Conversational AI English tutor LAURA
Looking at the AI Tutor LAURA project first will provide a better understanding of this project.
Summary
Internally, there was a recognition of the need to enhance the product's competitiveness based on the business context. LAURA includes various activities, some of which involve conversing with LAURA, which provides feedback on the user’s grammatical errors. However, in certain free conversation activities, grammatical errors are not corrected as they are perceived as part of a single ongoing conversation. This results in users having difficulty receiving feedback, and parents or teachers must correct grammatical errors, which is not always feasible immediately. Based on these issues, the goal to add a grammar correction feature was established. Based on this goal, we derived conclusions through competitor research and hypotheses about interaction methods. Users should be able to see grammar correction feedback in free conversation activities without necessarily involving teachers or parents. They should also have the option to choose whether to view the feedback or not. This approach aims to avoid disrupting user motivation and allows users to focus more effectively on their English learning.
With the framework we developed, users were able to study English much more efficiently using LAURA. However, as seen from the User Test results of LAURA version 1, when children used LAURA and engaged in conversations without the help of parents or teachers, they experienced discomfort due to the lack of feedback.
Provide Immediate Feedback
Offer real-time feedback to ensure smooth and effective conversations for users.
Enhance User Motivation
Increase user motivation and encourage continuous engagement with the learning process.
Advance Grammar Correction Features
Elevate LAURA’s grammar correction functionality to maximize learning effectiveness and improve the quality of the product.
Impact
Since the grammar correction feature was not applied to all activities, I closely monitored the usage frequency of activities where this feature was enabled. To assess whether users were re-engaging with the feature, I tracked the increase in the average number of utterances following the use of the grammar correction function. We also considered the usage frequency of active users as a key metric in assessing our primary goal of reducing a decline in learning motivation.
+155
%
+60
%
To evaluate user engagement and learning effectiveness differently, competitor research was conducted based on two criteria. The goal was to find a balance between providing immediate correction features and maintaining immersive learning experiences.
Step by step Learning
Although this approach does not use a chat format, it offers feedback in a structured, step-by-step manner. This allows users to progress through learning stages and systematically recognize their errors at each step. This method supports gradual learning and helps users address issues in a more organized way.
Instead of being optional, it is provided as a required step with more detailed content.
In services that focus on learning words and sentences, the system checks how well the user applied previously learned content and provides corrections based on that.
In services focused on listening and speaking, pronunciation correction is provided.
Chat-based Learning
Users are given the option to view corrections through a button, but the button is not intuitively recognizable.
Displaying corrections immediately after the user speaks may cause hesitation or fear in making the next attempt.
How users view grammar corrections on the screen?
If users are left to choose, they may overlook checking feedback. On the other hand, if the correction feature is activated and shown immediately after the user speaks, it can disrupt the flow of conversation and reduce their confidence in learning.
Formulating a hypothesis
New Issues!
What amount grammar-corrected feedback should be?
If there is too much content, it can reduce usability in a chat-based service, especially on small devices like mobile phones. This is because the amount of information that users can see at a glance is significantly reduced.
Considering that the button contains a label instead of an icon, it was necessary to choose a label that would not undermine the user's confidence in their learning experience while also encouraging them to accept feedback.
Solution
Limit the amount of text in prompts
Use clear and concise language.
Provide responses in a single sentence or a couple of sentences.
Choosing User-Friendly Button Labels
I led the discussion with the PM on button label ideation, focusing on the following two criteria.
