Replication Data for: From nineteenth-century letters to entities: "a NER pipeline for French correspondence and its methodological lessons" - Article for Digital Humanities Benelux Journal doi:10.34934/DVN/HNA7QO Social Sciences and Digital Humanities Archive – SODHA 2026-08-25 1 Zuzana Černáková; Fien Messens; Tess Dejaeghere; Julie M. Birkholz, 2026, "Replication Data for: From nineteenth-century letters to entities: "a NER pipeline for French correspondence and its methodological lessons" - Article for Digital Humanities Benelux Journal", https://doi.org/10.34934/DVN/HNA7QO, Social Sciences and Digital Humanities Archive – SODHA, V1, UNF:6:MemhfA5Fl4s1tvLWf4VR2A== [fileUNF] Replication Data for: From nineteenth-century letters to entities: "a NER pipeline for French correspondence and its methodological lessons" - Article for Digital Humanities Benelux Journal doi:10.34934/DVN/HNA7QO Zuzana Černáková Fien Messens Tess Dejaeghere Julie M. Birkholz Social Sciences and Digital Humanities Archive – SODHA Messens, Fien Messens, Fien 2026-06-22 2026-08-25 Arts and Humanities Named Entity Recognition (NER) Natural Language Processing (NLP) Historical NLP This dataset accompanies the article From nineteenth-century letters to entities: A Named Entity Recognition pipeline for French correspondence and its methodological lessons. It contains the input data, preprocessing scripts, analysis notebooks, evaluation outputs, and experimental results used to develop and evaluate Named Entity Recognition (NER) workflows for nineteenth-century French correspondence from the François-Joseph Navez corpus (KBR – Royal Library of Belgium). The dataset documents the complete experimental workflow, from the construction of a manually annotated gold-standard corpus to the evaluation of off-the-shelf spaCy models, a custom-trained spaCy model, and transformer-based models (CamemBERT, CamemBERTav2, D'AlemBERT and Europeana BERT). The repository is organised into four directories. Input contains the Label Studio annotation exports, the preprocessed gold-standard corpus, and the training, development and test splits used throughout the experiments. Notebooks provides the Jupyter notebooks implementing the preprocessing, training and evaluation workflows. Output contains the evaluation results of the spaCy experiments, including model predictions, overall and per-entity performance metrics, mismatch analyses and summary tables. Results_transformers contains the outputs of the transformer-based experiments, including hyperparameter optimisation, cross-validation results, learning curves, statistical significance tests, prediction files and qualitative error analyses. Together, these files provide the complete computational workflow and all intermediate and final outputs required to reproduce the analyses presented in the accompanying publication. These files contain training and evaluation data derived from nineteenth-century letters from the Navez Project. To protect access conditions associated with the source material, the files are restricted. Researchers may request access for non-commercial scholarly use. Requests will be evaluated on a case-by-case basis. combined_eval_scores3.tab 24 13 text/tab-separated-values UNF:6:IHdDRnEBZ4B81/KMQBqVKA== cv_results.tab 20 6 text/tab-separated-values UNF:6:72/ohGHOKcR5yXwDUXrubA== errors_camembertav2.tab 159 10 text/tab-separated-values UNF:6:dX1zhbyrPPk1DreziVEOHw== errors_camembert.tab 529 10 text/tab-separated-values UNF:6:LKJL6zjeIoC7+p0mefydwg== errors_dalembert.tab 185 10 text/tab-separated-values UNF:6:+TqzKZgt8g2h9mleixtmLw== errors_europeana.tab 161 10 text/tab-separated-values UNF:6:r0qXPIXovg15bXXLwViguQ== headline_counter_table.tab 16 12 text/tab-separated-values UNF:6:2raSn3MgN38lZjzDY60qZA== headline_cv_summary.tab 4 3 text/tab-separated-values UNF:6:LwsPP4NHhKrvsotyYYTR+w== hp_search.tab 72 6 text/tab-separated-values UNF:6:zu0B79FNEke0HpLSL1WIKw== learning_curve.tab 4 4 text/tab-separated-values UNF:6:fXKPhq8vFAp9d+v7Jlmpsg== length_stratified.tab 16 5 text/tab-separated-values UNF:6:/TrpFwgbWWGaD2ZI1dFDKQ== length_summary.tab 9 5 text/tab-separated-values UNF:6:XKylVdddSgIVxmbYCdrnFg== navez_spacy_custom_trained_evalall_no_dk-2.tab 4 1 text/tab-separated-values UNF:6:yq7KQEdVC0k1T86KGkSn0g== nervaluate_counters_overall.tab 16 14 text/tab-separated-values UNF:6:/3uciuaa/yLyO27c+p2Cyg== per_tag_test.tab 128 14 text/tab-separated-values UNF:6:YzGwFRM6nYl9P3yPhkiCEg== significance_tests.tab 6 6 text/tab-separated-values UNF:6:9YJ3w4uTtbYmfNKxJTh1UA== test_predictions_camembertav2.tab 17 3 text/tab-separated-values UNF:6:6PxO9ELAf/ENQQAY98ekCQ== test_predictions_camembert.tab 17 3 text/tab-separated-values UNF:6:wwYLN6EuL27q8HvnSckZ9A== test_predictions_dalembert.tab 17 3 text/tab-separated-values UNF:6:QOnGSP2nZCOfl3Bdo7Wayw== test_predictions_europeana.tab 17 3 text/tab-separated-values UNF:6:47AUFPWKuqAoGujhisXcrQ== test_scores_per_seed.tab 48 13 text/tab-separated-values UNF:6:BULzL9VsIcllUNnVwyPBzg== scenario UNF:6:guT9fyOClBsdG7Zs4lSk0A== workflow 0.7677718959499145 4.0 . 3.0 2.0 1.8 20.0 1.0 UNF:6:eeMzP9x4WHPanAuxyKNpAw== model UNF:6:d0yQGhpV/OO02tS5+gcRPQ== correct 99.5 136.0 4.0 20.0 15.553896110316744 . 79.0 104.85 UNF:6:Blvpg/AXxXBYEc3IwdTPgg== incorrect 33.5 0.0 25.736519210115254 . 69.0 4.0 34.0 20.0 UNF:6:UfJ7MHCo6S6kJPEg3QssYw== partial 20.0 4.0 22.698191441981503 . 12.449999999999998 60.0 0.0 0.0 UNF:6:Re+LPeoTtItmdi2Va3TWvg== missed 4.0 20.0 58.0 67.0 6.75589411278462 57.2 . 49.0 UNF:6:U1MneR7DFJPh+EpR/WiUOw== spurious 4.0 117.0 20.0 . 84.0 81.0 34.931511938589054 24.0 UNF:6:RANN6Xs+DsFn5pPPY5CQow== possible 208.0 20.0 4.0 208.0 0.0 208.0 . 208.0 UNF:6:1jcKcIAFo9D5xsnGnMRXrQ== actual 165.0 234.8 267.0 240.0 20.0 37.832873480472955 . 4.0 UNF:6:0z0ZtH4pe55vvV7hqvvuvA== p . 0.3 0.445 0.8 0.4915 0.14180323654535124 20.0 4.0 UNF:6:oVxpCCMQ7IZCSaEbqNJWGw== r 0.53 0.65 0.08449696350819516 20.0 0.38 . 4.0 0.5335 UNF:6:tvCGrcWM4wbP/JjqhzPl+A== f1 0.10457306687569926 4.0 0.5075 0.33 20.0 . 0.505 0.71 UNF:6:ev20S9tHslE2eGdfK2yWLQ== model UNF:6:58+a5/pKEXueFOgiC5eWsg== fold 2.0 . 0.0 4.0 2.0 1.4509525002200232 0.0 20.0 UNF:6:GoMYq/48z1yzRJdm+KjV6A== strict_f1 0.5282051282051282 0.6378132118451024 20.0 0.10142737546736787 0.5034070979186959 0.0 . 0.3561128526645768 UNF:6:FlHLjD6/xq2jgsQy7D3CIA== exact_f1 0.0 0.5864578037620742 0.550975257485731 20.0 . 0.10674705669159901 0.3711598746081505 0.6788154897494305 UNF:6:0Lq2TriqZ3omdmbWgXOI2w== partial_f1 0.0 . 0.6350428678276339 0.7653758542141231 0.6761101581856299 0.11121765369336904 0.4369905956112852 20.0 UNF:6:Q1gc/LuFZJn9uq9ktw6uOw== type_f1 0.46962616822429903 0.0 0.6204099419197274 0.7422360248447204 0.09864066985119124 0.6624840276003068 20.0 . UNF:6:pw9c7yAtimEjW4enlILOJQ== bucket UNF:6:Z3MAwCntARuUewxy3T02sw== ms UNF:6:F1NvBj1Da9keQfhGu+N2Ww== gold_type UNF:6:zqFPcWFuMecwoZ/1J9OM1Q== gold_start 875.0 245.99122807017534 45.0 243.9100550748577 . 152.0 114.0 0.0 UNF:6:Ia4QTUBFo2c5ta9NCKuTxA== gold_end 247.32456140350888 153.5 0.0 45.0 . 875.0 243.7784608522042 114.0 UNF:6:zzh45T7/ua+2BmFtiRh8uA== gold_text UNF:6:9ndUGZB4m71YtiyapEhxzw== pred_type UNF:6:kf+pTrblULZLBOwxE8h+bQ== pred_start . 213.77687418212741 250.25555555555556 90.0 214.0 841.0 0.0 69.0 UNF:6:o9WIrCkuY0IpbCNqmMI/gA== pred_end 69.0 214.5 90.0 213.78318120070776 841.0 250.74444444444444 0.0 . UNF:6:lJiV+kwXTIuXOIHhYU+dPA== pred_text UNF:6:MP66zkrN14v/ht+snq6SRA== bucket UNF:6:KcDaCRK/7GRB1I9EUcO0jA== ms UNF:6:GSOKLOrg1oNrv9CiHAzHJQ== gold_type UNF:6:Bxn47JiR2JOwLLm/lziA3w== gold_start 227.0 259.60811768527014 881.0 0.0 295.15151515151524 430.0 . 99.0 UNF:6:GoysvWcBtqd6mYNkqmWbtQ== gold_end 1.0 259.4338090100685 . 885.0 99.0 430.0 296.43434343434325 228.0 UNF:6:o9mFEoYa2oNVdbCrOtD+OA== gold_text UNF:6:xVV2UYufg/DmkW9lLJcylA== pred_type UNF:6:KuhNT2zgeJsB6Uwn46oqSg== pred_start 931.0 7.0 220.5 209.12801729521792 522.0 259.5555555555561 . 0.0 UNF:6:UXjPSh7OWN90OQtbVQen3w== pred_end . 7.0 932.0 522.0 221.0 0.0 209.20831706344615 260.1609195402293 UNF:6:+qQhY+spzNcqnXJIuI36Rw== pred_text UNF:6:N5PyURGsNR1iDfbJPSr72w== bucket UNF:6:73tm7wEsvFfq3kSDYt7+Sw== ms UNF:6:gMwbm5A0T6rQPZoFNac92Q== gold_type UNF:6:yTuvHak0oGRJjLDuX4Ruww== gold_start 195.5 875.0 67.0 118.0 0.0 253.4745762711865 233.1473878822464 . UNF:6:Jem/npq9hpaN0JiARU6toQ== gold_end 118.0 196.0 254.83050847457636 . 67.0 233.01943373284368 875.0 1.0 UNF:6:+uzsdjTlBOknRVE2vUCYRA== gold_text UNF:6:0whxKQIn/X5ygDF2N5zx8g== pred_type UNF:6:mXbcxKB0DWCNNbaK3VgdLA== pred_start 164.0 896.0 232.71774193548384 124.0 5.0 . 209.00645963814 61.0 UNF:6:NpgVOTTT0x0stRR7+oCHQg== pred_end 233.1290322580646 164.5 124.0 6.0 . 896.0 209.18210713878793 61.0 UNF:6:sqsTt1vY1Dfqz/sbldvEaA== pred_text UNF:6:MVEsfjv9Ar7AfhD5/Dglxg== bucket UNF:6:PaeyOc/UW7LMF4KErQks6w== ms UNF:6:P7qH4V/ki4p8vpGmiE1haA== gold_type UNF:6:DqdaC91rdqw6ofiFDlQlzg== gold_start 0.0 232.3513590405187 45.0 875.0 233.7672413793103 . 116.0 149.0 UNF:6:AmW5b2qYS9IWLBD8CfPdvw== gold_end 116.0 0.0 151.0 . 232.2088033125189 875.0 235.10344827586204 45.0 UNF:6:vXqQ5az7uSFWmWZFftUzNQ== gold_text UNF:6:UBzxQx45p2SglkHccJqpGA== pred_type UNF:6:Noag+J/T3aIA5JUtr3OukA== pred_start 2.0 . 231.1290322580646 93.0 197.0 852.0 68.0 198.8270619433125 UNF:6:1sPdzqobQbyWnaqkzCxLfQ== pred_end 93.0 2.0 198.8528144010826 231.5806451612904 . 197.0 68.0 852.0 UNF:6:Z4gW3zWoU9kjCOjLVtMsJQ== pred_text UNF:6:WibaiZu9PYvHfdRi5KBxTg== model UNF:6:kpTfvbfk7/sGgBrZhUeuwQ== scheme UNF:6:hB9xBQ4Ec6uBrjXZ1F9prw== correct 0.0 257.0 25.570816699250468 . 168.0 178.5 158.0 16.0 UNF:6:7LCZAzScqMmO/Ur9G8dMHg== incorrect . 16.0 0.0 96.0 39.5 33.378635881853135 0.0 38.5 UNF:6:mR6KSi/Lir/pyB659F3Grw== partial 16.0 0.0 16.0 93.0 0.0 0.0 29.97332147093478 . UNF:6:wXR5OHCpo+p8ksbczGCi4g== missed 51.0 16.0 36.0 . 2.0 20.668817092422103 45.5 0.0 UNF:6:MnhZ0XKsgM12DO8gg1Y/aA== spurious 458.0 . 171.78979403134906 16.0 61.0 0.0 170.5 81.5 UNF:6:YAxnTAVvz+GP0jXi9fGA6A== possible 16.0 0.0 269.0 269.0 . 269.0 269.0 0.0 UNF:6:Zd3BbOJcUPPgBrdwgkQ/Jg== actual 192.27549679214633 280.0 403.5 0.0 . 725.0 16.0 304.5 UNF:6:Nq+MqFdbwx23DNebsr0U6Q== precision 0.5279125 16.0 0.2359 0.15912954418753714 0.6966 . 0.0 0.5719 UNF:6:u924ES6xbvfa+zaVYS15tw== recall 0.6933 16.0 0.9554 0.5874 0.6784 . 0.0 0.10036617624146758 UNF:6:Zgyy3p/k9BLdi9xtUjfnrQ== f1 16.0 0.58835 0.12120868436983658 0.3441 0.5816125 . 0.7227 0.0 UNF:6:qeDQxY2E1IwyvpSmqPQbhQ== model UNF:6:C1Eh2OlGL4uDPzqYdbUAEg== cv_mean 0.5252987803863477 4.0 0.5920197775226445 0.3710110533794434 . 0.10717392702937042 0.5034070979186958 0.0 UNF:6:greybgZ9wXP+BT1ENd5cvQ== cv_std 0.018774522899703643 0.03843585919122774 4.0 0.00876821372661337 . 0.0 0.05197978066849702 0.03440492819439146 UNF:6:grHn+b35EE74Q0rm0w/IWg== model UNF:6:p5nC4LgNWL5cB7uA7OP3wQ== lr 2.0331755499467835E-5 3.5833333333333335E-5 72.0 1.0E-5 3.0E-5 0.0 . 8.0E-5 UNF:6:q9BhoZj9/N/iB3jlPk843g== wd 0.0 0.005 . 0.0 72.0 0.005035088149780136 0.01 0.005 UNF:6:FIb9pp4k2q89sVYUGmeWPA== epochs 0.0 7.666666666666667 . 5.0 72.0 2.069224526445854 10.0 8.0 UNF:6:y3+jqDE1GCAbyljL1nWHDw== dev_strict_f1 72.0 0.16179550119195255 . 0.3494407811435388 0.0 0.0 0.3425073433218967 0.5603217158176944 UNF:6:6hIZIx8dU4nF1n4K5ewi/g== dev_partial_f1 . 72.0 0.5168751359019228 0.19767326842942148 0.7131367292225201 0.0 0.0 0.4874900606173867 UNF:6:AXxtGlcep36N17NgCsw0hg== fraction 1.0 . 0.625 4.0 0.25 0.0 0.625 0.3227486121839514 UNF:6:DLZRzcOjKc8xswdgXBmAHA== n_letters 26.0 104.0 33.56585566713095 65.0 0.0 . 65.0 4.0 UNF:6:ZGH9W5nVkNFTmOJUgPrqoA== dev_strict_f1 0.3578947368421053 0.2618351841028638 0.3086546513594165 0.0 . 0.041012080996138206 0.3074443422463484 4.0 UNF:6:kJQ70Pcc4VJX7tgqMTfVCw== test_strict_f1 0.29050894490237267 0.03536753183573118 . 4.0 0.25136612021857924 0.0 0.3369890329012961 0.28684031324480763 UNF:6:5RL/vXaVfOVSuXO8gRgSIg== model UNF:6:CQC7U1vPPyLCg9YVfWF6wA== quartile UNF:6:SWQcFbZTCpN/MxnBK8qqng== n_letters 16.0 0.4472135954999579 4.0 5.0 . 4.0 0.0 4.25 UNF:6:GtAejoGVV+WDhTs4JdbXPQ== strict_f1 0.12956119373002548 0.48339628265171414 . 0.6712328767123288 16.0 0.4771460423634337 0.2238805970149254 0.0 UNF:6:bSdHz4OCNBA8CYjwH2c7CA== partial_f1 . 16.0 0.6104844560448139 0.7733333333333334 0.13181334952254545 0.34328358208955223 0.0 0.5884921777521834 UNF:6:j2rQ+f7glzbNHRtVz63mnA== model UNF:6:I/dhne68SAk/z1o+6U6Jdg== split UNF:6:8Dqbf2SbdKhIsYZb+93WCg== mean_subwords 519.9942810457517 . 662.1176470588235 428.5882352941176 473.4326923076923 96.26966497755558 9.0 0.0 UNF:6:XtLC2OQihvcSp/7PXyiuTA== max_subwords 2583.0 9.0 . 0.0 640.2983245678879 999.0 1824.7777777777778 1868.0 UNF:6:9dGfZrvzOr2JpwXPZLY15w== pct_over_512 0.0 5.883289508533153 37.14806435394671 35.294117647058826 . 47.05882352941176 9.0 29.807692307692307 UNF:6:m+ZZijwxapqFWZnrf9x3eA== scenario;workflow;model;correct;incorrect;partial;missed;spurious;possible;actual;precision;recall;f1 UNF:6:yq7KQEdVC0k1T86KGkSn0g== model UNF:6:CQC7U1vPPyLCg9YVfWF6wA== split UNF:6:iRyMpghfxnLr3qVv5JG2ww== tag UNF:6:fkdvJ8NfYrQfObPiDmfmNg== scheme UNF:6:N0JYu8wqGOOIZ9OT7moybg== correct . 158.0 257.0 0.0 168.0 178.5 25.570816699250468 16.0 UNF:6:Dio7pLgF3ersydZYV19JNw== incorrect 38.5 33.378635881853135 39.5 16.0 0.0 . 0.0 96.0 UNF:6:sRbO8gFJq9YPM7S8tRv5wg== partial 0.0 0.0 93.0 29.97332147093478 16.0 16.0 0.0 . UNF:6:SXr8n+VFqWiZUGRNJ6NZfQ== missed 20.668817092422103 45.5 2.0 51.0 36.0 16.0 0.0 . UNF:6:3kteUac9rkClfn30GDzahQ== spurious 458.0 170.5 . 16.0 0.0 61.0 171.78979403134906 81.5 UNF:6:H7feCSKfcQllDaXZIqYXAw== possible . 0.0 269.0 269.0 0.0 16.0 269.0 269.0 UNF:6:Zd3BbOJcUPPgBrdwgkQ/Jg== actual 192.27549679214633 725.0 0.0 403.5 16.0 304.5 280.0 . UNF:6:N3gy9oU1HgdX4II7XHYCyg== precision 0.0 16.0 . 0.5718596059113301 0.696551724137931 0.23586206896551723 0.5279098326242723 0.15913042446406475 UNF:6:qMd3lRIFO1Zx4hSrXVlfzQ== recall 0.10037174721189591 16.0 0.0 . 0.587360594795539 0.6784386617100371 0.6933085501858736 0.9553903345724907 UNF:6:QXL3Aa+uDDS3+npf6h8tWA== f1 0.5816155373025881 16.0 0.0 . 0.3440643863179075 0.7227191413237924 0.5883652502028408 0.12121821060484483 UNF:6:qVMQmbS5B9sMzXqTIVKJWQ== model UNF:6:X5BN8XJ5YIjhihmTv4sR2g== split UNF:6:JLMKKUcUrzlFi0DbdlfGeQ== tag UNF:6:Q1VURVVLykE6y4/vQt2qgQ== scheme UNF:6:mAwlvWGuGWYXdPiXVjhAPQ== correct 0.0 257.0 128.0 44.625 15.0 0.0 . 59.25522004315067 UNF:6:N3us5JBSOpzetds0ZTc0zA== incorrect . 0.0 0.0 128.0 2.5 96.0 17.61307674544527 9.625 UNF:6:sKNLVqo6o+U1tb05poJPlA== partial 12.647243015990158 . 0.0 4.0 93.0 0.0 128.0 0.0 UNF:6:IFKguEuxFMa8TR+bVhB2ug== missed 0.0 128.0 51.0 13.1400691521329 4.5 0.0 9.0 . UNF:6:j0B3BWLXbcg62NKxCqMmYg== spurious 0.0 458.0 16.0 . 128.0 82.08532146492453 0.0 42.625 UNF:6:pA8zRs5gYb4/m0E3C6Hplg== possible . 269.0 0.0 5.0 128.0 67.25 84.18464894569696 32.5 UNF:6:D/OcIdlkuYp3buwj/F2rJQ== actual 142.70585352367738 0.0 128.0 49.5 100.875 1.0 725.0 . UNF:6:0e7NAtaDqyNcldrExd/RuQ== precision 0.25868873594147107 0.0 . 0.0 0.5 0.457882607416167 128.0 1.0 UNF:6:frXeISLH9ZBFHe1OMLmBuA== recall 0.0 0.0 1.0 0.26905508190058236 . 0.6181012296253932 0.5741175292008275 128.0 UNF:6:wvzElCF4yUEGe0E/ptr4dg== f1 0.4720642287713962 0.8584070796460176 0.0 0.24287730198526605 . 0.5033333333333334 0.0 128.0 UNF:6:j8qEaNwnfJFQ2nS6aFWRmA== a UNF:6:9adg+TH/qgHDspO5AOYyJw== b UNF:6:5BgcMwjSJpXvHMYYgUA2GA== mean_diff 0.0 . -0.11643279139425317 -0.014253623730299734 -0.23286558278850633 -0.12106864964821236 6.0 0.10317161748169774 UNF:6:cfWL6ek4c4VcurE1jHkM1w== ci95_lo -0.15704353748089273 -0.15791745353253395 6.0 -0.037347367799293374 0.11358703502589378 -0.27599502851993896 0.0 . UNF:6:wvON5infQ75SobPbpuO65A== ci95_hi -0.17315830891823278 0.010631568937260551 -0.06860708155935627 6.0 0.0 0.085110869108066 -0.07528920974788772 . UNF:6:P6ZCuLLlf//3GDyQ7Li3Pw== p_2sided 6.0 0.1095694604653444 0.05766666666666667 . 0.0 0.0 0.274 0.001 UNF:6:pHciBQhg68ueWWkX5mjJoA== ms UNF:6:MVYbDZTCwKmFIiA58Vxx5A== true_iob UNF:6:PeolXhxV6ezA6LWOB05rFg== pred_iob UNF:6:lCLbDgK/6vfjzu3jQdEHIQ== ms UNF:6:MVYbDZTCwKmFIiA58Vxx5A== true_iob UNF:6:PeolXhxV6ezA6LWOB05rFg== pred_iob UNF:6:BarU6CyBJtiTJm/pH1TFuQ== ms UNF:6:MVYbDZTCwKmFIiA58Vxx5A== true_iob UNF:6:PeolXhxV6ezA6LWOB05rFg== pred_iob UNF:6:iCvgRAWB7JPtKj1WqMwudA== ms UNF:6:MVYbDZTCwKmFIiA58Vxx5A== true_iob UNF:6:PeolXhxV6ezA6LWOB05rFg== pred_iob UNF:6:ha1yGA/ct/ODcspLhPicPg== model UNF:6:wBX2UHiWVwwDugXRAFGw/w== seed 0.0 48.0 . 123.0 2024.0 42.0 925.5206546963191 729.6666666666667 UNF:6:1a3/v7xiN5JkO6ChFGSvzQ== scheme UNF:6:3hi6QXy5b0megqCl646e/A== correct 48.0 24.7998280564847 177.22916666666666 168.0 0.0 151.0 . 257.0 UNF:6:GUoKMTE3c8DxtB93Uq8jyA== incorrect 0.0 102.0 . 40.45833333333333 48.5 48.0 0.0 33.8802385967969 UNF:6:7U1iBKSkT8ecmvrQko8IJw== partial 0.0 48.0 0.0 100.0 16.3125 . 30.062457148197883 0.0 UNF:6:yFVwbAVnWkqYQBS83nHyTQ== missed 19.254178877722083 2.0 . 43.0 48.0 0.0 51.0 35.0 UNF:6:FGOCkp0MFVokYc8exdpSNA== spurious 503.0 61.0 74.0 48.0 0.0 . 173.25 178.23108356551103 UNF:6:vE89mIQe+NOqwbDfctq1GA== possible . 269.0 48.0 269.0 269.0 0.0 0.0 269.0 UNF:6:go4x4422IOs3AwH5YqYj0A== actual 48.0 280.0 407.25 300.0 . 769.0 0.0 197.17154211821895 UNF:6:ja92tg6ekM5jfs+23CCyaQ== precision 0.0 0.5717575187969925 . 0.1614041161638638 0.525891676189172 0.21326397919375814 0.7295373665480427 48.0 UNF:6:VgvhfGomm/8BEBQnNfQvyw== recall 0.5613382899628253 0.6891651177199505 0.0 0.9553903345724907 0.09838663123574758 48.0 0.6635687732342007 . UNF:6:0EcO3YLGW1FXvrf23hFmzQ== f1 0.1254759793685271 0.7454545454545454 0.578250574832892 . 0.5925530735985562 48.0 0.0 0.31599229287090563 UNF:6:wqujaaVc+rWC559pknk/tQ== 00_navez_gold_standard_preprocessing_json2df.ipynb Preprocessing notebook that converts Label Studio JSON exports into the structured gold-standard corpus used throughout the experiments. The workflow merges annotation projects, extracts entity annotations, aligns character offsets, generates IOB labels, removes unsupported nested entities, and creates the training, development and test datasets used for the NER experiments. application/x-ipynb+json 01a_navez_NER_spacy_off_the_shelf_no_dk_ner_pipe_sm.ipynb Implements Workflow 1 described in the article by evaluating the off-the-shelf spaCy fr_core_news_sm model using only the isolated Named Entity Recognition (NER) component. The notebook evaluates the model on the held-out test set without additional domain-specific training. application/x-ipynb+json 01b_navez_NER_spacy_off_the_shelf_no_dk_ner_pipe_lg.ipynb Implements Workflow 1 using the larger fr_core_news_lg model. The notebook evaluates the isolated NER component on the historical correspondence corpus and compares its performance with the smaller spaCy model. application/x-ipynb+json 02a_navez_NER_spacy_off_the_shelf_no_dk_nlp_pipe_sm.ipynb Implements Workflow 2 by using the complete spaCy fr_core_news_sm pipeline, including all NLP components. The notebook assesses whether embedding the NER component within the full pipeline influences recognition performance. application/x-ipynb+json 02b_navez_NER_spacy_off_the_shelf_no_dk_nlp_pipe_lg.ipynb Implements Workflow 2 using the complete fr_core_news_lg pipeline. Results are compared with Workflow 1 to evaluate the effect of the full NLP pipeline on NER performance. application/x-ipynb+json 03_navez_NER_spacy_custom-trained_dk.ipynb Implements Workflow 3 described in the article. The notebook fine-tunes fr_core_news_lg on the Navez gold-standard corpus by extending the default spaCy model with the domain-specific entity types ART, EXH, GRP and LETT, in addition to the standard PER, LOC and ORG categories. It trains and evaluates the custom NER model on the historical correspondence corpus. application/x-ipynb+json best_hps.json JSON file containing the final hyperparameter configuration selected for each transformer model (CamemBERT, CamemBERTav2, D'AlemBERT and Europeana BERT) after hyperparameter optimisation. application/json development_set-2.csv Development (validation) partition of the corpus used for hyperparameter optimisation, model selection and intermediate evaluation during NER model training. text/csv development_set-2.pkl Python Pickle version of the development dataset used in the training workflows. application/octet-stream headline_test_summary.csv Summary of the final evaluation results obtained on the held-out test set for all transformer-based models averaged across three different seeds. text/csv navez_gs_preprocessed-2.csv Preprocessed version of the manually annotated corpus after cleaning and conversion from Label Studio. Contains token- and entity-level information used for model training and evaluation. text/csv navez_gs_preprocessed-2.pkl Python pickle version of the preprocessed gold-standard corpus, preserving data structures used directly in the notebooks. application/octet-stream Navez_ner_finetuning_weightedcrossentropy.ipynb Notebook for fine-tuning and evaluating French transformer-based models for Named Entity Recognition on the Navez correspondence. It implements the experiments with CamemBERT, CamemBERTav2, D'AlemBERT and Europeana BERT, including weighted cross-entropy training, hyperparameter optimisation and evaluation using the Nervaluate framework. application/x-ipynb+json navez_spacy_custom_trained_evalall-2.csv Overall evaluation metrics for the custom-trained spaCy model, including the four project-specific entity types introduced during fine-tuning. Reports precision, recall and F1 under the Nervaluate evaluation scenarios. text/csv navez_spacy_custom_trained_evalpertype-2.csv Detailed evaluation metrics for each entity type recognised by the custom-trained model, including both standard and domain-specific categories (PER, LOC, ORG, GRP, ART, EXH and LETT). text/csv navez_spacy_custom_trained_mismatches.csv Token-level comparison between predicted and gold-standard IOB labels for the custom-trained spaCy model. Used for the qualitative error analysis discussed in the article. text/csv navez_spacy_custom_trained_results.csv Predictions generated by the custom-trained fr_core_news_lg model after fine-tuning on the Navez gold-standard corpus. Includes all project-specific entity types (PER, LOC, ORG, GRP, ART, EXH and LETT). text/csv navez_spacy_off_the_shelf_ner_lg_evalall-2.csv Overall evaluation results for the off-the-shelf fr_core_news_lg model. Reports precision, recall and F1 scores under the Strict, Exact, Partial and Type evaluation scenarios using the Nervaluate framework. text/csv navez_spacy_off_the_shelf_ner_lg_evalpertype-2.csv Evaluation metrics for each entity type recognised by the off-the-shelf fr_core_news_lg model (PER, LOC and ORG). Enables comparison of model performance across entity categories. text/csv navez_spacy_off_the_shelf_ner_lg_mismatches.csv Token-level comparison between predicted and gold-standard IOB labels for the off-the-shelf fr_core_news_lg model. Used for qualitative error analysis and identification of recurring recognition errors. text/csv navez_spacy_off_the_shelf_ner_lg_results.csv Model predictions generated by the off-the-shelf fr_core_news_lg model using the isolated NER component (Workflow 1). Contains predicted entities, spans, labels and corresponding gold-standard annotations for the held-out test set. text/csv navez_spacy_off_the_shelf_ner_sm_evalall-2.csv Overall evaluation results for the off-the-shelf fr_core_news_sm model. Reports precision, recall and F1 scores under the Strict, Exact, Partial and Type evaluation scenarios using the Nervaluate framework. text/csv navez_spacy_off_the_shelf_ner_sm_evalpertype-2.csv Evaluation metrics for each entity type recognised by the off-the-shelf fr_core_news_sm model (PER, LOC and ORG). Enables comparison of model performance across entity categories. text/csv navez_spacy_off_the_shelf_ner_sm_mismatches.csv Token-level comparison between predicted and gold-standard IOB labels for the off-the-shelf fr_core_news_sm model. Used for qualitative error analysis and identification of recurring recognition errors. text/csv navez_spacy_off_the_shelf_ner_sm_results.csv Model predictions generated by the off-the-shelf fr_core_news_sm model using the isolated NER component (Workflow 1). Contains predicted entities, spans, labels and corresponding gold-standard annotations for the held-out test set. text/csv navez_spacy_off_the_shelf_nlp_lg_evalall-2.csv Overall evaluation results for the complete fr_core_news_lg NLP pipeline. Reports precision, recall and F1 scores under the Strict, Exact, Partial and Type evaluation scenarios using the Nervaluate framework. text/csv navez_spacy_off_the_shelf_nlp_lg_evalpertype-2.csv Evaluation metrics for each entity type recognised by the complete fr_core_news_lg pipeline (PER, LOC and ORG). Enables comparison with the isolated NER workflow and the custom-trained model. text/csv navez_spacy_off_the_shelf_nlp_lg_mismatches.csv Token-level comparison between predicted and gold-standard IOB labels for the complete fr_core_news_lg NLP pipeline. Used for qualitative error analysis and comparison with the isolated NER workflow. text/csv navez_spacy_off_the_shelf_nlp_lg_results.csv Model predictions generated by the off-the-shelf fr_core_news_lg model using the complete spaCy NLP pipeline (Workflow 2). Contains predicted entities, spans, labels and corresponding gold-standard annotations for the held-out test set. text/csv navez_spacy_off_the_shelf_nlp_sm_evalall-2.csv Overall evaluation results for the complete fr_core_news_sm NLP pipeline. Reports precision, recall and F1 scores under the Strict, Exact, Partial and Type evaluation scenarios using the Nervaluate framework. text/csv navez_spacy_off_the_shelf_nlp_sm_evalpertype-2.csv Evaluation metrics for each entity type recognised by the complete fr_core_news_sm pipeline (PER, LOC and ORG). Enables comparison with the isolated NER workflow. text/csv navez_spacy_off_the_shelf_nlp_sm_mismatches.csv Token-level comparison between predicted and gold-standard IOB labels for the complete fr_core_news_sm NLP pipeline. Used for the qualitative error analysis and identification of recurring recognition errors. text/csv navez_spacy_off_the_shelf_nlp_sm_results.csv Model predictions generated by the off-the-shelf fr_core_news_sm model using the complete spaCy NLP pipeline (Workflow 2). Contains predicted entities, spans, labels and corresponding gold-standard annotations for the held-out test set. text/csv project-237-at-2025-05-25-08-36-f874dc2a-2.json Part 1 - JSON export from Label Studio containing manually annotated nineteenth-century French correspondence from the Navez Project. Includes entity annotations, transcriptions and project metadata. Used as source data for preprocessing and model training. application/json project-238-at-2025-05-25-08-34-694e26ec-2.json Part 2 - JSON export from Label Studio containing manually annotated nineteenth-century French correspondence from the Navez Project. Includes entity annotations, transcriptions and project metadata. Used as source data for preprocessing and model training. application/json split_entity_counts.csv Summary of the number of annotated entities per entity type in the training, development and test datasets. Used to document the composition of the gold-standard corpus and the experimental data splits. text/csv test_set-2.csv Held-out test split used exclusively for final evaluation of the trained NER models. text/csv test_set-2.pkl The file contains manually annotated letter transcriptions and their token-level entity labels. For each letter, it includes the manuscript identifier, full transcription, tokenized text, annotated entities and entity types, character and token spans, remapped annotations, and IOB tags. application/octet-stream training_set-2.csv Held-out test partition of the gold-standard corpus used exclusively for final evaluation of the NER models after training. text/csv training_set-2.pkl Python Pickle version of the held-out test dataset used in the evaluation pipeline. application/octet-stream