This project implements a subset of the syntax of Attempto Controlled English (ACE) version 6.7 in Grammatical Framework (GF) and ports it to ~20 natural languages (see the Makefile for the currently supported languages). Note that this project does not implement the mapping of ACE sentences to discourse representation structures.
The grammar should cover a reasonably large and interesting subset of ACE, specifically the OWL-compatible subset that is supported by AceWiki. Ideally, the grammar should cover the full ACE.
The grammar should not overgenerate, i.e. it should be possible to use the grammar in a look-ahead editor (which explicitly exposes the coverage of the grammar to the user).
The grammar should allow for bidirectional translations between ACE and a number of other controlled natural languages, i.e. ACE-like German, Italian, Finnish, etc.
Tested with:
The ACE-in-GF grammar can be turned into a PGF file in various ways depending on
Several targets are available in the Makefile, which start with the
pgf_ prefix. To build the grammar execute e.g.
make pgf_acewiki_aceowl # AceWiki grammar, ACE only, tiny test vocabulary
make pgf_ontograph_40 # AceWiki grammar, all languages, small test vocabularyTo build the full ACE grammar (ACE-only) with a lexicon of ~1000 general words, run
bash make-pgf.bash(Note that it is important that you use bash.)
The building should not take more than a couple of minutes. The GF libraries are expected to be found in a system-wide location, e.g.:
Example of translating an ACE sentence to other languages.
$ make pgf_Words300
$ gf Words300.pgf
Words300> p -lang=Ace -cat=Text "it is false that X is read by nothing but computers that Y does not see ." | l -treebank
Words300: sText (falseS ... see_V2))))))
Words300Ace: it is false that X is read by nothing but computers that Y doesn't see .
Words300Cat: és fals que X està llegit per nomÈs ordinadors que Y no veu .
Words300Dan: det er falsk at X bliver læst af kun datamaskiner som Y ikke ser .
Words300Dut: 't is onwaar dat X door slechts computers die Y niet ziet gelezen wordt .
Words300Fin: on epätotta että X luetaan vain tietokoneiden joita Y ei näe toimesta .
Words300Fre: il est faux que X est lu par seulement des ordinateurs qu' Y ne voit pas .
Words300Ger: es ist falsch dass X durch nur Rechner die Y nicht sieht gelesen wird .
Words300Ita: è falso che X viene letto da soltanto computer che Y non vede .
Words300Lav: nav tiesa , ka X lasa tikai datori , ko Y neredz .
Words300Nor: det er falsk at X blir lest av kun datamaskiner som Y ikke ser .
Words300Pol: jest źle , że X przez tylko komputery , których nie widzi Y jest czytane .
Words300Ron: este fals cã X este citit de către doar nişte calculatoare pe care nu le vede Y .
Words300Rus: неверно , что X читается с помощью единственных компьютеров , которые Y не видит .
Words300Spa: es falso que X está leído por solamente computadoras que Y no ve .
Words300Swe: det är falskt att X blir läst av bara datorer som Y inte ser .Some tools need to be installed and available on the PATH:
The commands
ghc --make -o Parser Parser.hs
bash make-pgf.bash grammars/acewiki_aceowl/ "words/acewiki_aceowl/TestAttempto{Ace,}.gf"
bash run-test.bash tests/acewiki_aceowl/sentences.txtdo the following:
The test script creates two output files
test_out.txt all the sentences classified as OK or FAIL, with ambiguity shown in case of OK, and successfully parsed prefix in case of FAILtest_out_fail.txt: frequency ranking of failed sentencesTo run a test with the full ACE grammar and the 1000-word vocabulary on all the test cases in the tests-directory, execute:
bash make-pgf.bash
bash run-all-tests.bash > tests/run-all-tests.out 2> tests/run-all-tests.errThe output files are created into the subdirectories of the tests-directory.
Additional test-targets are provided by the Makefile.
Use the following Makefile targets to generate and store the linearizations.
lin_ontograph_40_save
lin_ontograph_ext_save
lin_Geography_save
lin_Words300_save
lin_acewiki_aceowl_saveAll the linearizations except for acewiki_aceowl are also under version control.
See the Makefile targets, that have the prefix test_precision.
echo "gr -number=10" | gf --run TestAttempto.pgf | Roundtripper -f TestAttempto.pgf -l TestAttemptoAce | grep DIFF
echo "rf -lines -file=tests/ontograph_ext/sentences.txt | p -lang=Ace -cat=ACEText" |\
gf --run TestAttempto.pgf | Roundtripper -f TestAttempto.pgf -l TestAttemptoAce | grep DIFFCurrently this results in (TODO: fix these):
Tom buys a picture -> Tom buys the picture (article changes, in Fin)
Mary sees no man -> Mary doesn't see no man (negation added, several languages)
if X sees somebody who sees Y ... -> if X sees somebody who Y sees ... (word order changes, in Ger and Dut)
what does Tom buy ? -> what buys Tom ? (word order changes, all langs but ACE and Eng)coverage_acewiki_aceowl_savecoverage_ontograph_ext_saveChanges to the ACE-in-GF grammar can be done in 3 directories.
ACE resource grammar. Based on the English resource grammar. Describes
deviations from the English grammar, e.g. ACE uses who instead of whom.
Implementation of different ACE subsets and their ports to other languages.
Contains the common interface AttemptoI.gf.
Different domain vocabularies, most of which are automatically generated from existing external terminologies.
To add a new language that is supported by the GF Resource Grammar Library (RGL)
add these files into the grammars-directory (Xyz is the ISO code of the new language):
grammars/acewiki_aceowl/AttemptoXyz.gfgrammars/acewiki_aceowl/LexAttemptoXyz.gfgrammars/acewiki_aceowl/NumeralXyz.gf (this is only needed if you get a "cannot unify the information" error, in this case copy this file from the RGL and change the first line to contain CatXyz [Numeral,Digits])Then add various lexicon files to be able to run the tests:
words/Words300/Words300Xyz.gfwords/acewiki_aceowl/TestAttemptoXyz.gfwords/ontograph_40/TestAttemptoXyz.gfFinally, register the new language in the Makefile and run the linearization tests.
We measure the status of this project in various ways, see more: