Possible inflection errors

[ryskina]

I've been comparing Apertium-generated paradigms with the ones in Iskhakov & Pal'mbakh 1961 grammar book (Ф. Г. Исхаков, А. А. Пальмбах. Грамматика тувинского языка: Фонетика и морфология.) and found some mismatches.
Disclaimer: I am not a speaker of Tuvan.

1. Some Apertium-generated imperative forms for кел:

   келеалыңар:кел<v><iv><imp><p1><pl>
   келейн:кел<v><iv><imp><p1><sg>
   келеалы:кел<v><iv><imp><p1><du>
   

   I&P book has келиилиңер, келийн, келиили respectively (pp. 391-392).

2. Some <p3><pl> forms have a double -лер. I haven't seen this in the literature and it looked suspicious.

   келдилер:кел<v><iv><ifi><p3><pl>
   келдилерлер:кел<v><iv><ifi><p3><pl>
   

   I&P has келдилер for this analysis (I&P 365), and Harrison, 2000 has keldi(ler). The same pattern in other tenses:

   келгеннер:кел<v><iv><ger_past><nom>+э<cop><aor><p3><pl>
   келгеннерлер:кел<v><iv><ger_past><nom>+э<cop><aor><p3><pl>
   келгендирлер:кел<v><iv><ger_past><nom>+э<cop><aor><evid><p3><pl>
   келгендирлерлер:кел<v><iv><ger_past><nom>+э<cop><aor><evid><p3><pl>
   ...
   

[ryskina]

Update on 1: the analyzer inflects номчу and ал incorrectly:

номчуайн/номчу<v><tv><imp><p1><sg>
алайн/ал<v><tv><imp><p1><sg>

I&P have номчууйнand алыйн.

[jonorthwash]

2. Some <p3><pl> forms have a double -лер. I haven't seen this in the literature and it looked suspicious.
келдилер:кел<v><iv><ifi><p3><pl> келдилерлер:кел<v><iv><ifi><p3><pl>

This is an alternative form in the transducer, available for analysis only. If you don't want to see it, you should use the generator, not the analyser.

[ryskina]

Thanks! I've generated the forms using

echo "[ к е л  %<v%> [ ? - [ %+ | %<subst%> ] ]* ]" | hfst-regexp2fst -o prefix.hfst
hfst-invert .deps/tyv.LR.hfst | hfst-compose-intersect -1 - -2 prefix.hfst | hfst-fst2strings 

Is there a different way I should use that would exclude analysis-only forms?

[ftyers]

You can try using .deps/tyv.RL.hfst if it exists.

[ryskina]

Sorry, do you mean using it without the hfst-invert?

[jonorthwash]

I think @ftyers means to use the RL file instead of the LR file with hfst-invert, yes. The RL file has all the analyses removed that we would normally want to analyse and not generate, and constitutes a stage in the creation of the generator.

[ryskina]

Thank you! Sorry, I misread @ftyers's comment. The forms are correct now, and changing from LR to RL was helpful!