In the context of the digitization of manuscripts, transcription and annotation are often distinct, sequential steps. This could lead to diculties in improving the transcribed text when annotations have already been dened. In order to avoid this, we devised an approach which merges the two steps into the same process. Text Encoder and Annotator (TEA) is a prototype application embracing this concept. TEA is based on a lightweight language syntax which annotates text using Semantic Web technologies. Our approach is currently being developed within the Clavius on the Web project, devoted to studying the manuscripts of Christophorus Clavius, an inuential 16th century mathematician and astronomer.
Within the eld of Digital Humanities, several projects are devoted to preserving, analyzing and studying the large amounts of manuscripts, books, newspapers, maps, photos and paintings stored in archives, museums and libraries around the world.
The Clavius On the Web 3 project [
so that personalities such as RenØ Descartes, Marin Mersenne and Johannes Kepler built their knowledge on it.
In this context, Semantic Web technologies, such as the Resource Description Framework (RDF), make it possible to approach text annotation in an innovative way. RDF-based annotations provide a method for enriching texts using structured data already described in details and maintained by the Semantic Web community (i.e., Linked Data sets). In addition, due to the interlinked structure of Linked Data, RDF-based annotations produce valuable annotated documents, characterized by a strong connection with external resources. 1.1
Text annotation consists in attaching additional information such as comments, tags or links, to specic portions of a text. Annotations can be mainly performed using two methods: inline and stando . The former directly includes annotations within the text, while the latter denes them in a dierent location. Inline markup keeps the annotations and the annotated text close together, but it has the drawback of weighing the document down. Moreover, depending on the complexity of the markup language, the text could become hard to read. This aspect is crucial and must be taken into account when developing manual annotation tools, as users need to be able to read the annotated text with ease. Last but not least, a complex and heavyweight markup language could make the manual annotation process even more dicult since users have to rstly know all the syntax rules and secondly write a considerable amount of additional markup.
In contrast, the stando approach does not have markup overloading problems due to its total independence from the resource text. Annotations are in fact separately dened in a dierent location where the relative text osets are stored and kept up to date. In addition, stando markup has the advantage of allowing overlapping annotations. Nevertheless, this approach has some drawbacks related to the sequential process of transcribing and annotating. Typically, the available tools of this type separate the transcription and the annotation phases. However, if a transcription error is found during the subsequent annotation phase, it is necessary to recompute the osets in order to reect the changes in the transcribed text. This implies an automatic recomputation of the osets, a process that could be complex and costly. The logical conclusion is therefore to make transcription and annotation a joint process. 1.2
The core idea of this work is to combine transcription and annotation of text, thus streamlining the workow process. Hence, we devised a lightweight language that enables this continuous and mixed process. Another key-point is that we propose to treat every textual phenomenon as an annotation independently from its specic type (e.g., semantic, syntactic, lexical). Every portion of text is treated in the same way, and RDF-based annotations are used to describe their content. RDF supports our purpose by allowing any possible annotation to be specied using the enormous amount of ontologies, vocabularies and Linked Data sets available on the Web.
To the best of our knowledge, none of the existing tools include this approach.
For instance, Pundit [
In the light of the above, this article presents the Text Encoder and Annotator, a Web application, which provides an editor to transcribe texts and a lightweight language for annotating them with RDF (all within the same environment). It was envisaged for linguists, historians and more generally for scholars and students. We devised a layout composed of three main views, horizontally placed along the interface of the application (Figure 1): 1. Image box : it displays the digitized image of a specic manuscript to be transcribed and annotated. 2. Editor box : it is the main component of the tool. It is used to write text and enrich it with RDF-based annotations, which follow the specic language syntax described below. Text highlighting identies the markup and improves its legibility. Moreover, a top bar contains shortcut buttons for inserting some basic annotations, characterized by common RDF predicates, such as rdfs:seeAlso, which can be used to link to external resources, rdfs:comment to specify text comments and foaf:page to include hyperlinks related to the topic the annotation is about. 3. Diagram box : it is an optional view that can be activated on-demand if the user needs a summary of their annotations. The node-link diagram contains a white node representing the whole text of the document, blue nodes identifying the portion of text referred to in an annotation, orange nodes displaying the identiers of the annotation and gray nodes describing the objects of the RDF triples specied. The edges between orange and gray nodes represent the predicates of the triples dened in the annotations. 2.1
We propose a Lightweight Markup Language 4 (LML) employed only within
the interface of TEA, in order to provide an easy and quick way of transcribing
and annotating text using RDF. The main reason behind the adoption of an
LML is that common markup languages based on XML (e.g., TEI) are not easy
to write and read in their raw form, due to their complex syntax. Moreover, the
use of LMLs has already proven to be benecial in other systems such as the
Leiden-plus5 language employed in the papyri.info editor. It is worth clarifying
that we do not propose our approach as a format for the representation and
interchange of texts, consequently it cannot be compared to standards such as
the Text Encoding Initiative (TEI) [
Our language is used for marking portions of text and assigning them identiers that will be used in a dierent, reserved and stando-like" section of the text where the details of the annotations are specied (Figure 2). More precisely, a portion of text can be annotated by enclosing it in a span using angle brackets, while round brackets specify a string identifying the annotation (i.e., hannotated portion of texti(identier) ). This inline syntax uses a very limited amount of characters and does not weigh the text down too much, keeping it easy-to-read. Identiers are then used in a distinct part of the text, called directive section, where the annotation body is specied. Three plus signs (i.e., +++) are used both to open and close this section that can be repeated within the text more than once. Inside this block, annotations can be specied as RDF triples with the identier of a certain span of text as subject. The choice of predicates and objects 4 http://en.wikipedia.org/wiki/Lightweight_markup_language 5 http://papyri.info/editor/documentation?docotype=text is totally free, although some default predicates are suggested in order to perform the most common and basic annotations (e.g., rdfs:seeAlso, rdfs:comment, foaf:page). Currently, the syntax used in the directive section denes triples as three text values separated by a space. 2.2
We here provide an example of annotation performed on a portion of text
extracted from the Euclidis Elementorum Libri XV. Accessit XVI [
7 http://claviusontheweb.it/lexicon/math/ 8 http://www.w3.org/TR/rdf-schema/
This article describes an approach for merging the distinct steps of transcription and annotation as a single process. We implemented a tool based on a lightweight syntax language that allows RDF annotations to be performed. We conducted some preliminary tests, which involved 50 students, who were asked to use the prototype, and provide feedback. Future works will consist in developing an improved language with a syntax capable of handling nested as well as overlapping (i.e., not hierarchically nested) annotations. New syntax elements will be introduced: milestone elements, for annotating a single location in the text (e.g., a gap), partition elements, for the identication of phenomena such as line, page or sentence breaks. Additional syntax will be introduced to provide shortcuts ("syntactic sugar") to the most common annotations. The Turtle syntax 9 will also be taken into account for the RDF triples specication. Finally, dierent formats (e.g., turtle, json, csv, xml) will be chosen for exporting annotations according to various data models (e.g., Open Annotation, NLP Interchange Format (NIF)).