Expanding Duolingo to connect classroom learning with out-of-class practice

Learning Media Design Project F24 Collaborator: City of Bridges High School, PA Team: Christina Ma, Pragati Maheshwary, Eadin Wang

Overview

Original Design Brief

How Might We enable students to stay organized, value the learning materials they create, and continue to enjoy studying the language?

Reframed Problem

In Spanish language-learning classrooms, students receive various learning resources including vocabulary list handouts, grammar rules worksheets, and media resources from both online and offline sources. They also engage in a variety of activities targeted to improve their Spanish reading, listening, speaking, writing and cultural (RLSWC) understanding.

However, students have limited opportunities to practice different RLSWC skills or reference their learning materials regularly during or out of class. In order to support sustained learning efficacy, engagement, and motivation, there could be a better alignment between the learning objectives, in-class activities, and out-of-class practices.

How Might We provide more practice opportunities for RLSWC skills after class that are in line with class teaching?

Envisioned Solution

A curriculum-aligned extension for Duolingo (students’ current tool for Spanish practice) with new language practice opportunities including weekly “Real-World Challenges” that provides AI-Simulated conversation practice on real-world scenario with an option to re-enact it in real-life, and daily interactive practice notifications; Because we believe language learning is most effective when practices are aligned with in-class learning material, allowing students to feel that their learning is supported & contextualized to real-life.

Learning Indicators

- Students demonstrate better performance during in-class practices such as sentence translation
- Students show more interest in Spanish learning materials
- Students ask questions about Spanish in real-world situations
- Students use the new Duolingo extension more, with continued practice streaks on daily notifications and real-world challenges
- The instructor indicates a better sense of student progress than before
- The instructor sense more customization of student practice

Discovery Phase:

Design Research & Synthesis

Before diving into solution development, the discovery phase is critical for thoroughly understanding the problem and its context. Effective design research and synthesis provide the foundation for crafting impactful solutions by uncovering user needs, challenges, and opportunities. In our project, we engaged in stakeholder interviews, contextual inquiries, and focus groups with students and educators at City of Bridges High School. Our study objectives were to understand how students were interacting with the language learning resources (vocabulary list, grammar rules, completed worksheets, etc.), identify key frustrations and pain points in their utilization and organization, and explore the potential for more engaging and organized language learning materials. We aimed at learning how students engage with different types of resources, and how that compares and contrasts with teachers’ intentions and challenges. This helped us uncover areas for innovation, ensuring that our eventual solution was deeply informed and targeted to real user needs.

Stakeholder Discovery Interview

Dr. Neil Brown is an educator and VP of the CoBHS director board. By interviewing him, we got deeper insights into experiential learning and CoBHS's position and uniqueness as a school: e.g., it needs something simple but innovative.

A challenge for us then, is how our design could enhance the school's resilience, sustainability, and inclusivity, through experiential learning experience.

Focus Group Findings

During class: Students mostly take notes in English, searching for the notes is not difficult, but students may easily forget about class content.

Out of class: Students value after class practice like Duolingo, but they want them to be better aligned with classroom learning and more interactive and collaborative. Also they do not look back to their notes out of class.

Class structure: Students like the activities that helped them engage with the language in a fun way, e.g., songs, telenovela, and role play, but they find the class to be fast-paced.

Contextual Inquiry Findings

During class: Activities are used to engage students, but students may not look at or effectively search through notes during practice.

Out of class: Extra practice is not aligned with class content.

Class structure: Students do not have a lot of time for practicing RLSWC skills and note taking during or after class.

Synthesis Model: Learning Goals Evaluation Matrix

We mapped all our findings from the contextual inquiry (which involved the teacher interview and classroom observation) in a model categorized by instructional activities, to the RLSWC (Reading, Listening, Speaking, Writing, Culture) goals from the course syllabus provided to us by the CoBHS's Spanish teacher.

This approach helped us identify pain points in students' learning experience, along with the strengths and shortcomings of different instructional activities. This approach in turn assisted us in reframing the problem and realizing the opportunity areas for innovation.

Ideation Phase: Concept Enactment

Photovoice Design Probe

We provided a custom designed postcard activity to 5 students to answer the following question:

When and where do students most frequently engage in Spanish learning?
What triggers those learning moments? Are they planned or spontaneous?
How do students feel during these moments? Is there excitement, frustration, or curiosity?
What challenges or breakthroughs do they experience when learning Spanish?

The activity asked students to respond to 10 different prompts by text and take any pictures or videos that could support their response. We provided a QR linked to a drive folder to upload the image/videos. The other side of the postcards had an emoji scale to rate the difficult experienced in answering prompts and also provided empty space for students to express their thoughts through text or art.

Through this activity we want to tap into the emotional landscape of Spanish learners, understanding their excitement in "aha" moments and their frustration or confusion in more challenging instances. By identifying these emotions, we can design resources or interventions that better support learners emotionally and cognitively.

Speculative Design Propositions

We built over the student responses from the design probes and came up with 15 different inspiration for solutions. These inspirations were later converted into storyboards.

Storyboard Speed Dating

We engaged with CoBHS students and teachers to gather feedback on 5 storyboard concepts via speed dating. Speed dating enabled iterative, participant-driven feedback, allowing students and teachers to assess the ideas' relevance, usability, and potential impact. We wanted to validate these ideas, understand user needs, and uncover new/better opportunities to improve language learning. We had the students and teachers comment on what they "liked", were "doubtful about", and "disliked" about different aspects of the storyboards. We used sticky notes to keep track of those ideas. Presented below are the 5 storyboard and findings related to them:

Interpretations

The above quotes and findings suggested a strong preference for tools that emphasize authentic conversation practice (e.g., a real-world scavenger hunt) and personalized, contextualized prompts (e.g., notification-based reminders). Privacy concerns posed challenges to the social media applications and the timing of the notifications, as students were uncomfortable with devices listening to their conversations or tracking their activities. Students used terms like “a little freaky” and “unnecessary” to describe such features.

Insights from the speed dating feedback led to the finalization of our solution: combining real-world conversation practice with flexible notification systems to balance engagement with user autonomy. This phase was critical in understanding the users’ needs and preferences, which led to our iterated final Learner Experience Map with Storyboard below. It pivoted us from individual concepts to a hybrid solution that combines features that students and the teacher liked while maintaining the tools' practical alignment with in-class learning.

Learner Experience Map & Storyboard

We proposed "LingoLink" extension as our ultimate solution to connect in-class learning with Duolingo's in-app features, enabling learners to practice language skills—reading, writing, listening, and speaking—through daily notifications, streaks, and personalized challenges and a conversational AI-agent.

The map emphasizes the integration of digital (in-app or on-device) and human (teacher and community) touchpoints, showcasing the diverse interactions one can have with LingoLink. Through multiple iterations, we refined the backstage, channels, and possible data collection points in our project as demonstrated in this map. We also tried to showcase the alignment of different phases of our solution with learning goals to display the alignment between interaction, instruction, learning goals and assessment.

Realization Phase:

Prototyping & User Testing

Prototyping

We used Figma, React-Native, ExpoGo, and ChatGPT to develop our final prototype. A hi-resolution mid-fidelity figma prototype was used to provide an idea of what the UI of the app for an entire routine/scenario will look like for our users. We had a hi-resolution, hi-fidelity screenshot of an AI-simulated real-world challenge conversation where people can inspect and test the functionality of the bot. However, it can also be classified as low-resolution as this feature was not actually integrated into the Duolingo app and was acting as a standalone during the prototyping session. Lastly, we had a standalone low-resolution, hi-fidelity notification app that was set up to send 3 multiple choice type questions as notifications every 30 seconds to demo how this feature would work in real-life if the questions were pre-set by the teacher.

Visual Style: Adhere and Innovate

Daily Notifications for Vocab and Grammer Practice

Real-World Challenge for Advanced Weekly Practice

Interactive Conversational Practice with AI

User-testing

We conducted two Wizard-of-Oz user-testing sessions for our prototypes above: one with 4 adult Spanish learning speakers and the instructor who helped us recruit, and the other with CoBHS's Spanish teacher and 2 of her students in Spanish 2 class. The first prototyping session gave us useful feedback on scenario building and our Wizard-of-Oz method. We received suggestions for some improvements to our prototype such as language consistency (use “report” and “reflection” uniformly) and adding simpler customized scenarios such as letting students upload images and have a conversation to describe the image or just chat with your friend.

Method: Since we had a fairly complex setup in which think-aloud would have distracted the user from the overall experience of the system, we used Retrospective Review & System Debrief. We recorded audios and videos and took notes so students and teachers could verbally mark up screens or describe any disconnects between the interface and their expectations.

Final Reflection

"Super cool! I will actually use it on my phone."

"You need to pitch this to Duolingo right now. Just bang on their door."

Thoughts on Feedback

Our core strength lies in an accurate reframing of design challenge, which leads to curriculum-targeted learning experience innovation and integrated practice opportunities with personalized scenarios. People were mostly impressed by the design concepts and actual functionality we developed. For example, the ability of AI chat to provide immediate, contextual grammatical corrections creates a responsive learning environment, which is especially critical in language learning.

A key point of inquiry was how evaluation and assessment would be conducted within our approach. In language learning, we view real-time feedback and correction as an inherent form of ongoing assessment and improvement, and all the learning progress will be reported to the instructor for reviewing. In addition, our design builds upon CoBHS's less grade-centric philosophy, yet we recognize the potential need for quantifiable metrics. Should instructors and students desire a measurable score, we could further develop comprehensive rubrics—both for AI interaction and instructor evaluation.

Teachers' Customization Experience

Due to the time limit, we haven't dive deep into the teacher's end in the learning experience. We could imagine huge possibility to investigate the customization options for teachers, and the balance between teachers' and students' preference in customization. For example, how might we make it an easy experience for teachers to input desired knowledge components and questions, instead of doing a direct prompt engineering (which needs a learning curve)?

We believe it's useful and meaningful to create a more accessible experience empowering teachers to personalize learning without needing to understand underlying AI mechanics.

Device and AI Access Considerations

From a parent's perspective, more focus was mentioned on student access to phones and AI technologies. Although we observed digital and AI usage in CoBHS classrooms, such parental concerns sustains broadly.

Potential solutions might include ensuring transparent reporting on interactions, developing age-appropriate AI interaction guidelines, and providing non-AI/digital alternatives. Nevertheless, they all remind us to acquire enough parents' input, other than students' and teachers', in our future work!

Acknowledgement

We want to say thank you to all people contributed to the project!

From City of Bridges High School: Head of school Dr. Randy Bartlett, Instructor Xochitl Martinez-Lopes, Anika and other helpful students.

From CMU: Prof. Marti Louw, TA Siyun Wen, and everyone who gave valuable feedback to us.

Page updated

Report abuse