2018 Fiscal Year Research-status Report
Multilingual Socio-Emotional Communication Support for Improving Foreign Student Adjustment and Mental Health Outcomes
Project/Area Number |
18K18085
|
Research Institution | The University of Tokyo |
Principal Investigator |
ハウタサーリ アリ 東京大学, 大学院情報学環・学際情報学府, 助教 (70752236)
|
Project Period (FY) |
2018-04-01 – 2021-03-31
|
Keywords | socio-emotional / multilingual / communication |
Outline of Annual Research Achievements |
Development and testing of a prototype software tool for supporting foreign student’s socio-emotional communication was completed. The initial approach taken in the prototype development was to design for a method to display an abstraction of the polarity and strength of message sentiment to non-native speakers who may have difficulties detecting the emotional nuances in second language messages. The developed system automatically detects the emotional valence (i.e., positive/negative) and the strength of the sentiment in text chat messages, matches the valence information to an emotionally arousing font with similar valence, and displays the original message written in the emotional font to the receiver.
The system was evaluated with Japanese non-native English speakers who perceived second language text messages as more emotionally arousing when displayed with the emotional font selected by our proposed system. Details of the system design and evaluation results were published in an IEICE Technical Report.
Development of a multilingual sentiment corpus is also underway, and non-native English speakers' evaluations of English emotional words will be gathered using crowdsourcing through Mechanical Turk. This corpus will be used to improve the accuracy of emotional valence detection and font matching based on user's native language in the developed socio-emotional communication support system.
|
Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The socio-emotional communication support system development is proceeding more smoothly than anticipated thanks to effective collaboration with other researchers and developers. The initial plan was to use a multilingual sentiment corpora to detect the emotional valence in text-based messages, but a bidirectional-LSTM trained with human-annotated emotional sentences was implemented first, and the plan is to instead improve the detection accuracy with the multilingual sentiment corpora. Through collaborative research, a novel method for displaying the abstraction of polarity and strength of message sentiment by using emotionally arousing fonts was implemented.
|
Strategy for Future Research Activity |
- Conduct a multilingual communication experiment using our developed system in order to evaluate the efficacy of the proposed approach on interpersonal relationship building in a controlled laboratory setting. - Continue the creation of the proposed non-native sentiment corpora for further system development by utilizing crowdsourcing on Mechanical Turk, and analyze the discrepancies in emotional word evaluations between native and non-native speakers of English. - Longitudinal user studies are planned to be conducted at two Japanese universities. - Development of a cross-cultural training program based on the research results.
|
Causes of Carryover |
The research results were presented in a domestic conference, which reduced travel costs for the fiscal year. The next publications are planned to be presented in international conferences, and the travel costs will be allocated for this purpose. Personnel and article costs are allocated to be used for software development and a laboratory experiment, both of which will be conducted this fiscal year.
|