Ontolex-Lemon in Wikidata and other Wikibase instances

Ontolex-Lemon in Wikidata and other Wikibase instances David Lindemann UPV/EHU University of the Basque Country Unibertsitateko Ibilbidea, 5 01006 Vitoria-Gasteiz [email protected] Abstract This paper provides insight into how the core elements of the Ontolex-Lemon model are integrated in the Wikibase Ontology, the data model fundamental to any instance of the Wikibase software. This includes Wikidata lexemes, which today is probably the largest OntolexLemon use case, a dataset collaboratively built by the community of Wikidata users. We describe how lexical entries are modeled on a Wikibase, including the linguistic description of lexemes, the linking of lexical entries, lexical senses and lexical forms across resources, and links across the domain of lexemes and the ontological part of a Wikibase knowledge graph. Our aim is to present Wikibase as a solution for storing and collaboratively editing lexical data following Semantic Web standards, and to identify relevant research questions to be addressed in future work. 1 Introduction Wikibase,1a set of extensions to MediaWiki2, is a software solution for storing, collaboratively editing and exhibiting structured data on the web, in the shape of a knowledge graph. The software is used, first and foremost, by Wikidata (Vrandeˇ ci´ c and Krötzsch,2014;Erxleben et al.,2014). 3 Many other instances of the software have emerged since the software packages (and hosting solutions) are freely available. 4 The types of entities described in a Wikibase include items, which represent all kinds of real-world objects and ontological concepts, and lexemes, describing words. As it will be explained in section 2.1, in a Wikibase, lexemes are described following the core of the Ontolex-Lemon model (McCrae et al.,2017), and so they are on Wikidata, which is today probably the largest open and 1See https://wikiba.se. 2See https://mediawiki.org. 3See https://www.wikidata.org. 4 See https://wikiba.se/showcase/ and https:// wikibase.world, a catalogue of Wikibase instances. collaboratively editable Ontolex-Lemon use case. By May 2025, Wikidata described 1.42 Million lexical entries in 1,379 different languages, which compared to earlier figures (Nielsen,2020) means an exponential growth. German, Russian, Danish, Estonian, English and Malayalam are, in this order, the languages with the most described lexemes. The potential uses of the linguistic descriptions contained in this growing resource, which we present in more detail in section 2.2, are manifold. Section 2.3 is devoted to Wikibase as a linking hub for descriptions of lexical entries,senses and forms across different resources; cross-resource links are encoded as external ID properties. 5 External identifiers do not only constitute hyperlinks for a user to jump between the resources presented in different web portals, but also enable SPARQL federation, 6 that is, a Wikibase’s content may be integrated also with the lexical and ontological content of any other Wikibase, including Wikidata, or an RDF database of other kind. 7 In this regard, it is important to point out how RDF is used on a Wikibase, which will be explained in section 2.4. Related to this, an important and distinguishing feature of Wikibase is its multi-layer integration of lexical and ontological entities inside the same database, which will be discussed in detail in section 2.5. While the lexemes collection on Wikidata is open for continued enrichment, some use cases may require a separate Wikibase instance, in a first phase of a contribution project, or even for a whole project lifetime: A language may be already described on Wikidata, so that any addition would involve lexeme, sense and form disambiguation 5 See https://www.wikidata.org/wiki/Wikidata: External_identifiers. 6 See https://www.w3.org/TR/ sparql11-federated-query/. 7 See https://www.mediawiki.org/wiki/Wikibase/ Federation. and deduplication tasks, which when working on an own instance could be left to the final phase of a project. Also, a dataset to be curated may be regarded too noisy (e. g. in a legacy dictionary digitisation project), or too esoteric (e. g. when dealing with dialectal or historical language data) to be included in Wikidata without previously ensuring relevance and quality. In addition, to work on an own Wikibase instance means freedom in data modeling and community management. Licensing might also be an issue, since Wikidata content is obligatorily licensed according to CC0. Exploring these potentials, we briefly discuss the Wikibase ecosystem for lexeme descriptions in section 2.6. In the closing section 3, we provide an outlook for open research questions to be worked on, focusing on the relation between the fine-grained modeling proposals made by the Ontolex Community, on the one hand, and the conventions emerging in the community working on Wikidata lexemes, on the other. 2 Ontolex-Lemon on Wikibase 2.1 Lemon core classes The core of Ontolex-Lemon, 8 that is, the classes ontolex:LexicalEntry , ontolex:LexicalSense , and ontolex:Form , is reused in the Wikibase Ontology, 9 the backbone model of any Wikibase instance. Wikibase treats the lexical entry, with its own numeral identifier preceded by letter L, as primary entity describing a lexeme, with forms and senses as sub-entities, and presents the instances of those three Ontolex-Lemon core classes together on one editable lexeme page. 10 This structure is pre-set in the data model fundamental to any Wikibase instance and cannot be modified by the user, and the same is true for a small number of properties to be attached to the three core classes, listed in table 1; the three obligatory properties describing a lexical entry must have a value, a sense must have a gloss (a short sense-disambiguating text, represented in RDF using skos:definition ), and aform must have a representation (a value for ontolex:representation).11 8 For the original Lexical Model for Ontologies, see https: //lemon-model.net/. 9See https://wikiba.se/ontology/. 10 See https://www.wikidata.org/wiki/Lexeme:L1 for the lexeme page describing lexeme wd:L1. 11 Although ontolex:writtenRep would be the best match here, the Wikibase Ontology uses ontolex:representation ; ontolex:LexicalEntry wikibase:lemma wikibase:lexicalCategory dct:language ontolex:sense ontolex:lexicalForm ontolex:LexicalSense skos:definition ontolex:Form wikibase:grammaticalFeature ontolex:representation Table 1: Obligatory basic classes and properties in their domain for describing lexemes in Wikibase In detail, this obligatory structure entails the following restrictions: • A lexical entry must point to exactly one item in the same Wikibase as value for lexical category. • A lexical entry must point to exactly one item in the same Wikibase as value for its language. • A lexical entry must have at least one lemma. Several lemmata can co-exist for the same entry for covering different spelling variants (e. g. British and American English). 12 In the RDF representation, lexeme lemmata appear attached to the entry using a property named wikibase:lemma . Lemmata are indexed for the MediaWiki text search index, 13 so that a user can search for lexemes, manually in the interface, or via API. • A lexical sense must have at least one sense gloss, in any language, i. e. not necessarily or not only in the language associated to the entry. Purpose of the gloss is to provide the information necessary to discriminate word senses. • Aform must have at least one written representation. Alike different values for lemma in Ontolex-Lemon, the former is defined as subproperty of the latter. 12 See an example at https://www.wikidata.org/wiki/ Lexeme:L1347 , where the English color/colour is described, an example for an entry with lemmata in distinct scripts at https://kurdi.wikibase.cloud/wiki/Lexeme:L3447. 13 See https://www.mediawiki.org/wiki/ Elasticsearch. on entry level, form representations in several spelling variants can be attached to the same form entity. A written representation in most cases will be a string as found in text of that language, but written representations also include code transcriptions, such as those describing sign language forms. 14 To describe a form without providing any type of written representation is not foreseen. • Aform can have zero or more items in the same Wikibase as values for grammatical feature. Lemmata, sense glosses ( skos:definition ), and also form representations ( ontolex:representation ) are associated to a language code. The available codes are the same that are available throughout the mediaWiki instance, e. g. for labels, descriptions, and monolingualtext strings.15 Figure 1: The Wikibase lexeme, as illustrated in the Wikibase documentation These restrictions guarantee interoperability on a basic level. Beyond that, any additional relation involving entry,sense or form objects is not 14For an example, see wd:L991786-F1. 15 See https://www.wikidata.org/wiki/Help: Monolingual_text_languages , and for a list of all codes implemented in a Wikibase instance, e. g. https://www.wikidata.org/w/api.php?action=query& meta=wbcontentlanguages. predefined and can be modeled according to the use case using self-defined properties in Wikibase statements, 16 as illustrated in Fig. 1. 17 The RDF classes and properties mentioned above are part of the Wikibase Ontology, that is, they are used in the RDF representation of an entity (an item, a lexeme, aproperty), and accordingly, they appear in RDF entity data dumps. 18 In opposition to that, all other relations added to lexeme,sense or form as Wikibase statements (see section 2.4) always involve a property defined in the namespace of the Wikibase instance itself, identified by a number preceded by the letter P, and that has a range restricted to one Wikibase datatype.19 2.2 Wikidata lexemes Wikidata lexemes is an open and editable collection, where everybody is invited to collaborate. A documentation of the data model based on the three Ontolex-Lemon core classes is given on the Wikidata documentation pages. 20 Concerning advanced modeling questions, contributors get support from each other. 21 A core group of more experienced and active users provides advice to newcomers and occasional contributors, also through dedicated outreach events. 22 A manually curated list of lexical entries provides examples in several languages of good and complete modeling practice.23 Instead of following prescribed models concerning morphology, etymology, multilingual equivalents, etc., on Wikidata, a bottom-up grown set of properties is used for describing entries, senses, and forms; see table 2for the ten most frequent properties for each of the classes, pointing to other 16 See https://www.wikidata.org/wiki/Help: Statements. 17 The picture is used at https://www.mediawiki.org/ wiki/Extension:WikibaseLexeme/Data_Model. 18 In Turtle serialization, see, for example, https://www. wikidata.org/wiki/Special:EntityData/L1347.ttl. 19 See https://www.wikidata.org/wiki/Help: Data_type. 20 See https://www.mediawiki.org/wiki/Extension: WikibaseLexeme/Data_Model. 21 Discussions take place on-wiki (see https://www. wikidata.org/wiki/Wikidata_talk:Lexicographical_ data) and in a dedicated channel on the Telegram platform. 22 The pages dedicated to the 2021 and the 2024 Lexicodays provide video recordings, presentation slides and links to other pages containing lexicographical guidelines and descriptions of tools related to Wikidata lexemes, see https://www.wikidata.org/wiki/Wikidata:Events/ Lexicodays_2024 and https://www.wikidata.org/wiki/ Wikidata:Events/30_lexic-o-days_2021 23 See https://www.wikidata.org/wiki/Wikidata: Showcase_lexemes. entities on Wikidata, or to a data value (in table 2, excluding external id properties). Without going into much detail, we point out some of them; in some cases, the modeling is unambiguous (and straightforwardly alignable to Ontolex), and sometimes alternative modeling approaches coexist: • As it will be explained in section 2.5, translation equivalence is expressed using wd:P5972 , a property set (in both directions) directly between senses; the property used in Ontolex-Lemon for this purpose is vartrans:translation . In May 2025, Wikidata contained 119,847 translation links between senses of different languages. • For representing etymology, a lexeme is linked directly to the lexical entry describing the etymon using wd:P5191 . 24 In May 2025, Wikidata contained 40,540 etymology links from lexeme to lexeme. • For representing decomposition, wd:P5238 , the combines lexeme property links a compound or multiword entry to its constituents, in the same way decomp:subterm is used according to Ontolex. 25 This property is used in May 2025 207,799 times. • Pronunciation is described in multiple ways, but always associated to forms; the value of a general pronunciation property wd:P7243 is in general identical to the form representation, but may include stress indicators and other diacritics, e. g. to indicate vowel length. It is recommended that pronunciation audio files and/or IPA transcriptions are attached as qualifiers to the pronunciation claim, but audio files can also be directly attached to form. 26 Today, 74% of 252,652 pronunciation audio files are linked directly to a form. Only audio files hosted on Wikimedia Commons can be used.27 • As seen in table 2, properties devoted to certain transcription and transliteration sys24 See, for example, https://www.wikidata.org/wiki/ Lexeme:L630740#P5191. 25 See wd:L625224 for the German phrase "frohes neues Jahr". 26 For an example, see wd:L3338-F2 .Qualifiers as part of astatement are explained in section2.4 27 The property is of datatype Commons Media File, see https://www.wikidata.org/wiki/Help:Data_type# commonsMedia. tems exist; each of them is, like all Wikidata properties, described through its entity data, and on an own discussion page. 28 In parallel, a general property transliteration or transcription ( wd:P2440 ) can be used, and its value qualified using the property determination method or standard ( wd:P459 ). The former, defined as subproperties of wd:P2440 , are heavily used on Wikidata in general, 29 but in only about 1.2% of its use cases (i. e., not more than 3,808), that property is attached to forms, 30 on which the general property is used around ten times more often: The 70,165 uses of wd:P2440 , with a wd:P459 qualifier pointing to the transliteration system, are almost half-divided between items and forms.31 • Usage examples ( wd:P5831 ) are recommended to be attached to entry, and not to sense. However, about 7% of the 31,271 usage examples in Wikidata lexemes today remain attached to sense; the distribution across languages shows a diverse picture, but a clear preference for entry as subject of the property. 32 If attached to entry, one or more senses can be declared subject sense ( wd:P6072 ); this is done using that property as qualifier in the example statement. The advantage of this modeling, apart from being able to declare a usage example to be pertinent to more than one sense, lies in the ability to have examples attached to the entry also if in the moment of upload and without or before whatever sense disambiguation procedure it is not clear which sense should be marked as subject sense, e. g. when dealing with examples stemming from corpora, or if (still) no senses are described for the lexeme: As soon as the correct sense can be determined, the wd:P5831 claim is enriched with with a wd:P6072 qualifier, without having to delete and re-write the whole statement (with its references), and attach it to sense. Another strong reason for attaching 28 For example, about the property wd:P5825 ISO 15919 transliteration, a documentation is available at https://www. wikidata.org/wiki/Property_talk:P5825 , and about wd:P4187 Tibetan to Latin transliteration, https://www. wikidata.org/wiki/Property_talk:P4187. 29See https://w.wiki/EBWF. 30See https://w.wiki/EBYi. 31 See https://w.wiki/EBZ2 for usage counts according to the different transliteration systems. 32 See https://w.wiki/ECBm for the use of this property across languages. usage examples to entry is to enable their annotation with a subject form, i. e. the word form appearing in the example. 33 In addition, having examples at both levels complicates their retrieval in queries. The Wikidata lexemes collection is quantitatively described on dedicated pages, 34 and it can be explored using the Ordia tool (Nielsen,2019), 35 which generates dictionary-like exhibitions of lexical data; it also features a tool to look up Wikidata forms matching to tokens in text. Ordia also provides statistics on the Wikidata lexeme collection, such as, for example, lists of the most frequently used properties in the domains of entry (a. k. a. lexeme), sense and form. 36 The Synia tool also shows statistics on lexemes, e. g. counts of values for wd:P6191 language style in different languages.37 Table 3lists overall counts for the ten languages with best absolute coverage at the three levels. 38 Asking for relative coverages, without data about the total amount of corpus lemmata, corpus types, and dictionary senses in a language, we might ask for a relation between the coverage on Wikidata and the number of speakers of a language. A query like that (taking into account languages with more than 100,000 speakers) results in a different ranking, with Estonian, Breton, Basque and Danish leading lexemes, 39 and Basque, Breton, Dagbani and Norwegian Bokmål as top four languages for senses.40 Apart from those already mentioned, a range of tools 41 has emerged around Wikidata lexemes, designed to help creating entries and append senses, 42 33See, for example, the usage examples for wd:L87. 34 See https://www.wikidata.org/wiki/Wikidata: Lexicographical_data/Statistics 35Accessible at https://ordia.toolforge.org/. 36 Accessible at https://ordia.toolforge.org/ property/ . Wikidata SPARQL queries as used in Ordia can be modified for custom searches, e. g. for listing the property statistics for a single language, as in the following queries derived from those in Ordia:entry, https://w.wiki/E4pg ,sense, https://w.wiki/E4pw , Form, https://w.wiki/E4p$ ; note the filter for external id properties. 37 See https://synia.toolforge.org/ #languagestyle. 38 See up-to-date statistics at https://w.wiki/ECLq for lexeme, at https://w.wiki/EDkc for sense, and at https: //w.wiki/EDjh for form. 39See https://w.wiki/EDna. 40See https://w.wiki/EDnQ. 41 See also https://www.wikidata.org/wiki/ Wikidata:Tools/Lexicographical_data. 42 For a lexeme creation UI, see https://hangor. toolforge.org ; for python, in addition to the genLexicalEntry wd:P5185 grammatical gender wd:P5911 paradigm class wd:P5238 combines lexeme wd:P31 instance of wd:P5187 word stem wd:P1552 has characteristic wd:P5402 homograph lexeme wd:P2348 time period wd:P5191 derived from lexeme wd:P5186 conjugation class wd:P5831 usage example wd:P7486 grammatical aspect LexicalSense wd:P5137 item for this sense wd:P5972 translation wd:P5973 synonym wd:P1343 described by source wd:P18 image wd:P9488 field of usage wd:P6191 language style wd:P8394 gloss quote wd:P9970 predicate for wd:P6271 demonym of wd:P6084 location of sense usage wd:P10339 semantic gender Form wd:P7243 pronunciation wd:P898 IPA transcription wd:P443 pronunciation audio wd:P5279 hyphenation wd:P5825 ISO 15919 transliteration wd:P8881 ITRANS (Indic scripts) wd:P8530 alternative form wd:P5276 Slavistic Phonetic Alphabet transcr. wd:P10822 homophone form wd:P1721 Hanyu Pinyin transliteration wd:P11950 appears before phonological feature wd:P11951 appears after phonological feature Table 2: The 12 most frequently used properties describing Wikidata lexemes, senses and forms (May 2025) for linking senses to ontological references, 43 for creating forms collections through templates, 44 searching for usage examples and adding them to eral wikibaseintegrator library ( https://github.com/ LeMyst/WikibaseIntegrator ), specially for lexemes, tfsl ( https://gitlab.wikimedia.org/toolforge-repos/ twofivesixlex.) 43See https://lexica-tool.toolforge.org/. 44 See https://www.wikidata.org/wiki/Wikidata: Wikidata_Lexeme_Forms. LexicalEntry language LexicalSense language Form language 1,422,331 all Wikidata 585,893 all Wikidata 14,396,207 all Wikidata 239,465 German 48,308 Bokmål 2,802,106 Estonian 102,150 Russian 41,973 English 1,257,083 Basque 96,753 Danish 30,756 Basque 1,246,745 Russian 83,218 Estonian 29,790 Nynorsk 1,199,194 Latin 68,758 English 24,127 Czech 871,813 Czech 67,367 Malayalam 24,055 Swedish 753,118 Malayalam 64,445 Italian 22,000 Italian 692,671 Spanish 63,128 Spanish 21,036 Japanese 641,454 Danish 56,296 Latin 20,484 Persian 571,091 German 48,214 Swedish 18,851 Egyptian 522,107 Italian Table 3: Absolute coverage of some languages in Wikidata lexemes (May 2025) entries, 45 for massively recording pronunciation audios, 46 or for graph visualisations of lexical relations.47 One goal of the Wikidata lexemes collection is to enable natural language generation for drafting Wikipedia article text from abstract knowledge representations (Vrandeˇ ci´ c,2021;Morshed,2024), 48 and another possible application is corpus annotation (Lindemann and Alonso,2024); all these depend on the degree the lexemes, senses and forms in the collection cover the languages to process. 2.3 Wikibase lexemes as linking hub: The case of Wikidata In addition to these and other properties for the linguistic description of the lexeme, Wikibase lexical entries contain pointers to external resources encoded as external id properties. These lead a human user to an entry or sense description in a dictionary web portal. In some cases, federated database queries can access content in several graph databases at a time using such external ID. For example, a query can involve Wikidata and the LiLa Latin Knowledge Base (Passarotti and Mambrini,2022), 49 exploiting the LiLa URI attached to Wikidata Latin lexemes, and calling the LiLa SPARQL endpoint from within the Wikidata Query Service, or vice versa. 45 The Luthor tool uses Wikisource content as corpus, see https://luthor.toolforge.org/. 46See https://lingualibre.org/. 47 For an example, an etymological network, see https://lucaswerkmeister.github.io/ wikidata-lexeme-graph-builder/?subjects=L184995& predicates=P5191. 48 See https://meta.wikimedia.org/wiki/Abstract_ Wikipedia. 49 See https://lila-erc.eu ; the LiLa ID property is wd:P11033. Wikidata lexemes, on entry level, by May 2025 count 2.3 Million external id statements. Table 4 shows usage counts for the most frequent 20 properties. 50 As for sense, all lexical resources aligned to Wikidata on sense level today still lack significant coverage, 51 although we point out the fact that e. g. English Wordnet synset identifiers are aligned to Wikidata items (McCrae and Cillessen,2021), which, as explained, are referenced by a significant number or senses. An alignment of forms to external resources such as corpus-based form repositories is at large still not present, although it would be interesting, for instance in rich-morphology languages, where the morphologically possible forms outnumber the forms that are actually attested in corpora, so that forms attestation is very valuable information, with a similar value lemma attestation has in languages with a comparably reduced number of different inflected forms, like English. 2.4 Reification in Wikibase Wikibase statements include by default a mechanism for further describing the main claim of a statement using qualifiers,ranks, and references. A graphical model of a Wikibase statement is given in figure 2, where "entity" represents an item,lexeme, sense,form, or property node, each of which has its own URI in the main entity namespace of the Wikibase instance, 52 and where the blue-colored "value" nodes, depending on the property datatype, represent entities of the same Wikibase, or data val50For up-to-date counts, see https://w.wiki/E9j3. 51 An example for one of the most used external identifiers is DWDS sense ID, see https://www.wikidata.org/wiki/ Property_talk:P12550. 52 On Wikidata, e. g. wd:Q1 for an item, wd:L1 for a lexeme, wd:L1-S1 for a sense, wd:L1-F1 for a form, and wd:P31 for a property. Count Property Property label (en) Lang. (ISO-639-1) 231,859 wd:P9940 DWDS-Lemma-Identifikator de 153,322 wd:P8376 Duden-Lexem-Identifikator de 139,067 wd:P11519 elexiko ID de 122,350 wd:P11138 Sõnaveeb entry ID (45 lang.) 84,951 wd:P9947 WDG-Lemma-Identifikator de 66,512 wd:P13258 Presisov veˇcjeziˇcni slovar ID fr de en sq sh 65,599 wd:P9529 Den Danske Ordbog article ID da 61,333 wd:P5912 Oqaasileriffik online dictionary ID da en kl nb 54,755 wd:P12630 Aragonario ID (6th version) es an 51,566 wd:P11033 LiLa Linking Latin URI la 49,987 wd:P9385 DWB Lemma ID de 45,751 wd:P5275 OED Online ID en 39,052 wd:P9962 Ordbog over det danske sprog ID da 37,925 wd:P12420 Il Nuovo De Mauro ID it 37,137 wd:P10042 Bokmålsordboka-ID nb nn 36,535 wd:P11838 Svenska Akademiens ordlista ID sv 35,124 wd:P9387 GWB Lemma ID de 31,014 wd:P12690 New Oxford American Dictionary ID en 29,803 wd:P11319 Little Academic Dictionary ID ru 29,316 wd:P12828 DAKA Danish-Greenlandic Dictionary ID da Table 4: Most frequently used external id properties on Wikidata lexemes, and lexeme languages Figure 2: Graphical model of a Wikibase Statement ues, including strings, external identifiers, date objects, globe coordinates, et cetera. Each statement node may be linked to several qualifier values, and to several reference nodes (in the RDF representation attached to the statement blank node using prov:wasDerivedFrom ), which make up blocks of references. Ranks are used for annotating multiple statements for the same property with one of the three values normal,preferred, or deprecated.53 In order to keep all data pertaining to the same lexical entry "together", so that it would all get stored in the same entity data JSON blob, 54 and displayed on the same lexeme entity page, it is convenient to keep the modeling of the linguistic description as shallow as a list of Wikibase statements. Unlike in the Ontolex modules, where the linguistic description often involves several reification layers, the model followed in Wikidata, and, for the same reason, in any other Wikibase, will try, wherever possible, to stay with certain reification approaches, which can be combined, but will be limited to the following: • Using statement qualifiers, i. e. semantic triples describing the main claim of the statement. • Using references for provenance annotations. • Using subproperties such as wd:P1721 53 See https://www.wikidata.org/wiki/Help: Ranking. 54 At https://www.wikidata.org/wiki/Special: EntityData/L1.json , the entity data JSON for wd:L1 as stored in the database, and from which the entity page display is generated, can be obtained. Hanyu Pinyin transliteration, subproperty of wd:P2440 transliteration or transcription. If we compare, for example, how a translation relation between two senses can be further described, according to Ontolex this can be modeled introducing a blank node of class vartrans:Translation into the dataset (option A), linking that to both sense nodes using typed properties (source and target), and attaching to that node supplemental information about the translation relation (Bosque-Gil et al.,2015). As a shallower alternative (option B), which does not involve any additional blank node in the structure, and consequently allows no additional description of the translation relation, Ontolex uses the vartrans:translation property. 55 The second option is more suitable for a Wikibase, because the additional node in option A, since it does not fit into the Wikibase statement structure, would have to be created as named individual entity, i. e. a Wikibase item with its own Q-identifier, its labels to be indexed in the Wikibase ElasticSearch, its class declaration, and descriptions. However, in a Wikibase, option B caters for a description as rich as option A in Ontolex: In the discussed example, the translation relation type, its direction, or any other translation restriction feature can be expressed using a qualifier on the translation claim. This, in turn, is not possible when using option B in Ontolex, since the semantic triple describing the translation relation there cannot be further described or qualified. 2.5 Lexicon-Ontology interface and multilinguality According to Ontolex-Lemon, a property named ontolex:reference links a lexical sense to an ontological concept (Bosque-Gil et al.,2015). On Wikidata, wd:P5137 item for this sense has the equivalent function. For example, Wikidata’s lexeme wd:L3549 , describing the English noun foot, has three senses, each of them pointing to a different conceptual item node in the graph, using that property. wd:L3549-S1 , the first listed sense, is linked to an item describing a unit of length, while the second sense points to an item describing a furniture part, and the third sense, wd:L3549-S3 , glossed as “anatomical structure found in vertebrates”, links to the anatomical entity. Each of the three linked Wikidata items is 55 See examples with figures for both at https://www.w3.org/2016/05/ontolex/ #translation-as-a-relation-between-lexical-senses . itself further described, for example, with links to Wikipedia articles in multiple languages (the property used here is schema:about ), which have a title in that language, and which provide encyclopaedical descriptions of the concept. That means in general terms that Wikibase provides a framework where lexical and ontological (conceptual) descriptions converge, and where text pages (for Wikidata, Wikipedia articles) about concepts also have their habitat. Since, in addition, Wikibase items are annotated with multilingual labels ( rdfs:label and skos:altLabel ) and descriptions ( schema:description ), the wd:P5137 reference of a lexeme sense into the ontological part of the Wikidata graph already provides three facets of multilinguality: The labels, textual concept descriptions, and entire text pages attached to an ontological item referenced by lexeme senses may cover multiple languages. Since several languages’ word senses can be linked to the same item, wd:P5137 provides translation (and, inside the same language, synonymy) information. In Ontolex, this way of modeling translation relations is referred to as translation as shared reference. 56 As of May 2025, Wikidata contains 227,908 item for this sense claims that link senses to items. 57 Calculating the number of translation links through shared references for every item, that sums 5.23 million translation (and intralingual synonymy) links between lexeme senses, counting the connecting links twice, i. e. as translation link in both directions. 58 This by far outnumbers wd:P5972 sense translation relations (see section 2.2), which constitute an alternative without leaving the domain of lexemes, as needed for senses without an ontological reference in Wikidata (most prominently, senses of lexemes with a lexical category other than noun). As described above, multilingual sense glosses (very short sense descriptions attached using the built-in skos:definition ) provide another facet of multilingual sense description. Fig. 3shows an example for a lexeme’s relations in the graph, including monolingualtext values in two languages: the lexical entry describing a German noun Pferd ("horse"), linked to entities of dif56 See https://www.w3.org/2016/05/ontolex/ #translation-as-shared-reference. 57 See the distribution of item for this sense across languages at https://w.wiki/ECCA. 58 See https://w.wiki/EcDh for a list of all items linked to from senses, and the corresponding number of translation links (includes intralingual translation, i. e. synonymy). Figure 3: Wikidata entry wd:L34708 describing the German lemma Pferd and some relations ferent types: ontological concepts (items), other lexical entries (lexemes), and lexical senses. Entities of type item are used to represent the language, which allows querying for lexemes according to language features such as the language family or countries where languages are spoken or native to, to represent the lexical category, 59 and, as explained above, for ontological sense references. Label,definition and lemma values are always strings associated to a language code (figure 3only shows English and German, while the cited entities on Wikidata cover more languages here). 2.6 Wikibase as an infrastructure for lexical datasets In terms of the FAIR Guiding Principles for scientific data management and stewardship 60 , a lexical dataset on a Wikibase can safely be called to be state of the art, since permanent URI on the level 59 On Wikidata, a range of 320 different items is used here, some defined as subclasses of others, e. g. wd:Q1166153 intransitive verb subclass of wd:Q24905 verb; see https: //w.wiki/ECKP for use counts. 60See https://www.go-fair.org/fair-principles/ of the three Ontolex-Lemon core classes assure findability, answering human user calls with the display of an editable entity page, and programmatical calls with machine-readable data in various formats. 61 Accessibility is furthermore given through the graphical query service, 62 and through a SPARQL endpoint. Interoperability is sustained by re-using W3C-recommended RDF vocabularies for a set of basic classes and properties defined in the Wikibase Ontology, such as Ontolex-Lemon for lexical data. Finally, reusability is ensured by open licenses.63 Wikimedia Deutschland, the WMF chapter responsible for Wikidata, is providing the Wikibase software for self-hosting, 64 and also provides a 61 Entity data dumps are available in JSON and TTL format. 62 For Wikidata, accessible at https://query.wikidata. org/. 63 Wikidata declares its terms of use at the bottom of every displayed page, and in detail at https://foundation.wikimedia.org/wiki/Policy: Terms_of_Use , and any other Wikibase may contain declarations in a similar form. 64 See https://www.mediawiki.org/wiki/Wikibase/ Docker.