Version classiqueVersion mobile

Proceedings of the Seventh Italian Conference on Computational Linguistics CLiC-it 2020

Felice Dell'Orletta
Johanna Monti
Fabio Tamburini

Contributed Papers

Græcissāre: Ancient Greek Loanwords in the LiLa Knowledge Base of Linguistic Resources for Latin

Greta Franzini, Marco Passarotti, Francesco Mambrini, Giovanni Moretti et Federica Zampedri


This paper describes the addition of an index of 1,763 Ancient Greek loanwords to the collection of Latin lemmas of the LiLa: Linking Latin Knowledge Base of interoperable linguistic resources. This lexical resource increases LiLa’s lemma count and tunes its underlying data model to etymological borrowing.

Texte intégral

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme - Grant Agreement No 769994.

1. Introduction

  • 1 “Captive Greece captured her savage conqueror” (our translation).

“Graecia capta ferum victorem cepit”1
Horace, Epistles, II, 1, 156

  • 2 In Egypt, for instance.

1Boasting over two thousand years’ worth of written attestation, Latin’s evolutionary history is among the longest in existence. The diachronic and geographical reach of the Roman Empire exposed Latin, an Indo-European Italic language, to many regional dialects and languages, including Ancient Greek. The mutually profitable linguistic contact between Latin and Ancient Greek2, facilitated by their similar morphosyntactic structures and characteristic syntheticity (Ledgeway 2012, 10–28), is most evident in their vocabulary, chiefly calques and loanwords. Both lexemes presuppose a certain knowledge of the donor language, but while the former takes from the donor with translation, the latter does not (Hock and Joseph 2009, 252).

2Examples of Latin words calqued from Ancient Greek are unicornuus “unicorn” (unus “one” + cornu “horn”) from μονόκερως (μόνος “one” + κέρας “horn”), and infans “infant” (in- “not” + fans “speaking”) from νήπιος (negative prefix νη- + > ἔπος “speech”). Calques can also involve affixes, as is the case of Latin’s suffix -us being substituted for the Greek -os (Hock and Joseph 2009, 253). The adjective “dramatic”, for instance, is attested as both dramaticos and dramaticus.

3Example loanwords in Latin are crocodilus “crocodile”, imported from the Ancient Greek κροκόδειλος, and liquiritia “liquorice” from γλυκύρριζα. Adams identifies three categories of Greek loans in Latin :

  1. words for which there existed a Latin equivalent; the writer was so familiar with the local Greek term that he adopted it in response to local conditions;

  2. local Greek technical terms for which it might have been difficult to find a Latin equivalent; and

  3. transfers determined by a writer’s lack of fluency in Latin, as a result of which he either adopted Greek words because he was unaware of their Latin equivalents, or did so unconsciously because of his poor command of Latin.

4For each category, Adams provides a handful of examples, including (1) (h)amaxa from < ἅμαξα “wagon”, (2) buneurum from βούνευρον “whip of oxhide” and (3) arura from > ἄρουρα “land”.

5Over the course of its long history, Latin lexicography has produced a plethora of lexical resources, notably dictionaries, thesauri and lexica. Many are available in machine-readable form but their differing annotation schemes and formats are seldom interoperable. In an effort to offset the issue, the LiLa: Linking Latin project is leveraging Linked Data technology to dovetail a wide range of Latin resources into an interoperable whole, producing an ever-growing lexically-based data model capable of accommodating etymological, morphological, syntactic and semantic information, and more besides (Passarotti et al. 2020)3. In LiLa, glossaries, lexica, treebanks, textual resources and tools intersect and interact through their common denominator, the lemma (itself, incidentally, a loanword from the Ancient Greek λῆμμα). Indeed, the LiLa Knowledge Base hinges on a lemma bank of approximately 130,000 lemmas largely derived from the lexical basis of LEMLAT (Passarotti et al. 2017). As textual and lexical resources are added to the Knowledge Base, LiLa’s lemma bank and coverage of the Latin lexicon grow in size.

6Though chiefly targeting readily available lemmatised resources on the web, LiLa also creates linguistic resources in-house as a means of further developing its underlying data model. Examples of these are the Index Thomisticus Treebank (Passarotti 2019) and Latin VALLEX (Passarotti, González Saavedra, and Onambele 2016). Here, we describe the addition of a new homegrown lexical resource, the Index Graecorum Vocabulorum in Linguam Latinam (Saalfeld 1874), to the LiLa Knowledge Base of Linguistic Resources for Latin.

2. Data and Methodology

  • 4 The Latin verb græcissō used in the title of this paper is a nativised version of the Greek graikiz (...)

7Etymological data is not new to LiLa. Mambrini and Passarotti describe the inclusion of 1,391 entries from the Etymological Dictionary of Latin and the other Italic Languages (De Vaan 2008) modelled against the lemonETY etymological extension (Khan 2018) of the OntoLex Lexicon Model for Ontologies (lemon) (McCrae et al. 2017), which have provided LiLa with 1,465 Proto-Italic and 1,393 Proto-Indo-European reconstructed forms. Whereas those entries came to Latin via inheritance, the work described here targets (nativised) loans from Ancient Greek4.

  • 5 Crude estimate of an average ten to fifteen entries per page, for a total 592 pages.

8The Index Graecorum vocabulorum in linguam Latinam translatorum quaestiunculis auctus (hereafter IGVLL) is a list of 1,763 Ancient Greek loanwords in the Latin language published in 1874 by classical scholar Günther Alexander E. A. Saalfeld. An extended edition of the Index, published in 1884 as Tensaurus Italograecus: Ausfürliches historisch-kritisches Wörterbuch der Griechiscen Lehn- und Fremdwörter im Latenischen, is the most comprehensive lexicographic collection of its kind, counting roughly six to eight thousand entries (Saalfeld 1884)5.

9Of the two, the size and Optical Character Recognition (OCR) quality of the 1874 edition best suited a first development of a derivative linguistic resource, conducted as part of a Master’s internship at the CIRCSE Research Centre in Milan6.

  • 7 The specific uncertainty remains unclear as no editorial documentation is provided.

10IGVLL is structured into three columns of information: the Latin loanword (occasionally accompanied by variants), the Ancient Greek source lemma(s) (multiple lemmas include graphical, morphological and dialectal variants), and a record of attestations (see Figure 1). Explanatory notes at the bottom of the page provide additional context. In thirteen cases, question marks indicate some level of uncertainty7, and, as is convention, asterisks are used to identify thirty-nine unattested –and thus reconstructed– Ancient Greek forms.

Figure 1: Three lexical entries in IGVLL, translating to “Bear’s Foot (plant)”, “without smoke”, and “acatalectic (line of verse)”, respectively

Figure 1: Three lexical entries in IGVLL, translating to “Bear’s Foot (plant)”, “without smoke”, and “acatalectic (line of verse)”, respectively

2.1 Data Preparation

11Judging by the illegible Greek, the engine used to produce the OCR’d text available from Internet Archive (ABBYY FineReader 8.0) was set to recognise the Latin alphabet only (see Figure 2).

Figure 2: Latin OCR of Fig. 1, Internet Archive

Figure 2: Latin OCR of Fig. 1, Internet Archive

12The OCR quality of the text written in the Latin alphabet, however, was sufficient to automatically isolate and tabulate the Latin lemmas, which were then manually cleaned. Next, this list was automatically mapped against the LiLa lemma bank to measure the degree of lexical overlap, which came up at 1,488 unique matches (84.40%), 207 ambiguous matches (11.74%) and 68 unmatched lemmas (3.85%). Unique matches inherited their respective LiLa identifier, ambiguous matches were manually disambiguated, and unmatched lemmas were added –once again, manually– to the LiLa lemma bank. Ambiguities were caused by homography between lemmas belonging to different categories, be those morphosyntactic (the lemma philosophus, for instance, matched against LiLa’s adjective philosophus “philosophical” and common noun philosophus “philosopher”) or inflectional (the common noun er might refer to the masculine er, eris “hedgehog” or the invariable er (graphical variant of R) “seventeenth letter of the Latin alphabet”). Of the 68 unmatched lemmas, 33 were graphical variants of lemmas already present in LiLa and 35 were new additions.

  • 8 For the most recent overview of Ancient Greek optical character recognition, see Robertson and Bosc (...)

13Next, we OCR’d IGVLL with Tesseract v. 4.1.1 set to Ancient Greek recognition (Smith 2007)8. As Figure 3 shows, contrary to the Latin OCR the noise affecting Greek lemmas required heavy manual intervention for clean tabulation, e.g. the rectification of instances of κ (cappa) misread as χ (chi) or of π (pi) misread as ττ (double tau) and viceversa, missing breathings and incorrect accents, to mention but a few.

Figure 3: Ancient Greek OCR of Fig. 1, Tesseract

Figure 3: Ancient Greek OCR of Fig. 1, Tesseract

14In LiLa, a lemma can have one or more graphical variants, known as “written representations” (e.g. the verb sacrifico “to sacrifice” is also attested as sacrufico), as well as inflectional variants, with which it holds a symmetric “lemma variant” property or relation in the Knowledge Base (the active sacrifico, sacrufico vs. the deponent sacrificor, sacruficor).

15Therefore, for the purposes of LiLa, where the editor provides multiple Ancient Greek lemmas for a single Latin loanword, e.g. burrus “red” πυρρός (πυρσός); cyperum “rush (botany)” κύπειρον (κύπειρος), these were distinguished into written representations of the same lemma (i.e. πυρρόςvs. πυρσός) and lemma variants (i.e. the neuter κύπειρονvs. the masculine κύπειρος).

16Compounds such as authepsa “an urn, boiler”, (αὐτός & ἕψω) were tabulated as two separate words, and entries followed by a question mark (13 in total) were marked as “uncertain”.

2.2 Data Model

17The transformation of IGVLL into an RDF lexicographic resource bound for LiLa relied on a combination of vocabularies. In line with previous etymological work, we integrated the aforementioned lemon and lemonETY modules of OntoLex to represent lexical entries in IGVLL. The example lemma abacus “sideboard” shown in Listing 1 is treated as an ontolex:LexicalEntry linked to LiLa’s own abacus (lemma ID 86829) through the property ontolex:canonicalForm.

18We employed the Simple Knowledge Organization System (SKOS) (Miles and Bechhofer 2009) to point Ancient Greek lemmas to their corresponding canonical forms in a machine-readable version of the Greek-English Liddell-Scott Jones (LSJ) lexicon (Blackwell 2018). As Listing 2 shows, we modelled the Ancient Greek source lemma of abacus, >’abax, as an etymon, which, in the absence of a Linked Data Knowledge Base for Ancient Greek, currently points to a blank node.

19The skos property stores the LSJ identifier of ἄβαξ as an exactMatch to denote an exact correspondence between the Ancient Greek lemma of IGVLL and that of LSJ. Failing an exact match, the property skos:broadMatch is used to indicate that the IGVLL lemma is incorporated in a different entry of LSJ (e.g. the IGVLL noun φυσική “science of nature, physics” does not have its own entry in LSJ but is listed as a nominalised adjective under the adjectival entry φυσικός “natural”); further, failing both exact and broad matches, the property skos:relatedMatch is used to indicate a loose relation between IGVLL and LSJ (e.g. IGVLL’s πορφυρίζον “purple dye pigment”, neuter present participle of πορφυρίζειν, and LSJ’s verb πορφυρίζω “to be purplish”). As LSJ is not currently equipped with a URN resolver, no actionable link can be made between LiLa and LSJ.

20If multiple written representations of a Greek word are listed in the IGVLL, those are all assigned to the canonical form of the related etymon, for instance ontolex:canonicalForm [ontolex:writtenRep “πυρρός”, “πυρσός”].

21In the case of multiple Ancient Greek variant lemmas, these are all treated as individual etyma, with the difference that the primary etymon points to the URI(s) of the other etyma –classed as both lemonEty:etymon and lemonEty:cognate– via the additional property lemonEty:cognate (Listing 3).

22Latin composite words in IGVLL never point to an Ancient Greek compound but to the two constituent lemmas. In contrast, in the LSJ lexicon seven of the total thirteen multi-word lexical entries in IGVLL are traced back to a Greek compound lemma, e.g. authepsa (IGVLL: αὐτός & ἕψω; LSJ: αὐθέψης9). In keeping with the IGVLL, we employed the decomp:subterm property of lemon10 to point the Latin lexical entry to its two constituent Ancient Greek etyma and reconciled these with LSJ using the skos:relatedMatch property.

23The etymology of abacus is expressed with the CIDOC Conceptual Reference Model (CRM) class E89 Propositional Object11 as a borrowing by way of the lemonEty:etyLinkType property. This set-up is also valid for calques, should these become available in future.

24The CRMinf extension of CRM and the Open Vocabulary (Davis 2004) were used to represent uncertainty as a “belief” or confidence value (Stead et al. 2019; Doerr 2003; Mambrini and Passarotti 2020). Specifically, we coded uncertainty as a CRMinf Belief class (crminf:I2) carrying an arbitrary Belief Value (crminf:I6) of 0.5 (Listing 4).

25Additionally, we employed the Dublin CoreTM Metadata Terms vocabulary to supply the resource with descriptive metadata, such as publisher and licence (DCMI 2020).

26All editorial notes in IGVLL were excluded from the data model.

27As previously mentioned, with this development LiLa’s etymological purview now covers both direct inheritance and borrowing. Figure 4, for example, shows all etymological information in the Knowledge Base associated with LiLa’s common noun muscus “moss, musk” (top row, centre node). LiLa’s “muscus” is connected to the “muscus” lexical entries of both IGVLL and the Brill Etymological Dictionary via the bidirectional OntoLex property canonicalForm. These lexical entries point to their respective etyma via the directed lemonETY etymology and etymon properties.

3. Conclusion

28This paper describes the preparation and integration of Saalfeld’s Index Graecorum Vocabulorum in Linguam Latinam (1874) in the LiLa Knowledge Base of Linguistic Resources for Latin. This first list of 1,763 Latin loans from Ancient Greek adds 68 new Latin lemmas to LiLa, stretches its data model to include borrowing and has been mapped to the digitised Greek-English Liddell-Scott-Jones lexicon. Beyond LiLa, this linguistic resource might be integrated in other resources, such as dictionaries (Bowers and Romary 2016) or digital scholarly editions. Future improvements might acquire a list of calques (Detreville 2015; Fruyt 2011) and the extended edition of Saalfeld’s Index (1884).

29The data and code for the project are available at:​CIRCSE/​index-graecorum-vocabulorum.

Marco Passarotti, Berta González Saavedra, and Christophe Onambele. 2016. Latin Vallex. A Treebank-based Semantic Valency Lexicon for Latin. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016), pages 2599–2606. https://www.

Marco Passarotti. 2019. The Project of the Index Thomisticus Treebank. In Monica Berti, editor, Digital Classical Philology. Ancient Greek and Latin in the Digital Revolution, volume 10 of Age of Access? Grundfragen der Informationsgesellschaft, pages 299–319, Berlin-Boston. De Gruyter GmbH.​10. 1515/9783110599572.

Bruce Roberston and Federico Boschetti. 2017. Large-Scale Optical Character Recognition of Ancient Greek. Mouseion, 14:341–359.​10.3138/​mous.14.3-3.


James Noel Adams. 2003. Bilingualism and the Latin Language. Cambridge University Press.

Christopher W. Blackwell. 2018. “CITE LSJ Browser, v. 1.4.0.”

Jack Bowers and Laurent Romary. 2016. “Deep Encoding of Etymological Information in TEI.” Journal of the Text Encoding Initiative, no. 10.

Ian Davis. 2004. “, V. 20040205.”

Usage Board DCMI. 2020. “DCMI Metadata Terms.”

Eleanor Detreville. 2015. “An Overview of Latin Morphological Calques on Greek Technical Terms: Formation and Success.” Master’s thesis, Asheville, NC: University of North Carolina.

Michiel De Vaan. 2008. Etymological Dictionary of Latin: And the other Italic Languages. Vol. 7. Leiden Indo-European Etymological Dictionary Series. Amsterdam: Brill.

Martin Doerr. 2003. “The CIDOC conceptual reference module: an ontological approach to semantic interoperability of metadata.” AI Magazine 24 (3): 75–75.

Michèle Fruyt. 2011. “Latin Vocabulary.” In A Companion to the Latin Language, edited by James Clackson, 144–56. Wiley-Blackwell.

Hans Henrich Hock and Brian D. Joseph. 2009. Language History, Language Change, and Language Relationship: An Introduction to Historical and Comparative Linguistics. 2nd revised. Trends in Linguistics. Studies and monographs, 218. Mouton de Gruyter.

Anas Fahad Khan. 2018. “Towards the Representation of Etymological Data on the Semantic Web.” Information 9 (12): 304–20.

Adam Ledgeway. 2012. From Latin to Romance. Morphosyntactic Typology and Change. Oxford: Oxford University Press.

Francesco Mambrini and Marco Passarotti. 2020. “Representing Etymology in the LiLa Knowledge Base of Linguistic Resources for Latin.” In Proceedings of the Globalex Workshop on Linked Lexicography (LREC 2020), edited by Ilan Kernerman, Simon Krek, John P. McCrae, Jorge Gracia, Sina Ahmadi, and Besim Kabashi, 20–28. Paris, France: European Language Resources Association (ELRA).

John P. McCrae, Julia Bosque-Gil, Jorge Gracia, Paul Buitelaar, and Philipp Cimiano. 2017. “The Ontolex-Lemon model: development and applications.” In Proceedings of the Electronic Lexicography in the 21st Century Conference (eLex 2017), 237–51.

Alistair Miles and Sean Bechhofer. 2009. “SKOS Simple Knowledge Organization System Reference.” W3C Recommendation 18.

Alistair Passarotti. 2019. “The Project of the Index Thomisticus Treebank.” In Digital Classical Philology. Ancient Greek and Latin in the Digital Revolution, edited by Monica Berti, 10:299–319. Age of Access? Grundfragen Der Informationsgesellschaft. Berlin-Boston: De Gruyter GmbH. https://www.w3. org/TR/skos-reference/#L4858

Marco Passarotti, Marco Budassi, Eleonora Litta, and Paolo Ruffolo. 2017. “The Lemlat 3.0 Package for Morphological Analysis of Latin.” In Proceedings of the Nodalida 2017 Workshop on Processing Historical Language, 24–31.

Marco Passarotti, Berta González Saavedra, and Christophe Onambele. 2016. “Latin Vallex. A Treebank-based Semantic Valency Lexicon for Latin.” In Proceedings of the Tenth International Conference on Language Resources and Evaluation (Lrec 2016), 2599–2606.

Marco Passarotti, Francesco Mambrini, Flavio Massimiliano Cecchini, Greta Franzini, Eleonora Litta, Giovanni Moretti, Paolo Ruffolo, and Rachele Sprugnoli. 2020. “Interlinking through Lemmas: The Lexical Collection of the LiLa Knowledge Base of Linguistic Resources for Latin.” Studi E Saggi Linguistici 58 (1): 177–212.

Günther Alexander E. A. Saalfeld. 1874. Index graecorvm vocabvlorvm in lingvam latinam translatorum qvaestivncvlis avctvs. Berlin: apvd F. Berggold.

Günther Alexander E. A. Saalfeld. 1884. Tensaurus Italograecus: Ausführliches historisch-kritisches Wörterbuch der griechischen Lehn- und Fremdwörter im Lateinischen. Wien: Carl Gerold’s Sohn.

Smith, Ray. 2007. “An Overview of the Tesseract OCR Engine.” In Ninth International Conference on Document Analysis and Recognition (ICDAR 2007), 2:629–33.

Stead, Stephen, Martin Doerr, Christian-Emil Ore, and Athina et al. Kritsotaki. 2019. “CRMinf: the Argumentation Model, Version 0.10.1 (draft).”\%20ver\%2010.1.pdf


1 “Captive Greece captured her savage conqueror” (our translation).

2 In Egypt, for instance.


4 The Latin verb græcissō used in the title of this paper is a nativised version of the Greek graikizw “to imitate the Greeks; speak Greek”. For a detailed overview of linguistic “nativisation”, see Hock and Joseph (2009, pp. 247-57).

5 Crude estimate of an average ten to fifteen entries per page, for a total 592 pages.


7 The specific uncertainty remains unclear as no editorial documentation is provided.

8 For the most recent overview of Ancient Greek optical character recognition, see Robertson and Boschetti (2017).




Table des illustrations

Titre Figure 1: Three lexical entries in IGVLL, translating to “Bear’s Foot (plant)”, “without smoke”, and “acatalectic (line of verse)”, respectively
Fichier image/jpeg, 90k
Titre Figure 2: Latin OCR of Fig. 1, Internet Archive
Fichier image/jpeg, 80k
Titre Figure 3: Ancient Greek OCR of Fig. 1, Tesseract
Fichier image/jpeg, 76k
Titre Listing 1
Fichier image/jpeg, 43k
Titre Listing 2
Fichier image/jpeg, 62k
Titre Listing 3
Fichier image/jpeg, 80k
Titre Listing 4
Fichier image/jpeg, 38k


Università degli Studi di Pavia, Pavia, Italia –

Le texte et les autres éléments (illustrations, fichiers annexes importés) sont sous Licence OpenEdition Books, sauf mention contraire.


Open access

Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search