Ekspertni sustav za podučavanje japanskog slikovnog pisma kanji

Expert system for teaching Japanese Kanji characters. In this doctoral thesis, we presented the model of an expert system for teaching or learning Japanese logographic characters kanji. Expert system is a field of artificial intelligence, a rule-based system that helps users in decision-making and consists of an inference engine and a knowledge base. In this thesis, we designed the model of an expert system that helps the user decide the optimal kanji learning order. A user inputs kanji-learning-related parameters and is given a character list as output. The input parameters include a textbook used, goal number of characters, students' level, student's goals, and omitted characters. We learned that the most common kanji in Japanese texts are more useful to know earlier in one's studies because they follow Zipf's law and one is able to understand more characters in text by knowing more frequent characters. In addition to that, many kanji characters are made from components that are characters themselves. It is easier to memorise the parts before the whole, or components before the complex characters. Therefore we used two basic principles in optimisation of learning order. Firstly, the system takes into account the “part before whole” principle and uses an adapted topological sort algorithm. Secondly, it computes the relative weights of characters based on their frequency in corpora, such as literature, newspaper, Wikipedia, and Twitter web corpus. The expert system was evaluated by using one of the system’s output throughout a two-semester study at a university-level Japanese language course. In a kanji-focused module, the students (N ? 43) evaluated the kanji learning order made by the system. Comparing it to the textbook order, they rated it more favourably (3.023 out of 5 compared to 4.027 out of 5) and 72% agreed or strongly agreed that the new order improved their learning process. Additionally, the participants were very happy with this learning method, rating it 4.476 out of 5 on average, and passed the module with high marks (60% of students achieved A+, while 77.5% achieved either A+ or A). Designing the model of this expert system and evaluating it in the practical teaching experiment lead us to believe that expert systems do have a role in education, specifically in curriculum design and class planning. We assert that the application of an expert system in the field of teaching kanji can help both teachers and learners to better organise and plan their learning and achieve desired results quicker.