Bleaching Text: Abstract Features for Cross-lingual Gender Prediction

TL;DR

This paper introduces 'bleaching text', an abstract feature transformation that improves cross-lingual gender prediction, demonstrating comparable human and model performance and surpassing lexical approaches.

Contribution

It proposes a novel text transformation method that enhances transferability in cross-lingual gender prediction and provides the first comparison with human performance.

Findings

Bleached features outperform lexical models in cross-lingual transfer.

Humans perform similarly to bleached models in gender prediction.

Bleached features enable better cross-lingual transfer than embeddings.

Abstract

Gender prediction has typically focused on lexical and social network features, yielding good performance, but making systems highly language-, topic-, and platform-dependent. Cross-lingual embeddings circumvent some of these limitations, but capture gender-specific style less. We propose an alternative: bleaching text, i.e., transforming lexical strings into more abstract features. This study provides evidence that such features allow for better transfer across languages. Moreover, we present a first study on the ability of humans to perform cross-lingual gender prediction. We find that human predictive power proves similar to that of our bleached models, and both perform better than lexical models.