Aspect Flow Representation and Audio Inspired Analysis for Texts

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1469–1477 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1469 Aspect Flow Representation and Audio Inspired Analysis for Texts Larissa L. Vasconcelos12, Claudio E. C. Campelo1, Caio L. M. Jeronimo1 Federal University of Campina Grande1, Federal Institute of Paraiba2 882, Aprigio Veloso, Campina Grande - PB - Brazil1 PB-264, Monteiro - PB - Brazil2 larissa.vasconcelos@ifpb.edu.br, campelo@computacao.ufcg.edu.br, caiolibanio@copin.ufcg.edu.br Abstract For better understanding how people write texts, it is fundamental to examine how a particular linguistic aspect (e.g., subjectivity, sentiment, argumentation) is exploited in a text. Analysing such an aspect of a text as a whole (i.e., through a summarised single feature) can lead to significant information loss. In this paper, we propose a novel method of representing and analysing texts that consider how an aspect behaves throughout the text. We represent the texts by aspect flows for capturing all the aspect behaviour. Then, inspired by the resemblance between these flows format and a sound waveform, we fragment them into frames and calculate an adaptation of audio analysis features, named here Audio-Like Features, as a way of analysing the texts. The results of the conducted classification tasks reveal that our approach can surpass methods based on summarised features. We also show that a detailed examination of the Audio-Like Features can lead to a more profound knowledge about the represented texts. Keywords: Text Representation, Text Analysis, Aspect Flows 1. Introduction A challenge in Natural Language Processing (NLP) is to reproduce the way humans perceive a certain linguistic as- pect in a text and, with that, achieve a better understanding of how different kinds of texts are written. For instance, analysing how fake news make use of subjective language can lead to meaningful knowledge about them (Jeronimo et al., 2019). In this study, we employ the term linguistic aspect (or just ‘aspect’, for short) to refer to a distinguishable linguis- tic characteristic expressed in a text, such as sentiment, argumentation, or subjectivity. Observing the particulari- ties of how a text utilises some aspect requires not only a global analysis of the entire text or individual analysis of each word or sentence, but also an analysis of its behaviour throughout the text. Generally, for executing NLP tasks, the applied techniques model local representations of an aspect to extract sum- marised features that represent the input text as a whole, frequently by an average or median of that local represen- tations. This sort of representation can lead to relevant in- formation loss, especially for large texts, as they can be ig- noring significant aspect singularities present in any part of the text that could be decisive on text type identification and characterisation (Aker et al., 2019). An approach to avoiding representing a text in a globally summarised way is modeling a text as a flow, which is de- fined by Mao and Lebanon (2007) as a sequence of infor- mation collected from the words, sentences, or paragraphs of …

Content truncated to 500 words