Dataset origin: https://zenodo.org/records/3383658

Automatic TEI encoding of manuscripts catalogues with GROBID-Dictionaries

Credits

GROBID dictionaries is developed by Mohamed Khemakhem (GitHub).

More info on GROBID technologies can be found here.

Corpus

The corpus is based on scans of catalogues published by Auguste Laverdet, the Revue des autographes, directed by Gabriel Charavay (data.bnf) and different auction sales catalogues published by the Maison Charavay.

Methodology

PDF are OCRised with Transkribus.

The GROBID dictionaries model is trained on four different types of catalogues. For each type, 15 annotated pages are available : 10 as training data and 5 as evaluation data.

The training data are extracted from the following catalogs issues:

Gabriel Charavay, Revue des Autographes, first series : 25, 35, 50, 80.
Gabrielle Charavay Revue des Autographes, second series : 24, 56.
Auguste Laverdet, Catalogue de lettres autographes et manuscrits : 1, 22.
Etienne Charavay, Catalogue d’une intéressante collection de lettres autographes… (14 décembre 1908).

The source of the training files are references in the file names using the following acronyms:

RDA_1 for Revue des Autographes, first series
RDA_2 for Revue des Autographes, second series
LAV for catalogues written by Auguste Laverdet
ETI for auction catalogues written by Etienne Charavay

In compliance with the PDF are OCRised with recommendations of Grobid⁻Dictionaries, the separators, whether there are composed of one or two punctuation signs, are tagged with the neutral label .

Examples :

        <entry>
            <num>350</num>
            <pc>.</pc>
            <form>
                <name>STANISLAS</name>
                <pc>,</pc>
                <desc>roi de Pologne N. 1677. M. 1766</desc>
            </form>
            <pc>. -</pc>
            <sense>
                <subSense>L. aut. sig. à madame de Boufïlers. Lunéville, 5 janvier 1741. 1 page in-4, 10</subSense>
            </sense>
        </entry>
        <pc>»</pc>
        <entry>

Files

Training data and evaluation data are available in ToyData

Licence

Regarding GROBID, cf. here.

Regarding the corpus: extracted data is CC-BY.