Introduction
Here we summarize some natural language processing tasks. Compare their semantic levels, difficulty, and similarity. We rank and group them into a table.
NLP Tasks
| Task | Input | Type | 
|---|---|---|
| Word tagging | ||
| Word segmentation | sentence | tagging/generation | 
| Shallow syntax - Chunking | sentence | tagging/generation | 
| Named entity recognition | sentence | tagging/generation | 
| Part-of-speech tagging | sentence | tagging/generation | 
| Semantic role labeling | sentence | tagging/generation | 
| Sentence parsing | ||
| Constituency parsing | sentence | parsing/generation | 
| Semantic parsing | sentence | parsing/generation | 
| Dependency parsing | sentence | parsing/generation | 
| Text classification | ||
| Sentiment analysis | sentence | classification | 
| Text classification | sentence or doc | classification | 
| Text pair matching | ||
| Semantic textual similarity | sentence or doc pair | classification/regression | 
| Natural language inference | sentence or doc pair | classification | 
| Text information extraction or analysis | ||
| Temporal Processing | sentence or doc | tagging/generation | 
| Coreference resolution | sentence | tagging/classification | 
| Word Sense Disambiguation | sentence or doc | ranking/matching/classification | 
| Relation Prediction | sentence or doc | classification/generation | 
| Information Extraction | sentence | tagging/generation | 
| Text generation | ||
| Language modeling | sentence or doc | generation | 
| Machine translation | sentence | generation | 
| Simplification | sentence | generation | 
| Summarization | doc | generation | 
| Dialogue | sentence seq | generation | 
| Question answering | sentence or doc | generation/span prediction/matching/ranking/classification | 
Summarize
The main NLP tasks include word level tagging, sentence level parsing, text classification, text pair matching (classification, ranking, etc), fine-grained disambiguation or information extraction, and generation.
Comments
- We can group tasks according to their semantic-level requirements.
 - We can unify most tasks into generation.
 
