This paper proposes a design of a facilitator assistant aimed at supporting children speaking low-resource languages (LRL) in intercultural children's group work. Our previous studies have highlighted the challenges faced by LRL children in understanding conversations and expressing their ideas in such workshop settings. In response, our ongoing research focuses on exploring effective ways to provide support for LRL children. The design of the facilitator assistant function presented in this paper aims to address these challenges and enhance the participation of LRL children. By offering necessary assistance, we seek to create an inclusive and empowering environment that enables LRL children to effectively engage and contribute during the group activities. This paper provides a summary of our previous works and proposes facilitator assistant functionalities based on the findings of those studies.
To address the complexities of global issues and foster international cooperation, it is
crucial for Global Citizenship Education (GCED) programs aimed at children to embrace the
diverse aspects of societies beyond mere language and cultural differences. In this context,
machine translation emerges as a valuable tool that enables children to collaborate and
engage in meaningful exchanges despite lacking a shared language. By leveraging machine
translation within GCED initiatives, children can overcome language barriers and actively
participate in cross-cultural collaboration, contributing to a deeper understanding and
resolution of various international problems. However, cultural differences can still hinder
effective collaboration. Furthermore, existing MT systems often struggle to provide accurate
translations for low-resource languages (LRL), limiting the participation of LRL children in
conversations [
In this paper, our objective is to summarize our previous works and discuss the functionalities of an automated facilitator assistant specifically designed to provide support to children speaking low-resource languages during group work within the context of problem-based learning based on our previous works. We aim to present a comprehensive overview of the proposed features and capabilities of the facilitator assistant, highlighting its potential in enhancing the engagement and learning experience of LRL children in collaborative problem-solving activities. By analyzing the specific needs and challenges faced by LRL children in group work scenarios, we aim to contribute to the development of effective tools and approaches that can facilitate their active participation and promote their educational outcomes.
An extensive study was carried out previously, with the specific objective of formulating a comprehensive design for an automated system that would effectively aid in facilitating group work.
In our previous study [
The analysis focuses on the correlation between facilitator utterances and children's responses in adjacency pairing. Through annotation and statistical analysis of the tagged data, it is found that certain types of facilitator messages have a significant influence on the engagement of children speaking low-resource languages.
By analyzing the facilitator's utterances based on Sarle’s [
In addition, another significant research endeavor focused on introducing a concept: the
virtual facilitation agent [
The outcomes of an experimental study revealed a statistically significant effect, with a significance level of 5%, in the subjective evaluation regarding whether the facilitator's communication elicited additional utterances. The strategies that trigger new utterances from LRL speakers are request for a summary, response to the utterance from LRL speaker and response positively to the utterances from LRL speakers.
In our previous work, we introduced a virtual facilitation agent. However, in real-world workshop settings, human facilitators play a crucial role in leading conversations and activities. Therefore, our proposal aims to introduce an assistant that complements rather than replaces the facilitator, with a specific focus on supporting conversations involving children speaking low-resource languages (LRL).
Currently, this agent can be utilized within a multilingual chat system. However, there is potential for further development, envisioning a physical agent that can respond to voice commands in the future. 4.2.
The functionalities of this assistant have been designed based on our previous research efforts.
Based on the research mentioned in section 2, it has been observed that children tend to respond more positively to soft directive utterances or requests. In contrast, when facilitators use "tell" utterances, the response rate tends to decrease. Building upon this finding, we propose that the assistant agent should be programmed convert "tell" utterances into "request" utterances before delivering them to children who speak low-resource languages (LRL).
For instance, an utterance such as "Tell me your ideas" can be transformed into "Please share your idea" or "Do you have any idea?" by employing indirect requests. By adopting this approach, we aim to enhance the likelihood of receiving responses and encouraging active participation from LRL children.
Based on the research discussed in section 3, it has been found that asking for a summary from the group, including both LRL (low-resource language) and HRL (high-resource language) children, encourages children to engage in the conversation. This approach proves beneficial for LRL children to catch up with the ongoing discussion. However, the appropriate timing for implementing this strategy is not yet clear, as asking for summaries too frequently may not be ideal.
One potential method to address this issue is to periodically check with LRL children if they are following the conversation when they haven't sent a message for a certain period of time. Another approach could involve employing physical-based detection mechanisms, such as eye tracking or facial emotion recognition systems, to identify any child who may be experiencing difficulty in understanding the ongoing conversation.
Upon detecting signs of difficulties, the agent can then prompt the group to provide a summary, ensuring that LRL children have the opportunity to comprehend and actively participate in the discussion. By utilizing these methods, we aim to strike a balance between facilitating LRL children's understanding and avoiding excessive reliance on summary requests.
Similar to the proposal for the facilitator agent, it has been observed that responding neutrally and positively to LRL children tends to be effective. Instead of relying solely on the facilitator agent, we suggest that the assistant agent should adopt this strategy. In previous research, the facilitator agent sent response messages to the chat room for all utterances by LRL speakers.
To implement this strategy, we propose utilizing a language model such as BERT (Bidirectional Encoder Representations from Transformers) to identify whether a message expresses an idea or not. By applying BERT or a similar model, the assistant agent can determine if an utterance from an LRL child is indeed an idea.
When the agent identifies an idea from an LRL child, it should respond with an appropriate automated message. This response could be neutral, for example, "I see," or positive, such as "That is a good idea!" This approach aims to encourage and validate the contributions of LRL children within the conversation.
By leveraging the capabilities of language models like BERT, the assistant agent can effectively identify idea expressions from LRL children and provide suitable automated responses that contribute to a supportive and inclusive environment.
In the context of discussion and future work, although this work initially aimed to support children speaking low-resource languages (LRL), it is evident that the strategies proposed can be beneficial for all children, as some may struggle to participate irrespective of their language proficiency. In our future implementation, we intend to extend these strategies to encompass all children in the group, aiming to enhance their engagement and inclusion. To facilitate this, we are interested in exploring the use of technologies such as facial expression detectors to identify children who may be experiencing difficulties and require assistance from the assistant agent.
Furthermore, an important consideration is whether the agent should privately send messages to individual children or send them to the chat room for all participants to see. To determine the most effective approach, we plan to conduct experiments in both situations, evaluating the impact on children's engagement and participation.
As this paper primarily presents a design proposal, our next steps involve implementing these functionalities within our multilingual chat system. Additionally, we aim to further develop this proposal by creating a physical agent (robots) that can be utilized in face-to-face conversations. By combining technology with real-world interactions, we anticipate providing comprehensive support for children in intercultural group work settings in the future.
In conclusion, this paper has proposed the design of a facilitator assistant function aimed at supporting children speaking low-resource languages (LRL) in intercultural children's group work. Through our previous studies, we have identified the challenges faced by LRL children in understanding conversations and expressing their ideas in workshop settings. To address these challenges and enhance the participation of LRL children, we have presented an overview of the facilitator assistant functionalities. By offering necessary assistance, we aim to create an inclusive and empowering environment that enables LRL children to effectively engage and contribute during group activities. The proposed functionalities, including converting "tell" utterances to "request" utterances, asking for conversation summaries, and responding to LRL children's ideas, are based on the findings of our research.
Acknowledgements. This research is partially supported by a Grant-in-Aid for Scientific Research (A) (17H00759, 2017–2020), a Grant-in-Aid for Scientific Research (B) (21H03561, 2021–2024) and a Grant-in-Aid for Early-Career Scientists (21K17794, 2021– 2024) from the Japan Society for the Promotion of Sciences (JSPS).