Translation with user computation

Abstract

Human Computation aims to solve problems by means of combining the power of human and computer computation. The typical approach is to assign tasks to computers and humans based on what they are better capable of performing. The aim is to bring human-computer relationship into a symbiotic mutualism, where humans compute what they are good at and vice versa. Games with a purpose introduced a concept that harvests human brain power. They try to solve a large scale complex problem with a game structure. In our work, we present a model with an objective of improving machine translation. TURC, Translation with User Computation, is a web-based multiplayer online game. TURC focuses English-Turkish translation. As there are large amounts of English data available in the web, the lack of Turkish is disappointing. There are available machine translation efforts between these languages, such as Google Translate. However a speaker of both languages can easily tell that these efforts are insufficient. Our model is designed to generate output to help improving machine translation while making people enjoy translation with the element of competition. Two non-communicating players are presented a machine translation of an English sentence. They have the options of adding, editing, deleting or moving a word available. Their aim is trying to make the same actions. A score value is generated according to the correspondence of these actions. A prototype of our game model is hosted on a server for two weeks and the results are discussed.