论文信息 - Haiku Generation Using Deep Neural Networks

Haiku Generation Using Deep Neural Networks

Japanese Haiku is a cultural heritage with a history of more than six hundred years. It is a traditional form of Japanese poetry that expresses facts, seasons, emotions, and reasoning in a compact way. One Haiku poem consists of 17 音 (on, also known as morae though often loosely translated as “syllables”) which are separated into 3 columns and is frequently written in a right-toleft way (Figure 6). The beginning and ending columns of a Haiku have 5 syllables (i.e., Hiragana ons) and the middle line has 7 syllables. The ending words of each columns are not limited to be rhymed. For example, the following famous Haiku:

[1] Fei-Fei Li,et al. Visualizing and Understanding Recurrent Networks , 2015, ArXiv.

[2] Mirella Lapata,et al. Chinese Poetry Generation with Recurrent Neural Networks , 2014, EMNLP.

[3] Jürgen Schmidhuber,et al. Long Short-Term Memory , 1997, Neural Computation.

[4] Lantao Yu,et al. SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient , 2016, AAAI.

[5] Wojciech Zaremba,et al. An Empirical Exploration of Recurrent Network Architectures , 2015, ICML.

[6] Lawrence D. Jackel,et al. Handwritten Digit Recognition with a Back-Propagation Network , 1989, NIPS.

[7] Yoshua Bengio,et al. Generative Adversarial Networks , 2014, ArXiv.

[8] Yoshua Bengio,et al. Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation , 2014, EMNLP.

[9] Alexander M. Rush,et al. Character-Aware Neural Language Models , 2015, AAAI.