Flammie Pirinen

> However, the change [a40be44#diff-1d15f6f673fdf9fb7bf3586260e12791a1169fca41cc91cc4532587124792f97R917](https://github.com/hfst/hfst/commit/a40be44f9205df89b2f7b9cc0051466413933690#diff-1d15f6f673fdf9fb7bf3586260e12791a1169fca41cc91cc4532587124792f97R917) of `sigmasizes+3` to `sigmasizes+30` should probably be upstreamed. Surely stock Foma would run into that issue? It is possible there is a bug in hfst-related...

I think this would be a very useful feature for the tokenizers in giellalt. If you want test cases, see files like https://github.com/giellalt/lang-sme/blob/main/tools/tokenisers/tokeniser-disamb-gt-desc.pmscript

> One very important thing to understand about character classes like this is that they don't work with respect to eg. composition and subtraction (like many other pmatch features of...

I like the idea of just putting these behind config flag, it's definitely something I've seen other software do with bundled libraries when packaging for gentoo. Eventually if we test...

tr1 path for headers was some stuff while c++0x or some other old standard was still under construction, so some really old compilers might need it, probably from the time...

> @flammie @eaxelson @frankier I have one question that is above my paygrade for this project. Python 2.7 hit EOL almost a year ago, and many python packages have dropped...

Are you sure the digit etc. lexicon combinations don't allow combinatorics that are really large but finite? I was testing with the hfstol version and it seems to take most...

it depends on the use case too, the digit combinations can be reduced for even full-coverage testing but only to the extent that there are no morphological dependencies that need...

hfst-reweight should still have simple weight replacement functionality, using the -l and -u switches, though I have never had use for them so never tested thoroughly. For the related question...

I used 1.8.1 and tried updating to 1.8.2 too