Les apports du Traitement Automatique du Langage (TAL) à l'évaluation des facteurs de lisibilité du néerlandais en tant que langue étrangère

Details

Supervisors

Degand, Liesbeth

Faculty

Faculté de philosophie, arts et lettres

Degree label

Master [120] en linguistique, à finalité spécialisée: traitement automatique du langage

Abstract

This thesis explores the contributions of Natural Language Processing (NLP) to the assessment of the readability of Dutch as a foreign language. Given the growing importance of language learning and the status of Dutch as an official language of the European Union, which is particularly relevant in the Belgian context, this study aims to fill a gap in research. The main objective is to identify the most relevant linguistic variables (lexical and syntactic) for predicting the difficulty of a text for French-speaking learners. The methodology is based on a quantitative and empirical approach. A corpus of 178 texts, classified according to CEFR levels, was compiled from textbooks and authentic sources. NLP techniques were used to extract 18 linguistic variables, the relevance of which was assessed using statistical analyses (ANOVA, multiple linear regression). The results show that lexical variables, in particular the proportion of words not found in reference lists and concreteness, are robust predictors of difficulty. In terms of syntax, the proportion of sentence-final verb (SOV) structures is also significant. The multiple linear regression model developed explains approximately 65% of the variance in text difficulty, confirming the validity of a multidimensional approach. This research constitutes the first systematic modeling of the readability of Dutch as a foreign language and lays the foundations for the future development of objective and appropriate assessment tools.