Tagging Tools: Useful set of tools for POST (Part-of-speech tagging)

Tagging Tools allows you to perform common tasks such as visual confusion matrix generation and complex file split.

Confusion matrix generator

Define 2 files to compare (gold standard and tagged file) and build the confusion matrix. The output can be set as LaTex or plain text.

You can see below an example of a pdf confusion matrix output. By default the cells with the biggest differences are shown.

Pdf output example:

In the next example you can see a plain text confusion matrix output. Notice that for each cell the file shows the words whose tags differ.

Plain text output example:

Tags translation:

You can also use a tags translation file in order to convert tags from one tagset into another before the comparison occurs. Penn Treebank-C5 translation is included as an example.

Command:

tt -compare <goldStandard> <fileToCompare> <output> [options]

Compares goldStandard file against fileToCompare generating a confusion matrix as output.

Where [options] is one or more of the following:

• -l: Set latex output, default is text plain output.

• -t=<matrixTitle>: Matrix title to be written on output file.

• -rt=<rowTitle>: Row title to be written on output file.

• -ct=<columnTitle>: Column title to be written on output file.

• -SC=<specificCellsFile>: Use a custom file to set specific cells to be shown.

• -T=<translationFile>: Specify a custom tags translation file to be used on comparison. The translation is applied to fileToCompare tags in order to convert them to goldStandard tags.

• -s=<size>: If latex output is set, size sets matrix size. If plain text output is set, size sets quantity of different words for each cell.

Files splitter:

This tool splits a file in several parts, preserving sentences. It optionally generates the complementary file for each extracted part.

Command:

tt -split <file> <parts> [options]

Where [options] is one or more of the following:

• -c[=complementPrefix]: Generates a complement file for each part. If specified, complement file names will be generated using * * complementPrefix as the base name. Default complementPrefix is 'Comp'

• -r: Random selection of senteneces.

File format:

Sentences are composed of tokens (words and symbols).

Each line should contain a token. Empty lines will be used to denote sentence break.

For comparison operations, each line must contain the token followed by a tab and the POS tag.

Example:

Binaries:

Linux and Windows binaries are available here

License

Tagging Tools is released under the MIT License.