How to use the morphological parsers

Setting up the environment (You probably did this already):

1. Set up the computer and download the source files
2. Build the programs for analysis (= compile the transducers).

1. Analysis, write u and the 3-letter language code using the (basic) xfst program, e.g. for North Saami, write usme, and <ENTER>. Then write the words that shall be analysed, one word at a time, followed by <ENTER>. To leave analysis mode, press ctrl C.
2. For generation, use d instead of u, e.g. dsme. Then write lemma and grammatical tags (in the same form as was given as output, followed by <ENTER>.
3. For testing, you may also write a file with one wordform on each line, and then feed that to the analyser (example here is for Inari Saami, with a file testfile.txt):

cat testfile.txt | usmn | less

If you want to use hfst instead of xfst, write husme and hdsme. The actual command behind these aliases are revealed by writing alias usme etc. If this commands gives no answer, you should run command sh gtsetup.sh. See the documentation in the preamble.

We have set up aliases for text analysis (xfst only), the aliases contain a pipeline combining perl pre- and postprocessing with xfst transducers and constraint grammar. These aliases may be written anywhere (replace "sme" with your own language code).

smedis

Gives a sentence analysis of North Saami

smedep

Gives a dependency analysis of North Saami

smedist

Gives a sentence analysis of North Saami, in trace mode (showing which dis rules work)

smedept

Gives a dependency analysis of North Saami, in trace mode (showing which dep rules work)

These aliases may be used in two ways: either write the alias followed by a sentence in quotes

smedis "Mun lean boahtán."

Or, alternatively, pipe a file through it:

cat testfile.txt | smedis

For hfst, we have an alternative procedure for preprocessing text, using transducers instead of perl. The command to tokenise, analyse and print the output in a CG compatible format is:

cat testfile.txt | hfst-tokenise --giella-cg tools/tokenisers/tokeniser-disamb-gt-desc.pmhfst

In case the transducer contains weights, the constraint grammar may make use of them, as follows

cat text | hfst-tokenise --giella-cg tools/tokenisers/tokeniser-disamb-gt-desc.pmhfst | ...

Please note that the file tools/tokenisers/tokeniser-disamb-gt-desc.pmhfst is not built by default. To enable building it, configure as follows:

./configure --with-hfst --enable-tokenisers

This setup is experimental, so there are no aliases yet

(Warning: this part contains basic unix commands only.)

Instead of just showing the result on the screen as running text (as above), much can be done to manipulate it. Here are some examples, all the textstrings should be added after the smedis etc. above.

| grep '+N+Pl' > plnouns
(to get all plural nouns and save them to the file plnouns)

| grep -v '\?' | cut -f2 | sort | uniq -c | sort -nr | less
(to get a frequency list of the lexemes that the parser recognizes.

| grep '\?' | sort | uniq -c | sort -nr | less
(to get a frequency list of the words that the parser does not recognize)

| grep '\+\?' | sort | uniq -c | sort -nr | less
(to get a frequency list of the word forms that the parser does not recognize)