Dependency Parser
Last updated
Last updated
A natural language parser is a program that works out the grammatical structure of sentences, for instance, which groups of words go together (as "phrases") and which words are the subject or object of a verb.
Each sentence consists of multiple tokens, and those tokens are syntactically/grammatically dependent on each other. For example a sentence like this one: I love you have a nominal subject I, a direct object you and a verbal phrase love. So the verb love can be said to be a parent of I and you, and since it doesn't have a parent, we describe it as the root of the sentence.
A little bit more elaborate example is : The boss gave me a white pen. This sentence also have a root gave and this root has children which are parents of their own children. You can imagine a visual tree-like representation of this sentence as follows:
You can notice from the above chart that:
The is a determiner, it's parent is boss.
Boss is a nominal subject, it's parent is gave.
gave is a verbal phrase, it has no parent (this it's the master of the sentence).
me is a direct object, it's parent is gave.
a is a determiner, it's parent is pen.
white is an adjective, it's parent is pen.
pen is an indirect object, it's parent is gave.
Dependency parsing have a set of rules that every tree adheres to:
Sentence root has no parents.
Sentence root is usually a verb that dictates the meaning of the sentence.
A sentence can only have one root.
A token can only have one parent.
A token can have any number of children, or no children at all.
So it really is a tree-like structure.
Each parent-child relationship has an annotation for it. Take the sentence above for example, me is a child of gave and the relationship between the two can be annotated as DOBJ which means Direct Object.
For a full list of the annotation specifications, refer to the Dependency Parsing Annotations Specifications article.
