NATURAL LANGUAGE PROCESSING TECHNIQUES USED FOR AN AUTOMATIZED TEST GENERATION PROCESS FOR TURKISH

Educational systems force the students to deal with numerous exams. It can be really difficult to keep motivation up while handling those exams. In today’s world, information technologies have been spread over to all levels of business sectors and industries, including education. Therefore, to make the studying more instructive and enjoyable, it is a good approach to use information technologies on this issue.

Brief Description and Project Goals:
This paper is about a research that aims to suggest NLP (Natural Language Processing) methods for Turkish language to develop a user-friendly and goal-oriented educational software. Primary purpose of this research is to provide a computer mediated self-study opportunity for students. In line with this target, the project includes the processing of text-based Turkish lecture notes of secondary education students and automatic test generation. Under favour of the generated questions, students will get the chance to test and evaluate themselves and make progress on the particular courses they need. Users will also be able to keep track of their progression in time and compare their results with other users. By collecting the meaningful and qualified questions in time, to create a question bank for further usage and allow users to benefit from this service is the long-term goal of the project.

Method of the Project:
The working principle includes basically two steps: Students load their lecture notes and choose a test type (Three different test types are offered to the users which are true-false, fill in the blanks and multiple choice). Thus, a proper test is generated. Test generation process is divided into four main tasks; which are mostly in the scope of NLP: Classification of the input lecture notes, using the “glossary of terms” structure for related course (Geography and History are the courses included to be worked on), identification and conversion of positive-negative verbs and finding phrases of sentences. 1200 proper documents (600 for geography and 600 for history) are collected to construct the glossary of terms structures and perform experiments. Both statistical and rule-based methods are used to deal with the problems encountered because of the agglutinating nature of the Turkish language.

Advantages and Innovator Value of the Project:
This project offers the user a decent time gain; as it simplifies an examination process by isolating students from the question preparation. The choice of the lecture notes is in the user’s hands; so possible changes on syllabus won’t affect the validity of the project in a negative way. As the generation and storage of the test questions are done in the electronic environment, it offers a paperless self-education to the students. Besides the educational software developed, this research draws conclusions about some of the major NLP tasks for Turkish; like sentence boundary detection, morphological analysis, POS (Part-of-Speech) tagging, finding phrases of sentences and document classification.