Classifying Temporal Relations Between Events by Deep BiLSTM

Neural networks illustrate their advantages in comparison with traditional classifier-based methods for event temporal relation classification. However, most of them may not be able to explore the deep semantic representation in the larger hypothesis space because of the shallow architectures (e.g., one-layer CNN or RNN). To address this issue, we propose to use deep bidirectional long short-term memory networks (DBiLSTMs) to classify event temporal relations in this paper, where we concatenate the outputs of all prior layers together as the input for the subsequent layer. The experimental results on TimeBank-Dense and Richer Event Description indicate that the proposed DBiLSTMs has outstanding performance over the state-of-the-art methods.