CL1.101 - Introduction to Linguistics - 1 | Intro to Linguistics Guest Lecture

This is a guest lecture by Prof. Dipti Misra

Sentence Structure

Sentences consists of words, each of which belong to certain word classes and have functions within the structure of the sentence.
Words in sentences often form constituents, which can be analysed as single units in addition to their individual members.

Constituent Analysis and Dependency Grammar

Constituent analysis is based on Phase Structure Grammar (PSG) Here the constituent structure can be represented as a tree, with a heirarchy displaying the relations between the constituents.

For the sentence:
Hari eats two bananas every morning
Phrase Structure Tree

graph TD
    S --> NP_1
    S --> VP_1
    NP_1 --> N_a
    N_a --> Hari
    VP_1 --> V
    V --> eats
    VP_1 --> NP_2a
    VP_1 --> NP_2b
    NP_2a --> det_b
    NP_2a --> N_b
    det_b --> two
    N_b --> bananas
    NP_2b --> det_c
    NP_2b --> N_c
    det_c --> every
    N_c --> morning

In contrast, we have another way of looking at the sentence and the relations between the words in it known as Dependency Grammar
If we were to parse the same sentence that we saw previously, we would get the following:
Dependency Tree

graph TD
    eats -- subj --> Hari
    eats -- obj --> bananas
    eats -- time --> morning
    bananas -- det --> two
    morning -- det --> every

If we look at the PSG parsing, we note that every word falls under a constituency and the sentence itself forms the root of the tree.
Each leaf is a word.
However in dependency grammar, the words form nodes at all levels of the trees and there is one single word which is the root of the sentence.
Phrase Structures Trees preserve the structure, i.e the word order of the sentence, however Dependency Trees do not.

The nature of the dependencies is decided on the basis of meaning. The dependencies capture semantic relations between the words and their dependent.

English has a relatively strict word order that lends itself to working well with PSG parsing, however in recently DG has grown more favourable with those in the field of natural language processing.
This is because DG firstly works better with languages with more flexible word order, including many ILs, and since it focuses on the relations of meaning.
Probablistic and Statistical methods which were used to infer data from large databanks have now largely been replaced by machine learning methods.