TrUDucer: "session" Tool
Part of the TrUDucer tool.
For creating new conversion rules or checking coverage and precision of existing rules, there is this graphical tool available.
- displays percentage of sentences which where not converted, as well as percentage of each relation to cause a sentence to not be converted
- visualizes trees, the conversion process and compares them with a validation tree
- easy navigation to find trees, by sorting and filtering them
- convert treebank in graphical interface and see results
When the tool is started a list of the not converted trees grouped by blocking relation is shown on the left hand side. If no trees are showing up, probably your output directory is empty (the treebank is not converted?).
Usage Information from the software itself:
usage: TrUDucer session [-h] [-t TRANSDUCER] [--treecount TREECOUNT] [--validation VALIDATION] input_dir output_dir positional arguments: input_dir The directory containing the CoNLL files to be converted. output_dir The directory where the converted files should be generated. optional arguments: -h, --help show this help message and exit -t TRANSDUCER, --transducer TRANSDUCER The file containing the transformation rules. --treecount TREECOUNT how many trees of the treebank to convert ("0" to convert all) --validation VALIDATION the directory containing a validation set of converted CoNLL files
To convert a treebank, a rule file has to be given. This can be done either when
starting the tool with the command line parameter
TRANSDUCER or by clicking
on set Transduer in the Transducer menu.
Then you can click on convert in the Conversion menu.
By default, only not converted trees are listed and sorted by blocking relations. To list all trees regardless of conversion result, the remember rules used checkbox in the Conversion menu has to be checked and the treebank has to be converted again. Afterwards, a list with all the sentences grouped by the used rules is also available.
by clicking on a tree in the filelike tree explorer on the left side, the tree will be rendered on the right side.
To see the original tree before conversion, you can click on show original tree in the Tree menu or right click on a tree in the tree picker on the left side and select show original tree. If you have selected a rule file, you can step through the conversion process with the < and > buttons.
You can also visualize rules. For doing so, the treebank needs to be converted with the remember rules used checkbox checked at least once. Afterwards, you can right click on the rule you want to visualize and click on visualize rule. This will show a popup window with the matcher tree structure on top and the replacement tree structure below. This will also show frontier nodes in the matcher structure, which is usually only implicitly contained in the rule syntax.
To check a converted treebank against a validation treebank, you need to set a
validation directory for the tool. This can be done either as a program argument
validation or with set validation directory in the File menu.
Then you can click on validate in the conversion menu. You can now compare
the converted trees with validation trees in a compare window by clicking .
You can filter the trees for rules used or tree structure in the original and generated trees. To do that, right click on conversions by rule and click on the desired filter menu. That will open an input dialog. An empty input will reset the filter.
The rule filter will always reset all filters. After filtering, only trees using a rule containing the input phrase will be shown.
The structure filter will take a while for big treebanks. Be patient or use a subset of the treebank. Afterwards, only trees where either a subtree of the source tree structure or a subtree of the generated tree matches the input tree structure will be shown. Input needs to be given in the TrUDucer rule syntax.