[radoslove.isv-latn.wordlist] - Initial creation of Interslavic lexical model - #351
Conversation
First Interslavic lexical model in the catalogue. Companion to the `isv_latin` keyboard. 39,777 inflected forms, not lemmas. In a language this heavily inflected the distinction decides whether prediction is useful at all: you type `slovami`, not `slovo`. Derived from the 248,845-form wordlist generated from the medzuslove dictionary database — the same source as the Interslavic HeliBoard dictionary. Trimmed to frequency >= 100 because the full list compiles to a 33 MB trie, past what an iOS keyboard extension can hold. Vocabulary specific to Interslavic itself is kept regardless of frequency: the source corpus scores `medžuslovjansky` at f=26 and `tipkovnica` at 99, so a plain cut-off would drop precisely the words its users type most. Standard orthography only, matching the keyboard — the suggestion bar never proposes a spelling the standard does not use. Sources and the generator: https://github.com/radoslove/keyboard-interslavic
|
Thank you for your pull request. You'll see a "build failed" message until the Keyman team has reviewed the pull request and manually initiated the build process. Every change committed to this branch will become part of this pull request. When you have finished submitting files and are ready for the Keyman team to review this pull request, please post a "Ready for review" comment. |
|
Thanks for the submission. Do you know if there's any plans to make an Interslavic lexical model using Cyrillic script? I wonder if we should rename the paths (and references) to |
Both from review feedback. Rename: plain `isv` claimed the whole language while the model only covers Latin, and Interslavic is genuinely written in both scripts. `isv-latn` also lines up with the companion keyboard, which is already `isv_latin`. Paths, project file, package, model source and all references renamed together. Licence: `Copyright © 2026 Radoslove` — the bare `©` came from the `kmc generate` template.
|
Agreed on the rename — pushed as On Cyrillic: not in the immediate plans. Cyrillic layouts for Interslavic do exist in my repo, but they've never been audited to the same standard as the Latin one, and I have no way to test a Cyrillic model properly right now. I'd rather submit it as a separate PR once it's actually verified than add an unchecked script to this one. |
|
Ready for review |
…oth READMEs All four come from review feedback or from checking what the docs actually promise. Rename (keymanapp/lexical-models#351 review): `radoslove.isv.wordlist` -> `radoslove.isv-latn.wordlist`. Plain `isv` claimed the whole language while the model only covers Latin, and Interslavic is genuinely written in both scripts. Renamed together: directory, .kpj, .kps, .model.ts, the published .kmp, every reference in the generators and docs, and the upstream_prep staging. Licence: `Copyright © 2026 Radoslove` — the bare `©` came from the `kmc generate` template, so the same slip was in both submissions and both are fixed. README, Windows section: it told users to run `setup.exe` for the Latin layout. That layout has no setup.exe — it installs via `install.ps1`. The claim was false for the one layout we actually recommend, in both language versions. README, contributing section: it described an opt-in on-device collection system — pseudonyms, filters, withdrawable consent — as though it existed. It does not. The section now documents the channel that does exist (GitHub Issues) and marks the rest as planned, including why collection has to live in the app rather than the keyboard on iOS. Both packages rebuilt warning-clean; check_docs.py passes 53 checks.
darcywong00
changed the title
darcywong00
left a comment
darcywong00
left a comment
There was a problem hiding this comment.
lgtm (looks good to me).
Thanks for the submission!
4 participants
First Interslavic lexical model in the catalogue. Companion to the
isv_latinkeyboard.39,777 inflected forms, not lemmas. In a language this heavily inflected the distinction decides whether prediction is useful at all: you type
slovami, notslovo.Derived from the 248,845-form wordlist generated from the medzuslove dictionary database — the same source as the Interslavic HeliBoard dictionary. Trimmed to frequency >= 100 because the full list compiles to a 33 MB trie, past what an iOS keyboard extension can hold. Vocabulary specific to Interslavic itself is kept regardless of frequency: the source corpus scores
medžuslovjanskyat f=26 andtipkovnicaat 99, so a plain cut-off would drop precisely the words its users type most.Standard orthography only, matching the keyboard — the suggestion bar never proposes a spelling the standard does not use.
Sources and the generator: https://github.com/radoslove/keyboard-interslavic