AUTHOR=Viola Lorella , Verheul Jaap 

TITLE=One Hundred Years of Migration Discourse in The Times: A Discourse-Historical Word Vector Space Approach to the Construction of Meaning

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=3

YEAR=2020

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2020.00064

DOI=10.3389/frai.2020.00064

ISSN=2624-8212

ABSTRACT=<p>This study proposes an experimental method to trace the historical evolution of media discourse as a means to investigate the construction of collective meaning. Based on distributional semantics theory (Harris, <xref ref-type="bibr" rid="B42">1954</xref>; Firth, <xref ref-type="bibr" rid="B31">1957</xref>) and critical discourse theory (Wodak and Fairclough, <xref ref-type="bibr" rid="B104">1997</xref>), it explores the value of merging two techniques widely employed to investigate language and meaning in two separate fields: neural word embeddings (computational linguistics) and the discourse-historical approach (DHA; Reisigl and Wodak, <xref ref-type="bibr" rid="B80">2001</xref>) (applied linguistics). As a use case, we investigate the historical changes in the semantic space of public discourse of migration in the United Kingdom, and we use the <italic>Times Digital Archive</italic> (TDA) from 1900 to 2000 as dataset. For the computational part, we use the publicly available TDA word2vec models<xref ref-type="fn" rid="fn0001"><sup>1</sup></xref> (Kenter et al., <xref ref-type="bibr" rid="B49">2015</xref>; Martinez-Ortiz et al., <xref ref-type="bibr" rid="B65">2016</xref>); these models have been trained according to sliding time windows with the specific intention to map conceptual change. We then use DHA to triangulate the results generated by the word vector models with social and historical data to identify plausible explanations for the changes in the public debate. By bringing the focus of the analysis to the level of discourse, with this method, we aim to go beyond mapping different senses expressed by single words and to add the currently missing sociohistorical and sociolinguistic depth to the computational results. The study rests on the foundation that social changes will be reflected in changes in public discourse (Couldry, <xref ref-type="bibr" rid="B22">2008</xref>). Although correlation does not prove direct causation, we argue that historical events, language, and meaning should be considered as a mutually reinforcing cycle in which the language used to describe events shapes explicit meanings, which in turn trigger other events, which again will be reflected in the public discourse.</p>