I’ve recently created a Keyman keyboard package for the Shavian alphabet (en-Shaw), using Keyman Developer 12.0.64.0 on Windows 10. This is working well across all devices I’ve tested the keyboards on.
I wanted to create a companion lexical model, and have followed the instructions in the documentation. It is a simple word list lexical model. It builds without any errors. But when I install it on my iOS devices, the lexical model only suggests one word (out of the thousand or so in the list), “𐑻𐑔”. Always that word and no other. There is nothing special about this word in the word list so I’m baffled as to why it is being singled out as the only word for predictive text. The source files for the lexical model can be accessed here.
The keyboard package can be downloaded via this page.
Finally, a small thing, but whenever I type in any Shavian letter into the touchscreen keyboard templates in Developer, the app produces an error message saying it has to close. However, it works fine if I just ignore it and close these messages. Just a minor issue.
I rebuilt your lexical model package with Keyman Developer 12.0.64.0 and I noticed the compiled shavian_info.en-shaw.shavian1000.model.js file contained split surrogate pairs for the SMP characters. (Possibly connected to issue 2374?)
I then built your lexical model project with the lexical-models repo build.sh script to make this kmp file. It installed and seems to work fine on my Android device.
Our software team will return from Christmas break next week, and will look into this…
Thanks darcy. I’m not sure if it’s related—I installed the lexical model you built from my source files on my iPhone and iPad (iOS 13.3) and about half of the time when I select a predicted word in the Shavian alphabet it deletes the preceding one or two characters. This makes the predictive text a challenge to use (deleting spaces and sometimes the end of the preceding word).
I recognise the Shavian alphabet probably isn’t a priority in terms of underserved scripts, but I mention it as it may affect other scripts and could be looked into at the same time as the issue above.
@joshua_horton can you check out the reported issue with iOS? I am investigating the problem with Keyman Developer (I think it’s probably the bundled Node version).
The compiler issue is resolved in Keyman Developer 13.0 alpha (https://keyman.com/alpha). I have not currently back-ported this fix to 12.0; 13.0 is due out in February so would prefer not to do that additional work if you are able to work with version 13.0.
Yes, I just tried it again with my mock-up lexical model on iOS and it seems to be working. Thanks for following up. I’ll develop a full lexical model now.