Investigation on the Use of Mora in Assessment of L2 Speakers’ Japanese Language Proficiency

This paper presents the analysis on the language use of non-native (L2) Japanese speakers with focus on their proficiency in group discussion conversations. Due to the demographic development the necessity to close the resulting gap with foreign employees/workers and thus, non-native speakers is increasing. This work is based on a corpus collected in an experiment which acquired multimodal sensory data in collaborative tasks with unbalanced mixed setup, composed of one none-native speaker and three native (L1) speakers. Each group was given the task to discuss two topics and find a joint decision. This work aims to find the insights how the proficiency of the speakers and the difference in being a native or non-native speakers changes the used vocabulary and decision-making process, which will later be a major issue to ensure an efficient work of such mixed groups. The analysis is based on findings of a total number of seven groups and thus seven L2 speakers. The analysis is on the mora duration of the participants during the discussion, the stability of L1 and L2 speakers are compared with overall statistical information. We also introduced the use of one-dimensional convolution neural networks to analyze micro characteristics of the sequence of moral durations. The results show that it is positive to adopt the characteristics in automatic assessment on the proficiency level of L2 speakers.