Assessing Semantic Annotation Activities with Formal Concept Analysis

Abstract

The enormous efforts to digitize physical resources (documents, books, museum exhibits, etc.), along with recent advances in information and communication technologies, have democratized access to a cultural, scientific and academic heritage previously available to only a few. Likewise, the current trend is to produce new resources in a digital format (e.g., in the context of social networks), which entails an in-depth paradigm shift in almost all the humanistic, social, scientific and technological fields. In particular, the field of the humanities is one which is going through a significant transformation as a result of these digitalization efforts and the paradigm shift associated with the digital age. Indeed, we are witnessing the emergence of a whole host of disciplines, those of Digital Humanities (Berry 2012), which are closely dependent on the production and proper organization of digital collections. As a result of the undoubted importance of digital collections in modern society, the search for effective and efficient methods to carry out the production, preservation and enhancement of such digital collections has become a key challenge in modern society (Calhoun, 2013). In particular, the annotation of resources with metadata that enables their proper cataloging, search, retrieval and use in different application scenarios is one of the key elements to ensuring the profitability of these collections of digital objects. While the cataloging and retrieval of resources (whether digital or non-digital) have been the object of study in library sciences for decades (Calhoun, 2013), modern applications require annotating resources in semantically richer and more flexible ways, in many cases allowing multiple alternative annotations in the same collection. In consequence, the tendency is to introduce the use of ontology-based semantic technologies, in addition to conventional metadata schemas (Keyser, 2012). While in recent years we have witnessed significant advances in the automatic annotation of resources, in particular of those with heavy text content (see section 6), there are multiple scenarios in which resource annotation cannot be inferred from the contents of these resources (e.g., scenarios involving resources in which the content is not directly related to the meta-information required). In these cases it is necessary to involve human annotators in the semantic annotation of the resources. The resulting activities are referred to as semantic annotation activities in this paper. Some examples of semantic annotation activities are the annotation of digital educational resources (e.g., learning objects) in the eLearning domain (Aroyo & Dicheva. 2004, Devedzic et al. 2007, Tiropanis et al. 2009, Kurilovas et al. 2014), the annotation of media content in the multimedia domain (Labra et al. 2010, Mu 2010, Hunter & Herber, 2010, Šimko et al. 2013), or the one chosen as a case study in this paper: the annotation of digitized literary texts (Azouaou & Desmoulins, 2006; Koivunen 2005; Rocha et al. 2009; Schroeter et al. 2006; Tazi et al. 2003; Gayoso et al. 2012, 2013; Donato et al. 2013).