Hi Steve,
I have an idea for a new feature you could add to Readlang in the Practice section. Instead of only using sentences we’ve read for vocabulary practice, you could use AI to generate short texts and leave a few blanks in them. The user would then guess the missing words.
This way, learning becomes more engaging and we also get to see how words are used in different contexts.
I’m not against the idea, but with ‘blanks exercises’ sometimes multiple answers are correct:
That would make it harder for the system to mark the exercises. How would you deal with this issue?
In your suggestion, would the available answer options be words from your words list?
True, I totally agree if there are no provided answers. However, the style that I meant was having answers on top of the text (Reference Image).
Yes, the answers should be from the word list
Good idea! 
As long as you don’t use too many words in the exercise, otherwise it’ll start to resemble a a sudoku puzzle 
Here’s the exercise with the keywords I mentioned before (going, walking, cycling, traveling, hopskipping). I generated this story with a simple prompt:
It was a bright Saturday morning, and Mia was excited to start her day. First, she put on her sneakers because she was … to the park to meet her friends. Once there, they decided to warm up by … around the lake trail. After a while, they rented bikes and spent an hour … through the countryside. Later, they talked about … to different cities together one day. Before heading home, Mia laughed as the younger kids were … in a line, playing their favorite game.
You can sort of do it already, although it doesn’t look as pretty as yours. In the Generate story section I asked it to write a short version of the Ugly Duckling in Portuguese. I asked it to include ten verbs in the Past Simple, replace these with blanks, and list them in random order at the beginning. all of which it did. Note - the random bolding doesn’t appear in the base text, and seems to be something to do with how Readlang deals with the blanks (which are all the same length in the base text).
This was just a very quick test to see it would work, and I’m sure with a bit of refinement we could get something acceptable (for example numbering the blanks). For example I often get it to generate a story based on a given set of words. These could presumably also be replaced by blanks.
What I didn’t do was ask for the answers, but I don’t suppose that would be a problem.
