ID No.: 2015HZ67004
NAME OF THE STUDENT: Shalaka Vishnudas Malu
DISSERTATION TITLE: Distributional and Compositional Semantics
ABSTRACT
Language has been important step in the process of evolution. With time, technology has become an integral part of human life, which necessitates us to be able to communicate with this technology. Hence, it is necessary that the machines also are able to process and communicate using natural language. Semantics is the field that deals with the study of meaning; it is the theory of understanding the meaning of the words and sentences used in natural language. Computational semantics is the automation of language. Among the various methods used to compute the meaning of the language, distributional semantics has become increasingly popular. Distributional semantics concentrates on the meaning of individual words. It is based on a hypothesis that the meaning of a word depends on its context. Hence, words that share the same context are assumed to be similar. This helps in many applications in the field of natural language processing like synonym detection, automatic generation of thesaurus, information retrieval, categorization of words, solving analogy problems, etc. The benefits of distributional semantics over the traditional models of semantics are that it is language independent, involves simple representation, and is obtained completely automatically.
Compositional semantics deals with how the words combine to form the sentences and how the sentences get their meanings. There are many different models to combine different words in an expression like addition, multiplication, regression, recursive neural network (RNN) model, etc. of which RNN model has been found to perform the best on various applications.
There are still many challenges in the computational semantics and a long way for the machines to understand the meaning of natural language as accurately as human beings. Among all the current models, compositional and distributional semantics have been found to give the best result.
