“Le masculin l’emporte sur le feminin” - Pourquoi est-ce qu’il importe: Using Word Embeddings to Quantify the Effect of French “L’Écriture Inclusive”

Abstract

The French language is characterized by grammatical domination of the masculine gender over the feminine gender. It is widely posited that this grammatical gender inequality manifests itself as social gender inequality in French society. As a result, many feminist thinkers advocate for the widespread use of “écriture inclusive,” a collection of grammatical practices that seek to improve the visibility of women in the written language. Word embedding language models capture social and cultural biases in their mathematical encodings of words as vectors. These embeddings can be used to quantify social ideologies and social change. The goal of this work is to quantify the effects of “écriture inclusive” at improving gender equality. This is done through the construction of a French corpus dataset in which these writing practices appear frequently. This dataset is then used to train a word embeddings language model. The values of these word embeddings are then evaluated on the basis of their mathematical alignment with either the masculine or feminine gender. The results quantify the effects that “écriture inclusive” has on the status of women in French society and show that less masculine values are encoded in word vectors when using inclusive language word forms.