Predictive Text for Language Learners

[wcjord]

Should we be doing predictive text that's designed to use the user's vocabulary while occasionally introducing new words as needed slash level appropriate?

Flow:

• From the user's vocabulary, select which words they might say next and offer these as hints. They’ll enter into the text field as just lemmas. You collect new words by doing activities on messages that are sent to you.
• After the user has constructed the sentence, run IGC on the text to suggest potential changes. Possibly extend IGC with a version that more heavily questions whether the meaning is appropriate for the context.

Questions:

• How can we make service for predicting the next best word from a limited set ie the learners vocab? Can it be made to be clientside?
• Should we take over native predictive text? Should we build our own or use a third-party sdk

[wcjord]

@linhtphung Thoughts?

[linhtphung]

I suggest we think of potential activities to suggest learners do with the bot or in a chat. After the conference and talking with other colleagues, I have a set of ideas that I'd like to share regarding better UI to organize chats, activities to learners and teachers, and analytics for teachers. Where will be the best ways to share those?