LinguoExpansion

LLM-powered language learning is still limited to single AI interactions. Our app, LinguoExpansion, enables multi-party conversations, speech-based interfaces, and multi-dimensional detailed feedback.

Comment

Modern language learning applications missed one key thing: "many people".

Business occasions, research group meetings, drinking gatherings… virtually all conversations we make are at all >2-agent occasions. Modern language learning apps default practices on 2-agent conversations. Therefore, we built LinguoExpansion, which provides realistic and immersive experience of talking to many LLM-powered agents in infinite settings, with a speech-and-text-enabled interface that provides in-time multi-dimensional detailed feedback for users.

A screenshot of our working interface with multiple AI characters

Freely customizable conversation scene with up to 8 AI-characters and 1 human user.

Scenes are customizable upon user input for number of characters, activity of the scene, location, and many more details. They can also be easily randomly generated. The construction of a scene brings upon automatic building of AI-characters. Here is an example scene:

Activity: Discuss possible hackathon topics
Time: Saturday afternoon, around 2 PM
Location: A cozy coffee shop called "Java Junction," known for its creative atmosphere and free Wi-Fi.
Number of people: 4
Extra information: The coffee shop has a relaxed vibe with soft jazz music playing in the background. The group has gathered around a rustic wooden table in the corner. They're all holding their favorite beverages, brainstorming ideas for an upcoming hackathon.

An example of our scene setup screen

AI-directed conversation between all constructed characters, plus contextually fitting cues to human users.

Our AI directs conversation between all AI-characters and the human user, ensuring every AI character talks around the same amount, while the human user gets more talking opportunities to practice their language skills more! AI Characters are speaking to you, speak back to them!

Beyond just text! Use speech-enabled interface to converse with AI-characters.

Speaking is a crucial skill in communication, especially for the most of us language-learners who study to teach, host meetings, and present great ideas at an international crowd of audience. Therefore, practice opportunities must also envelop speaking experiences. This philosophy leads to the development of our speech-enabled interface, which supports speech-to-speech conversation with supportive transcriptions in the learning experience. Speaking, not typing! Our speech interface supports a speech-to-speech communication.

Multi-dimensional detailed feedback for AI and human-generated messages.

We offer detailed feedback for user and AI-generated messages alike, across dimensions of grammatical correctness, contextual recommendations for messages, and improvement alternatives for existing sentences. We believe this will positively reinforce the learning experiences of human users when enjoying our service.

LinguoExpansion’s Directions.

We stand proud with the current state of LinguoExpansion, to our knowledge the first-of-its-kind multi-agent conversation practice application with the richest form of feedback at present. The web interface is easily accessible, the speech interface is intuitive, and the agents’ conversational responses and redirections are finely tuned towards human expectation.

To reiterate the significance of our product, we:

  1. Expanded the Language-Learning Experience’s World. From ****the default two-party interactions in language learning apps to customizable multi-party conversations.
  2. Expanded Experience’s Modality Beyond Text. From traditional text-based windows to speech-enabled interface.
  3. Expanded Learning Opportunities, making any interaction a chance to learn. From only feedbacks regarding correctness of user response to generated suggestions on all AI, human messages across directions like composition, diction, and conversation tone alignment.

Built With

Share this project:

Updates