Poetry (almost) From Scratch - Teaching versification to a neural network
summary
Describes using neural-network techniques inspired by “Natural Language Processing (almost) from Scratch” to detect and generate poetic verse with minimal hand-crafted assumptions. Covers applying deep learning to learn meter and rhyme from raw text data, and contrasts simple rule-based notions of versification with interpretable and hard-to-interpret learned models.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:25Features of poetry?
- 4:12No Second Troy - Yeats
- 7:20Happy Feet
- 8:55lambic pentameter
- 14:10The Algorithm
- 17:10Stress
- 17:45Model Woes
- 21:17Enter Machine Learning
- 24:48Law of diminishing returns
- 28:36Word Embeddings as a Language Model
- 35:08The unreasonable effectiveness of Recurrent Neural Networks
- 36:27What we have
- 37:53Everything is moderately awesome
- 39:15Raymond Queneau to the rescuel
- 43:02Credits
related talks
38mMachine Learning for Poets: The eMiLy App
03h 22mNonsense verse with Python and machine learning
01h 39mNatural Language Processing (NLP) in Python - From Zero to Hero
24mPyambic Pentameter: generating rhyming and metered poems
49mDeep Learning Applied to Natural Language Processing (NLP)
01h 30mIntroduction to Deep Learning & Natural Language Processing
39mExperimental Creative Writing with the Vectorized Word
28mGPT or how I learned to stop worrying and love the generative AI