A Complete NLP Mind Map: Core Concepts and Techniques

1. Overview

Natural Language Processing (NLP) is a branch of artificial intelligence that focuses on enabling computers to understand, process, and generate human language.

2. Core Branches

NLP consists of two core branches:

Natural Language Understanding (NLU) : aims at representation learning and information extraction to support downstream tasks.

Natural Language Generation (NLG) : converts structured data into human‑readable text.

3. NLU Sub‑modules

The main NLU tasks and typical implementation approaches are:

Word Segmentation – sub‑tasks: ambiguity resolution, out‑of‑vocabulary detection; approaches: dictionary‑based / statistical.

Part‑of‑Speech Tagging – sub‑tasks: POS ambiguity resolution, unknown word tagging; approaches: rule‑based / statistical.

Syntactic Analysis – includes dependency parsing, phrase‑structure parsing, deep‑grammar parsing, and deep‑learning‑based parsing.

Text Classification – sub‑tasks: text representation, machine‑learning classification; approaches: rule‑based / machine learning / neural networks.

Information Retrieval – sub‑tasks: query understanding, resource quality measurement, ranking, evaluation; applications: personalized search, semantic search.

Information Extraction – includes named‑entity recognition, relation detection/classification, event detection/filling, coreference resolution, entity linking.

Proofreading

4. NLG Sub‑modules

Typical NLG tasks include:

Machine translation – rule‑based (dictionary, linguistic rules), statistical (large corpora), end‑to‑end neural models (seq2seq, NMT, attention).

Question‑answering – question understanding, text information extraction, knowledge reasoning; types: retrieval‑based, community QA, knowledge‑base QA.

Automatic summarization – extractive and generative approaches.

5. NLG Pipeline (Six Steps)

The standard NLG process consists of:

Content Determination – decide which information from the data should be included.

Text Structuring – organize the selected information into a logical order (e.g., time, location, teams, overview, result for a sports report).

Sentence Aggregation – combine multiple pieces of information into fewer sentences for fluency.

Lexicalisation – convert the aggregated content into natural‑language expressions, adding connective words.

Referring Expression Generation (REG) – generate domain‑specific expressions, ensuring the vocabulary matches the content domain.

Linguistic Realisation – assemble the final well‑formed sentences.

6. Typical NLG Objectives

Mass‑scale generation of personalized content.

Helping humans gain insight from data.

Accelerating content production.

7. Three Analysis Layers

NLP processing can be viewed through three analytical layers:

Lexical Analysis – includes word segmentation and POS tagging. Segmentation splits input text into individual words; POS tagging assigns categories such as noun, verb, adjective, etc.

Syntactic Analysis – parses sentences to reveal structural relationships. Three mainstream methods are:

Phrase‑structure parsing: identifies phrase structures and hierarchical relations.

Dependency parsing (shallow syntax): captures word‑to‑word dependencies; simple and multilingual but provides limited information.

Deep‑grammar parsing: uses deep grammatical formalisms (e.g., lexicalised tree adjoining grammar, combinatory categorial grammar) to obtain rich syntactic and semantic information, though computationally expensive for large data.

Semantic Analysis – aims to understand the true meaning of sentences. Current mature shallow techniques include Semantic Role Labeling, which builds on syntactic analysis. Joint models that learn multiple tasks together can significantly improve quality but increase complexity and inference time.