Cross-lingual Emotion Intensity Prediction

摘要: Emotion intensity prediction determines the degree or of an emotion that author expresses in a text, extending previous categorical approaches to detection. While most work on this topic has concentrated English texts, other languages would also benefit from fine-grained classification, preferably without having recreate amount annotated data available each new language. Consequently, we explore cross-lingual transfer for detection Spanish and Catalan tweets. To end annotate test set tweets using Best-Worst scaling. We compare six approaches, e.g., machine translation embeddings, which have varying requirements parallel – millions sentences completely unsupervised. The results show data, methods with low parallel-data perform surprisingly better than use more explain through in-depth error analysis. make dataset code at https://github.com/jerbarnes/fine-grained_cross-lingual_emotion.