Machine Translation Systems Analysis and Development Prospects

Abstract

In the age of data and reactive knowledge sharing, when information generation time has been reduced from months and weeks to days and hours and the volume of information is growing at a blistering pace, the process of obtaining and systematizing relevant knowledge is rapidly becoming more complicated. Modern tools are required for searching, selecting and processing relevant data as well as for systematizing the acquired knowledge. The authors describe the TRIZ-evolutionary approach as a tool for describing and revealing trends of artificial systems development on the example of machine translation systems. Development of any system is headed to the increase of ideality, i.e. increase of utility while reducing costs. At that, contradictions in the system prevent the ideality from increasing. Development occurs after such contradictions are eliminated. As per the TRIZ-evolutionary approach, the evolution of systems is being described from contradiction to contradiction and indicates TRIZ tools that allowed eliminating the contradictions. Such a description allows not only to systematize knowledge on relevant systems, but also to offer new high-performance solutions. The article covers main generations and paradigms of machine translation such as word-for-word machine translation, rule-based translation, statistical and neural machine translation. The analysis by means of the TRIZ-evolutionary approach allowed to trace the evolution of these systems, define main performance parameters, reveal trends and prospects of development taking into account ideality increase. The process of obtaining new effective solutions by eliminating subsequent contradictions is shown. Tasks for further research into machine translation systems have been set.