Deep Multi-task Learning with Cross Connected Layer for Slot Filling

Abstract

Slot filling is a critical subtask of Spoken language understanding (SLU) in task-oriented dialogue systems. This is a common scenario that different slot filling tasks from different but similar domains have overlapped sets of slots (shared slots). In this paper, we propose an effective deep multi-task learning with Cross Connected Layer (CCL) to capture this information. The experiments show that our proposed model outperforms some mainstream baselines on the Chinese E-commerce datasets. The significant improvement in the F1 socre of the shared slots proves that CCL can capture more information about shared slots.