Från ordfrekvenser till generativa modeller

Metodologisk reflektion kring datadrivna analyser av litteraturkritik

Abstract

In this article, we discuss the possibilities and limitations of three different data-driven methods for replicating a prior study in literary criticism: Lina Samuelsson’s Kritikens ordning: Svenska bokrecensioner 1906, 1956 och 2006 (‘The order of criticism: Swedish book reviews 1906, 1956 and 2006’, 2013). The methods we employ can be said to represent three generations of language technology and we use them to analyse a corpus of early-twentieth-century book reviews according to the same principles as in Samuelsson’s study, but on a larger scale. 

Drawing on the Russian Formalist Viktor Shklovsky’s concept of “defamiliarisation”, we show how data-driven methods can produce a distancing effect that opens up new perspectives on otherwise familiar material. First, we perform a word frequency analysis, indicating how such a simple quantitative approach yields interesting, if methodologically rigid perspectives on the corpus. Next, we show how topic modelling produces results that appear analytically flat, yet still sharpen attention to certain neglected aspects of the review texts. Finally, we demonstrate how generative models can contribute substantial analytical insights into evaluative patterns in literary criticism, while also posing new methodological challenges. We conclude by undescoring the need for critical reflection both of data-driven methods as such within literary studies and the perspectives these methods seem to open up on traditional approaches.