Releases: AceCentre/Correct-A-Sentence
v0.1.0-beta
Preview of CorrectASentence on Device
This is a preview of the CorrectASentence demo. It's very rough and indeed NOT ready for production. There are many use cases where this tool is NOT recommended as it can infer the wrong intention of a user. We need to understand better how these tools may be useful in the future.
How to use it.
- Run the installer. Note it is large and can take a while to install it.
- Once installed, you should find an example Gridset on your desktop suitable for the Grid. For other software, see below for some basic details
- When trialling this software, You should run 'correctsentence.exe' and leave the command window open. Note: On install it will run this for you. If you want to ever get this working again you will need to run this.
- Open the example Gridset and trial writing using autocorrect on speak.
If you have other software the basic principle is this:
- You need a function to copy your prepared message
- You then need to call the
client.exe
software. You can find the path to this following the shortcut on your desktop. Or use this%LOCALAPPDATA%\Ace Centre\CorrectSentence\client\client.exe
- You then paste the corrected message back to your message bar
You may need to create a pause or wait command to deal with the time it takes to correct the sentence. For your average AAC device I recommend a good 3-6 seconds. Go longer if you see no change.
Caveats
- It's English Only, but you may have some luck with German, French etc too.
- It's trained right now on QWERTY layouts. If you make lots of positional errors on other layouts - or access methods like switch scanning your mileage will undeniably vary.
- It does require a considerable amount of RAM/CPU to run. Your mileage may vary
What we would LOVE in return for trialling this
- We would love an example of sentences that don't work for you. Ideally, these must be realistic sentences that an AAC user would say or have said.
- Please remove any identifiable information.
- We would like to share this data more widely and help train a future model to improve correction and prediction
Send us your sentences here: https://forms.office.com/e/FMwMaHA3zP Note it is anonymous. Please remove any identifiable information before submitting