Machine Learning und Korpuspragmatik. Word Embeddings als Beispiel für einen kreativen Umgang mit NLP-Tools

Racist discourse in a German far-right blog: A corpus-driven approach using word embeddings

“…The contextualist approach to semantics unites collocation analysis with the calculation of word embeddings (Kozlowski et al, 2019;Mikolov et al, 2013) which builds on a systematic comparison of collocation profiles and has already been shown to be a promising approach for discourse-analytical purposes (Bubenhofer et al, 2019;Knuchel and Bubenhofer, 2023). It is based on the machine learning-based construction of a language model which represents the relationships between words in a corpus, that is, their contextual embeddedness in a high-dimensional vector space by assigning every word to a word vector based on its typical contexts or collocates (Lenci, 2018: 153).…”

“…Methodologically, this study is at the interface of corpus linguistics and discourse analysis. Following a corpus pragmatic approach (Aijmer and Rühlemann, 2014), it is focussed on the analysis of linguistic patterns which are interpreted as traces of recurrent and socially situated speech actions and seeks to explore the form, meaning and function of these patterns as indices of their social contexts (Knuchel and Bubenhofer, 2023). From the range of corpus linguistic methods, collocation analysis and the calculation of word embeddings are particularly relevant here.…”