Modern Practical Natural Language Processing

This course will cover how you can use NLP to do stuff.

There are four videos

1. Overview and Converting Text to Vectors

• For finding similar documents
• "I have this document or text, what others talk about the same stuff?"
• Video

2. Learning with Vectors and Classification

• For classifying documents
• "I need to put these documents into buckets."
• Video

3. Visualizing

• For seeing what document vectors look like in 3D space
• "I need to quickly see what looks similar to what."
• Video

4. Sequence Generation and Extracting Pieces of Information from Text

• For translation and document summarization, and for pulling out sentences and documents that talk about specific things
• "I need every mention of a street address or business in Garland, Texas; and I need each document translated to Urdu."
• Video

Additional Details

The idea is we make short videos that focus on the aspects of NLP that currently work well and are useful.

Speech-to-text now works pretty well, so these methods will also be useful for the audio portions of videos.

All code will be available on GitHub here https://github.com/jmugan/modern_practical_nlp

About Me, Jonathan Mugan

• PhD in Computer Science in 2010 from UT Austin
• Thesis work was about how a robot could wake up in the world and figure out what is going on
• Work at DeUmbra where we build AI for the DoD
  • We also work in healthcare, which I can talk about. A future video (not in this series) will cover how we use graph neural networks to identify who is at risk for opioid overdose
  • Hands-on technical advisor at our sister company Pulselight
• Wrote The Curiosity Cycle: Preparing Your Child for the Ongoing Technological Explosion
• Advisor at KUNGFU.AI
• Also do independent consulting work
• Can find me here jonathanwilliammugan@gmail.com or on Twitter at @jmugan

The Limits of NLP

Computers can't read

• Reading requires mapping language to internal concepts grounded in behaving in the same general environment as the writer.
  • Computers don’t have those concepts.
  • Example: “I pulled the wagon.” Computers don’t know that wagons can carry things or that pulling exerts a gentle tension to the arm and leg muscles as one walks.

Computers can't write

• Writing requires mapping internal concepts grounded in behaving in the same general environment as the expected reader.
  • Computers don’t have those concepts

NLP Works Around Computers Not Having the Experience or Conceptual Framework to Read and Write

• NLP is about how to make natural language amenable to computation even though computers can’t read or write.
• Representing text as vectors has transformed NLP in the last 10 years.
• There are also symbolic methods that are practically useful; we will cover those too.

Additional Information on NLP, AI, and Their Limits and Promise

• Microsoft Flight Simulator 2020 is an inflection point for virtual worlds and our own
• Generating Natural-Language Text with Neural Networks
• Why Is There Life? and What Does It Have to Do with AI?
• Chatbots: Theory and Practice
• You and Your Bot: A New Kind of Lifelong Relationship
• Computers Could Understand Natural Language Using Simulated Physics
• The Two Paths from Natural Language Processing to Artificial Intelligence
• DeepGrammar: Grammar Checking Using Deep Learning
• Deep Learning for Natural Language Processing
• What Deep Learning Really Means
• My O'Reilly course on NLP: Natural Language Text Processing with Python