Toward a Multilingual Conversational Agent: Challenges and Expectations of Code-Mixing Multilingual Users

Title of the Paper

Toward a Multilingual Conversational Agent: Challenges and Expectations of Code-mixing Multilingual Users

Paper Information

• Subject Area: Multilingual Human-Computer Interaction, User Experience Design, Conversational Systems
• Keywords: Conversational Agent, Multilingual Users, Code-mixing, Code-switching, Language Mixing, User-Centered Design, User Experience, Participatory Design

Research Background and Issues

• Problems and Challenges:

  • Multilingual users (especially those accustomed to code-mixing) face limitations in language comprehension and communication when using conversational agents.
  • Current conversational systems primarily support monolingual modes, and even those with multilingual capabilities lack natural recognition and processing of code-mixing.
  • Multilingual users often feel excluded from the "mainstream user" group when using language assistants, negatively impacting their user experience.

• Significance:

  • The proportion of multilingual users worldwide is steadily increasing, and the importance of multicultural and multilingual contexts in communication is becoming more prominent.
  • Understanding the needs of multilingual users is crucial for designing smarter, more user-friendly conversational systems, enhancing user experience, and increasing system adoption.

• Research Motivation and Related Work:

  • Existing research mainly focuses on technical optimization of agent systems, such as speech recognition and language generation technologies.
  • However, few studies address user needs from their perspective, particularly the specific experiences and expectations of multilingual users who frequently engage in code-mixing.
  • Given that code-mixing is a common linguistic strategy, this study aims to fill the academic gap concerning user experience and design directions.

Proposed Solution

• Proposed Methods or Solutions:

  • The authors employed focus group interviews and participatory design methods, involving 16 bilingual users proficient in Korean and English.
  • Users reflected on their daily code-mixing behaviors and provided feedback on current conversational systems to envision ideal future multilingual agent scenarios.

• Innovative Aspects:

  • The study not only explored the challenges users face with existing conversational systems but also revealed potential needs for future multilingual agents through participatory design methods.
  • It emphasized the impact of code-mixing contexts on language technology design, offering design guidelines from the perspective of user psychology and emotional formation.

• Implementation Steps and Techniques:

  • A two-phase research approach was adopted:
    1. Focus Group Interviews: To explore daily language-mixing behaviors and interaction challenges.
    2. Participatory Design Workshops: Using sketches to simulate ideal future conversational scenarios.
  • Qualitative coding analysis was used to extract themes and design directions related to user experience.

Research Outcomes

• Specific Findings:

  1. Identified the primary needs of code-mixing users: a relaxed and non-judgmental "safe conversational environment."
  2. Highlighted challenges multilingual users face with current conversational systems, including language switching, cultural background differences, and auditory recognition issues.
  3. Proposed interaction use cases accommodating users' code-mixing behaviors, such as casual chatting, language learning assistance, and professional work support.

• Advantages Over Existing Solutions:

  • Combined user experience with cultural contexts, providing new directions for designing multilingual conversational agents.
  • Extracted key characteristics of code-mixing conversational agents in terms of user-friendliness and human-centric features.

• Experimental or Evaluation Results:

  • Focus group interviews revealed that users tend to set agents to monolingual modes due to low expectations of current systems' code-mixing capabilities.
  • Users expressed that future multilingual agents should mimic their language-mixing patterns and be sensitive to cultural and linguistic contexts.
  • The participatory design experiments generated three major use scenarios: social chatting, professional support, and language learning feedback.

• Limitations and Future Directions:

  • The study sample primarily consisted of Korean-English bilingual users, presenting regional and linguistic specificity; future research should expand the sample to include a broader range of languages and cultural backgrounds.
  • The current study relied on user recollections, suggesting future studies incorporate observations of real interaction data.
  • Multimodal user interaction forms, such as conversational robots with humanoid appearances, require further exploration.

Conclusion

This study identified the intrinsic needs of code-mixing users when interacting with conversational agents, clarifying design principles and future directions for multilingual conversational systems. The research not only expands the understanding of the interaction between code-mixing and user experience but also lays a foundation for developing language technologies that are more socially adaptive and cognitively inclusive.