This paper presents the publication as Linked Open Data of a set of coreference and anaphora annotations (called CorefLat) performed on a set of Latin texts. Annotations are made on texts already available as Linked Open Data as part of the LiLa Knowledge Base of interoperable linguistic resources for Latin. By adopting a lemma-centered architecture and established guidelines for annotation inspired by those of the GUM corpus, CorefLat systematically identifies and tags entities and mentions, creating relational links. The annotated corpus covers multiple periods and genres, including Augustine's Confessiones, Plautus' Curculio, Caesar's De Bello Gallico, and Seneca's Medea, ensuring a balanced dataset for broader linguistic analysis. The publication of CorefLat as Linked Open Data relies on an OWL ontology that extends the POWLA framework, thus enabling interoperability with diverse linguistic resources within LiLa. We detail how coreference relations, including phenomena such as anaphora, cataphora, split antecedents, and multiword units, are encoded through specialized classes and object properties.
Coreference Resolution and Anaphora Resolution (CR and AR) have been central to Natural Language
Processing (NLP) since the 1960s, but were long considered complex tasks requiring advanced knowledge
and inference tools. In 1983, Roberto Busa highlighted the lack of research work for pronoun resolution,
asking if tools existed to “automatically link pronouns to their antecedents” [
By the 1990s, following the empirical turn that hit the NLP world in those years, research on automatic
CR and AR shifted to stochastic approaches based on machine learning algorithms. Such turn was
possible thanks to the development of corpora enriched with CR/AR annotations, through initiatives
such as the Message Understanding Conference (MUC) [
While most resources focus exclusively on practical or informational texts, some are specifically
dedicated to the study of literary texts, such as DramaCoref, a neural network system for CR on German
theater plays [
For Classical languages, fundamental resources are the Ancient Greek and Latin Dependency Treebank (AGLDT),3 which includes excerpts from Ancient Greek and Latin texts of the Classical era, and the Index Thomisticus Treebank (IT-TB),4 which features Medieval Latin texts of Thomas Aquinas.
The syntactic annotation of both corpora was originally based on a scheme resembling that of the analytical layer of the Prague Dependency Treebank (PDT) [15]. Both the treebanks feature a small subset of data annotated at the so-called tectogrammatical layer of the PDT [16, 17, 18]. This annotation layer captures the underlying syntactic structure of sentences (while the analytical layer represents surface syntax). Through tectogrammatical annotation, the treebank is enhanced with a range of annotation tasks, including semantic role labeling, information structure, and ellipsis and coreference resolution.
As for Latin, which is the focus of this paper, approximately 45,000 tokens out of the AGLDT and the IT-TB are available enhanced with tectogrammatical annotation, covering excerpts from Sallust, Caesar, Cicero (AGLDT), and Thomas Aquinas (IT-TB). Despite its significance, this set of CR/AR annotations for Latin remains unbalanced, with more than half of its tokens originating from Aquinas’ Summa contra Gentiles (approx. 27,000 words) and Sallust’s In Catilinam (approx. 10,936 words) [19]. To mitigate this imbalance, we developed CorefLat, a more diverse and balanced set of CR/AR annotations for Latin, which will ultimately incorporate a broader selection of Classical and Late Latin texts.
Since the CR/AR annotations provided by CorefLat are intended for publication as Linked Open Data (LOD), they were performed on a corpus of Latin texts that is already available as LOD. This corpus is part of the LiLa Knowledge Base of interoperable linguistic resources for Latin published as LOD.5 Annotating directly within LiLa ensures seamless interoperability with the (meta)data of other resources published therein.
This paper outlines the process of publication as LOD of the CR/AR annotations provided by CorefLat. The paper is structured as follows. Section 2 ofers a brief introduction to the LiLa Knowledge Base; Section 3 presents an overview of CorefLat’s annotation guidelines. Section 4 describes the ontology used to describe the data and for the publication of CorefLat as LOD (4.1), details a few examples that informed its development (4.2), and illustrates a case study showing the research potential of integrating CorefLat into LiLa (4.3).
The LiLa - Linking Latin project [20] was awarded an ERC Consolidator Grant (2018-2023) to integrate existing linguistic resources for Latin in a Knowledge Base to ensure their online interoperability.
The LiLa Knowledge Base (KB) was developed upon established standards for the publication of data in the Semantic Web, fitting the principles of the so-called Linked Data paradigm [ 21]. Accordingly, each data point in the linguistic resources interlinked in the KB is assigned a unique and persistent URI (Uniform Resource Identifier) published on the Web as URL using the HTTP protocol, to ensure its findability and accessibility. By employing web standards such as the RDF (Resource Description Framework) data model [22] and the SPARQL query language,6 LiLa facilitates the creation of links between distinct URIs and the reuse of data.
The LiLa KB leverages a few existing ontologies to represent the (meta)data of the Latin resources interlinked therein. Key ontologies integrated into the KB include POWLA for corpus data (Portable Linguistic Annotation with OWL, an ontology designed to express any textual data and metadata as LOD) [23], OLiA for linguistic annotation (Ontologies of Linguistic Annotation, a set of ontologies that allow to express and map linguistic categories) [24], and Ontolex-Lemon for lexical data [25].
In the architecture of the LiLa KB, lemmas play the core role, as the pivotal connection points among both lexical and textual resources. Such a highly lexically-based architecture is based upon a simple yet eficient assumption: textual resources consist of word occurrences (tokens) and lexical resources describe word properties in lexical entries. Following the lemma’s foundational role in LiLa, the KB is based on the so-called Lemma Bank, a collection of approximately 200,000 Latin lemmas (canonical citation forms of lexical items) published as LOD. This collection originates from the lexical base of the LEMLAT 3.0 morphological analyzer for Latin [26] and it is constantly extended as far as new resources are added to the KB. Interoperability is achieved by linking all lexical entries and corpus tokens to their corresponding lemma in the Lemma Bank, thus enabling seamless integration across resources.
This section provides an overview of the CorefLat data set, with a particular focus on the guidelines that informed the annotation process.7
The annotation process was performed following the guidelines of the GUM corpus,8 which are also employed in the Universal Anaphora (UA) project,9 aiming for consistency across linguistic resources enhanced with CR/AR annotation. In coreference annotation, a distinction is made between Entities, which are referred, and Mentions, which refer back or forward to an Entity. CorefLat’s approach emphasizes relationships rather than chains of Entities and Mentions. The diference between coreference relations and coreference chains lies in their scope and structure: • coreference relations refer to the specific linguistic connection between two or more expressions that refer to the same entity in a discourse. For example, in the sentence “Maria loves her cat. She takes good care of it”, the pronoun she is in a coreference relation with Maria, and it refers to her cat. • coreference chains are sequences of multiple referring expressions that all refer to the same entity throughout a text. A chain is made up of multiple coreference relations. For example, in “Maria loves her cat. She takes good care of it. The feline enjoys playing with her”, the chain consists of (Maria → she → her), and (her cat → it → the feline). 6https://www.w3.org/TR/rdf-sparql-query/. LiLa’s SPARQL endpoint can be accessed at https://lila-erc.eu/sparql/. 7Annotation was performed manually using the customizable Content Annotation Tool (CAT). (Meta)data were first saved in XML and then converted automatically into the CoNLL-U Plus format (https://universaldependencies.org/ext-format.html), following the recommendations provided by the UA project [27]. 8https://wiki.gucorpling.org/gum/entities. 9https://universalanaphora.github.io/UniversalAnaphora/.
In CorefLat, we annotate coreferences as relations and we select a limited set of fundamental types of coreference. Such a limited set aligns with our objective of building a foundational Latin corpus enhanced with coreference annotations. In examples (1) to (4) we provide a detailed review of the types of annotations available in CorefLat. (1)
Anaphora: a mention referring back to an entity. This type of coreference constitutes the most frequently represented category in our corpus, accounting for 1,222 instances out of the 1,520 annotated coreferences within the texts. 10 a. domine qui et semper vivis.
‘Lord (you) who live for ever’.11 (Aug. Conf. 1.6.8) b. Laudes tuae, domine, laudes tuae per scripturas tuas suspenderent palmitem cordis mei. ‘Your praises, Lord, your praises throughout your Scriptures would have supported the vine shoot of my heart’.
(Aug. Conf. 1.17.27) c. Quo usque, quaeso, ad hunc modum / inter nos amore utemur semper surrupticio? ‘How much longer, please, will we always conduct our love afair in secret?’ 12 (Pl. Curc. 1, 204-205) Pronominal anaphora are the prototypical case of anaphora, where the mention is represented by a pronoun, like the relative pronoun qui in (1a). Beside this type of anaphora, we annotate one type of anaphoric relation involving two (identical) content words, where the latter refers back to the former, as in (1b). Thus, the first utterance of the content word functions as the entity, while the second serves as the mention.
During the annotation process we got through cases of coreference relation where the entity is implicit, leading to the mention being anaphorically linked to an external entity (cf. 4.1). The personal pronoun nos in (1c) refers to Planesium and Phaedromus, two of the main characters from Plautus’ comedy Curculio. These characters, however, are not mentioned in close textual proximity to the pronoun. This is an instance of long-distance coreference, a phenomenon that presents a challenge in CR/AR, as there is no strict upper limit on the number of sentences after which a mention can no longer be linked to its entity [30]. However, modern NLP methods have proven highly efective in addressing this issue, successfully linking mentions to their entities across spans exceeding 200 sentences [31]. Literary texts, the primary focus of CorefLat, seem to exhibit long-distance coreference more frequently than other textual genres [32], thus making it crucial to devote particular attention to this phenomenon. To ensure consistency in annotation, we set a threshold: when a mention exceeds five sentences from its entity, it is connected to an external entity. This threshold is sentence-based rather than token-based, aligning with the standard practice in CR/AR studies, where sentences serve as the primary unit of analysis.
Cataphora: a mention referring forward to an entity. This type of coreference is the second most frequent in our corpus, with 177 occurrences, out of the total of 1,520 coreferential relations annotated in the texts. invocat te, domine. ‘invokes you, Lord’ (Aug. Conf. 1.1.1) Split antecedents: the mention has multiple antecedents, so one mention refers back or forward to more than one entity. In CorefLat, this linguistic phenomenon is underrepresented, with only 87 occurrences of the 1,520 annotated coreferential relations in the texts. However, it might (2) a. (3) 10Mentions referring to external entities are excluded from these counts, see (1c) and the corresponding discussion for further details. 11All translations of Augustine’s Confessiones are taken from [28]. 12All translations of Plautus’ Curculio are taken from [29]. (4) manifest in two structural patterns: as mentions referring to either conjoint (3a) or disjoint noun phrases (3b) and as mentions referring to previously listed nouns (3c). a. An vero caelum et terra, quae fecisti et in quibus me fecisti, capiunt te? ‘Heaven and earth, which you made, and in which you made me, encompass you?’ (Aug. Conf. 1.2.2) b. Nec mater mea vel nutrices meae sibi ubera implebant, sed tu mihi per eas dabas alimentum infantiae. ‘Neither my mother nor my nurses filled (their) breasts for themselves, but you gave the nourishment of infancy to me through them.’ (Aug. Conf. 1.6.7) c. Gallia est omnis divisa in partes tres, quarum unam incolunt Belgae, aliam Aquitani, tertiam qui ipsorum lingua Celtae, nostra Galli appellantur. Hi omnes lingua, institutis, legibus inter se diferunt. ‘Gaul is a whole divided into three parts, one of which is inhabited by the Belgae, another by the Aquitani, and a third by a people called in their own tongue Celtae, in the Latin Galli. All these are diferent one from another in language, institutions, and laws.’ 13 (Caes., Gal., 1.1.1) Multiword antecedents: the entity involved in the coreference relation consists of more than one token. In CorefLat, this type of coreference is the least represented, accounting for only 34 occurrences of the 1,520 coreferential relations annotated in the texts. However, it primarily occurs in two contexts: when the entity is a proper noun following the Roman onomastic system (tria nomina), as illustrated in (4a), or when the entity is linguistically realized as a noun phrase consisting of a noun and a modifier or specifier that semantically restricts the reference of a noun by specifying a subset of possible referents (4b). a. Itaque prius quam quicquam conaretur, Diviciacum ad se vocari iubet et, cotidianis interpretibus remotis, per C. Valerium Troucillum, principem Galliae provinciae, familiarem suum, cui summam omnium rerum fidem habebat, cum eo conloquitur ; ‘Therefore, before attempting anything in the matter, Caesar ordered Diviciacus to be summoned to his quarters, and, having removed the regular interpreters, conversed with him through the mouth of Gaius Valerius Trocillus, a leading man in the Province of Gaul and his own intimate friend, in whom he had the utmost confidence upon all matters.’ (Caes., Gal., 1.19.3) b. Comprecor vulgus silentum vosque ferales deos et Chaos caecum atque opacam Ditis umbrosi domum. ‘I invoke the thronging silent dead, and you the gods of the grave, and sightless Chaos, and the shadowy home of dark-enshrouded Dis.’14 (Sen. Med., 740-741) In (4b), ferales deos functions as multiword antecedent of the pronoun vos. In this instance, the semantics of the noun phrase is constrained by a modifier. In addition to illustrating a case of a multiword antecedent, (4b) also demonstrates a cataphoric relationship between an entity and a mention, as the pronoun vos precedes the noun phrase.
To ensure the applicability of the annotation guidelines explained above across diferent linguistic
and stylistic contexts, a diversified set of texts was selected, comprising works from various genres and
periods to provide a balanced representation of Latin traditions.15 So far, CorefLat includes a passage
from a Late Antique philosophical work (the first book of Augustine’s Confessiones), an archaic comedy
13All translations of Caesar’s De Bello Gallico are taken from [33].
14Seneca Medea’s translations are taken from [34].
15The decision to diversify the corpus by including texts from diferent genres is also driven by the interest in analyzing how
the coreference phenomenon occurs and manifests across various textual domains. For initial observations on this topic, see
Delfino et al. [
The workload was equally distributed between two annotators. To assess inter-annotator agreement,
however, both annotators annotated the final 50 sentences of the first book of Augustine’s Confessiones.
Agreement was measured using the Dice coeficient, a similarity metric widely employed in NLP ([
This section explains how we modeled the information in the coreference annotation. The adopted solutions aim to link the annotated text to the LiLa KB, and to ensure the semantic interoperability of the coreference annotation with the other Linguistic Linked Open Data in LiLa.
The LiLa CorefLat Ontology is an OWL ontology that extends the POWLA framework [23] shared
also by the other annotated corpora in LiLa [
The POWLA ontology defines four primitive concepts to describe corpora: documents, layers, nodes
and relations. While the former two are used in the LiLa corpus ontology to structure the texts and model
the structural metadata [
In POWLA, every unit of linguistic analysis is defined as an instance of the class Node. In our framework, coreference units often span across multiple tokens, as in (4a) and (4b), and are therefore best conceptualized as phrases encompassing one or more tokens in the text. The class of powla:Nonterminal, the subclass of Node used for phrases or chunks disjoint from the class of Terminal (used for actual base segments of texts), is once again very well suited to express the concept. The classes of Coreference Relation and Coreference Units are thus defined as subclasses of both CorefLat’s top concept Coreference Element and of the POWLA classes Node and Relation.
The classes of Entity and Mention, on the other hand, do not align with concepts in POWLA, and were abstracted from the guidelines described in Section 3. In our ontology, they are defined as disjoint subclasses of Coreference Element only that allow annotators to further specify the role of each coreference unit in the coreference relation, whether they serve as referred (entity) or the referring element (mention).
The object properties of the CorefLat ontology serve the purpose of expressing the relations between the corpus tokens and the coreference units, as well as the role of the units with the reified coreference 16The Classical Latin data originate from the Opera Latina corpus by LASLA, which contains over 1.7 million words from both Classical and Late Latin texts (https://lasladb.uliege.be/OperaLatina/), while Late Latin examples are sourced from The Latin Library http://www.m.thelatinlibrary.com/. 17The ontology is available at https://lila-erc.eu/lodview/ontologies/lila_coref/. 18http://lila-erc.eu/ontologies/lila_coref/CoreferenceElement. 19http://lila-erc.eu/ontologies/lila_corpora/. 20http://purl.org/powla/powla.owl#Relation. relation. The property hasCoreferenceTerminal,21 in particular, connects the coreference unit with each of the corpus tokens that make up the phrase.22 The properties hasCoreferenceSource and hasCoreferenceTarget,23 subproperties of powla:hasSource and powla:hasTarget, link the reified coreference relation to, respectively, the source and the target of the directed edge. Moreover, the ontology defines two additional properties hasMention and hasEntity as subproperties of, respectively, hasCoreferenceSource and hasCoreferenceTarget. These two properties define, respectively, the classes Mention and Entity as their range and presuppose and enforce a stricter interpretation of a coreference relation as a directed edge going from a mention to an entity. Users of the ontology are free either to adhere to the stricter interpretation or to adopt a looser model of relation that only involves coreference units.
Finally, in order to facilitate the harmonization and the recognition of the various entity-type coreference units, while also enabling cross-document coreference tasks in the future, we decided to introduce an extra-textual node linked to the entities via the itsrdf:taIdentRef property of the Internationalization Tag Set (ITS) Ontology.24 This node serves as an aggregator for all the entities present within the text, and provides a connector between the textual element and the extralinguistic entities and concepts referred to in texts.
Indeed, in addition to simplifying queries within the document by grouping all entity-type Coreference Units under distinct nodes, this extra-textual node also enables the expansion and interoperability of the annotated resource with other knowledge sources. The extra-textual entity nodes created by the project form a separate knowledge base of entities referred to in the annotated texts. These entities can be mapped to other encyclopedic resources, like DBpedia,25 or Wikidata,26 via e.g. the mapping properties of the Simple Knowledge Organization System (SKOS), such as skos:exactMatch. In this way, it becomes possible to connect the various “supernodes” to external gazetteers, and to enrich the nodes with properties that are transitively inherited from these external resources.
This Section illustrates how the modeling of the data by CorefLat applies, by detailing the representation of the examples (1)-(4) presented above.27 To begin with, in Example (1a), we annotated the token qui as an anaphoric reference to the token domine. Figure 1 visualizes how this relation is modeled using the classes and properties defined in our ontology. The visualizations are generated using an instance of the web application LodLive running on a server of the LiLa project.28
The token domine (yellow in Figure 1) is the object of the property hasCoreferenceTerminal, whose subject is the CoreferenceUnitEntity for domine (orange in Figure 1). The same applies to the token qui (yellow as token domine29), which is the object of the property hasCoreferenceTerminal, whose subject is the CoreferenceUnitMention for qui (lilac in Figure 130). Those coreference units are related through the reification of their relationship, which is represented by a node. This node is of type CoreferenceRelation (burgundy in Figure 1) and serves as the subject of two properties: (i) hasCoreferenceSource, which has as its object the CoreferenceUnitMention for qui, and 21http://lila-erc.eu/ontologies/lila_coref/hasCoreferenceTerminal. 22Note that the linear order of the tokens in the textual resources published in LiLa is captured thanks to POWLA’s symmetric relations next and previous that connect the text nodes in a chain. The sequence of the tokens within the coreference units can thus be expressed using these two properties of POWLA. 23http://lila-erc.eu/ontologies/lila_coref/hasCoreferenceSource; http://lila-erc.eu/ontologies/lila_coref/hasCoreferenceTarget. 24https://github.com/w3c/itsrdf. 25https://www.dbpedia.org/. 26https://www.wikidata.org/. 27Examples (1b) and 1 will not be explained in detail, as they follow the same modeling as Example (1a). Likewise, Examples (3b) and (3c) mirror (3a), while Example (4b) aligns with (4a). 28https://lila-erc.eu/lodlive/. 29In LODLive every class or property is represented through the same colour, although the colours are not pre-established, so they might change from a visualization to another. 30It has to be noted that this node has a diferent colour from the CoreferenceUnit for domine, because qui is also an instance of Mention, while domine is an instance of Entity. (ii) hasCoreferenceTarget, which has as its object the CoreferenceUnitEntity for domine. Finally, both tokens are linked to their corresponding lemma in the Lemma Bank via the property lila:hasLemma (both in purple in Figure 1).31
Example (3a) illustrates a case of split antecedent, in which a single mention (quae) refers to two distinct entities (caelum and terra). The challenge posed by this structure lies in the fact that the same mention establishes a coreference relationship with two diferent entities. The approach used in our ontology to model such cases is depicted in Figure 2.
In Figure 2, the token quae (yellow node) is the object of the property hasCoreferenceTerminal, whose subject is the CoreferenceUnitMention for quae (lilac in Figure 2). This coreference unit 31http://lila-erc.eu/ontologies/lila/hasLemma. is, in turn, the object of the property hasCoreferenceSource for two distinct coreference relations: CorefRelation quae → caelum and CorefRelation quae → terra. Since these coreference relations belong to the same class, they share the same color in Figure 2 (burgundy). The first coreference relation is the object of the property hasCoreferenceTarget, whose subject is the CoreferenceUnitEntity for caelum (orange in Figure 2). Similarly, the second coreference relation is the object of the property hasCoreferenceTarget, whose subject is the CoreferenceUnitEntity for terra (also orange in Figure 2). Both coreference units serve as subjects of the property hasCoreferenceTerminal, with their respective objects being the tokens caelum and terra (both yellow in Figure 2).
Example (4a) (per Caium Valerium Troucillum [. . . ] cui. . . ), instead, highlights the challenge of representing an entity consisting of multiple tokens within our ontological framework. This example demonstrates the importance of an abstract conceptualization of coreference relations based on coreference units rather than tokens, given that Caius Valerius Troucillus represents the typical three-element structure of the Roman onomastic system. If the coreference relation had been established between tokens, it would not have been possible to distinguish a ‘multiword’ case such as this one in (4a) from a split antecedent case like the one observed in (3a). This example is modeled in the same way as the one in (1a), with the only substantial diference being that the CoreferenceUnitEntity is the subject of the property hasCoreferenceTerminal three times: the object of the first one is the token Caium, the object of the second is the token Valerium, and the object of the third is the token Troucillum.
This section examines an example of research opportunities facilitated by linking our resource CorefLat to the LiLa Knowledge Base.
Linking CorefLat to LiLa enables interoperation between CorefLat and all the resources already integrated therein. For instance, it is particularly interesting to observe how CorefLat can interact with lexical resources available in LiLa, such as the Latin WordNet.32 32http://lila-erc.eu/data/lexicalResources/LatinWordNet/Lexicon
Figure 3 illustrates an example of how a token, domine, linked to the lemma dominus, from Augustine’s Confessiones, is the terminal belonging to a Coreference Unit of type Entity, which is the target of a Coreference Relation. Following the lemma-centred architecture of LiLa, lexical entries in the Latin WordNet are linked to the corresponding lemma in the LiLa Lemma Bank (as the canonical form of citation for the entry). As such, the lexical entry for dominus evokes a set of synsets.
Generalizing from this example, it is possible to formulate a SPARQL query to retrieve all synsets evoked by lexical entries associated with lemmas linked to tokens involved in a coreference relation. This query allows for the extraction of lemmas and their corresponding synsets in a two-column format, while also providing, for each lemma, the number of tokens involved in a coreference relation, as shown in Figure 4. See the appendix for a visualization of the SPARQL query output and the corresponding code listing.
This work has introduced CorefLat, a new resource designed to support coreference and anaphora analysis in Latin and to ensure smooth data integration within the LiLa Knowledge Base.
By leveraging existing standards for Linked Open Data, CorefLat enables wider interoperability and fosters cross-resource research. Future developments will center on (a) increasing the scope of the annotated corpus to include a broader range of textual genres and historical periods, and (b) exploiting the expanded dataset to train automatic CR/AR models for Latin, evaluating their performance both on in-domain and out-of-domain material.
To conclude, it would be advisable to annotate with coreference texts that have already been enriched with syntactic annotation in accordance with the Universal Dependencies guidelines. Specifically, regarding Classical Latin, we will make use of the UD Latin-Circse,33 a treebank repository currently under development by the CIRCSE Research Centre in Milan. The repository contains both prose and poetry texts from diferent periods. At present, it includes three texts taken from the Opera Latin corpus by LASLA, namely Seneca’s Hercules Furens and Agamemnon, and Tacitus’ Germania.
This contribution is funded by the PRIN-2022 project “Textual Data and Tools for Coreference Resolution in Latin” (CUP J53D23013680008), a project carried out jointly by the Università Cattolica del Sacro Cuore in Milan and by the University of Udine. 33https://github.com/UniversalDependencies/UD_Latin-CIRCSE
During the preparation of this work, the authors used X-GPT-4 for grammar and spelling check. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content. plays, in: M. Ogrodniczuk, S. Pradhan, M. Poesio, Y. Grishina, V. Ng (Eds.), Proceedings of the Fourth Workshop on Computational Models of Reference, Anaphora and Coreference, Association for Computational Linguistics, Punta Cana, Dominican Republic, 2021, pp. 36–46. URL: https: //aclanthology.org/2021.crac-1.4/. doi:10.18653/v1/2021.crac-1.4. [14] D. Bamman, O. Lewke, A. Mansoor, An annotated dataset of coreference in English literature, in: N. Calzolari, F. Béchet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, S. Piperidis (Eds.), Proceedings of the Twelfth Language Resources and Evaluation Conference, European Language Resources Association, Marseille, France, 2020, pp. 44–54. URL: https://aclanthology.org/2020.lrec-1.6/. [15] D. Bamman, M. C. Passarotti, R. Busa, G. Crane, The annotation guidelines of the Latin Dependency Treebank and Index Thomisticus Treebank. The treatment of some specific syntactic constructions in Latin, in: LREC 2008, ELDA, 2008, pp. 71–76. [16] F. Mambrini, Thucydides 1.89-118: A multi-layer treebank, CHS Research Bulletin 1 (2013). URL: http://nrs.harvard.edu/urn-3:hlnc.essay:MambriniF.Thucydides_1.89-118_ Multi-layer_Treebank.2013. [17] M. Passarotti, From syntax to semantics. first steps towards tectogrammatical annotation of latin, in: Proceedings of the 8th Workshop on Language Technology for Cultural Heritage, Social Sciences, and humanities (LaTeCH), 2014, pp. 100–109. [18] M. Passarotti, B. González Saavedra, The treebanked conspiracy. actors and actions in bellum catilinae, in: J. Hajič (Ed.), Proceedings of the 16th International Workshop on Treebanks and Linguistic Theories, Prague, Czech Republic, 2017, pp. 18–26. URL: https://aclanthology.org/ W17-7605/. [19] B. G. Saavedra, M. Passarotti, Using tectogrammatical annotation for studying actors and actions in sallust’s bellum catilinae, The Prague Bulletin of Mathematical Linguistics 111 (2018) 5–28. [20] M. Passarotti, F. Mambrini, G. Franzini, F. M. Cecchini, E. Litta, G. Moretti, P. Rufolo, R. Sprugnoli, Interlinking through lemmas. the lexical collection of the lila knowledge base of linguistic resources for latin, Studi e Saggi Linguistici 58 (2020) 177–212. [21] T. Berners-Lee, Www: Past, present, and future, Computer 29 (1996) 69–77. [22] O. Lassila, Resource description framework (rdf) model and syntax specification w3c working draft 08 october 1998, http://www. w3. org/1998/10/WD-rdf-syntax-19981008 (1998). [23] C. Chiarcos, S. Nordhof, S. Hellmann, Linked Data in Linguistics, Springer, 2012. [24] C. Chiarcos, M. Sukhareva, Olia–ontologies of linguistic annotation, Semantic Web 6 (2015) 379–386. [25] J. P. McCrae, J. Bosque-Gil, J. Gracia, P. Buitelaar, P. Cimiano, The ontolex-lemon model: development and applications, in: Proceedings of eLex 2017 conference, 2017, pp. 19–21. [26] M. Passarotti, M. Budassi, E. Litta, P. Rufolo, The lemlat 3.0 package for morphological analysis of latin, in: Proceedings of the NoDaLiDa 2017 workshop on processing historical language, 2017, pp. 24–31. [27] V. B. Lenzi, G. Moretti, R. Sprugnoli, Cat: the celct annotation tool., in: LREC, 2012, pp. 333–338. [28] W. W. Augustine, Confessions, Vol. 2: Books 9-13 (Loeb Classical Library, No. 27), 1912. [29] P. Nixon, et al., Plautus, Vol. II: Casina. The Casket Comedy. Curculio. Epidicus. The Two Menaechmuses (Loeb Classical Library), William Heinemann; GP Putnam’s Sons, 1917. [30] R. Simone, Fondamenti di linguistica, volume 9, Laterza Bari, 1990. [31] H.-L. Trieu, A.-K. D. Nguyen, N. Nguyen, M. Miwa, H. Takamura, S. Ananiadou, Coreference resolution in full text articles with bert and syntax-based mention filtering, in: Proceedings of the 5th workshop on BioNLP open shared tasks, 2019, pp. 196–205. [32] R. Thirukovalluru, N. Monath, K. Shridhar, M. Zaheer, M. Sachan, A. McCallum, Scaling within document coreference to long texts, Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 (2021) 3921–3931. [33] G. J. Caesar, The Gallic War, volume 72 of Loeb Classical Library, Harvard University Press,
Cambridge, MA, 1917. URL: https://www.loebclassics.com/view/LCL072/1917/volume.xml. [34] Seneca, Tragedies, Volume I: Hercules. Trojan Women. Phoenician Women. Medea. Phaedra,
SPARQL query to retrieve the synsets of those lexical entries of the Latin WordNet that are linked to lemmas in the Lemma Bank whose tokens are the terminals of a coreference unit entity involved in a coreference relation, together with the number of coreference units in which the tokens associated with such a lemma are involved. WordNet synsets are instances of the class ontolex:lexicalConcept. Endpoint: https://lila-erc.eu/sparql/.
PREFIX s k o s : < h t t p : / / www. w3 . o r g / 2 0 0 4 / 0 2 / s k o s / c o r e # > PREFIX o n t o l e x : < h t t p : / / www. w3 . o r g / n s / lemon / o n t o l e x # > PREFIX l i m e : < h t t p : / / www. w3 . o r g / n s / lemon / l i m e # > PREFIX l i l a : < h t t p : / / l i l a − e r c . eu / o n t o l o g i e s / l i l a / > PREFIX r d f s : < h t t p : / / www. w3 . o r g / 2 0 0 0 / 0 1 / r d f − schema # > PREFIX d c : < h t t p : / / p u r l . o r g / d c / e l e m e n t s / 1 . 1 / > PREFIX r d f : < h t t p : / / www. w3 . o r g / 1 9 9 9 / 0 2 / 2 2 − r d f − s y n t a x − n s # > PREFIX powla : < h t t p : / / p u r l . o r g / powla / powla . owl # > PREFIX l i l a _ c o r e f : < h t t p : / / l i l a − e r c . eu / o n t o l o g i e s / l i l a _ c o r e f / > SELECT d i s t i n c t ? l e m m a _ l a b e l ? s y n s e t _ d e f i n i t i o n ( c o u n t ( ? c o r e f _ u n i t ) a s ? n C o r e f U n i t ) WHERE { ? c o r e f _ r e l a t i o n r d f : t y p e l i l a _ c o r e f : C o r e f e r e n c e R e l a t i o n ; l i l a _ c o r e f : h a s C o r e f e r e n c e T a r g e t ? c o r e f _ u n i t ; r d f s : l a b e l ? c o r e f _ r e l a t i o n _ l a b e l . ? c o r e f _ u n i t r d f s : l a b e l ? c o r e f _ u n i t _ l a b e l ;
l i l a _ c o r e f : h a s C o r e f e r e n c e T e r m i n a l ? t o k e n . ? token r d f s : l a b e l ? t o k e n _ l a b e l ;
l i l a : hasLemma ? lemma . ? lemma r d f s : l a b e l ? l e m m a _ l a b e l . < h t t p : / / l i l a − e r c . eu / d a t a / l e x i c a l R e s o u r c e s / LatinWordNet / Lexicon > l i m e : e n t r y ? lex_entry_LWN . ? lex_entry_LWN o n t o l e x : c a n o n i c a l F o r m ? lemma ;
o n t o l e x : evokes ? s y n s e t . ? s y n s e t s k o s : d e f i n i t i o n ? s y n s e t _ d e f i n i t i o n .