Atril: an XML Visualization System for Corpus Texts

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5692–5695 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5692 Atril: an XML Visualization System for Corpus Texts Andressa Gomide, Conceição Carapinha, Cornelia Plag CELGA-ILTEC - Universidade de Coimbra Faculdade de Letras - 3004-530, Coimbra - Portugal {andressa.gomide, mccarapinha, cornelia.plag}@fl.uc.pt Abstract This paper presents Atril, an XML visualization system for corpus texts, developed for, but not restricted to, the project Corpus de Audiências (CorAuDis), a corpus composed of transcripts of sessions of criminal proceedings recorded at the Coimbra Court. The main aim of the tool is to provide researchers with a web-based environment that allows for an easily customizable visualization of corpus texts with heavy structural annotation. Existing corpus analysis tools such as SketchEngine, TEITOK and CQPweb offer some kind of visualization mechanisms, but, to our knowledge, none meets our project’s main needs. Our requirements are a system that is open- source; that can be easily connected to CQPweb and TEITOK, that provides a full text-view with switchable visualization templates, that allows for the visualization of overlapping utterances. To meet those requirements, we created Atril, a module with a corpus XML file viewer, a visualization management system, and a word alignment tool. Keywords: xml visualization, corpus software, utterance alignment 1. Introduction Many corpora are heavily marked with structural tags, normally adopting Extensible Markup Language (XML) formatting. This type of annotation benefits different types of corpora. It can mark boundaries within texts, allowing for searches within specific sections, as is the case with The English Scientific Text Corpus (SciTex) (Degaetano- Ortlieb et al., 2013). It also allows for the creation of subcorpora derived from these identified boundaries. For instance, for spoken corpora with rich speaker and text metadata such as the Spoken BNC 2014 (Love et al., 2017) you can create subcorpora by filtering the utterances by speakers’ profile and conversation context. XML mark-up is also useful to indicate events and situations occurring in parallel to the discourse, such as pauses in a speech and unintelligible words in a document. In the context of Corpus Linguistics (CL), this structural annotation, also known as markdown, is normally the first of two steps of text annotation. In small or carefully prepared corpora, this annotation tends to be done manually or semi-automatically. The semi-automatic way is often done by automatizing the identification of patterns and their respective replacement with XML tags. The markdown process is normally followed by a second and automatic step for the annotation of each token, being lemma and parts-of-speech the most common types of annotation. Some corpus analysis software has an annotation built-in system and this step is done concurrently with the corpus loading or installation, as it is the case with SketchEngine (Kilgarriff et al., 2014). Once the processes above are done, the usage and visualization of the XML tags is subject to the functions offered by the chosen corpus software. This restriction yields, especially for richly marked texts, two main …

Content truncated to 500 words