Exploiting the Human Computational Effort Dedicated to Message Reply Formatting for Training Discursive Email Segmenters

摘要: In the context of multi-domain and multimodal online asynchronous discussion analysis, we propose an innovative strategy for manual annotation dialog act (DA) segments. The process aims at supporting analysis messages in terms DA. Our objective is to train a sequence labelling system detect segment boundaries. originality proposed approach avoid manually annotating training data instead exploit human computational efforts dedicated message reply formatting when writer replies by inserting his response just after quoted text appropriate intervention. We describe approach, new electronic mail corpus report evaluation segmentation models built.