AI-Assistant Language
TIMELINE
• Dec 2023 (1 Month)
CONTRIBUTION
• Secondary Research
• Wireflow
• Wireframe & Prototype
• Usability Testing
Background
Help English language learners practice speaking through AI Assistance: Conducting research, designing user persona, voice user interface and flow, wireframe, high-fidelity prototype and testing.
As an Indonesian, learning how to speak in English is challenging. Duolingo stands out as a language-learning app in Indonesia and offers a Conversation: Speaking feature.
Objective
Revamp the speaking feature to help learners practice their English speaking with a better approach.
Solution
An AI English Speaking Assistance Feature with voice user interaction that usability tested and scored 100% using CSAT.
Table of Contents
✧ Problem Discovery
Understand Duolingo
During my English learning journey, I already tried different learning methods such as learning from a book, joining a class, watching a movie, and more. People near me are far from the English Language which makes it hard to learn English anywhere and anytime.
Based on data, the Duolingo app is the world's most downloaded education app and over 500 million people have benefits while learning a language, especially English. I tried it myself and found 3 advantages to using Duolingo to learn English:
Goals Personalized
When I first used the app, it helped me to set my goals and the time and willingness or some commitment that I need to learn English, day by day. This helped to boost my motivation and make me focus on learning languages, in this case, English.
Gamification: Game-based
Duolingo uses a game-based method for their users to learn language. It makes it more fun to learn. They make every exercise short and valuable for their users so they don't overwhelmed.
Interactive Feedback
We learned by making a mistake. Duolingo gives immediate feedback after the users answer the question. The feedback is also user-friendly, and short, but to the point.
Based on my research, Duolingo focus on their Daily Active Users Metrics as their north-star metrics. It means the goals personalized, gamification, and interactive feedback helped them to maintain daily engagement. More interesting is, Duolingo community helped them spread the word to their user's family and friends, some of whom eventually download the app and subscribe.
That's why one of their strategies is to make the user available to share their achievement with the world and help market Duolingo in an organic way.That's why one of their strategies is to make the user available to share their achievement with the world and help market Duolingo in an organic way.
Duolingo have 3 learning methods that users can try when learning a language. And one of them is what I needed the most which is the 'Learning how to speak' feature because I want to enhance my english speaking skills. And the 3 learning methods :
1. Learning to form a sentence
2. Learning through listening to a story
3. Learning how to speak
Speaking Practice Flow Issue
I expected too much from the speaking practice but after trying it myself, I don't find the feature valuable enough to practice my speaking. When practicing speaking, I always think I need to have some conversation or simple question and answer. And here's 4 issues I found on this feature:
The exercise only focused on the pronunciation
The feature only gives a sentence to the users and wants them to repeat it after Duolingo's character—no other question than repeating the sentence to perfect the pronunciation.
Random context seems to have no personalization
Many people reviewed how Duolingo is personalized. However, I found that the question they gave in this feature not related to me and that's made me curious about what kind of personalization this feature offers to me.
Users can't see if they pronounce it right immediately
After I speak to the mic, I don't know if I already speak it properly because sometimes I find some words hard to pronounce.
Lack of feedback
I mentioned before how Duolingo gave valuable immediate feedback. But in this feature, I don't think I found it. For example, when I make a mistake to repeat the sentence, It tells me to skip it for now. So, I don't know what is the mistake.
Identify Assumptions
Before defining the problem statement and creating the design solution, I define some of the assumptions that I need to validate for this project. Let's talk about the riskiest assumption for this project. This project is too subjective because I work by myself and that's why I need to validate these assumptions:
1. Users tend to learn how to have a conversation rather than pronunciation.
2. Users with pre-intermediate and higher proficiency in English level need to practice speaking more than listening, writing, and vocabulary.
3. Improving speaking skills is challenging and the best way to improve it is by having conversations with partners.
These 3 assumptions are the foundations of why I think the current speaking feature from Duolingo is not valuable enough. I think learning the pronunciation is not helping much in the context of learning how to speak in English.
I already make some assumptions about Duolingo users in Indonesia. Here are the assumptions:
1. Indonesians who live in big cities (like Jakarta, and Surabaya) commonly speak decent English.
2. Indonesians who learn English for professional work need to practice more in speaking.
If I can validate the assumptions I mentioned before, then I can dig deeper into the problems themselves. And it's my assumptions:
1. Users find it difficult to start a conversation
2. If users speak more they feel they improve more
3. Immediate feedback makes users feel they have improved more
4. Users prefer having some context before having a conversation
5. When having a conversation, users commonly forget about the words they want to say
Research
The objective of this research is to validate my assumptions and learn more about English learner students in Indonesia and their problems. I use online surveys and In-depth Interviews for the methodologies. The survey results were filled in by 15 participants.
Participants
Mostly the users the aged 18-23 years old (Possibly college students, or high school students). Most of them already reached in intermediate level. But I found that there are people that don’t know their English level. And I found that people think practicing English speaking skills is easy.
Urgency
In the context of urgency, 45% of users want to learn to speak English rather than others such as learning vocabulary, listening, or writing. And I validated that the current speaking feature did not excite them.
Motivation, Pain Points, and Preferences
Using affinity diagram, I categorized what motivates the users, their pain points and how their preferences about solution that might work related to speaking english feature.
Motivation: Career-Related, Communication and Media, School, and Specific Interest.
Pain Points: Communication Challenges, Vocabulary and Expressiveness and Social and Partner-related Issues.
Preferences: Communication Style and Correction, Adaptability and Learning Together, Support and Relationship Dynamics.
User Persona
Based on the research, the existing feature that focuses on pronunciation practice is not valuable enough. From the research, I defined the initial user persona that helped me focus on building the design solution later.
Problem Statement
Research indicates that 47% of users need to practice their English speaking skill rather than other skills such as listening, reading, and writing. While Duolingo app only provided practice pronunciation for speaking, based on survey that all users expressed strong desire to engage in speaking exercises with a partner or friend.
The current feature also seems to have random context with no personalization, the users can't see if they pronounce it right immediately and it gives a lack of feedback.
So, How might we improve the Duolingo app to better support users in practicing and enhancing their English speaking skills by providing interactive exercises with personalized context and immediate feedback?
✧ Proposed Solution
Design & Iterations
In this design process, I tried to design an AI partner that can help users practice their English speaking skills. In the first iteration, I tried to design how easily the users can interact with the AI. In the second iteration, I tried to make it more fun and enjoyable for the users.
Why is AI the right solution for the problem:
Personalization
AI should response based on user prompts. The more users interact the AI can learn and improve more to make it more personal to the users.
Natural and Immersive Experience
AI can be represented as a virtual partner and that makes the users feel they have a real conversation with a real person virtually. The AI also can understand human language so they can response in natural language that humans understand.
Availability and Accessibility
AI offers flexibility in the availability and accessibility, it means the users can interact with the AI and practice anywhere and anytime 24/7.
Voice User Interface Flow
The MVP flow concept of this VUI is to hook the users to express something while speaking. For example, the AI will ask about how is the user's day or ask about their feeling.
The user also learned new vocabulary and practiced how to use it with the provided and asked for some examples. Based on the research the users seem easy to learn new vocabulary. It should interest them to try and engage more with this feature.
Testing & Improvement
I'm using Maze to test the usability of the design. Maze can track the success rate from the design scenario, and the misclick rate and provide a heatmap so I can improve the design better. For this project, I also gathered customer satisfaction to validate the design based on user satisfaction.
I invited 3 participants to test the prototype including 1 person who already familiar with Duolingo app, 1 person who learn speaking for career purposes and 1 person who learn speaking for academic purposes.
There are 2 opportunities to improve the design. The first one, when the users start the speaking feature with AI, they tend to click the button to activate their mic immediately instead I expected them to listen to AI until finished. It shows from the heatmap. But after that, the users want to listen to AI to finish when it explains some answer. So the first improvement should be to optimize the first message so the users don't miss any information because they want to immediately ask the AI.
The survey asked if the users were satisfied with the feedback or response from the AI and the dictionary feature to gather what vocabulary the users learned from the conversation with AI. Based on CSAT calculation, it scores 100%.
The last improvement based on the users feedback, it's about the Voice User Interface Flow. For now, the AI will ask or respond related to learning new vocabulary and help the users to express how they feel or what happened recently to boost their confidence in speaking with something close to them which is about their daily routine or activities. So, I need to dig deeper and create a more complex voice user interface flow.
Final Design
Future Plans
Learn user behavior and dig dive into the voice flow
I would like to learn based on the AI's respond to the users and the user's prompt. With this data, the feature can be improved by providing more complex and real scenarios that help users practice their speaking better.
Explore the flow for another language learning
This project only focuses on the English language, instead, Duolingo provides a lot of languages. Every language has different treatment, for example, the numbers of the alphabet are different and also some cultural things need to be adapted.
Design more assets for visualization
Provide some images to help the user visualize and feel immersed in the conversation. In the future, after understanding words that might be asked and doing some categorization, I can create visual assets to help visualize the words.
✧ Lesson Learned
AI is challenging to learn and still open to explore
This is the first time I designed a voice user interface (semi-because it's not fully voice interaction) and it's challenging to design the flow and make it valuable to the users. In the research process, I also studied how machine learning works and why it can help people. After this project, I am excited to work collaboratively or have discussions related to AI and how we can design better experiences by implementing AI in the design.
