 previous article in this issue | next article in this issue  |
Preview first page |
Document Details : Title: De kracht van foute voorspellingen Subtitle: Een methodische reflectie op machine learning aan de hand van onderzoek naar genderrepresentatie in moderne Nederlandstalige romans Author(s): VITSE, Sven Journal: Spiegel der Letteren Volume: 66 Issue: 2 Date: 2024 Pages: 169-199 DOI: 10.2143/SDL.66.2.3293547 Abstract : This article presents a methodological reflection on supervised machine learning, a research design which is gaining ground in digital humanities research. In supervised learning computer models learn to classify data, i.e. to assign data to categories established by the researcher. The classification model is trained on manually annotated data and is then tested on data which were not used in the training. This article argues that so-called ‘wrong predictions’ should be studied carefully since they can help detect and understand ambiguities in the training data and in the categories for annotation. The data for this contribution are gathered from a series of experiments applying machine learning to the study of gender representation in modern Dutch fiction. Based on qualitative analysis of selected ‘wrong predictions’ this article presents methodological reflections on the technical process of annotation, on the reading process and on the ideological dimension of annotation. |
|
