Automatic generation of poetry

The automatic generation of poems is introduced as follows:

1. can automatically generate a "text similar to ancient poetry". It can even meet the metrical requirements. Meet the requirements of confrontation. That text is much better than this example.

2. For the regularization method, please see "The Poetic Machine of the Old Farmer in Xiang Tao". For the practice of pure deep learning, please refer to Tsinghua's paper on Natural Language Processing Group Yi (arxiv 1608+037).

However, in the eyes of people who have read the poems carefully, it is easy to see that the above two sentences were not written by people.

Even if one day the machine can generate "ancient poems" that readers can't distinguish, its meaning can't be equal to the poems written by people. Poetry is just a form. What is really lovely is the god the author wants to convey.

5. Things that are difficult for people are not necessarily difficult for machines; Or conversely, it is not necessarily difficult for machines to achieve the results that people in Dallas can't achieve in the audience; or vice versa, Dallas to the auditorium

6. Not for the subject. About a hundred years later, people will only think that what we say today is rare and strange.

7. In the eyes of people who know both old-style poems and nlp, the level of these poems is far from the level of generating news reports. Judging from the amount of data, it is much easier to make modern poems than old ones. Suppose there is an index to evaluate the degree of simulation. If the ordinary literary generation scores 5 points, the old poetry generation scores 2 points at most.

8. It's just that many people don't understand nlp, let alone old poems.