Chatbots allow a computer program to interact more naturally with a human being. However, they remain limited in empathy, as they are not aware of the user's state of mind and emotions. Yet, it would help them to provide answers that are more appropriate. EmoContext (SemEval Task 3) has already explored textual emotion detection in English discussions, but no satisfactory dataset is available in French to this date. We propose to translate the EmotionLines dialog dataset, whose English utterances are taken from the Friends sitcom, using its French version. Our translation approach is generalizable to any dataset created from series or movies available in French. Using this translation, we propose a classifier based on the BERT language model, which aims to detect the user's emotion from text. This classifier takes into account the context of the discussion to refine its predictions.