Dependency Parsing Tutorial Section

Dependency Parsing

Understand how to map direct grammatical relationships between words without hierarchical phrasing.

Dependency Parsing

Unlike Constituency Parsing (which builds nested phrase boxes), Dependency Parsing directly draws directional linkages between the words themselves. It maps out which word is the grammatical "Head" and which word is its "Dependent".

Dependency parsing is much more popular than constituency parsing in modern NLP (especially through spaCy) because it is faster, lighter, and easier to use for Information Extraction.

Head vs Dependent

Dependency links are directional (). In the phrase "clever dog":

"dog" is the Head word (ROOT concept).
"clever" is the Dependent word (it modifies the dog).
Link type: amod (Adjectival Modifier).

Visualizing Dependencies

The
det

clever
amod

dog
nsubj

chased ROOT

the
det

ball
dobj

Universal Dependency Tags

Tag	Meaning	Example Relation
nsubj	Nominal Subject	Provides the subject to the verb. (dog → chased)
dobj	Direct Object	The entity acted upon. (chased → ball)
amod	Adjectival Modifier	Modifies the noun. (clever → dog)
prep	Prepositional Modifier	Links preposition to its head. (sat → on)
ROOT	Root of Sentence	The central active verb. (chased)

Extracting Dependencies with spaCy

import spacy

nlp = spacy.load("en_core_web_sm")
doc = nlp("The clever dog chased the ball.")

# Print Subject-Verb-Object relationships!
print(f"{'Text':<10} | {'Dependent Tag':<15} | {'Head Word':<10}")
print("-" * 40)
for token in doc:
    print(f"{token.text:<10} | {token.dep_:<15} | {token.head.text:<10}")

'''
Output:
Text       | Dependent Tag   | Head Word 
----------------------------------------
The        | det             | dog       
clever     | amod            | dog       
dog        | nsubj           | chased     <-- We found the subject!
chased     | ROOT            | chased     <-- Core verb of sentence
the        | det             | ball      
ball       | dobj            | chased     <-- We found the object!
.          | punct           | chased    
'''

Previous: Constituency Parsing