MEHDIE: The Middle East Heritage Data Integration Endeavor Sinai Rusinek Tomer Sagi Moran Zaga Efraim Lev Moshe Lavee

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

 MEHDIE: The Middle East Heritage Data Integration Endeavor 2023 Digital Humanities 2023 (DH2023), Graz 10.5281/zenodo.8107495 rusinek2023-mehdie rusinek2023-mehdie.txt, the project's staged copy 629e64c3c0741e7d73e60d55d6051abe0a9dd4ce cc-by digital-humanities

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

 42 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them Can historical Middle Eastern datasets be linked by matching place names directly between the domain's own languages, rather than by transliterating them into Latin script or routing them through a general-purpose gazetteer? Argues, from a worked example, that a general authority file under-covers a specialist domain's own vocabulary, and that intra-domain matching beats routing through a central one. method cross-lingual toponym entity matching system-or-tool-buildingsystem-or-tool-building Matching uses three kinds of evidence at once - coordinates, dates and the text of the name - rather than name similarity alone Transliterating between two Semitic languages preserves the ambiguity of both instead of forcing a decision, which is why it beats going through Latin script Latin transliteration fails because it demands vowels the source scripts never wrote Every candidate match is returned with a confidence and an account of which method produced it, so a human can judge it General gazetteers barely cover the Semitic-script toponyms this domain runs onwhat this limits: any reliance on general gazetteers for this domain: GEONAMES AND THE GETTY THESAURUS ... CONTAIN FEW, IF ANY, TOPONYMS IN THE SEMITIC SCRIPTS relevant to the sources Even Wikidata, which is multilingual by design, matches poorly against these sourceswhat this limits: Wikidata reconciliation specifically: EVEN WIKIDATA, WHICH IS ESSENTIALLY MULTILINGUAL, SHOWS LOW RECALL WHEN MATCHING AGAINST THEM Where general resources do cover the domain it is the major cities, not the smaller settlements or the historic variantswhat this limits: the coverage that general resources do provide: they include forms that DENOTE MAJOR, WELL KNOWN CITIES, BUT VERY LITTLE HISTORIC VARIANTS OF THE LESSER KNOWN SETTLEMENTS The domain's own scripts make the problem harder before any tool is applied: missing vowels, uncertain pronunciation, and no settled naming practicewhat this limits: every matching result in the domain: THE SCARCITY OF VOWELS IN SEMITIC SCRIPTS, THE INDETERMINACY OF PRONUNCIATION AND TRAN[sliteration] make the problem hard before any tool is applied Even the successful transliteration is one of several possible forms, not a unique answerwhat this limits: any successful match treated as unique: the Hebrew-to-Arabic transliteration used was ONE OF TWO POSSIBLE TRANSLITERATIONS transliteration - into Latin script: the usual move, which fails here because it requires vowels the source never wrote transliteration - between Semitic languages: preserves the same indeterminacy on both sides, so it narrows candidates instead of multiplying them the place name - as written in a nineteenth-century edition: בלביש, with an archaic final consonant the place name - as written in contemporary Hebrew: בילביס, with an added vowel and a different final letter - and the authority holds only this one linked data strategy - centralised: reconcile every source to one external authority, which is the default assumption linked data strategy - bottom-up and intra-domain: match the domain's datasets against each other first, which this paper argues for phenomenon: cross-lingual toponym entity matchinga second reading reaching the same value. Recorded: cross-lingual toponym entity matching — note: a second reading reaching the same value from a different sentence. Not applied - a person decides. extends → World Historical Gazetteer (Grossner and Mostern 2021) builds-on → Historical Middle East Data Alliance (Hist-ME) extends → the World Historical Gazetteer, which the matching tools are intended to enhance supports → the Hist-ME vision of connecting monolingual and intra-religious data silos critiques → reconciliation to general-purpose external gazetteers as the default linked-data strategy transliteration —operationalised-as→ entity resolutionstrength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its span and kept, but transliteration enabled matching; 'entity resolution' is the task, and operationalised-as overstates it.transliterationentity resolution Geonames —is-a→ gazetteerstrength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but endpoint by anaphora: 'two such resources' = external Gazetteers (previous sentence).Geonamesgazetteer Getty Thesaurus of geographic names —is-a→ gazetteerstrength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but endpoint by anaphora: 'two such resources' = external Gazetteers (previous sentence).Getty Thesaurus of geographic namesgazetteer gazetteer —operationalised-as→ entity resolutionstrength of the relation: 0.9 Audit 2026-09-28: DOUBTFUL - read against its span and kept, but reconciling names against a gazetteer is described; 'entity resolution' is the tagger's term for it.gazetteerentity resolution World Historical Gazetteer —is-a→ gazetteerstrength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but is-a rests on the word 'Gazetteer' inside the name.World Historical Gazetteergazetteer General-purpose gazetteers carry almost none of the Semitic-script toponyms that Middle Eastern historical sources actually use. Wikidata, although multilingual, also matches these sources poorly. A concrete gap: Wikidata holds a modern Hebrew variant of the toponym but not the archaic spelling the historical source uses. transliteration between Semitic languages reduces rather than adds ambiguity The authors draw a general conclusion for linked open data: prefer a bottom-up, intra-domain strategy over a centralised one. general gazetteers including Wikidata have low recall against Semitic-script historical toponyms transliteration between Semitic languages reduces rather than adds ambiguity: neither Arabic nor Hebrew writes the vowel sounds that English spells with the first i and the e Wikidata holds a modern Hebrew variant of the toponym but not the variant with the archaic letter ש. the results may also argue for collaboration within the domain MEHDIEMEHDIE Hist-MEHist-ME

MEHDIE: The Middle East Heritage Data Integration Endeavor

Rusinek Sinai Haifa University, Israel sinai.rusinek@mail.huji.ac.il Digital Humanities 2023. Collaboration as Opportunity Graz

Sagi Tomer Haifa University, Israel tsagi@cs.aau.dk Digital Humanities 2023. Collaboration as Opportunity Graz

Zaga Moran Haifa University, Israel moran82@gmail.com Digital Humanities 2023. Collaboration as Opportunity Graz

Lev Efraim Haifa University, Israel elev@univ.haifa.ac.il Digital Humanities 2023. Collaboration as Opportunity Graz

Moshe Lavee Haifa University, Israel mlavee@univ.haifa.ac.il Digital Humanities 2023. Collaboration as Opportunity Graz

Zentrum für Informationsmodellierung - Austrian Centre for Digital Humanities, Karl-Franzens-Universität Graz 2023 Graz Creative Commons BY 4.0 Digital Humanities 2023: Book of Abstracts Walter Scholger University of Graz Graz, Austria walter.scholger@uni-graz.at Georg Vogeler University of Graz Graz, Austria georg.vogeler@uni-graz.at Toma Tasovac Belgrade Center for Digital Humanities Belgrade, Serbia ttasovac@humanistika.org Anne Baillot Le Mans Université Le Mans, France anne.baillot@univ-lemans.fr Data Management Elisabeth Raunig Data Management Martina Scholger Data Management Elisabeth Steiner Data Curation Johanna Ofenauer Data Curation Christina Burgstaller 10.5281/zenodo.7961822

Converted from a Word document

DHConvalidator

We are building a multi-modal entity resolution tool that utilizes the spatial, temporal, and textual information about a pair of place records to help identify the semantic relation between them. We plan to share our progress towards the integration of knowledge sources related to the history of the Middle East.

English

Paper

Poster

entity matching multilinguality gazetteers transliteration toponyms

information retrieval and querying algorithms and methods linked (open) data spatial & spatio-temporal analysis modeling and visualization Geography and geo-humanities History

MEHDIE , the Middle East Heritage Data Integration Endeavour is a project at Haifa University and Aalborg University, dedicated to facilitating semantic integration of knowledge sources related to the history of the Middle East, using innovative data integration methods.

The multi-modal entity resolution tool we are developing will utilise spatial (geo-coordinates), temporal (dates/periods) and textual information about a pair of place/person records to help identify the semantic relationship between them. In the demonstration we present tools related to spatial data, which we hope to integrate into the World Historical Gazetteer (Grossner and Mostern 2021) system to enhance its capabilities for inter-dataset matching. The poster will address our workflows for data extraction and assembly, as well as the results of matching experiments between various datasets . We describe how we leverage the linguistic affinities and continuities between the domain languages and various textual matching techniques.

In figure 1 we see the interface of the MEHDIE clone of the World Historical Gazetteer, with the added option to match two datasets against each other, in this case a dataset of the Hebrew toponyms from Benjamin of Tudela’s travelog collected in the framework of the TraveLab project (Rusinek and Shalit 2022), and Damast Multilingual Dataset from the Dhimmis and & Muslims project (Franke and Kock 2023).  Also visible in the figure are the confidence measure and the explanation of the matching methods that yielded the candidate.

The mission to connect data on the Middle eastern history domain is faced with special challenges: the scarcity of vowels in Semitic scripts, the indeterminacy of pronunciation and transliteration of written toponyms, and varying practices of naming and classifying places.

A common way to identify places, their locations and other information about them is to reconcile the source place names with unique identifiers from an external Gazetteer. Geonames and the Getty Thesaurus of geographic names are two such resources, but they contain few, if any, toponyms in the Semitic scripts relevant to the sources of Middle Eastern History. Even Wikidata, which is essentially multilingual, shows low recall when matching against them. Such general gazetteers may include contemporary Arabic, Hebrew or Syriac forms that denote major, well known cities, but very little historic variants of the lesser known settlements. With this obstacle in mind, one may turn to transliteration. The similarity between semitic languages, however, means that direct transliteration between semitic languages would be much more efficient in creating candidates for matching that transliteration into Latin script languages.

The example in figure 2 is a telling illustration for this:

The Hebrew toponym for the city of Bilbeis on the Nile’s delta (spelled “Bilbays” in Adler’s translation of Benjamin of Tudela) was mentioned in the 19th century edition of Benjamin of Tudela’s itinerary in the Hebrew form בלביש. In contemporary Hebrew it is spelled בילביס, with an additional vowel and a different letter to represent the final consonant. Wikidata currently includes a Hebrew alternative variant without the additional vowel, but not a variant with the more archaic letter ‘ ש’ (Wikidata contributors, "Q393495").

Using the transliteration tool in our system to match against Arabic datasets such as Usaybia , Damast , ‘Al-Ṯurayya gazetteer (Romanov and Seydi, 2019) and from a dataset of toponyms from Yaqut al-Hamawi’s Mu 'jam  ul-Buldān, we could easily find the match thanks to the transliteration of the Hebrew form to the Arabic form بلبيس, one of two possible transliterations. This shows the power of transliteration between semitic languages that reduces, rather than adds ambiguity: Neither Arabic nor Hebrew explicate the sound that is expressed in English by the first i and the e in the name. Therefore, transliteration to latin script could not deduce an acceptable transliteration.

Several projects have been dedicated in recent years to the aggregation of geographic and prosopographic data on Middle Eastern History. The Historical Middle East Data Alliance (Hist-ME) was initiated by Nathan P. Gibson at the Linked Pasts 2021 conference with a joint vision of connecting intra-religious and monolingual silos of disconnected data into a linked contextual wealth of knowledge. The system we are developing at MEHDI facilitates collaboration between Hist-ME ’s partners towards this vision.

The advantages of intra-domain transliteration for matching opens a meta-discussion on approaches for linked open data and the preference of a “bottom-up” intra-domain strategy over a centralised one. The results of our work with the different datasets of Hist-ME member projects may also constitute an argument in favour of intra-domain collaboration.