File size: 37,406 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T01:13:13.921806Z"
},
"title": "early printed books by utilizing cross fold training",
"authors": [
{
"first": "Ahmad",
"middle": [
"P"
],
"last": "Tafti",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Ahmadreza",
"middle": [],
"last": "Baghaie",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Mehdi",
"middle": [],
"last": "Assefi",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Hamid",
"middle": [
"R"
],
"last": "Arabnia",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Zeyun",
"middle": [],
"last": "Yu",
"suffix": "",
"affiliation": {},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "The Quechua linguistic family has a limited number of NLP resources, most of them being dedicated to Southern Quechua, whereas the varieties of Central Quechua have, to the best of our knowledge, no specific resources (software, lexicon or corpus). Our work addresses this issue by producing two resources for the Ancash Quechua: a full digital version of a dictionary, and an OCR model adapted to the considered variety. In this paper, we describe the steps towards this goal: we first measure performances of existing models for the task of digitising a Quechua dictionary, then adapt a model for the Ancash variety, and finally create a reliable resource for NLP in XML-TEI format. We hope that this work will be a basis for initiating NLP projects for Central Quechua, and that it will encourage digitisation initiatives for under-resourced languages.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "The Quechua linguistic family has a limited number of NLP resources, most of them being dedicated to Southern Quechua, whereas the varieties of Central Quechua have, to the best of our knowledge, no specific resources (software, lexicon or corpus). Our work addresses this issue by producing two resources for the Ancash Quechua: a full digital version of a dictionary, and an OCR model adapted to the considered variety. In this paper, we describe the steps towards this goal: we first measure performances of existing models for the task of digitising a Quechua dictionary, then adapt a model for the Ancash variety, and finally create a reliable resource for NLP in XML-TEI format. We hope that this work will be a basis for initiating NLP projects for Central Quechua, and that it will encourage digitisation initiatives for under-resourced languages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In recent years, Quechua has become more visible in the countries where it is spoken, partly as a result of measures to strengthen its use in institutions, but also of a growing interest in these lan- the Quechua-Spanish dictionary (Menacho L\u00f3pez, 2005) , published by the Ministry of Education, which contains 971 entries and can be queried through the online platform Qichwa 2.0 3 .",
"cite_spans": [
{
"start": 241,
"end": 253,
"text": "L\u00f3pez, 2005)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "An online cross-dialectal lexicon (Jacobs, 2006) , featuring about 1,800 entries for Ancash, is downloadable in spreadsheet format. This format can be easily used for NLP, but the lexicon contains some redundancies, discrepancies and formatting irregularities.",
"cite_spans": [
{
"start": 34,
"end": 48,
"text": "(Jacobs, 2006)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The largest Ancash Quechua-Spanish dictionaries are either not officially digitised or have been published under restrictive copyright that prevent their use for NLP purposes. The main dictionaries for our variety are: Swisshelm, 1972 , 399 pages, Parker et al., 1976 Carranza Romero, 2013 , about 8,000 entries, also available as an ebook.",
"cite_spans": [
{
"start": 219,
"end": 234,
"text": "Swisshelm, 1972",
"ref_id": null
},
{
"start": 235,
"end": 267,
"text": ", 399 pages, Parker et al., 1976",
"ref_id": null
},
{
"start": 268,
"end": 289,
"text": "Carranza Romero, 2013",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The main corpus is in paper format only. It consists of two volumes of narratives in both Quechua and Spanish (Cuentos y relatos en el Quechua de Huaraz, Ramos and Ripkens, 1974) , with a total of 698 pages. A digitised dictionary would be useful to automatically post-edit the OCR of this corpus (Poncelas et al., 2020) .",
"cite_spans": [
{
"start": 154,
"end": 178,
"text": "Ramos and Ripkens, 1974)",
"ref_id": "BIBREF10"
},
{
"start": 297,
"end": 320,
"text": "(Poncelas et al., 2020)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Corpora",
"sec_num": "2.1.2"
},
{
"text": "The importance of digitising lexical resources for under-resourced languages has been repeatedly expressed. For the languages of the Americas, two projects are particularly similar to ours.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "A off-the-shelf use of Tesseract is reported (Maxwell and Bills, 2017) (Parker, 1975) . It is an unpublished draft 137 of the Ancash Quechua to Spanish dictionary men-138 tioned above (Parker et al., 1976) This book is a (Budin et al., 2012) . The markup structure is built with the following rules :",
"cite_spans": [
{
"start": 45,
"end": 70,
"text": "(Maxwell and Bills, 2017)",
"ref_id": "BIBREF5"
},
{
"start": 71,
"end": 85,
"text": "(Parker, 1975)",
"ref_id": "BIBREF7"
},
{
"start": 184,
"end": 205,
"text": "(Parker et al., 1976)",
"ref_id": "BIBREF8"
},
{
"start": 221,
"end": 241,
"text": "(Budin et al., 2012)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "\u2022 Spanish loans, marked in the dictionary by an asterisk before the word, are indicated by insertion of the tag <etym>;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "\u2022 Homographs are grouped in a <superEntry>;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "\u2022 Cross-references are marked with <xr>;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "\u2022 Easily retrievable examples within the column corresponding to the translation or gloss are tagged with <cit>.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "Our XML-TEI lexicon contains 3626 entries, and is to date the largest digital resource for Ancash Quechua available for NLP and lexicometry.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "OCR of dictionaries",
"sec_num": "2.2"
},
{
"text": "The present work shows that it is relatively easy to train a new Tesseract model from an existing one, with very little data. The tests carried out on several OCRs show many that alternatives are available for this task depending on the desired output.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "Based on this work, we started the digitisation of a second dictionary and a corpus with the same characteristics.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "As the OSCAR corpus https://oscar-corpus. com",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "According to the 2017 census.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": ". Cropping: cutting the file to eliminate every-",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/tesseract-ocr/ tesseract",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "trast;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": ". Conversion to high definition PNG (between",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "and 390 dpi).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://gegl.org/ 6 https://imagemagick.org/script/ convert.php",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/impactcentre/ ocrevalUAtion",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/rumiwarmi/qishwar/ blob/main/Diccionario%20polilectal%20-% 20PARKER.ods 9 https://tei-c.org/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We warmly thank C\u00e9sar Itier for providing his copy of Parker's dictionary and allowing us to digitise it.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
},
{
"text": "The last two steps are automatically applied to the whole document thanks to a bash script, using These pages are segmented by lines with the",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "annex",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Creating lexical resources in tei p5. a schema for multi-purpose digital dictionaries",
"authors": [
{
"first": "Gerhard",
"middle": [],
"last": "Budin",
"suffix": ""
},
{
"first": "Stefan",
"middle": [],
"last": "Majewski",
"suffix": ""
},
{
"first": "Karlheinz",
"middle": [],
"last": "M\u00f6rth",
"suffix": ""
}
],
"year": 2012,
"venue": "Journal of the Text Encoding Initiative",
"volume": "",
"issue": "3",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gerhard Budin, Stefan Majewski, and Karlheinz M\u00f6rth. 2012. Creating lexical resources in tei p5. a schema for multi-purpose digital dictionaries. Journal of the Text Encoding Initiative, (3).",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Diccionario del Quechua Ancashino",
"authors": [
{
"first": "Francisco",
"middle": [],
"last": "Carranza Romero",
"suffix": ""
}
],
"year": 2013,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Francisco Carranza Romero. 2013. Diccionario del Quechua Ancashino.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Training & quality assessment of an optical character recognition model for Northern Haida",
"authors": [
{
"first": "Isabell",
"middle": [],
"last": "Hubert",
"suffix": ""
},
{
"first": "Antti",
"middle": [],
"last": "Arppe",
"suffix": ""
},
{
"first": "Jordan",
"middle": [],
"last": "Lachler",
"suffix": ""
},
{
"first": "Eddie",
"middle": [
"Antonio"
],
"last": "Santos",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
"volume": "",
"issue": "",
"pages": "3227--3234",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isabell Hubert, Antti Arppe, Jordan Lachler, and Ed- die Antonio Santos. 2016. Training & quality as- sessment of an optical character recognition model for Northern Haida. In Proceedings of the Tenth In- ternational Conference on Language Resources and Evaluation (LREC'16), pages 3227-3234, Portoro\u017e, Slovenia. European Language Resources Associa- tion (ELRA).",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Metrics for complete evaluation of ocr performance",
"authors": [
{
"first": "Romain",
"middle": [],
"last": "Karpinski",
"suffix": ""
},
{
"first": "Devashish",
"middle": [],
"last": "Lohani",
"suffix": ""
}
],
"year": 2018,
"venue": "IPCV'18-The 22nd Int'l Conf on Image Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Romain Karpinski, Devashish Lohani, and Abdel Be- laid. 2018. Metrics for complete evaluation of ocr performance. In IPCV'18-The 22nd Int'l Conf on Image Processing, Computer Vision, & Pattern Recognition.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Endangered data for endangered languages: Digitizing print dictionaries",
"authors": [
{
"first": "Michael",
"middle": [],
"last": "Maxwell",
"suffix": ""
},
{
"first": "Aric",
"middle": [],
"last": "Bills",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 2nd Workshop on the Use of Computational Methods in the Study of Endangered Languages",
"volume": "",
"issue": "",
"pages": "85--91",
"other_ids": {
"DOI": [
"10.18653/v1/W17-0112"
]
},
"num": null,
"urls": [],
"raw_text": "Michael Maxwell and Aric Bills. 2017. Endangered data for endangered languages: Digitizing print dic- tionaries. In Proceedings of the 2nd Workshop on the Use of Computational Methods in the Study of Endangered Languages, pages 85-91, Honolulu. As- sociation for Computational Linguistics.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Yachakuqkunapa Shimi Qullqa",
"authors": [
{
"first": "Leonel Alexander Menacho",
"middle": [],
"last": "L\u00f3pez",
"suffix": ""
}
],
"year": 2005,
"venue": "Anqash Qichwa Shimichaw. Ministerio de Educaci\u00f3n",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Leonel Alexander Menacho L\u00f3pez. 2005. Yachakuqku- napa Shimi Qullqa, Anqash Qichwa Shimichaw. Ministerio de Educaci\u00f3n, Lima, Per\u00fa.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Diccionario Polilectal del Quechua de Ancash",
"authors": [
{
"first": "Gary",
"middle": [
"J"
],
"last": "Parker",
"suffix": ""
}
],
"year": 1975,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gary J. Parker. 1975. Diccionario Polilectal del Quechua de Ancash. Universidad Nacional Mayor de San Marcos.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Diccionario quechua",
"authors": [
{
"first": "G",
"middle": [
"J"
],
"last": "Parker",
"suffix": ""
},
{
"first": "A",
"middle": [
"C"
],
"last": "Reyes",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Ch\u00e1vez",
"suffix": ""
}
],
"year": 1976,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G.J. Parker, A.C. Reyes, and A. Ch\u00e1vez. 1976. Dic- cionario quechua, Ancash-Huailas. Ministerio de Educaci\u00f3n.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "James Hadley, and Andy Way. 2020. A tool for facilitating ocr postediting in historical documents",
"authors": [
{
"first": "Alberto",
"middle": [],
"last": "Poncelas",
"suffix": ""
},
{
"first": "Mohammad",
"middle": [],
"last": "Aboomar",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2004.11471"
]
},
"num": null,
"urls": [],
"raw_text": "Alberto Poncelas, Mohammad Aboomar, Jan Buts, James Hadley, and Andy Way. 2020. A tool for facilitating ocr postediting in historical documents. arXiv preprint arXiv:2004.11471.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Cuentos y relatos en el quechua de Huaraz. Estudios culturales benedictinos",
"authors": [
{
"first": "S",
"middle": [
"P"
],
"last": "Ramos",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Ripkens",
"suffix": ""
}
],
"year": 1974,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S.P. Ramos and J. Ripkens. 1974. Cuentos y relatos en el quechua de Huaraz. Estudios culturales bene- dictinos.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"uris": null,
"text": "guages as a cultural element among citizens. At the same time, Quechua languages are gradually handled by NLP software. For Southern Quechua (variety of Quechua II, the most widespread linguistic family), resources already exist and many projects are experimenting large corpus digitisation to create Deep Learning models 1 . However, Quechua varieties are heterogeneous and available resources for the aforementioned variety are hardly usable for others, because of important differences in both morphology and lexicon. The present work aims at laying foundations for the development of NLP tools for another variety, the Ancash Quechua (variety of Quechua I).",
"num": null
},
"FIGREF1": {
"type_str": "figure",
"uris": null,
"text": "139list of Ancash lexemes along with their area of140 use (division by province), their POS, translation 141 or gloss in Spanish, and a set of internal cross-142 references indicating synonyms, related terms or 143 lectal variants. The overall structure is relatively 144 homogeneous. The elements mentioned above are 145 separated by blanks, but are not vertically aligned.146The typography is that of the old typewriters; some 147 typing errors remain in the document.",
"num": null
},
"FIGREF2": {
"type_str": "figure",
"uris": null,
"text": "written using Latin script. In 150 the particular case of our document, the author 151 used a phonemic spelling to represent charac-152 ters whose official modern spelling is a digraph. 153 The",
"num": null
},
"TABREF0": {
"text": "Hubert et al., 2016) for a large written corpus (100,000 words). Optimal settings discovery was conducted by training 12 models with distinct parameters. This work also experimented training the model with images generated from text using a font similar to the targeted documents, which did3 https://dic.qichwa.net/#/ not prove to be efficient. The best model, trained 123 on the original source, obtained 96.47% character 124 rate accuracy (CRA) and a 89.03% word rate accu-125 racy (WRA). Quechua in OCR tools 127 Both Tesseract 4 and ABBYY include a pretrained 128 Quechua model for OCR. ABBYY's model is 129 trained on Bolivian Quechua (Q.II). The training 130 corpus for Tesseract's model is not documented.",
"html": null,
"type_str": "table",
"content": "<table><tr><td>to digitise 3 bilingual dictionaries (Tzeltal-English, Muinane-Spanish, Cubeo-Spanish). More specifically, authors used Tesseract's hOCR function to preserve entry's structure and infer lexical entries with associated linguistic information. A finite state transducer was used to create the lexicon from this hOCR file. Tesseract can also be (re)trained to create ded-icated models. This has been experimented for an almost extinct Canadian language (Northern Haida) (126 2.3 131 3 OCR of the Ancash Quechua 132 Dictionary 133 3.1 Source Document 134 The document we digitised is a working document</td></tr></table>",
"num": null
},
"TABREF1": {
"text": "shows the special characters used 154 by Parker (in first column), their corresponding 155 phonemes, and the graphemes commonly used to-",
"html": null,
"type_str": "table",
"content": "<table><tr><td>156</td></tr></table>",
"num": null
},
"TABREF2": {
"text": "Special characters in the dictionary 2. Conversion to greyscale and increasing congegl 5 and convert 6 commands.",
"html": null,
"type_str": "table",
"content": "<table><tr><td>173</td><td/><td/></tr><tr><td/><td>3.5 OCR selection</td><td/></tr><tr><td>177</td><td/><td/></tr><tr><td>178</td><td colspan=\"3\">performing OCR software (Tafti et al., 2016), and</td></tr><tr><td>179</td><td colspan=\"3\">compared their output on a set of 5 pages of the</td></tr><tr><td>180</td><td colspan=\"3\">document, randomly extracted. For Tesseract's</td></tr><tr><td/><td colspan=\"3\">OCR, we used both Quechua and Spanish pre-</td></tr><tr><td/><td colspan=\"3\">trained models. For ABBYY's OCR, we used the</td></tr><tr><td>183</td><td colspan=\"3\">Bolivian Quechua model. GoogleDocs OCR does</td></tr><tr><td>184</td><td colspan=\"3\">not allow to control any parameter. Table 2 shows</td></tr><tr><td/><td colspan=\"2\">the error rates for each of them.</td></tr><tr><td/><td colspan=\"3\">Tesseract ABBYY GoogleDocs</td></tr><tr><td/><td>CER 6.64</td><td>6.43</td><td>5.26</td></tr><tr><td/><td>WER 25.5</td><td>27.5</td><td>20.7</td></tr></table>",
"num": null
},
"TABREF3": {
"text": "OCR comparison on our dictionaryThis evaluation shows that GoogleDocs OCR is the struc-ture of the document is not preserved. The opposite situation occurs in the case of ABBYY. It is worth noting that the output of the latter could be greatly improved by using the numerous settings the software offers.",
"html": null,
"type_str": "table",
"content": "<table><tr><td colspan=\"2\">In addition to performances, we also took</td></tr><tr><td colspan=\"2\">in consideration the possibility to distribute the</td></tr><tr><td colspan=\"2\">trained model with an open licence. According</td></tr><tr><td colspan=\"2\">to these considerations, we chose Tesseract, which</td></tr><tr><td colspan=\"2\">gives satisfying results and allows the model to be</td></tr><tr><td>shared.</td><td/></tr><tr><td colspan=\"2\">3.5.1 Preliminary tests with Tesseract OCR</td></tr><tr><td colspan=\"2\">In order to have a better view of Tesseract's per-</td></tr><tr><td colspan=\"2\">formance, we applied OCR on 10 PNG files,</td></tr><tr><td colspan=\"2\">randomly extracted from the pre-processed (Sec-</td></tr><tr><td colspan=\"2\">tion 3.4) document, using: Spanish model alone</td></tr><tr><td colspan=\"2\">(spa FAST); Quechua model alone (que FAST);</td></tr><tr><td colspan=\"2\">Spanish and Quechua models together (que+spa)</td></tr><tr><td colspan=\"2\">in their compressed (FAST) and uncompressed</td></tr><tr><td>(BEST) versions.</td><td/></tr><tr><td colspan=\"2\">OCR outputs per page are concatenated into</td></tr><tr><td colspan=\"2\">a single file, as well as corresponding gold stan-</td></tr><tr><td colspan=\"2\">dards. Resulting files are compared to measure</td></tr><tr><td colspan=\"2\">Character Error Rate (CER) and Word Error Rate</td></tr><tr><td colspan=\"2\">(WER) with the ocrevalUAtion tool 7 . Table 3 re-</td></tr><tr><td>ports those evaluations.</td><td/></tr><tr><td colspan=\"2\">The results show that OCR performance is rel-</td></tr><tr><td colspan=\"2\">atively poor, with a word recognition accuracy</td></tr><tr><td colspan=\"2\">(WRA) of less than 80%. The characters with di-</td></tr><tr><td colspan=\"2\">acritics presented in Section 3.2, which are absent</td></tr><tr><td colspan=\"2\">from the character set of Quechua and Spanish</td></tr><tr><td colspan=\"2\">models, are not recognised, making manual cor-</td></tr><tr><td colspan=\"2\">rection tedious. However, Tesseract offers the pos-</td></tr><tr><td colspan=\"2\">sibility to adapt a pre-trained model to additional</td></tr><tr><td colspan=\"2\">fonts and characters. In the next section, we de-</td></tr><tr><td colspan=\"2\">scribe the training of a model specific to our book,</td></tr><tr><td colspan=\"2\">based on Tesseract's Quechua model.</td></tr><tr><td/><td>CER WER</td></tr><tr><td>spa FAST</td><td>6.23 23.20</td></tr><tr><td>que FAST</td><td>7.40 27.55</td></tr><tr><td colspan=\"2\">que+spa FAST 5.89 21.82</td></tr><tr><td colspan=\"2\">que+spa BEST 6.05 21.29</td></tr></table>",
"num": null
},
"TABREF4": {
"text": "Tesseract performance with pretrained models ; each segment is then OCRised and the output is manually corrected to constitute the gold 232 standard. The training process is done from the",
"html": null,
"type_str": "table",
"content": "<table><tr><td>3.6 Model training</td></tr><tr><td>A training corpus was built from 30 pages of</td></tr><tr><td>the document (5,676 words, 33,687 characters).</td></tr></table>",
"num": null
},
"TABREF5": {
"text": "CER and WER of the Ancash Quechua modelTo get a better idea of the impact of the training,",
"html": null,
"type_str": "table",
"content": "<table><tr><td>254</td><td/><td/><td/></tr><tr><td>255</td><td colspan=\"4\">we also evaluated the error rate on the characters</td></tr><tr><td>256</td><td colspan=\"4\">with diacritics. Table 5 shows their volume in the</td></tr><tr><td>257</td><td colspan=\"4\">training corpus (\u016b,\u0159 and\u0109 having only one or two</td></tr><tr><td>258</td><td colspan=\"4\">occurrences, they are considered negligible) and</td></tr><tr><td/><td colspan=\"3\">corresponding error rates.</td></tr><tr><td/><td colspan=\"4\">Nb train Vol train (%) CER (%)</td></tr><tr><td/><td>\u0161</td><td>167</td><td>0.50</td><td>4.04</td></tr><tr><td/><td>\u00eb</td><td>157</td><td>0.47</td><td>7.83</td></tr><tr><td/><td>c</td><td>148</td><td>0.44</td><td>5.11</td></tr><tr><td/><td>a</td><td>130</td><td>0.39</td><td>63.7</td></tr><tr><td/><td>i,\u0113,\u014d</td><td/><td><0,1</td><td>100</td></tr></table>",
"num": null
},
"TABREF6": {
"text": "",
"html": null,
"type_str": "table",
"content": "<table><tr><td/><td>: Training volume and CER for special charac-</td></tr><tr><td/><td>ters</td></tr><tr><td>260</td><td>Empirically, manual correction of OCR output</td></tr><tr><td>261</td><td>is easier with the new model: the most frequent</td></tr></table>",
"num": null
}
}
}
} |