Many and diverse scientific communities, including digital textual scholars, have shown increasing interest in semantic web technologies and linked open data (LOD) as a means of knowledge representation since the early 2000s, when Tim Berners-Lee coined the concept of a web of interconnected data rather than simply documents (see [Spadini et al. 2021]). Nonetheless, there is a lack of user-friendly tools for working with digital scholarly editions and LOD.

Linked data from TEI (LIFT) is a Digital Humanities tool for generating linked open data (LOD) from TEI-encoded texts. LIFT, which is accessible at https://projects.dharc.unibo.it/lift, provides a set of TEI-to-LOD transformation scripts written in Python. The scripts, addressing the transformation of different types of entities to LOD, are thoroughly documented to facilitate their understanding and reuse. The purpose of this paper is to introduce LIFT as a teaching tool for supporting the adoption of linked open data in digital scholarly editing. Our plan is to use LIFT as part of a laboratory activity for the University of Bologna’s Master Degree in Digital Humanities and Digital Knowledge, which will combine the knowledge and skills acquired by the students from different modules: text encoding with the TEI, ontology development, Python programming, and linked open data for cultural heritage.[2]

This paper first provides a brief background on linked open data and their use in digital text representation (see Section 2). It then goes on to consider related work in the field of TEI to LOD transformation (see Section 3) as well as to introduce LIFT from both a technical and methodological perspective (see Section 4). Finally, it discusses the potential application of LIFT in digital humanities learning contexts (see Section 5 and Section 6).

The concept of linked open data refers to a method of publishing structured data – as well as organizing knowledge – on the web in such a way that they can be semantically interlinked to form open clouds of integrated information. As is already well known, LOD are based on the Resource Description Framework (RDF), a graph data model used for making statements about resources. Each RDF statement contains a subject, a predicate, and an object, represented by unique URIs. The subject is the resource being described, the object is either a characteristic of the subject (such as its type) or a resource related to the subject, and the predicate expresses the relationship between the subject and the object (see example 1 below). The meaning of RDF data is defined by machine-readable ontologies, formal descriptions of specific concepts within a domain and the relationships holding between them.

RDF statements can be used to represent information in digital scholarly editions as linked open data.[3] To cite just a few examples, the Henry III Fine Rolls edition experiments with linked data to express complex relationships between people in historical documents (see [Ciula et al. 2008]). The Sharing Ancient Wisdoms project, on the other hand, uses linked data to reconstruct text interaction (see [Jordanous et al. 2012]), while the Paolo Bufalini’s Notebook edition uses linked data to represent the interconnections between the various types of fragments – quotations, translations, and annotations – that make up the text (see [Daquino et al. 2019]). Similar information extracted from texts, in addition to administrative, structural, and descriptive metadata, can make a significant contribution to the development of a cultural heritage linked open data cloud: texts do indeed occupy a prominent space in cultural heritage representation. However, in order to reuse and integrate texts into LOD systems, information must first be expressed in RDF according to widely used ontologies in the digital humanities and cultural heritage domains.

The Text Encoding Initiative (TEI), an XML vocabulary for encoding texts in the humanities, is the de facto standard for digital scholarly editing.[4] The standard defines terms for representing a wide range of textual entities, including personal names, place names, dates, concepts, and events. Unlike RDF, the TEI has traditionally taken a document-centric approach to representing such entities, with texts processed as ordered hierarchies of content objects nesting neatly one inside the other (see [DeRose et al. 1990] and [Renear 1993]). However, transitioning from a document-centric paradigm to a data-centric paradigm, such as LOD, is not an easy task: converting TEI editions into graphs of RDF statements necessitates a conceptual shift in which texts are no longer viewed as documents but as collections of interconnected entities (see [Tomasi 2012]).

The movement beyond TEI towards linked open data began in 2004, when a TEI Special Interest Group on Ontologies was formed with the goal of investigating the feasibility of encoding information about persons, dates, events, places and objects outside the text, in parallel with the steps taken by the Museum community, which was working on the development of CIDOC CRM, an ontological vocabulary for the description of museum collections (see [TEI 2019]). The new P5 release of the TEI in 2007 introduced new elements for distinguishing real-world entities from their in-text occurrences [Wittern et al. 2009]. This change, along with the addition of new attributes that could be used to link an entity to external authority records (see example 2 below), made it easier to express relationships between entities in TEI and extract them as RDF statements.

Despite these meaningful developments, there are very few user-friendly tools for working with TEI editions and LOD [Pierazzo 2016, 121]. The fact that the TEI guidelines are not intended to be strictly followed complicates matters: annotators can encode the same textual features in different ways, depending on the project. Because no two texts or editions are identical, this flexibility is required; however, the unavoidable drawback is a certain loss of interoperability (see [Andrews 2013, 2]). Enhancing TEI-encoded texts with linked open data represents a possible way of overcoming this limitation.

Being willing to introduce DHDK students to the extraction of linked open data from TEI-encoded texts, we ran into a lack of user-friendly tools capable of raising awareness about both the methodology and the technology that underpin such operation. LIFT was created to fill this gap.

As digital humanities practitioners know, XSLT (Extensible Stylesheet Language Transformations) is a language for transforming XML documents into other XML documents. Because the TEI is an XML vocabulary, digital textual scholars have naturally adopted XSLT as their primary transformation language from TEI to other formats including RDF/XML, which is one of the possible serializations of LOD graphs. The TEI Consortium develops and maintains the TEI Stylesheets, a collection of ready-for-use XSL Transformations scripts that includes a TEI-to-RDF transformation. Despite providing a relatively simple method of extracting RDF statements from TEI documents, the Stylesheets have several shortcomings, particularly when used as a teaching tool. The Stylesheets were not, in fact, designed with a didactic aim in mind and do not offer TEI-to-RDF transformation guidelines: users need to possess good XSLT skills to prepare the TEI document, gain insight into the rationale of the transformation, and adapt it to their specific needs. Also, RDF can be serialized in a variety of ways, with Turtle being by far the most widely used due to its legibility. Since the TEI-to-RDF Stylesheet was created, more powerful methods of linked data production have been released. RDFLib, a Python package for working with RDF, is one of these. The key difference between using XSLT and RDFLib to create RDF triples is that the former requires developers to hardcode RDF triples into the transformation stylesheet (i.e., the developer must choose ahead of time a specific serialization and adhere to its syntax), resulting in a significant loss of flexibility and an increase in human errors. RDFLib, on the other hand, automatically serializes RDF triples, allowing you to add properties to an entity at any point along the way without worrying about syntax and structure (see example 3, which shows a comparison between XSLT and RDFLib as two different methods of generating RDF statements). Finally, while the TEI-to-RDF Stylesheet reuse some concepts from the CIDOC Conceptual Reference Model (see [Crofts et al. 2011]) and the Dublin Core (see [Weibel and Traugott 2000]), it does not make use of the expressive potential of ontologies such as FRBRoo (see [Bekiari et al. 2015]) and PRO (see [Peroni et al. 2012]), which have recently gained attention and application in both text and cultural heritage representation (see [Daquino et al. 2019]).

Another tool we considered is XTriples. XTriples is a webservice-based infrastructure for extracting RDF statements from collections of XML documents [Gruntgens and Torsten 2016]. Unlike the Stylesheets, XTriples can produce multiple serializations of the resulting RDF graph and comes with a documentation. The transformation is customizable via an XPath-based configuration file. LOD beginners, however, may find this operation difficult and the tool not immediately usable or replicable.

There are also tools, such as Recogito, that enable the extraction of specific features from TEI-encoded texts. Recogito, developed by the Pelagios Network, allows users to add annotations to specific passages of text, which they can then download as RDF statements (see [Simon et al. 2017]). The transformation is not configurable nor explained as this is not the focus of the tool.

Based on the shortcomings discussed above, we decided to develop a TEI-to-LOD transformation tool for our students that would simplify the transformation process while also openly and transparently documenting the workflow, with the goal of making such workflow modifiable and adaptable to new contexts. Linked data from TEI (LIFT) is a Python-based, open-source application for creating graphs of RDF statements from TEI documents. LIFT targets digital humanities students as well as textual scholars experimenting for the first times with linked open data in digital scholarly editing. The main component of LIFT is a collection of TEI-to-LOD transformation scripts. The scripts, written in Python, take as input a TEI document and generate an RDF graph. Each script addresses the transformation of specific TEI entities including persons, personal relations, places, and events. A transformation script for the TEI critical apparatus is also available (see Section 4.2). The application is extensively documented in order to facilitate the reuse and adaptation of the transformation workflow. The documentation walks users through the process of preparing their TEI document for transformation, describes the expected RDF graph, and provides step-by-step explanations of the transformation scripts in the form of an interactive Jupyter notebook (see Section 4.3).

Users can access LIFT through a browser or download the collection of scripts for local use. A basic understanding of the TEI and RDF is recommended (the ideal target audience are digital humanities students who are already familiar with digital scholarly editions, the TEI, RDF and ontologies, and basic Python programming).

LIFT’s web interface has a straightforward design. When using a browser to access LIFT, users are directed to the homepage, which contains instructions for getting started with the transformation (see Figure 1): by clicking on the 'Quick start' tab, an upload bar for the input TEI document becomes available (see Figure 2). After uploading the TEI document into LIFT, users are directed to another page where they can select a transformation option (see Figure 3). If the transformation is successful, multiple serializations of the resulting RDF graph become available for download (see Figure 4).

Digital humanities, as a discipline, lie at the crossroads of theory and practice. The learning-by-example paradigm, which entails presenting existing models that can be adapted to address new problems, is particularly effective in similar interdisciplinary contexts (see [Tomasi 2020]). LIFT was created in response to a perceived lack of teaching tools for TEI to LOD transformation based on such paradigm.

LIFT will be used in a text representation laboratory activity for the University of Bologna’s Master Degree in Digital Humanities and Digital Knowledge. As anticipated at the beginning of this paper, the activity aims to combine the knowledge and skills acquired by the students from various modules and, particularly, text encoding with the TEI, ontology development, Python programming, and linked open data for cultural heritage. The plan for the activity involves gathering the students in 3-member groups. Each group will choose an existing digital scholarly edition and work on the underlying TEI document to ensure that the encoding requirements of LIFT are met.[8] The groups will then use LIFT, either via browser or locally, to produce an RDF graph. Finally, each group will present their findings to the class, highlighting challenges and solutions in transforming each type of entity. Following activities will shift the emphasis away from the input TEI document as students will focus on modifying the transformation scripts to adapt them to the existing digital scholarly editions rather than the other way around. By experimenting with the shift from tree-structured to graph-structured data, and from the TEI vocabulary to RDF using multiple ontologies, students will gain a better understanding of the implications of annotating a text in one way versus another within the context of a real-world task that requires them to combine different skills and competencies that are typically taught separately. Furthermore, because the activity is structured as group work on a specific problem to be solved, students will be able to strengthen their ability to collaborate as a team, which is one of the pillars of digital humanities scholarship and, by extension, should be one of digital humanities pedagogy (see [Licastro et al. 2020]).

The laboratory activity will also allow us to evaluate the clarity and effectiveness of LIFT as a teaching tool both directly through an end-of-activity questionnaire and indirectly through the students’ presentations.

This paper introduced LIFT, an open-source tool written as a set of Python scripts for generating linked data from TEI-encoded texts. The use of linked open data for text representation facilitates interoperability with other cultural resources on the web and opens up editions to new areas of cultural heritage research. LIFT’s goal is to walk users through the transformation process from TEI to linked data step by step, as well as to promote a better understanding of the theoretical and methodological aspects that underpin the transformation.

Some of the key characteristics of the evolving web of data include open-source tools, well-documented applications, and shared ontologies. User-friendly resources that reduce the complexity of transforming digital scholarly editions from a document-centric to a data-centric model are also critical in the development and expansion of a linked open data cloud of cultural heritage. We envisage LIFT both as a teaching tool and a ready-for-use resource for TEI-to-LOD transformation, hoping it will serve as an example of a digital humanities teaching tool that raises awareness about the methodological and theoretical implications of the structural and semantic shift that occurs when transitioning from a TEI-based text representation to LOD.

With LIFT, we also aim to encourage further research into the development of open-source, user-friendly tools aiding the mutual integration of digital scholarly editions and the cultural heritage linked open data cloud. Such tools have the potential to make digital humanities, and especially knowledge representation, a more inclusive field of study and research. In order to test this idea, we will use LIFT in the context of the University of Bologna’s Master Degree in Digital Humanities and Digital Knowledge, as a teaching tool for students encountering linked open data for the first time as a method of publishing digital scholarly editions on the web in a data integration perspective. We expect that LIFT will help students understand the process not only practically but also, and primarily, in terms of methodology and theory.

LIFT is under active development. Features that will be provided in the near future include a transformation script for bibliographic references encoded using specialized elements from the TEI model.bibLike class. Furthermore, the documentation will be expanded with new sections on ontologies for text representation as well as useful resources, particularly examples of digital scholarly editions enhanced through the use of linked open data.