The Undesirable Dependence on Frequency of Gender Bias Metrics Based on Word Embeddings
This reveals a critical flaw in widely used bias quantification tools for NLP, potentially misleading researchers and practitioners in fairness assessments.
The study investigated how word frequency affects gender bias metrics based on word embeddings, finding that methods like Skip-gram and GloVe produce spurious male or female bias depending on frequency, even with shuffled data, while an alternative PMI-based metric shows less frequency dependence.
Numerous works use word embedding-based metrics to quantify societal biases and stereotypes in texts. Recent studies have found that word embeddings can capture semantic similarity but may be affected by word frequency. In this work we study the effect of frequency when measuring female vs. male gender bias with word embedding-based bias quantification methods. We find that Skip-gram with negative sampling and GloVe tend to detect male bias in high frequency words, while GloVe tends to return female bias in low frequency words. We show these behaviors still exist when words are randomly shuffled. This proves that the frequency-based effect observed in unshuffled corpora stems from properties of the metric rather than from word associations. The effect is spurious and problematic since bias metrics should depend exclusively on word co-occurrences and not individual word frequencies. Finally, we compare these results with the ones obtained with an alternative metric based on Pointwise Mutual Information. We find that this metric does not show a clear dependence on frequency, even though it is slightly skewed towards male bias across all frequencies.