_gemma":0.001193955,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03185959,"about_ca_topic_score_gemma":0.03417153,"domain_scores_codex":[0.9965785,0.0007991931,0.000483906,0.0003691439,0.001310989,0.0004583026],"domain_scores_gemma":[0.9977991,0.0005609579,0.0004945116,0.000758499,0.0001549533,0.0002319876],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004974287,0.0005728125,0.5438338,0.000008453961,0.00009432029,0.00001074415,0.4514106,0.00004990556,0.0008062328,0.0002336612,0.0006054001,0.002324369],"study_design_scores_gemma":[0.001506245,0.002071383,0.7565621,0.0002743815,0.00007929908,0.000001321347,0.2376144,0.0002856197,0.0004860076,0.0005585418,0.0001667983,0.0003938935],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9859685,0.00004082208,0.0001585789,0.0003260243,0.0005758933,0.0008030116,0.0000351297,0.0000722697,0.01201977],"genre_scores_gemma":[0.9961842,3.561883e-7,0.0008259438,0.0004821725,0.0002914397,0.00005246538,0.00004822072,0.0000272375,0.002087927],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2137962,"threshold_uncertainty_score":0.9986878,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3209416046","doi":"10.1080/15434303.2021.1992629","title":"The Relationship between Word Difficulty and Frequency: A Response to Hashimoto (2021)","year":2021,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":21,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Word lists by frequency; Correlation; Vocabulary; Rank (graph theory); Word (group theory); Linguistics; Psychology; Range (aeronautics); Mathematics; Statistics; Sentence; Philosophy","retraction":null,"screen_n_in":null,"score":{"opus":0.01621099591469831,"gpt":0.3205802786497973,"spread":0.304369282735099,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008984907,0.0001938085,0.0001912975,0.00009390563,0.0004288843,0.0009590951,0.0007568303,0.00009353749,0.00001103087],"category_scores_gemma":[0.0004595777,0.0001418333,0.00005721494,0.0006811222,0.00004796463,0.0004620012,0.0001692929,0.0003622269,0.00002167767],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001068833,"about_ca_system_score_gemma":0.00019927,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005300479,"about_ca_topic_score_gemma":0.0001576293,"domain_scores_codex":[0.9979206,0.0004893631,0.0002960157,0.000521683,0.0004005971,0.0003717886],"domain_scores_gemma":[0.9969036,0.001762222,0.0001013076,0.0009453963,0.0001204722,0.0001670124],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00006691663,0.0001317575,0.07032432,0.00006162352,0.0000910083,0.001077234,0.03220738,5.620614e-7,0.0735165,0.1457214,0.00256396,0.6742374],"study_design_scores_gemma":[0.0009096138,0.000871026,0.8794996,0.0002803844,0.0000504389,0.0001215748,0.004729796,0.0002866516,0.005626216,0.1043072,0.002288206,0.001029369],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6749615,0.002377786,0.3044482,0.0169088,0.0001730063,0.0003476464,0.00001356968,0.0004709179,0.0002985394],"genre_scores_gemma":[0.7433425,0.000001633015,0.255436,0.0002475899,0.00008125598,0.0000525474,0.000009437008,0.00001205259,0.0008169753],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8091753,"threshold_uncertainty_score":0.9248577,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2002561419","doi":"10.1080/15434303.2014.936603","title":"Using Lexical Profiling Tools to Investigate Children’s Written Vocabulary in Grade 3: An Exploratory Study","year":2015,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; University of Calgary","funders":"","keywords":"Vocabulary; Rubric; Lexical diversity; Salient; Psychology; Computer science; Linguistics; Profiling (computer programming); Exploratory research; Vocabulary development; Natural language processing; Lexical density; Trait; Artificial intelligence; Mathematics education; Lexical item; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.1277240996145899,"gpt":0.430229454825071,"spread":0.3025053552104811,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002637984,0.000147985,0.0002077686,0.0001803436,0.0002353262,0.0003643282,0.0002291485,0.00007915723,0.00001125822],"category_scores_gemma":[0.0001356636,0.0001580007,0.00003082362,0.0003914221,0.00006112197,0.0007167904,0.0000220987,0.0002436164,0.00001509819],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002959958,"about_ca_system_score_gemma":0.0005688856,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006299836,"about_ca_topic_score_gemma":0.01298977,"domain_scores_codex":[0.9972938,0.001050411,0.0003125855,0.000384343,0.0005367931,0.0004220735],"domain_scores_gemma":[0.9991072,0.00006832943,0.00008037413,0.0002718226,0.00007916812,0.0003931522],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000125865,0.001161682,0.43023,0.000008751249,0.00001833998,0.00007879752,0.5546758,0.00007963276,0.003168103,0.001595917,0.0001058807,0.00886455],"study_design_scores_gemma":[0.0007641609,0.0007511763,0.1161382,0.00007210406,0.00002328103,0.000003671978,0.8810411,0.0003250043,0.0001297832,0.0003811876,0.00003249125,0.0003378797],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9970941,0.00004538858,0.0002869081,0.0005533728,0.0002307511,0.0007520586,0.000005741676,0.000157382,0.0008743054],"genre_scores_gemma":[0.9917354,3.749739e-7,0.00706238,0.0003061135,0.000696584,0.00008802666,0.0000349129,0.0000223749,0.00005383479],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3263653,"threshold_uncertainty_score":0.952351,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2028824355","doi":"10.1080/15434300801934751","title":"Comments on “Evaluation of the Usefulness of the<i>Versant for English</i>Test: A Response”: The Author Responds","year":2008,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Institute for Christian Studies","funders":"","keywords":"Test (biology); Psychology; English language; Language assessment; Mathematics education; Linguistics; Computer science; Cognitive psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.0786983967731293,"gpt":0.3273672114131697,"spread":0.2486688146400404,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002996123,0.0001388345,0.0001776122,0.00005501841,0.0007511618,0.0000536849,0.0004478805,0.00004148886,0.0001992077],"category_scores_gemma":[0.0004549111,0.00006923619,0.0001822421,0.00007175727,0.0002083104,0.00009493726,0.0000283828,0.0002999836,0.000002143958],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007069464,"about_ca_system_score_gemma":0.0001481035,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001898412,"about_ca_topic_score_gemma":0.0002238977,"domain_scores_codex":[0.9972039,0.0014268,0.0003188625,0.0001655149,0.0007014611,0.0001834243],"domain_scores_gemma":[0.9974958,0.001301713,0.0003446569,0.0006080061,0.0002288604,0.00002102096],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003670759,0.0003783925,0.002387921,0.00004656475,0.00009628521,0.000001561647,0.9611493,0.00009622127,0.002109016,0.02116338,0.005474928,0.006729292],"study_design_scores_gemma":[0.007172287,0.004920913,0.2214613,0.0009073002,0.0007476123,0.00001209707,0.6172599,0.006174871,0.006508079,0.001656918,0.1324107,0.0007680353],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9917682,0.00004268778,0.00004743382,0.00150682,0.0006475321,0.0006282856,0.00007133525,0.00002983928,0.005257904],"genre_scores_gemma":[0.9951643,3.475844e-7,0.00004016339,0.0003280049,0.0002196172,0.00009087246,0.000008211459,0.00001950724,0.00412892],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3438895,"threshold_uncertainty_score":0.5777406,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2955380721","doi":"10.1080/15434303.2019.1628238","title":"“Be a Machine”: International Graduate Students’ Narratives around High-Stakes English Tests","year":2019,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Second Language Learning and Teaching","field":"Arts and Humanities","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Test of English as a Foreign Language; Narrative; Test (biology); Agency (philosophy); Language proficiency; Mathematics education; Language assessment; Psychology; Pedagogy; Study abroad; Medical education; Sociology; Linguistics; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.02159823643195352,"gpt":0.2981795875042351,"spread":0.2765813510722815,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0003910792,0.0002812064,0.0002883778,0.000170238,0.0002366649,0.001075395,0.0005170614,0.00005018563,0.009127589],"category_scores_gemma":[0.00002465464,0.0002372752,0.0001219798,0.00004213826,0.00007196659,0.0006191853,0.00004053348,0.0005262446,0.0001888761],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007840367,"about_ca_system_score_gemma":0.00003267734,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002052845,"about_ca_topic_score_gemma":0.001934968,"domain_scores_codex":[0.9980704,0.0001928726,0.0003211245,0.0004213761,0.0006477235,0.0003465639],"domain_scores_gemma":[0.9990793,0.0001502657,0.0001890679,0.000382428,0.0001171422,0.00008182456],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002832333,0.0003646926,0.01924434,0.00004193885,0.0003328171,0.0001133751,0.9229043,0.00001030571,0.0006247049,0.03789573,0.001087817,0.01735167],"study_design_scores_gemma":[0.00185901,0.0008046326,0.01574284,0.00009490437,0.00004205905,0.00000856458,0.9430533,0.0002798982,0.00002568546,0.0003666711,0.03711516,0.0006072807],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.92681,0.0001323384,0.00005877423,0.0004261753,0.002479571,0.0002558429,0.00007167429,0.0003038565,0.06946173],"genre_scores_gemma":[0.9751648,0.000001772397,0.0003771827,0.0007979156,0.001335034,0.00003433993,0.0003271212,0.00004460079,0.02191726],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04835474,"threshold_uncertainty_score":0.9999616,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4220982091","doi":"10.1080/15434303.2022.2038172","title":"Investigating the Effects of Task Type and Linguistic Background on Accuracy in Automated Speech Recognition Systems: Implications for Use in Language Assessment of Young Learners","year":2022,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Speech Recognition and Synthesis","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Christian Studies; University of Toronto","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Task (project management); Computer science; Natural language processing; Artificial intelligence; Meaning (existential); Task analysis; Language proficiency; Test (biology); Psychology; Speech recognition","retraction":null,"screen_n_in":null,"score":{"opus":0.03373161378752211,"gpt":0.3391697548696433,"spread":0.3054381410821213,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007901873,0.0001376777,0.0002578178,0.0002632455,0.00009964306,0.0001197882,0.0003134268,0.00004047323,0.000007160274],"category_scores_gemma":[0.0004106816,0.0001208793,0.00004151708,0.000671074,0.00003731088,0.0002187025,0.00005566734,0.0002094251,7.998246e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000149962,"about_ca_system_score_gemma":0.0001346493,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001161538,"about_ca_topic_score_gemma":0.0002669494,"domain_scores_codex":[0.9983001,0.0004852442,0.0004469017,0.0003127513,0.0002515895,0.0002034236],"domain_scores_gemma":[0.9968898,0.002262396,0.0003245895,0.0003838889,0.00009759433,0.00004167946],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00007194033,0.002177156,0.09075653,0.002364651,0.0002893207,0.0002100774,0.07732934,0.0003813506,0.2397976,0.01711576,0.0005045187,0.5690017],"study_design_scores_gemma":[0.003811826,0.003415609,0.5500887,0.001022584,0.000107409,0.00008704417,0.05579581,0.3804032,0.003116983,0.001361287,0.00004726583,0.0007422416],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9951417,0.00009034952,0.003044118,0.0001155248,0.0002428165,0.0009524125,0.00005021445,0.00008451298,0.0002783866],"genre_scores_gemma":[0.9772757,0.000005544649,0.02219416,0.00009742351,0.00002048904,0.0003024883,0.00007095152,0.0000138024,0.00001940587],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5682595,"threshold_uncertainty_score":0.4929315,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4290958425","doi":"10.1080/15434303.2022.2073886","title":"Developing a Scenario-Based English Language Assessment in an Asian University","year":2022,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Language assessment; Linguistics; Language proficiency; Psychology; Mathematics education; Sociology; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.01712562869737202,"gpt":0.2795077938743099,"spread":0.2623821651769379,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008401234,0.0002346399,0.0002734024,0.0003467374,0.0008733291,0.0003067803,0.0004307335,0.00004170007,0.002972894],"category_scores_gemma":[0.000009299368,0.0002585398,0.00009371968,0.0001467065,0.00006157136,0.0004276185,0.00006055425,0.0009889649,0.00001006102],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007123305,"about_ca_system_score_gemma":0.0003705552,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002901461,"about_ca_topic_score_gemma":0.009161063,"domain_scores_codex":[0.9975241,0.0008623251,0.0002697662,0.0004425501,0.0004500706,0.0004512009],"domain_scores_gemma":[0.9992201,0.00008864541,0.0001543348,0.0003898806,0.00004854992,0.00009853939],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004240245,0.000570468,0.01077061,0.00006672284,0.00004902412,0.0009867682,0.8673573,0.0003600421,0.0003492221,0.08566462,0.0002832735,0.03349951],"study_design_scores_gemma":[0.001443496,0.0008800743,0.009062035,0.00005364723,0.00002917051,0.000003068094,0.9456292,0.003114318,0.000007014879,0.00007223843,0.03919638,0.0005093745],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9560176,0.00004801408,0.001063788,0.0006381067,0.0005079249,0.0002812499,0.00005653458,0.0003691141,0.04101768],"genre_scores_gemma":[0.9937437,3.553579e-7,0.002901983,0.0004564562,0.0003092686,0.00002449106,0.0003074609,0.00003902467,0.002217246],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08559238,"threshold_uncertainty_score":0.9999867,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4396694688","doi":"10.1080/15434303.2024.2346089","title":"The Academic Achievement of Undergraduate Students with Different English Language Proficiency Profiles","year":2024,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Second Language Learning and Teaching","field":"Arts and Humanities","cited_by":5,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"York University","funders":"York University","keywords":"Language proficiency; Mathematics education; Academic achievement; Psychology; Language assessment","retraction":null,"screen_n_in":null,"score":{"opus":0.009969094360222603,"gpt":0.2888573473264172,"spread":0.2788882529661946,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005094784,0.0002734727,0.0002492467,0.0001187554,0.0003813168,0.0006480062,0.0004930673,0.00004987644,0.000258537],"category_scores_gemma":[0.0000149429,0.0001474507,0.0001073542,0.00007820329,0.0001815839,0.0002429361,0.00004465547,0.0007018984,0.00001753969],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007014588,"about_ca_system_score_gemma":0.00005725858,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004517772,"about_ca_topic_score_gemma":0.000853608,"domain_scores_codex":[0.9979455,0.0002030755,0.000389423,0.0003599794,0.0006883228,0.0004137194],"domain_scores_gemma":[0.9990898,0.0002614463,0.000147379,0.0003786317,0.00005433141,0.0000683566],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000386411,0.0002681976,0.006061791,0.0003299014,0.0004170667,0.0001368745,0.726674,0.000004670115,0.001260241,0.1622972,0.0005211754,0.1019902],"study_design_scores_gemma":[0.001128374,0.002421236,0.006847472,0.000861092,0.000276224,0.00001061203,0.9754286,0.0006163207,0.0009117102,0.0005649865,0.01022713,0.0007062203],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9869882,0.002633921,0.0001966714,0.000551855,0.0006808381,0.0004715573,0.00003385007,0.0003219009,0.008121135],"genre_scores_gemma":[0.9882254,0.00001638685,0.0000450051,0.00008479388,0.0005992431,0.0001083218,0.00005838456,0.00004637696,0.01081607],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2487546,"threshold_uncertainty_score":0.6248739,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4323544073","doi":"10.1080/15434303.2023.2184266","title":"Aligning Language Frameworks: An Example with the CLB and CEFR","year":2023,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Second Language Learning and Teaching","field":"Arts and Humanities","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Rasch model; Benchmarking; Dimension (graph theory); Computer science; German; Argument (complex analysis); Linguistics; Vocabulary; Certificate; Natural language processing; Computational linguistics; Language proficiency; Artificial intelligence; Psychology; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.02005334755489001,"gpt":0.2752021690098073,"spread":0.2551488214549173,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005484178,0.0002133799,0.0001969639,0.0001149605,0.000624606,0.0007486553,0.0002250085,0.00007595622,0.001131914],"category_scores_gemma":[0.000008406101,0.0001323814,0.00004451845,0.00009667943,0.0001385073,0.0003650283,0.00002646692,0.0006493533,0.00006871385],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001981689,"about_ca_system_score_gemma":0.00002470066,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006839872,"about_ca_topic_score_gemma":0.00741013,"domain_scores_codex":[0.9986045,0.0002143442,0.0001621862,0.0003529482,0.0002666618,0.0003993184],"domain_scores_gemma":[0.9990612,0.0002692427,0.00009699659,0.0004512317,0.00003096972,0.00009036184],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001080775,0.00003964519,0.0004961847,0.00003001023,0.00007648129,0.0002267584,0.8525257,0.00001097249,0.0005724882,0.02496104,0.00137938,0.1196706],"study_design_scores_gemma":[0.0004785633,0.0006584169,0.002713481,0.00006172082,0.00005636114,0.00002070967,0.9796958,0.0009874803,0.000025679,0.0001303006,0.01483835,0.0003331432],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9849994,0.000317055,0.0003732293,0.0007247393,0.0001626314,0.0001842617,0.000016343,0.0005715318,0.01265087],"genre_scores_gemma":[0.9922587,0.000002258946,0.0003904813,0.001166112,0.0008330718,0.00004671986,0.000131703,0.00005089961,0.005120045],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1271702,"threshold_uncertainty_score":0.9997812,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4391927924","doi":"10.1080/15434303.2024.2311724","title":"The Development and Initial Validation of O-WSVLT, a Meaning-Recall Online L2 Spanish Vocabulary Levels Test","year":2024,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology; Meaning (existential); Vocabulary; Linguistics; Test (biology); Language proficiency; Recall; Vocabulary development; Cognitive psychology; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.02360608309281445,"gpt":0.3639875715569394,"spread":0.3403814884641249,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005859068,0.0001832388,0.0001926645,0.0001142033,0.0001516738,0.0001901663,0.0001693395,0.0000960251,0.004454937],"category_scores_gemma":[0.0000334276,0.0001400422,0.00006041227,0.0001817123,0.00006484816,0.0001436255,0.0000262795,0.0003095969,0.00007111816],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004589388,"about_ca_system_score_gemma":0.00009852782,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008839235,"about_ca_topic_score_gemma":0.00009513569,"domain_scores_codex":[0.9984726,0.0002013338,0.0004479366,0.0003368974,0.0002600162,0.0002812023],"domain_scores_gemma":[0.9988405,0.0006298001,0.0001155227,0.0002907131,0.00004946917,0.00007407299],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000452397,0.0003899719,0.001138133,0.0001586872,0.0004256057,0.0007384956,0.2016637,0.000002973693,0.01096854,0.009408382,0.001514887,0.7735454],"study_design_scores_gemma":[0.005472224,0.004472151,0.3194526,0.001172982,0.0005094934,0.0008546462,0.5836545,0.002372616,0.01213329,0.001243798,0.06651432,0.002147364],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9802194,0.003948024,0.002686016,0.0003912582,0.0006382292,0.0002564985,0.00005378405,0.0001768407,0.01162991],"genre_scores_gemma":[0.9949679,0.000002268969,0.002162643,0.0005738481,0.0002987536,0.00003676175,0.0001724728,0.00003400696,0.001751311],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.771398,"threshold_uncertainty_score":0.9964551,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4405148704","doi":"10.1080/15434303.2024.2438142","title":"The Academic Achievement of Undergraduate Students with Different TOEFL iBT Score Profiles: A Replication Study","year":2024,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"York University","funders":"York University","keywords":"Test of English as a Foreign Language; Replication (statistics); Psychology; Mathematics education; Academic achievement; Language assessment; Statistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.01246481434015625,"gpt":0.349279442253663,"spread":0.3368146279135067,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006087375,0.0001795869,0.000184372,0.0000829928,0.0001384244,0.0003819167,0.001024307,0.00003848883,0.000001599297],"category_scores_gemma":[0.000008750815,0.0001005638,0.00005236595,0.0004311984,0.0000371502,0.0002062565,0.0001001279,0.0004261221,0.000007986389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007472916,"about_ca_system_score_gemma":0.00009430547,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005202865,"about_ca_topic_score_gemma":0.00005835441,"domain_scores_codex":[0.9978987,0.0001915785,0.0003501787,0.0005156952,0.0007971546,0.0002467186],"domain_scores_gemma":[0.9985532,0.0001471957,0.0001489745,0.001026401,0.0000648014,0.00005941532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0000558437,0.003123515,0.351705,0.0003883801,0.001308943,0.0002647144,0.06044709,0.0001842119,0.005221935,0.08809556,0.000703849,0.488501],"study_design_scores_gemma":[0.003173164,0.01934534,0.7479639,0.001436942,0.0005008889,0.00003749141,0.0498673,0.1651597,0.001135501,0.00935493,0.0007807754,0.001244067],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9491369,0.0002421828,0.0436085,0.006020373,0.0001382099,0.0005741208,0.000003515794,0.000195934,0.00008030186],"genre_scores_gemma":[0.9977668,0.00001660896,0.0009135941,0.00004804061,0.00007220569,0.00008784634,0.00001154122,0.00001580267,0.001067565],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4872569,"threshold_uncertainty_score":0.4100872,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4406605978","doi":"10.1080/15434303.2024.2448963","title":"Test Takers’ Attitudes Toward Varieties of Accents in Listening Tasks of the Duolingo English Test (2021 test version)","year":2025,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Linguistics, Language Diversity, and Identity","field":"Arts and Humanities","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Brock University","funders":"","keywords":"Test (biology); Active listening; Psychology; Communication","retraction":null,"screen_n_in":null,"score":{"opus":0.01004874064719094,"gpt":0.2588661768107817,"spread":0.2488174361635908,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003020353,0.0002055422,0.0003669812,0.0001861711,0.0001919918,0.0001939405,0.0005582641,0.00007767183,0.0008006198],"category_scores_gemma":[0.002744091,0.000169721,0.0001630989,0.0001541451,0.0002745546,0.0002430094,0.0001541908,0.0002672081,0.000009213869],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009768789,"about_ca_system_score_gemma":0.0001364775,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004852038,"about_ca_topic_score_gemma":0.004575317,"domain_scores_codex":[0.998526,0.0000697779,0.0004679498,0.0002806179,0.000376977,0.000278703],"domain_scores_gemma":[0.9977451,0.0006314099,0.0002726576,0.0004438197,0.0008655202,0.00004149758],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001308859,0.0009491231,0.7698497,0.0004329456,0.0000952595,0.00002744258,0.2203466,0.000003576118,0.0003105421,0.003817083,0.003164022,0.0009905933],"study_design_scores_gemma":[0.002227938,0.0006180283,0.5072176,0.0008696498,0.0003499447,3.663901e-7,0.4821713,0.0001852671,0.0009926008,0.0006478268,0.00421149,0.0005079823],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9665318,0.0003597569,0.00003463438,0.0000760426,0.005715716,0.0003676447,0.0004678673,0.0000544208,0.02639209],"genre_scores_gemma":[0.9956277,0.0000131174,0.0001515512,0.0001228256,0.001104946,0.00000719656,0.00004360715,0.00001325414,0.002915775],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2626321,"threshold_uncertainty_score":0.8766225,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4385361752","doi":"10.1080/15434303.2023.2237487","title":"The Canadian English Language Proficiency Index Program (CELPIP) Test","year":2023,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Index (typography); Language proficiency; Language assessment; Test (biology); Linguistics; Test of English as a Foreign Language; Psychology; Mathematics education; Computer science; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.01261532384898779,"gpt":0.2915964155539256,"spread":0.2789810917049378,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0008037756,0.0002058801,0.0001594096,0.0001778392,0.001804934,0.001391422,0.0004122182,0.00006504244,0.0005489207],"category_scores_gemma":[0.0001066867,0.0001458548,0.00009081815,0.0001639108,0.0001609546,0.0002040959,0.00002243013,0.0005807744,0.000318577],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000129097,"about_ca_system_score_gemma":0.0002212818,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1012706,"about_ca_topic_score_gemma":0.6043105,"domain_scores_codex":[0.9981971,0.0001457784,0.0002538466,0.0002966456,0.0003821367,0.0007244915],"domain_scores_gemma":[0.9989436,0.0002684548,0.00009821641,0.0004298486,0.00008449742,0.0001754316],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000003353329,0.0001289232,0.008477188,0.00003782659,0.00004499299,0.0002095472,0.7134807,0.000005758194,0.000082322,0.02696617,0.01038941,0.2401739],"study_design_scores_gemma":[0.0004213448,0.0008257196,0.01532376,0.00005514741,0.00002885233,0.000003600545,0.5988744,0.002268922,0.000009045413,0.0001687332,0.3815727,0.0004478384],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7016331,0.0002385344,0.000009463056,0.001335995,0.001224501,0.000737383,0.00005404128,0.001629131,0.2931379],"genre_scores_gemma":[0.9685302,0.00000234604,0.00005380716,0.0001836718,0.001121726,0.0001759952,0.00011911,0.00004315991,0.02976993],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5030398,"threshold_uncertainty_score":0.9996452,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3111103222","doi":"10.1080/15434303.2020.1846190","title":"“Follow Your Interests Because Those Will Motivate You to Excel”: An Interview with Alister Cumming","year":2020,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of British Columbia; York University","funders":"","keywords":"Psychology; Library science; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.07021481758870055,"gpt":0.395323558554342,"spread":0.3251087409656415,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006396985,0.0001962878,0.0002422146,0.0000747466,0.0002876946,0.000470607,0.0003921152,0.0000685698,0.0003009987],"category_scores_gemma":[0.00005930223,0.0001736469,0.0000658861,0.0003112649,0.00005718819,0.000684138,0.00002804325,0.0002110954,0.00005752145],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001092551,"about_ca_system_score_gemma":0.0001363558,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001753144,"about_ca_topic_score_gemma":0.004123657,"domain_scores_codex":[0.997992,0.0004444034,0.0002727716,0.000449281,0.0004126303,0.0004288896],"domain_scores_gemma":[0.9990129,0.00006310768,0.0001218257,0.0002899341,0.0001015471,0.0004106613],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004508517,0.0002938773,0.009821521,0.0001155275,0.00006884174,0.0001177505,0.8265554,0.00000487225,0.003437066,0.001915282,0.001828543,0.1557962],"study_design_scores_gemma":[0.001459226,0.003620596,0.02337172,0.001162622,0.0001406201,0.000008629941,0.928805,0.0001885402,0.0004243159,0.0001851129,0.03956272,0.001070882],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9738936,0.00007826358,0.003582457,0.01472564,0.0002279276,0.0003911186,0.00001504094,0.0002851115,0.006800854],"genre_scores_gemma":[0.9903906,0.000004129167,0.004063496,0.003225372,0.0008862569,0.00005051612,0.00004478266,0.00003492544,0.001299898],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1547253,"threshold_uncertainty_score":0.7081116,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4407098810","doi":"10.1080/15434303.2025.2455196","title":"Differential Item Functioning Due to Cultural Familiarity on a Large-Scale Reading Test: Does the Length of Residence Matter?","year":2025,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Differential item functioning; Psychology; Reading (process); Scale (ratio); Test (biology); Residence; Item response theory; Psychometrics; Developmental psychology; Linguistics; Geography; Sociology; Demography","retraction":null,"screen_n_in":null,"score":{"opus":0.07156230965493464,"gpt":0.4314743672554329,"spread":0.3599120576004983,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002635869,0.0001976091,0.0004161496,0.000381679,0.0003488715,0.0004043621,0.0008954201,0.0000760258,0.0004235713],"category_scores_gemma":[0.006917705,0.00009908403,0.0001464125,0.001785621,0.0000580519,0.0002242498,0.0001269074,0.0003146392,0.00004380498],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006266213,"about_ca_system_score_gemma":0.00003928794,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003421109,"about_ca_topic_score_gemma":0.0001998626,"domain_scores_codex":[0.9967728,0.0006052325,0.0007084486,0.0005900934,0.0009425812,0.000380809],"domain_scores_gemma":[0.9796287,0.01891887,0.0003037312,0.0008863478,0.0001903543,0.00007202741],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009287456,0.0002988652,0.710614,0.00004427918,0.00006531377,0.00003628416,0.01502431,0.00003176865,0.02495592,0.002064525,0.01152146,0.2352504],"study_design_scores_gemma":[0.0003476424,0.0004401426,0.934936,0.0001673914,0.00002672681,0.000005533338,0.05745522,0.002297265,0.00114404,0.002765784,0.0002525607,0.0001616946],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9002083,0.0000388525,0.08772122,0.00143408,0.0008612484,0.0002754668,0.00002986952,0.00005955232,0.009371401],"genre_scores_gemma":[0.9884573,0.000001683877,0.007358666,0.0005704284,0.0001180259,0.00003359832,0.000003722086,0.000007793355,0.003448737],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2350887,"threshold_uncertainty_score":0.8281639,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4410015948","doi":"10.1080/15434303.2025.2497818","title":"On the Interplay Between Conceptions of Assessment and Assessment Agency: Perspectives of Iranian EFL Preservice Teachers","year":2025,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Mathematics education; Pedagogy; Agency (philosophy); Psychology; Alternative assessment; Semi-structured interview; Qualitative research; Sociology; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.01747647657889719,"gpt":0.4119940254646767,"spread":0.3945175488857796,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001743367,0.0002825011,0.000517161,0.0002304628,0.0004996376,0.0001572197,0.0007528146,0.0001433695,0.0006956308],"category_scores_gemma":[0.00004457453,0.0002212195,0.0001846963,0.000578091,0.0005743835,0.0002949994,0.0001143729,0.0004928942,0.000003487786],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000330875,"about_ca_system_score_gemma":0.0005849473,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002279508,"about_ca_topic_score_gemma":0.0009944428,"domain_scores_codex":[0.9966862,0.0009070963,0.000596832,0.0004637791,0.0008885299,0.0004575244],"domain_scores_gemma":[0.9974739,0.001275573,0.0003910088,0.0005611354,0.0001931315,0.000105254],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00002433273,0.0007222304,0.2293775,0.00009805688,0.0007415653,0.000006567496,0.3489155,0.000004347066,0.003569879,0.3921085,0.001256981,0.0231746],"study_design_scores_gemma":[0.0007276802,0.0005283807,0.560868,0.0001753325,0.0001718709,1.443843e-7,0.4341747,0.00007823697,0.00004469846,0.002535828,0.0005005131,0.0001946642],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8479688,0.0002231862,0.00201482,0.003887486,0.000215559,0.0009416142,0.00005198371,0.00006195569,0.1446346],"genre_scores_gemma":[0.9957344,0.00006212567,0.002184031,0.0001414676,0.0001139922,0.00009997735,0.00002124134,0.00001787504,0.001624895],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3895727,"threshold_uncertainty_score":0.9021069,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4399443232","doi":"10.1080/15434303.2024.2364172","title":"Adapting to a New Normal: A Review of <i>Technology Assisted Language Assessment in Diverse Contexts</i>","year":2024,"lang":"en","type":"review","venue":"Language Assessment Quarterly","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Language assessment; Psychology; Computer science; Linguistics; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.05299991408977252,"gpt":0.4712575397084361,"spread":0.4182576256186636,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001219577,0.0009945481,0.00318065,0.00120721,0.00005701721,0.00008725081,0.001265385,0.0006757113,0.005208087],"category_scores_gemma":[0.00005081854,0.0007797022,0.0008569763,0.002627925,0.00008206622,0.0001396408,0.0001992962,0.001857733,0.0009801629],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004264192,"about_ca_system_score_gemma":0.0009391439,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001096198,"about_ca_topic_score_gemma":0.000255758,"domain_scores_codex":[0.9937788,0.000838128,0.002273635,0.001448543,0.0007373664,0.0009235708],"domain_scores_gemma":[0.9966918,0.0004179949,0.0009867125,0.001429248,0.0001127428,0.0003615312],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001018197,0.0007659027,0.00007668431,0.02602696,0.0005116272,0.0007819039,0.003198119,5.286821e-8,0.00001203882,0.003031937,0.02528999,0.9402946],"study_design_scores_gemma":[0.001004525,0.001543324,0.0008703185,0.141155,0.00191417,0.0002414291,0.01584509,9.595494e-7,3.777366e-7,0.0001956572,0.8359751,0.001254094],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002474536,0.9700064,0.0003376279,0.001013495,0.001764538,0.00266862,0.0003326955,0.0002344303,0.02339474],"genre_scores_gemma":[0.0007106119,0.9723353,0.008967584,0.002149634,0.0005942453,0.001815903,0.000927054,0.0001885249,0.01231116],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9390405,"threshold_uncertainty_score":0.9997977,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4406929912","doi":"10.1080/15434303.2025.2458599","title":"From Global Dependence to Local Expertise: An Interview with Rama Mathew","year":2025,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Education and Islamic Studies","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Psychology; Regional science; Economic geography; Sociology; Geography","retraction":null,"screen_n_in":null,"score":{"opus":0.01536946584694614,"gpt":0.3964436286729267,"spread":0.3810741628259806,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002560745,0.0001200312,0.0001694081,0.00003593292,0.0002018933,0.0001700442,0.0003338421,0.00004786521,0.0005557929],"category_scores_gemma":[0.00001384534,0.00009907395,0.00003449775,0.0003177922,0.0001014086,0.0002115067,0.00002293259,0.00007492613,0.00006289251],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002762907,"about_ca_system_score_gemma":0.0002616861,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01328942,"about_ca_topic_score_gemma":0.02737999,"domain_scores_codex":[0.9987579,0.0002353429,0.000161364,0.0002953969,0.0002887861,0.0002612032],"domain_scores_gemma":[0.9994535,0.00005322058,0.00004252728,0.0002749134,0.00006750633,0.0001082739],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000179556,0.0002825261,0.005664908,0.00001437997,0.00008029569,0.00002742881,0.2626165,0.000001325613,0.0001145823,0.02789152,0.005082428,0.6982061],"study_design_scores_gemma":[0.0004671528,0.0004263719,0.05648994,0.0002285635,0.00004403656,9.056248e-7,0.9219869,0.00006244091,0.00004190131,0.002159105,0.01781141,0.0002812515],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8027202,0.0007819665,0.1090062,0.01384096,0.0007542603,0.0004702231,0.00002582673,0.0002074239,0.07219286],"genre_scores_gemma":[0.9912912,0.00001078678,0.003483512,0.002826566,0.0001619084,0.00007356388,0.00001029258,0.000005379371,0.002136775],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6979249,"threshold_uncertainty_score":0.9932812,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}