Wikidata as a Knowledge Base for People of the Greco-Roman World

Repository location

https://doi.org/10.7910/DVN/42QMWG

Repository name

JOHD Dataverse

Object name

folders: Annotation, Query, Entities

Format names and versions

CSV, TSV, TXT

Creation dates

01/09/2024–01/11/2025

Dataset creators

Margherita Fantoli, dataset curator and data annotator; Valeria Irene Boano, dataset curator and data annotator; Evelien de Graaf, dataset curator and data annotator; Camillo Carlo Pellizzari di San Girolamo, dataset curator and data annotator; Herbert Verreth, data annotator.

Language

English for metadata; Latin; Ancient Greek

License

CC BY-NC-SA 4.0

Publication date

2025-11-03

Sections 3 and 4 detail the contents of the Dataverse folders.

(3.1) Query development

Formulating a SPARQL query to retrieve “ancient people” from Wikidata is complex. Several criteria, such as time and place of living, nationality, language, religion, and cultural contributions, must be considered, and these are spread across multiple properties in the Wikidata ontology. Combining them effectively requires carefully balancing precision (to avoid noise) and recall (to include relevant entries).

Because the final query is lengthy, we provide its full text in the Dataverse (“Query” folder) and summarize its components here. It relies on multiple inclusion and exclusion criteria and was iteratively refined based on result analysis. We discuss the most complete version,¹⁸ though evaluation used an earlier version with minimal differences;¹⁹ both are available in the Dataverse. Criteria added later are marked [new].

• – “Instance of” criteria: items with best-rank values of P31 (instance of) that include at least one of:

  • – Q5 (human)

  • – Q21070568 (human whose existence is disputed)

  • – Q64643615 (prosopographical phantom) [new]²⁰

• – “Inclusion” criteria (see Table 1): The retrieved items need to match at least one of the following criteria:

  • – I1–I3: Temporal filters based on birth, death, floruit.

  • – I4–I5: Conditional inclusion based on ancient world properties or languages.

  • – I6–I7: Citizenship, political entities, and time periods.

  • – I8: Source-based inclusion.

• –“Exclusion” criteria (see Table 2): items are excluded if they match at least one of the following criteria, relative mainly to Chinese ancient people:

  • – E1: Property related to Chinese history databases

  • – E2–E4: Chinese names, birthplace, citizenship

  • – E5: being associated with Jainism

Some specifications are needed to fully describe the query: first, for each property, we always use only best-rank values: deprecated-rank values are always discarded; if a property has preferred-rank values, they are the only ones included, whilst if there are no preferred-rank values, all normal-rank values are included.²²

Moreover, we applied some additional filters on some of the criteria:

• – I4: if the item has best-rank values of P569 (date of birth) and/or P570 (date of death) and/or P1317 (floruit) and they are not unknown values, it must fall into the criteria I1, I2, I3; this is to exclude, for instance, medieval and modern authors receiving some of the identifiers related to the ancient world, such as a Pinakes author ID (P6831).

• – I5: if the item has best-rank values of P569 and/or P570 and/or P1317 and they are not unknown values, it must fall into the criteria I1, I2, I3; this is to exclude later people speaking or writing ancient Greek and Latin, especially modern philologists.

Some criteria deserve a broader discussion, since their application impacts the overall evaluation:

• – Dates: Our goal is to create a prosopographical resource for the Greco-Roman world, whose temporal boundaries are, of course, inherently uncertain. The chosen dates roughly correspond to the rise of the Mycenaean civilization and the fall of the Western Roman Empire. Naturally, this latter boundary is highly conventional and results in the inclusion or exclusion of individuals who lived relatively close in time and were undoubtedly influenced by Greco-Roman culture—not to mention those living in the Eastern Roman Empire. This criterion, therefore, should be seen as flexible and adaptable to the specific needs of each project or resource.

• – Exclusion criteria: It is striking that our exclusion criteria primarily target individuals associated with Chinese civilization. This choice is motivated by two factors. First, in our initial query tests, the majority of irrelevant results fell into this category. Not excluding explicitly China-related records falling within our time-frame results in ca. 5000 additional results to query, i.e. a significant drop in precision. Second, we are reasonably confident that this exclusion would omit only a minimal number of individuals relevant to our focus (if any), as Sino-Roman contacts at the time were very shallow and mostly limited to the silk trade.²³ In contrast, for several other civilizations—whose members, as discussed below, are sometimes incorrectly retrieved by our query—it is more difficult to apply strict exclusion criteria. This is due to the fluid boundaries of Greek civilization and the Roman Empire. Such complexity affects, for instance, Egyptian, Babylonian, and Jewish civilizations, whose interactions with the Greco-Roman world evolved significantly over time and whose histories were often subjects of interest for ancient Greek and Latin authors. This aspect is further discussed in Section 4, when commenting on the results of the query.

(3.2) Annotation

In order to evaluate how well Wikidata represents named individuals associated with literary sources of the Greco-Roman world, we conducted a specific case study: we compared the query results to the individuals mentioned in a selected corpus of ancient Greek and Latin texts. This section describes the annotation process of linking named individuals occurring in it to both Wikidata and the Wikisource version of the RE. The file containing the annotation data and a description of the format used is found in the Dataverse folder Annotation.

The corpus comprises two subsets of differing sizes, each annotated according to slightly different criteria. The first subset (Subset 1, henceforth) was developed within the context of the NIKAW project²⁴ and includes ancient Greek and Latin texts by authors from different cultural and historical contexts (de Graaf, Forthcoming). The second subset (Subset 2) is significantly smaller and consists exclusively of books II-VI of the Naturalis Historia by Pliny the Elder. Its annotation was performed with the aim of studying the canon of people created by Pliny, in the context of the MECANO doctoral network (Boano, Forthcoming).²⁵ Together, these two datasets contain both Greek and Latin texts spanning a time period from the 4th century BCE to the 4th century CE. Different genres and text types are included, such as letters, (philosophical) treatises, speeches, biographies, histories, dialogues, miscellanies. The diversity within the data means that, although not exhaustive, it encompasses a wide spectrum of individuals referenced in Greco-Roman literature, making it appropriate for assessing the relative coverage of knowledge bases.

The annotation process consisted of two steps: identifying named individuals and linking them to both the RE and Wikidata. In both corpora, the first step followed the general principle that a person is defined as “any identifiable single individual, including deities and anthropomorphic mythological figures” (Palladino et al., 2024). The second step entailed linking each mentioned individual to the corresponding RE article and Wikidata item. For Wikidata, the linking was performed by pointing to the item identifier (e.g. M. Vipsanius Agrippa was linked to the item Q48174). Linking to the RE required extracting relevant data from the Wikisource register and importing it into a separate database, where each RE entry was assigned a unique identifier.²⁶ The linking process differed slightly between the two subsets. In Subset 1, the named entities were first manually linked to RE identifiers and subsequently semi-automatically matched to their corresponding Wikidata items. Any RE entries already linked to Wikidata items in the Wikisource register were linked automatically. All remaining entities were linked manually. When a Wikisource page already exists for a RE article, it is also connected to the corresponding Wikidata item, improving the alignment between the two resources (Figure 2).

Figure 2

Manual linking involved checking for variant spellings across different languages and was prioritised for individuals appearing in multiple works, followed by those associated with a RE identifier. For individuals mentioned in only one work and not present in the RE, only a small alphabetical sample, approximately 40% of the total, was annotated. In Subset 2, instead, the linking to the two knowledge bases was performed in parallel and entirely manually.

In both subsets, tokens were annotated individually, without merging multi-token entities. Names like Gaius Julius Caesar were linked per token, so a single name counts as multiple attestations of the same person—for example, three for Gaius Julius Caesar. With regard to nested multi-token entities, the two subsets followed different annotation criteria. In Subset 1, each token referencing a different individual was annotated with a link to the respective person: for instance, in the sequence Africani fratris nepos (Cic., Tusc. Disp. 1, 81), each token was linked to a different individual (i.e. the RE Wikisource entities Cornelius 335,²⁸ Fabius 109²⁹ and Fabius 107³⁰). In Subset 2, instead, the annotation was highly dependent on proper names and did not take into account nested entities: for instance, in the sequence Daedali filio (Plin. Nat. Hist. 3, 102), only Daedali was tagged and linked to the respective entity.³¹

As previously mentioned, the annotation guidelines we followed led to the inclusion of deities and mythological figures in the dataset. To accurately evaluate the coverage of the Wikidata query over our corpus, it was necessary to modify the dataset. The query includes only instances of “human”, “human whose existence is disputed” and “prosopographical phantom”. However, our dataset, as illustrated in Figure 3, included entities from many other Wikidata categories. Retaining these additional categories would have skewed the comparison results, artificially inflating the number of unmatched entities. To avoid this bias, we removed all the entities that were not “instance of” (P31) the three categories listed above, by leveraging the established links to Wikidata. The entities that did not have a link to Wikidata were annotated manually as human or not human. After removing all the non-human entities, the dataset contained 19,262 attestations out of the original 24,807 (77.65%), corresponding to 2,279 unique entities out of the original 3,004 (75.87%). Table 3 shows the distribution of attestations and unique entities, including and excluding non-human entities, across the Greek and Latin works in our corpus.

Figure 3

(4.1) Accuracy of the query

To evaluate the accuracy of the query results, we randomly selected 1,000 items and assessed whether they fell within the scope of our dataset. The assessment was conducted by two scholars specializing in the ancient world. Of the sample, 835 items were deemed relevant, while 165 were not (yielding an accuracy of 83.5%).

The non-relevant items fall into the following categories:

• – Individuals with missing or incorrect metadata which caused the filters to malfunction.³² Most of these cases involve people outside the temporal scope of the query, but whose Wikidata items lacked proper dating.³³ Similarly, in some instances, Chinese nationality was absent or not adequately marked.³⁴

• – Individuals within the temporal scope but unrelated to the Greco-Roman world, such as Vikings,³⁵ Japanese,³⁶ or Ancient Egyptians.³⁷ As discussed above, we chose not to implement strict exclusion rules in the query, with the exception of the case of Chinese entities, in order to keep the query flexible and generalizable. However, such rules could be added if needed.

When establishing which items fall within the scope of our dataset and which do not, we inevitably engage with the contested concept of ‘Western civilization’, traditionally rooted in the Greco-Roman world and early Christian Europe.³⁸ Excluding ancient eastern civilizations like China or Japan emphasizes this “western” focus, yet boundaries remain fluid.³⁹ For instance, figures from the Sasanian Empire were included due to frequent interactions with Rome.⁴⁰ Our approach is pragmatic: we target Wikidata items relevant for annotating Greek and Latin sources. Consequently, non-western individuals mentioned by these authors – such as Aktisanes, a Nubian king cited by Hecataeus⁴¹ – were also included, along with figures near the dataset’s temporal limits, even when dating is uncertain.⁴²

Our guiding principle was inclusivity within and around the Greco-Roman context, even at the cost of some inconsistencies. Through our discussions, we defined a set of empirical inclusion criteria to assist with decisions on temporal and geographical edge cases. These criteria are not intended as necessary conditions, meaning a person does not need to meet all of them, but rather as sufficient ones for resolving ambiguous cases: if at least one criterion applies, the individual is included:

• – The person is mentioned in the RE.

• – The person is identified in Wikidata, based on primary and secondary sources, as an early Christian martyr, bishop, monk, nun, saint, or another religious category within the Christian tradition.

• – The person is cited by an ancient Greek or Latin author relevant to our scope.

• – The person travelled to a location within the Greek world or the Roman Empire.

• – The person was involved in battles against Greek or Roman armies.

(4.2) Coverage of the query and of Wikidata

(4.2.2) General coverage of Wikidata compared to RE and annotated texts

Analysis of the annotated data allows for several observations regarding Wikidata’s coverage of individuals mentioned in our texts. Figure 4 illustrates how linkable attestations are distributed among entities covered by both the RE and Wikidata, those covered by only one of the two, and those that cannot be linked, represented by a separate bar. Figure 5 presents the same distribution but at the level of unique individuals. In general, most attestations in our corpus are covered by both knowledge bases. However, the proportion of mentions not covered by either is noteworthy, especially when considering unique individuals, where they account for 8.8% of the total. A crucial note in correctly interpreting these results is the presence of 1,586 attestations of Jesus Christ,⁴⁹ who does not appear in the RE: this significantly skews the overall distribution of attestations covered only by Wikidata and explains the shift in proportions observed when comparing attestation-level data to individual-level data.

Figure 4

Figure 5

From this perspective, we observe that, at the time of writing, the RE still provides better coverage of the individuals mentioned in our corpus. However, the RE, in terms of included entities, is a closed project, and its degree of representation of ancient individuals cannot be extended. On the other hand, Wikidata is an ongoing project, thus leaving the door open for an enhancement of its coverage in relation to the ancient world. In Section 5, we discuss how such extensions could be covered by community initiatives.

A more detailed examination of the data shows how individuals who are either represented exclusively in Wikidata or not linkable at all (i.e., absent from both Wikidata and the RE) are distributed across the works in our corpus (see Table 4). In Aristotle’s Metaphysica, Seneca’s Epistulae Morales, and Lactantius’s Divinae Institutiones, all mentioned individuals are fully represented in Wikidata. Interestingly, these texts span both ancient Greek and Latin as well as different time periods. The texts least covered by Wikidata, with a substantial number of individuals lacking representation, are: Diogenes Laertius’s Vitae Philosophorum and Apuleius’ Apologia. However, the two cases differ significantly. In the case of Diogenes Laertius, in fact, the number of attestations not covered by Wikidata constitutes only around 10% of the total, whereas for Apuleius’ Apologia the percentage rises to approximately 50%. The latter can then be considered the least represented work in terms of the mentioned individuals.

Table 4

The number of attestations and unique entities not covered by Wikidata or not linkable at all, distributed per work.

AUTHOR_WORK	ATTESTATIONS_NO_WIKIDATA	ENTITIES_NO_WIKIDATA	NOT_LINKABLE_ATTESTATIONS
Ael_VH	67	49	14
Amm_Marc	26	16	2
Diog_Laert	505	340	174
Strabo	4	4	0
Cic_Tusc	15	10	4
Clem_Al_Strom	43	42	15
Tac_Hist	73	35	0
Gal_PHP	2	2	0
Origen_C_Cels	5	3	4
Plut_Mor_Quaest_conv	174	61	43
Lactant_Div_Inst	0	0	0
Plato_[Epist]	21	7	14
Sen_Ep	0	0	0
Apul_Apol	326	32	32
Arist_Metaph	0	0	0
Dion_Hal_Dem	18	9	5
Tert_De_anim	12	8	9
Pliny_NH	14	11	2

In Apuleius’s Apologia, the most frequently cited individual not represented in Wikidata is his wife Pudentilla, followed by his accusers Herennius Rufus, Sicinius Aemilianus, Sicinius Pontianus, and Iunius Crassus. Interestingly, the historicity of the events narrated in this work remains unverifiable, despite numerous scholarly efforts (Hunink, 1997). The critical factor, however, is that all these individuals are known solely from this text. This pattern is common among figures not linkable to Wikidata, who frequently appear in only a single work.

However, in some cases, even people who appear in more than one text are not represented by Wikidata. This is often true for family members of well-known individuals. Table 5 lists a sample of such individuals, with their own RE label and links to their more prominent relative.

Table 5

A sample of individuals not appearing in Wikidata and their relationship with well-known and represented people.

RE LABEL	RELATIONSHIP	RELATIVE’S RE LABEL	RELATIVE’S WIKIDATA ITEM
Babys 3	father of	Pherekydes 3	https://www.wikidata.org/wiki/Q311485
Bloson 2	father of	Herakleitos 10	https://www.wikidata.org/wiki/Q41155
Damasos 6	brother of	Demokritos 6	https://www.wikidata.org/wiki/Q41980
Exekestides 1	father of	Solon 1	https://www.wikidata.org/wiki/Q133337
Gryllos 2	father of	Xenophon 6	https://www.wikidata.org/wiki/Q129772
Peirithoos 3	father of	Alkmaion 6	https://www.wikidata.org/wiki/Q188332

In conclusion, the analysis of Wikidata’s coverage within our corpus demonstrates a generally strong representation of ancient individuals, though there remains room for improvement. In particular, the investigation revealed a clear correlation between the lack of representation and the isolated or infrequent occurrence of individuals in ancient texts.

Within the annotated items that were represented in Wikidata, a small number (ca. 40) had not been retrieved by the query. Some items were missed due to incomplete data in Wikidata items, while other items were missed in the original query but were included in the improved one. This observation reveals that the quality of the entries is a key-factor for the construction of the desired knowledge base.