أساسيات تحليل التبعية

POS tags name each word's job, but not how words connect. Dependency parsing reveals which word depends on which, building a sentence map. 🔗

A parser arranges words into a tree. One word becomes the root, and every other word hangs off it through a labeled link.

The root is usually the main verb that anchors the sentence. In The cat chased the mouse, the root is chased.

Every word has a head, the word it attaches to. The words attached to it are its children, forming the branches of the tree.

Each link has a label, the dependency relation. A subject is tagged nsubj, while a direct object is tagged dobj.

The same model that tags words also parses them. After nlp(), every token already carries its dependency information, ready to read.

Use token.dep_ for the relation and token.head for the word it attaches to. Together they describe the token's place in the tree.

To get who does the action, look for the token whose dep_ equals nsubj. That word is the grammatical subject of the verb.

Unsure what a label means? Pass it to spacy.explain() and get a clear description, just like you did for POS tags.

Render the parse with displacy. Set the style to dep and it draws labeled arrows from each head to its children.

Dependencies unlock real meaning, like linking an action to its actor and target. This structure powers extraction, search, and question answering.

Recall how subjects are marked in a parse.

You saw sentences as trees of heads and children, read dep_ in spaCy, and drew the parse with displacy. Now you can map how words relate. 🎉