Skip to content
NLP Foundations

Meaning as Vectors

The NLP Task Zoo

9 min+85 XPStep 1/10

A small zoo of canonical tasks

Classic NLP organized itself around a handful of repeatable jobs:

  • Text classification — assign a label: spam/ham, positive/negative, topic
  • Token classification (NER) — label each word: 'Apple hired Sam in Paris' → ORG, PERSON, LOCATION
  • Extractive QA — find the answer span inside a passage
  • Seq2seq tasks — translation, summarization: text in, different text out

For decades each had bespoke models and bespoke datasets — a whole subfield per animal.

warning

'Apple' in that sentence is an ORG only because 'hired' demands it — NER is where context became non-negotiable, years before attention.