In this paper, we proposed and evaluated a novel method to find mistypes in documents based on Bayesian theory. We considered that the characters entered just before mistyping are factors causing input errors. First, log data of key input is acquired for each user. Next, we obtain only the characters entered just before mistyped characters from the log data then analyze them. Finally, with these parameters, using the Bayes' theorem formula, the probability of mistaking the character immediately after the character which becomes the factor of mistype is calculated for each character. Under the cooperation of students, we confirmed there is a habit of keyboard input that cause mistype for each user, which we supposed for our method. Using documents from students that were made by themselves, we verified our method. Comparing characters that were placed just before mistypes found by eyes in a document and the characters with high probabilities that were calculated for that student by our method produced match rate of 93% in a realistic case.

View full abstract

Preventing truancy and expulsion (in this paper, referred to collectively as “dropouts”) is an extremely important task for educational institutions. Individual conference is a realistic measure for preventing such dropouts. However, there are various issues with holding individual conferences, such as the skills of the conference holder and the personnel cost. This paper presents the anomaly detection method as a method for predicting which students will drop out using data provided by an educational support system. This method is expected to reduce the issue of personnel cost and heighten the overall effectiveness of individual conferences.

View full abstract

We propose a Japanese utterance education support system for non non-native speakers using fundamental frequencies and original phonetic segments. In the system, after automatically extracting syllables using phonetic segment labels, accent and intonation are evaluated by comparing the estimated F0 of each syllable to the criteria which were obtained from the Japanese speech database. The system also provides suggestions for improvement for the learners. An experiment was conducted to evaluate accent kernels of long vowels and syllabic nasals, vowel devoicing, intonation in questions, and tonal phrases uttered by the learners. Of the 120 data items in the experiment, 31 were erroneous utterances. Of such 31 erroneous items, however, improvement was observed among 29 items with the help of the system. Since the system quantitatively evaluates utterances of Japanese learners, it was shown that it is effective for teachers and learners.

View full abstract

This study explored the challenges faced by Japanese language teachers in conducting online classes for groups of students at the beginner level. Unstructured interviews with three Japanese language teachers were conducted to examine this issue. An analysis of the data collected from the interviews revealed certain factors causing teachers’ dissatisfaction with overall class management in online classes, such as in handling practices that they normally provide in a face-to-face class, structuring the learning environment, or understanding the motivation of individual leaners, as they cannot watch over all students. The teachers expressed that simply getting accustomed to conducting online classes for groups of students could reduce their dissatisfaction, but felt that this may be akin to giving up. They also voiced that technological support offered by IT specialists, such as changing the Internet network or the online communication system, could offer limited solutions．

View full abstract

Among learning management systems (LMSs), there are proprietary learning-support systems that have been developed by researchers at different universities. Often time, these researchers also manage both research and education in the laboratories and lectures for which they are responsible. However, when these LMSs are rolled out for an entire university, it can be difficult for individual faculty members to adequately assess and respond to the high costs associated with the infrastructure or to manage the costs. This study addresses this type of virtual private server and proposes a framework for a personal LMS that is able to manage a proprietary LMSs, without assuming an immediate large-scale rollout and while keeping the scope of the system restrained, such that the costs and labor for the system can be handled by an individual. As a result, while maintaining focus on the scale of e-learning in addition to operating the system accounting for cost and management aspects, I was able to avoid running the risks anticipated earlier on.

View full abstract

For the practice of an extensive reading of English texts, it is thought that keeping providing texts with learners' suitable levels seems desirable. Because the difficulty of texts is more or less influenced by the percentage of unknown words in them, recording the status of "known / unknown" of words for individual learners is one of the key issues. On the other hand, it is obviously burdensome and time-consuming to record the information of a large number of English words. In this paper, the authors have devised an algorithm to estimate the "known / unknown" status of words in SVL12000 wordlist for individual learners. A simulation was also conducted using the real data for the comparison with other algorithms. The result shows that it worked out well for some learners though there still exist drawbacks of the algorithm.

View full abstract