User guide

> desr [options] [file]
For a full list of options, type:
> desr -h

Options (unambiguous abbreviations accepted):

-h, --help                  : Print this help message
-a, --algorithm alg  : Learning algorithm [ME]
-c, --conf  : Configuration file [desr.conf]
-C, --conv  : Convert from input to output format
-d, --dump  : dump training features
-R, --right-to-left  : reading direction
-f, --input-format format  : Input file format [CoNLL]
-F, --output-format format  : Output file format [CoNLL]
-l, --lang lang  : Language to be used [en]
-m, --model modelFile  : Model file
-r, --revise model/actions  : Revise parse tree according to
revision model or actions file
-t, --train  : Train parser from input file

Train a parsing model

To produces a left-to-right parsing model from a training corpus in CoNLL-X format:
> desr -t -m modelFile trainFile
Different input formats can be specified with -f format, where format is an XML file describing the column layout. Additional configuration parameters, including the feature model, are read from configuration file -c confFile, which defaults to desr.conf, if present.
Several classifiers are available, including: Multi Layer Perceptron (-aMLP), Maximum Entropy (-aME), Perceptron (-aAP) or SVM (-aSVM). The algorithm can be specified in the configuration file.

Parse sentences with the model

To parse sentences in CoNLL format:
 > desr -m modelFile parseFile > parsedFile
Subpages (1): Feature Templates