MaltParser 0.4: sweMaltSVM

The archive sweMaltSVM contains the files necessary to create a running parser for Swedish text together with MaltParser 0.4. The parsing model has been trained (using the feature model specified in swe/swedish.par) on 90% of the Bruksprosa section of Talbanken (section 1-9 in Nivre 2006), converted to dependency trees as described in Nivre (2006). The parser presupposes that the input is in Malt-TAB and tagged with the Granska part-of-speech tagset. (The part-of-speech tagset can be found in swe/swedish.pos and the dependency type set in swe/swedish.dep.)

Running sweMaltSVM

Before running the parser, you need to download and unpack MaltParser 0.4. Then download sweMaltSVM.tar.gz into the directory containing the executable file maltparser* and unpack it:

> gunzip sweMaltSVM.tar.gz
> tar xvf sweMaltSVM.tar

This will create a directory swe containing all the sweMaltSVM files. Run the parser with the following command:

> ./maltparser -f swe/option.dat

This parses the test file swe/swedish_input.tab and stores the result in swe/swedish_output.tab. To change input or output file, simply edit the file swe/option.dat and change the value of the parameters $INFILE$ and $OUTFILE$. To change the output format from Malt-TAB to Malt-XML (or TIGER-XML), change the value of the parameter $OUTFORMAT$ from TAB to MALTXML (or TIGERXML).

NB: In principle, any of the options in the option file can be changed, although we cannot guarantee how the parser behaves. In particular, changing the value of the parameters $ALGORITHM$, $FEATURES$ and $LEARNER$ (without retraining the parsing model accordingly) will make the parser either crash or produce garbage.