For a serious NLP pipeline, you usually separate the analysis into layers.

A sentence contains multiple kinds of information simultaneously:

1. Lexical Layer

“What are the words?”

• tokenization
• normalization
• lemmatization
• stemming
• subword splitting
• vocabulary lookup

Example:

running -> run

2. Grammatical Layer

“How do the words function grammatically?”

Includes:

• POS tagging
• morphology
• tense
• plurality
• gender
• person
• case
• agreement

Example:

running = VERB, present participle
dogs = plural noun

3. Syntactic Layer

“How are words structurally connected?”

Includes:

• dependency parsing
• constituency parsing
• phrase extraction
• subject/object detection

Example:

John ate pizza

subject = John
verb = ate
object = pizza

This is sentence structure.

4. Semantic Layer

“What does it mean?”

Includes:

• word sense disambiguation
• semantic roles
• semantic parsing
• entity linking
• synonym resolution
• concept extraction
• ontology mapping
• semantic similarity

Example:

bank = financial institution

vs

bank = river edge

5. Pragmatic Layer

“What is implied by context?”

Includes:

• intent detection
• sarcasm
• politeness
• indirect meaning
• conversational implication

Example:

Can you open the window?

Literal: question

Pragmatic meaning: request

6. Discourse Layer

“How do sentences connect together?”

Includes:

• coreference resolution
• topic tracking
• discourse relations
• context continuity
• memory linking

Example:

John entered the room.
He sat down.

"He" = John

7. Knowledge Layer

“How does this connect to world knowledge?”

Includes:

• knowledge graphs
• ontologies
• factual relations
• commonsense reasoning
• temporal knowledge
• causality

Example:

doctor -> medical profession
hospital -> place of work

8. Statistical Layer

“How common or probable is this?”

Includes:

• language modeling
• frequency analysis
• perplexity
• n-grams
• probability scoring

Example:

strong tea

is more probable than

powerful tea

9. Vector / Embedding Layer

“How is meaning represented numerically?”

Includes:

• word embeddings
• sentence embeddings
• contextual embeddings
• transformer hidden states
• semantic vector spaces

Example:

king - man + woman ≈ queen

10. Logical Layer

“Can meaning be represented formally?”

Includes:

• symbolic logic
• predicate logic
• theorem representations
• formal semantics

Example:

All humans are mortal.
Socrates is human.
=> Socrates is mortal.

11. Emotional / Affective Layer

“What emotion or tone exists?”

Includes:

• sentiment analysis
• emotion classification
• toxicity
• urgency
• mood detection

Example:

I love this

= positive sentiment

12. Phonetic / Speech Layer

“How does it sound?”

Includes:

• phonemes
• stress
• prosody
• pronunciation
• speech recognition

Useful for:

• TTS
• ASR
• speech NLP

13. Information Extraction Layer

“What structured data can be extracted?”

Includes:

• named entities
• dates
• relationships
• events
• triples
• slot filling

Example:

Apple acquired Beats in 2014

Extract:

(Apple, acquired, Beats)

14. Reasoning Layer

“What conclusions can be inferred?”

Includes:

• deduction
• induction
• abductive reasoning
• causal reasoning
• chain-of-thought
• planning

Modern LLM systems combine many of these implicitly.

Traditional NLP systems usually build them explicitly as separate pipelines.

A full advanced NLP stack often looks like:

Raw Text
→ Tokenization
→ POS Tagging
→ Parsing
→ NER
→ Coreference
→ Semantic Parsing
→ Embeddings
→ Knowledge Linking
→ Reasoning
→ Generation

WordNet mainly helps in:

• lexical semantics
• ontology
• synonym relations
• semantic disambiguation

SpaCy mainly helps in:

• grammatical
• syntactic
• entity extraction

Transformers mainly help in:

• contextual semantics
• embeddings
• reasoning
• generation
• statistical language modeling