Ho lam Yu Undergraduate Dissertation 2017/18

Gathering Visually Descriptive Language From Corpora

Supervised by R.Gaizauskas

Abstract

Visually descriptive language(VDL), is a language which describes propositions about an entity which can be confirmed as truth by vision alone. After performing a literature review, I collected a corpus of text and applied pre-processing, feature extraction and various methods of supervised learning to find the best method to find VDL. By comparing the different supervised learning results, for example, F-measures with different feature extraction approaches and choosing the most accurate method, we establish a program that used Support Vector Machine that can analyze a sentence and decide if it is VDL, with an accuracy of around 80%.