attempto / ace-in-gf Goto Github PK

Rewrite make target lin_acewiki_aceowl_save to measure each language separately.

i.e.

p -lang=Ace "John asks Mary and John asks Mary or John asks Mary ."

gives 2 trees, in one or is on the top and in the other and.

The solution used in mgl/finnishMathFin.gf is:

mkVar : Str -> MathVar = \s -> P.mkPN (M.nForms2N (M.dSDP s)) ;

Testcases:

• jos X sisältää Y:n, niin Y ei sisällä X:ää .

Convert a set of ACE sentences to trees. Assume that the ACE implementation does not have ambiguity, i.e. that we get 1 tree per each sentence.

To evaluate the ambiguity in a (non-ACE) concrete grammar, linearize each tree (but generate only one variant), and parse the obtained linearization, obtaining a list of one or more parse trees. One element of this list is the original parse tree, the other trees stand for alternative meanings.

Display the cases where the roundtrip ACE->LangX->ACE goes wrong, e.g.:

# list of test sentences
ACE sentence 1
  # list of languages where roundtripping fails
  Ger
    # list of ACE sentences that demonstrate the ambiguity
    ACE sentence 1-Ger-1
    ACE sentence 1-Ger-2
    ...
  Spa
    ACE sentence 1-Spa-1
    ...
ACE sentence 2
 ...

(Write a Haskell program to perform this analysis/visualization. Such a tool is actually not specific to ACE-in-GF, e.g. one could apply it to Phrasebook, where the role of ACE is played by DisambEng.)

Likely choices:

• Danish
• Norwegian
• Bulgarian
• Russian
• Romanian
• Polish

Other options:

• Latvian
• Maltese
• Estonian

This works:

p -lang=Ace "John asks Mary and asks Mary ." 

This does not work:

p -lang=Ace "John asks Mary and asks Mary and asks Mary ."

The exhaustive test did not find this problem because the AceWiki Codeco testset contains sentences of max length 8 tokens.

$ cat sentences.txt | sed 's/ *$//' | sed "s/  /|/g" | sed "s/[^|]//g" | sort | uniq -c | sort -nr
15254 |||||||
 3158 ||||||
  740 |||||
  223 ||||
   46 |||
    1 ||

Both of the commands given in readme for setup fail:

make pgf_acewiki_aceowl
make pgf_ontograph_40

Both fail with:

"File Numeral.gf does not exist."

Remove grammars/acewiki_aceowl/NumeralDut.gf given that it becomes available as part of the RGL.

See also commit b206a08

Investigate if this change allows us to remove the copies: GrammaticalFramework/GF@1ea9f7c