Automatic Generation and Evaluation of Chinese Classical Poetry with Attention-Based Deep Neural Network

The computer generation of poetry has been studied for more than a decade. Generating on human level is still great challenge the computer-generation process. We present novel Transformer-XL based classical Chinese model that employs multi-head self-attention mechanism to capture deeper multiple relationships among characters. Furthermore, we utilized segment-level recurrence learn longer-term dependency and overcome context fragmentation problem. To automatically assess quality generated poems, also built automatic evaluation contains BERT-based module checking fluency sentences tone-checker evaluate tone pattern poems. poems using our obtained an average score 9.7 10.0 pattern. Moreover, visualized attention mechanism, it showed learned tone-pattern rules. All experiment results demonstrate can generate high-quality