We present a new challenging stance detection dataset, called\nWill-They-Won't-They (WT-WT), which contains 51,284 tweets in English, making\nit by far the largest available dataset of the type. All the annotations are\ncarried out by experts; therefore, the dataset constitutes a high-quality and\nreliable benchmark for future research in stance detection. Our experiments\nwith a wide range of recent state-of-the-art stance detection systems show that\nthe dataset poses a strong challenge to existing models in this domain.\n