hfst-tokenise

How to use the hfst-tokenise pipeline to tokenise-as-you-analyse, using giella-sme on Mac as an example:

Prerequisites for Mac

First off, update your HFST+vislcg3 by running

  wget http://apertium.projectjj.com/osx/install-nightly.sh
  sudo bash install-nightly.sh

This should give you the most recent SVN versions (as of last night) of HFST and vislcg3.

(Packages exist for pretty much all Unix operating systems; the Prequisites links under http://wiki.apertium.org/wiki/Installation#If_you_want_to_add_language_data_.2F_do_more_advanced_stuff should give the right URL’s.)

For now, you’ll also need to get the program cg-mwesplit (which will later be included in vislcg3). To compile cg-mwesplit, first ensure you have a recent version of Xcode with the Command Line Tools (e.g. 7.3, available from https://developer.apple.com/services-account/download?path=/Developer_Tools/Xcode_7.3/Xcode_7.3.dmg ). Then do

  export CXX=clang++
  export CC=clang
  git clone https://github.com/unhammer/cg-mwesplit
  ./autogen.sh
  ./configure
  make
  sudo make install

Build sme

Now, svn up in giella-core and langs/sme, and run ./configure in langs/sme with the option –enable-tokenisers; e.g. if you want both the CG rules and the tokenisers, you would do

  ./configure --enable-tokenisers --enable-syntax

(If you use Apertium, you’d also want –enable-apertium –with-hfst, etc.)

Finally, run “make” (currently, this requires >8GB of RAM).

Test

To run just the raw tokenisation+morphological analysis:

echo 'sánit, jna. Leago' \
|  hfst-tokenise --giella-cg $GTHOME/langs/sme/tools/tokenisers/tokeniser-gramcheck-gt-desc

To include disambiguation of ambiguous multiwords:

echo 'sánit, jna. Leago' \
|  hfst-tokenise --giella-cg $GTHOME/langs/sme/tools/tokenisers/tokeniser-gramcheck-gt-desc.pmhfst \
|  vislcg3 -g $GTHOME/langs/sme/tools/tokenisers/mwe-dis.cg3

To include splitting disambiguated multiwords into their own cohorts:

echo 'sánit, jna. Leago' \
|  hfst-tokenise --giella-cg $GTHOME/langs/sme/tools/tokenisers/tokeniser-gramcheck-gt-desc.pmhfst \
|  vislcg3 -g $GTHOME/langs/sme/tools/tokenisers/mwe-dis.cg3 \
|  cg-mwesplit

To include regular morphological disambiguation:

echo 'sánit, jna. Leago' \
|  hfst-tokenise --giella-cg $GTHOME/langs/sme/tools/tokenisers/tokeniser-gramcheck-gt-desc.pmhfst \
|  vislcg3 -g $GTHOME/langs/sme/tools/tokenisers/mwe-dis.cg3 \
|  cg-mwesplit \
|  vislcg3 -g $GTHOME/langs/sme/src/syntax/disambiguation.cg3

To include regular syntax tagging:

echo 'sánit, jna. Leago' \
|  hfst-tokenise --giella-cg $GTHOME/langs/sme/tools/tokenisers/tokeniser-gramcheck-gt-desc.pmhfst \
|  vislcg3 -g $GTHOME/langs/sme/tools/tokenisers/mwe-dis.cg3 \
|  cg-mwesplit \
|  vislcg3 -g $GTHOME/langs/sme/src/syntax/disambiguation.cg3 \
|  vislcg3 -g $GTHOME/giella-core/giella-shared/smi/src/syntax/functions.cg3

etc.

If you use these steps often, you’ll probably want to make an alias. Open ~/.bashrc in your editor and enter for example

alias hsme='hfst-tokenise --giella-cg $GTHOME/langs/sme/tools/tokenisers/tokeniser-gramcheck-gt-desc.pmhfst'
alias hsmemwe='hsme | vislcg3 -g $GTHOME/langs/sme/tools/tokenisers/mwe-dis.cg3 --trace'
alias hsmesplit='hsmemwe | cg-mwesplit'