Ideas for testing

Goal: We want to test our transducers better

Existing tests

1. Paradgim testing against predefined answers: yaml tests
2. Tests written in the lexc and twolc code
3. Testing whether we generate the lemma or not
4. Tests using the lemma list as gold standard (do we generate the lemma)

Ideas for new tests

• Test for Multichar Symbols on the lower side
• Test for phonotactically illegal strings

Elaborating the test ideas

Test for Multichar Symbols on the lower side

Now and then Multichar Symbols slip through twolc and give “words” like
Suome^Vn pro correct Suomeen.

How to test for this:

• Read the set of multichar symbols from root.lexc
• Make a transducer multichar.fst of them
• Compose LANG.fst .o. multichar.fst in xfst
• list the result (should be empty)

This test one should be able to set up language-independently. In case we get

Test for phonotactically illegal strings

Example, from fkv (this must be adjusted to a script):

• Make a regex accepting strings in Vowel + Vowel + e: regex [ ?* [a|e|i|o|u|ä|ö] [a|e|i|o|u|ä](ö] e ) ;
• Compose it with the main fst: xfst -e "regex @\"src/analyser-gt-desc.xfst\" .o. @\"VVe.fst\" ; "
• print the result with xfst xfst[1]: print lower-words > lower-words.txt