CONTEXT SENSITIVE TRANSLATION OF ENGLISH WORDS AND PHRASES TO SINHALA IN WEB CONTENT

Abstract

ABSTRACT Online resources are most popular knowledge acquiring technique available in today‟s world. Several authors around the world publish articles which are in different languages in different literacy levels based on the author of the content. However, the readers may not get the full benefit of the online published resource such as article or documentation because an ordinary reader which makes a gap and uncomfortable when reading the particular web page content if it is include many difficult words which the reader does not know. In any language there is an ambiguity present but they are resolve based on the context of the sentence. Homonyms words are having different meanings according to the context for same word. If the context ambiguity is present for a sentence, the readers may frustrated or understand wrong meaning according to his literacy level. Because of this language barrier, readers may avoid reading some of the good contents as well. There are several online translation tools and plugins are available around the world. From this thesis, design and implementation of browser plugin is presented for English based web pages /sites and it is focused on the Sri Lankan people. Non- English user community has been left out because the author assumes that readers having at least mid-level understanding of English language are reading the English based web sources to acquire knowledge. As mentioned earlier there are many online translations tools available, but these are capable of translating whole sentence to the reader according to the preferred language they select. These types of tools do not help readers to learn particular language. But by using the proposed system the readers can mouse-hover to the particular English word that he/she wants to get the Sinhala definition while learning the language unintentionally. Natural language processing with Bidirectional Encoder Representations from Transformers (BERT) is used with machine learning approach in order to make this system success. In this thesis a computer enabling browser plugin with good network connection is only the simplest requirement the particular user need. Google Colab is used to implement the system prototype. User mouse-hovered word with the sentence will input to the system and based on the training model the system present it predicts the Sinhala definition and display as a tool tip. With this approach the reader not frustrated if the context ambiguity is present in the web page because within few milliseconds it will display the output and they can enjoy and acquire knowledge from the content and at the same time they can learn the English language