2024
CAVA: A Tool for Cultural Alignment Visualization & Analysis
EMNLP 2024system demonstrations
It is well-known that language models are biased; they have patchy knowledge of countries and cultures that are poorly represented in their training data. We introduce CAVA, a visualization tool for identifying and analyzing country-specific biases in language models.Our tool allows users to identif…