1. Natural Language Understanding

Broadly: any computational problem where the input is natural language, and the output is structured information that a computer can store (e.g. in a database) or execute (e.g. a command to a digital assistant).

1.1. Digital Assistants


1.2. Question answering


1.3. Sentiment analysis


1.4. Syntactic parsing


1.5. Semantic parsing


2. Natural Language Generation

Broadly: any computational problem where the input is non-linguistic data (e.g. data, images, sound) and the output is a natural language description of the input.

2.1. Data-to-text generation


2.2. Image caption


3. Machine Translation

Both input and output are text that convey the same meaning, but written in a different language or style.
Philosophically and technically, machine translation requires both NLU and NLG.

3.1. Language Translation


3.2. Text simplification


3.3. Summarization


4. About Deep Learning in NLP

4.1. Idea: Deep learning simplifies machine learning

Why has deep learning taken over NLP?

  • Deep learning simplifies the design of probabilistic models, by replacing complex dependencies and independence assumptions with universal function approximators.
  • Deep learning gives us representation learning: data representations are learned rather than engineered.
  • Learned representations are easy to obtain and reusable, enabling multi-task learning.
  • Deep learning provides a uniform, flexible, trainable framework that can easily mix and match different data types: strings, labels, trees, graphs, data, and images.

In short: deep learning solves the difficulties of applying machine learning to NLP… But it does not solve NLP!

4.2. Problem: Deep learning technology is energy intensive


4.3. Problem: Ethical practice lags technical practice

Modern NLP originated in laboratory experiments with machine learning methods on linguistically annotated text. But NLP has escaped the lab, and can have a direct effect on people’s lives

There are many wider ethical concerns about ML/data science, e.g., privacy. We’ll focus on NLP in the course, but specific problems in NLP often reflect more general problems.

🌸唠嗑QQ群兔叽の魔术工房 (942848525)