Profiling-UD: a Tool for Linguistic Profiling of Texts

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 7145–7151 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 7145 Profiling–UD: a Tool for Linguistic Profiling of Texts Dominique Brunato, Andrea Cimino, Felice Dell’Orletta, Simonetta Montemagni, Giulia Venturi Istituto di Linguistica Computazionale “Antonio Zampolli” (ILC–CNR) ItaliaNLP Lab Via G. Moruzzi, 1, Pisa, Italy {name.surname}@ilc.cnr.it Abstract In this paper, we introduce Profiling–UD, a new text analysis tool inspired to the principles of linguistic profiling that can support language variation research from different perspectives. It allows the extraction of more than 130 features, spanning across different levels of linguistic description. Beyond the large number of features that can be monitored, a main novelty of Profiling–UD is that it has been specifically devised to be multilingual since it is based on the Universal Dependencies framework. In the second part of the paper, we demonstrate the effectiveness of these features in a number of theoretical and applicative studies in which they were successfully used for text and author profiling. Keywords: Linguistic Profiling, Universal Dependencies, Computational Language Variation Analysis 1. Introduction The availability of large-scale corpora representative of real language use, together with the development of sophisti- cated Natural Language Processing (NLP) pipelines that make explicit the linguistic structure underlying a text, has encouraged a paradigmatic shift towards a more extensive use of data–driven methods to tackle traditional topics of linguistic research, going from the study of language varia- tion across textual genres, registers as well styles associated with particular authors or historical periods, to the investi- gation of linguistic complexity from a user–dependent per- spective. They have also favoured the development of lan- guage technologies focusing on modeling the ‘form’, rather than the content, of texts. By relying on different levels of linguistic annotation, it is possible to extract a large number of features modeling lexical, grammatical and semantic phenomena that, all to- gether, contribute to characterize language variation within and across texts. These are the prerequisites of linguis- tic profiling, a methodology in which – in van Halteren’s words – “the occurrences in a text are counted of a large number of linguistic features, either individual items or combinations of items. These counts are then normalized [...]” in order to detect and quantify differences and simi- larities across texts representative of distinct language va- rieties (van Halteren, 2004). Following this approach, the linguistic structure of a text is analyzed to extract relevant features, and a representation of the text is constructed out of occurrence statistics of these features, be they absolute or relative frequencies or more complex statistics. In linguis- tic profiling, each text or collection of texts is thus assigned a feature–based representation covering different levels of linguistic description. Nowadays, linguistic profiling as described above lies at the heart of current research in different areas, which share the purpose of reconstructing the linguistic profile underlying linguistic productions originating in specific contexts, e.g. in socio–culturally defined demographic groups or individ- ual authors (also across time), …

Content truncated to 500 words