Abstract: This study tackles the language-dependent bias in Visual Question Answering (VQA) models, where models tend to rely heavily on the semantic relationship between the question and predefined ...