Getting a (big) data-based grip on ideological change. Evidence from Belgian Dutch

Abstract In this paper we introduce a computationally enriched experimental tool designed to investigate language ideology (change). In a free response experiment, 211 respondents returned three adjectives in reaction to the labels for five regional varieties, one ethnic variety and two supra-regional varieties of Belgian Dutch, as well as the standard accent of Netherlandic Dutch. Valence information (pertaining to the positive/negative character of the responses) and big data–based distributional analysis (to detect semantic similarity between the responses) were used to cluster the response adjectives into 11 positive and 11 negative evaluative dimensions. Correspondence analysis was subsequently used to compute and visualize the associations between these evaluative dimensions and the investigated language labels, in order to generate “perceptual maps” of the Belgian language repertoire. Contrary to our expectations, these maps unveiled not only the dominant value system which drives standard usage, but also the competing ideology which frames the increasingly occurring non-standard forms. In addition, they revealed a much richer stratification than the “one variety good, all other varieties bad” dichotomy we had anticipated: while VRT-Dutch remains the superior (albeit increasingly virtual) standard for Belgian Dutch, the stigmatized colloquial variety Tussentaal is gradually being accepted as a practical lingua franca, and the Ghent-accent is boosted by modern prestige (dynamism) features. Even more crucially, separate perceptual maps for the older and younger respondents lay bare generational change: there is a growing conceptual proximity between VRT-Dutch and Tussentaal in the younger perceptions.