Idioma: Español
Fecha: Subida: 2021-04-13T00:00:00+02:00
Duración: 23m 02s
Lugar: Conferencia
Visitas: 951 visitas

The great recession’s influence in financial terminology: a longitudinal sentiment analysis

Javier Fernández Cruz (Universidad de Málaga)

Descripción

Do the consequences of major events, such as crises, influence the way economic language is used? After the outbreak of the Great Recession in 2008, the sentiment of multiple event words (Moirand, 2007) in the field of economics and finance varied rapidly. All these factors emerge in the manner that evaluations and opinions are expressed. The pressures behind the zeitgeist are also perceptible in economy news (Alim, 2011), as terminology may be subject to changes in its sentiment or semantic orientation. Even if these changes are undoubtedly difficult to trace only through introspection (Crowley & Bowern, 2010; Cook & Stevenson, 2010; Cavallin, 2012), such observation is now facilitated by the development of NLP solutions, such as Sentiment Analysis (SA) algorithms.
This work aims to analyze diachronically (2007-2015) the fluctuations in the sentiment of economic terms (e.g. “debt” or “credit”) in a large corpus of daily-news coverage. To do this, we used the Lingmotif (Moreno-Ortiz, 2017) SA suite in combination with SentiEcon a large financial-domain sentiment lexicon designed ad-hoc. Quantitative data were triangulated though observation of short-term variations in the semantic prosody and qualitatively with the correspondence to the turbulent sociopolitical events.
In relation to semantic changes in sentiment, semasiological changes are crucial here, highlighting the concepts of amelioration and pejoration (Bloomfield, 1933; Traugott & Dasher, 2002) in which the meaning of a word become more positive or negative, respectively. In addition, the semantic prosody of words may vary as they collocate with certain lexical elements during a timespan (Morley & Partington, 2003).
SA (Liu, 2015) deals with the computational treatment of opinion in texts. Through different approaches, it is possible to calculate the sentiment of texts. Basically, a SA system classifies a text as positive/negative or with a score within a scale (Pang & Lee, 2005). SA is highly sensitive to the specific domain (e.g., economy) (Loughran & McDonald, 2011; Malo et al., 2014; Van De Kauter, et al., 2015). To do this, it is necessary to tune up SA systems for the requirements of different specialized languages.
The study of evaluative language provides the main linguistic foundations of SA. It is a multidisciplinary subject that involves psychological and sociological factors. Even if research on this field can be traced back to the 1970s (Labov, 1972, 1985; Chafe, 1986; Biber & Finnegan, 1988, 1989), Martin and White’s (2005) Appraisal Theory is the most complete approach that, in a highly structured manner, covers most aspects related to the evaluative use of language: intrapersonal attitude, interpersonal engagement, along with levels of gradation.
Semantic factors such as sentiment were left behind in traditional models of terminology (e.g., Wüster, 1998) where terms were considered strictly univocal and synchronous. According to this, a term’s definition will be eminently neutral, and no evolution of its meaning was considered. More recent models (Cabré, 1998; Temmerman, 2000) opened the door to new theoretical possibilities, considering pragmatic, social and cognitive aspects of terms, such as affective meaning or asynchrony.
After the analysis of the large datasets two main facts stand out: (1) after major events, terms become event words after a great rise in their relative frequency and, (2) sudden sentiment fluctuations tend to cooccur with changes in semantic prosody.

Propietarios

Congreso Cilc 2021

Comentarios

Nuevo comentario

Serie: CILC2021: Lingüística computacional basada en corpus / Corpus-based computational linguistics (+información)