INCEpTION - Digital Athenaeus

Digital Athenaeus

Named Entity Annotation for Ancient Greek

Source: This use-case was kindly contributed by Monica Berti, Department of Digital Humanities, Universität Leipzig, Germany.

The Digital Athenaeus project is focused on annotating quotations and text reuses in the Greek text of the Deipnosophists of Athenaeus of Naucratis (2nd-3rd century CE). The goal is to provide an inventory of authors and works cited by Athenaeus and implement a data model for identifying, analyzing, and citing uniquely instances of text reuse in ancient Greek literature.

INCEpTION is used by the Digital Athenaeus project to extract and annotate named entities in the Greek text of the Deipnosophists, which is a rich collection of proper names pertaining to a wide variety of typologies like personal names, peoples, places, groups, languages, festivals, astronomical and meteorological phenomena, chronological data and currencies. Athenaeus’ work is also a huge mine of references to more than 900 authors of Classical literature and their writings.

NER has been performed semi-automatically on the text of the Teubner edition of the Deipnosophists by Georg Kaibel, which contains 264,750 tokens distributed in 15 books for a total of 1,328 paragraphs and 21,460 sentences. The result is the extraction of ca. 23,000 inflected forms of single named entities corresponding to more than 8,000 unique forms and more than 4,400 lemmata. Lemmata have been used to query external authority lists to disambiguate annotations. Data is stored in an SQL database, whose entries can be publicly interrogated in the Named Entities Digger and in the Named Entities Concordance of the Digital Athenaeus project.

INCEpTION is currently used to visualize, correct, and nest annotations of single ancient Greek named entities that have been semi-automatically extracted. Data has been imported into INCEpTION as TSV files generated according to the WebAnno TSV 3.2 file format. Each file includes the text of single paragraphs of the Deipnosophists with sentences split in separate lines. Figure 1 shows an example of a TSV file with annotated named entities and corresponding lemmata (Ath., Deipn. 1.7):

my alt text
Figure 1. TSV 3.2 file format (NEs and lemmata in Ath., Deipn. 1.7)

Figure 2 shows how pre-annotated data is visualized in INCEpTION in separate numbered lines. Single entities are annotated with two layers: one for the named entity tag, and the other for the lemma (Ath., Deipn. 1.7):

my alt text
Figure 2. INCEpTION: pre-annotated data (Ath., Deipn. 1.7)

A layer called Ancient Greek Catalog has been created in INCEpTION to annotate named entities of the Deipnosophists that correspond to names of ancient authors and to descriptions/titles of ancient works. The reason for this layer is to disambiguate names of ancient authors and titles of works in order to produce a text-based catalog of Greek literature with annotations of ancient Greek inflected forms and their corresponding lemmata.

Figure 3 shows the same paragraph of figures 1 and 2 with the Ancient Greek Catalog layer, whose values correspond to CTS URNS that uniquely identify authors and works. Also, individual entities have been linked together in spans corresponding to real entities, as for example Ἀρχέστρατος ὁ Συρακούσιος ἣ Γελῷος (Archestratus from Gela or Syracuse), who is identified as urn:cts:greekLit:tlg1175. In the same line, different forms of the title of Archestratus’ work are identified with a CTS URN that also includes a reference to the author: urn:cts:greekLit:tlg1175.tlg0020.

my alt text
Figure 3. INCEpTION: Ancient Greek Catalog layer (Ath., Deipn. 1.7)

Future work will include complete NEs disambiguation and linking, and coreference resolution with a focus on entities related to ancient Greek authors and works.

The final goal is to use INCEpTION for linking entity mentions to knowledge bases and structured vocabularies for ancient Greek authors and works that will enable scholars to annotate other texts and generate a text-based catalog of ancient Greek literature.