Poetry - Teaching Versification to a Neural Network

Explore the intersection of poetry and artificial intelligence in this Strange Loop Conference talk. Delve into the challenges of teaching versification to neural networks, examining the delicate balance between constraints and expression in poetry. Learn about approaches to cracking the structure and meaning of text, from reductionist methods to modern machine learning techniques. Discover how to identify poetic elements in unexpected textual sources and generate poetry using minimal assumptions. Investigate the tension between classical rules and deep learning models, touching on topics such as the elusive definition of poetry and the emergence of meter and rhyme rules. Follow along as the speaker discusses various poetic features, algorithmic approaches, machine learning models, and the effectiveness of recurrent neural networks in poetry generation. Gain insights into the challenges and successes of applying artificial intelligence to the art of poetry, with references to works by Yeats and Raymond Queneau.

Intro
Features of poetry?
No Second Troy - Yeats
Happy Feet
lambic pentameter
The Algorithm
Stress
Model Woes
Enter Machine Learning
Law of diminishing returns
Word Embeddings as a Language Model
The unreasonable effectiveness of Recurrent Neural Networks
What we have
Everything is moderately awesome
Raymond Queneau to the rescuel
Credits