What NLP Can Do
The map of tasks.
NLP isn't one thing. It's a whole toolbox of different jobs. When you get a new problem, the first thing to ask is, which of these jobs is it? Name the job correctly, and you've done half of the work.
- NLP is a toolbox
- Name the job first
- That's half the work
The main jobs of NLP
Here are the main tasks. Classification sorts text into categories, such as complaints and refunds. Extraction pulls out facts, like order numbers and names. Search finds the right document. Summarising makes long text short. Translation moves between languages. And generation writes new text. Scribe will do every one of these before this course is over.
Scribe, job by job
Here's our plan for Bright Books. To route each message, we need classification. To find the order number, extraction. To answer from store policies, search and generation, in other words, RAG. To shorten long complaints for staff, summarising. And to serve Spanish speaking customers, translation.
| Need at Bright Books | NLP job |
|---|---|
| Route each message | Classification |
| Find the order number | Extraction |
| Answer from policies | Search + generation |
| Shorten long complaints | Summarising |
| Serve other languages | Translation |
Summary
Six jobs, and one assistant. Now you've got the map. Before any clever model can read, text has to be prepared. Real customer messages are a mess. Next, your first real tool for handling text, and it's over fifty years old.
Key takeaways- Name the job first
- Six main tasks
- Scribe will do them all