One of the most recent developments of TRACER includes two preprocessing options to convert the output of the TreeTagger and Stanford CoreNLP tools to TRACER's required input format, thus saving the user considerable preprocessing work. The figure below shows the corresponding preprocessing sections in TRACER's configuration file.
As you can see, each category has five properties. If you wish to import TreeTagger and Stanford CoreNLP files into TRACER you need to change the first two properties: the first property defines the file path, while the second defines the input file's extension. The third property, which must not be changed, contains the mapping instructions TRACER needs to convert input into output.To improve the usability of TRACER, a script is being developed to drastically simplify the text preparation process.
The experimental development generates TRACER-compatible
.lemma files from TreeTagger output, which means that the user does not have to prepare every text (s)he wants to analyse according to the 4-column TRACER format.
First, users run TreeTagger on the texts they wish to analyse. The output file of TreeTagger shoud have
.tagged as a suffix, e.g.,:
author-title.tagged The preferred file-naming convention is
00-author-title_of_work.tagged. The naming convention helps TRACER generate the
.txt file to be used for the analysis (see below). If there are multiple works to analyse, the numbering of the files should be sequential (e.g.,
02-, etc.); digits, author and title should be separated by hyphens, and any white-space in the work title must be represented with an underscore.
.tagged files must be deposited in a tagged folder under TRACER's
Next, in TRACER's
tracer_config.xml file, users must change the
strDataDirectory path to the corresponding directory:
<category name="eu.etrap.tracer.preprocessing.external.lemmatisation.TreeTaggerLemmatiserImpl"><property name="strDataDirectory" value="data/corpora/FOLDERNAME/tagged" /><property name="strFileSuffix" value="tagged" /><property name="strTagSetMappingFile" value="data/pos-mappings/lat-treetagger-mapping.csv" /><property name="intScaleFactor" value="1000000" /><property name="boolRemoveNonLetters" value="true" /></category>
Then, in the main TRACER folder, the following command must be typed:
java -Xmx1g -Deu.etrap.medusa.config.ClassConfig=conf/tracer_config.xml -cp tracer.jar eu.etrap.tracer.preprocessing.external.lemmatisation.DeveloperTestClassLemmatisationMain
This command generates three files in the
/data/corpora/Bible/ folder: a
.txt file containing all of the texts for TRACER to study (already tokenised by sentence and sequentially ID'd), a
.lemma-list file containing all of the unique lemmas of all the texts under study, and a
.lemma file containing all of the lemmas+pos tags of all texts under study. TreeTagger uses a pipe to indicate lemma ambiguity, e.g.,
wordForm PoS lemma1 | lemma2. In these cases, the TRACER heuristic picks the most likely option based on the frequency in the corpus of
lemma2 and on the number of incoming links from the inflected word form (
Next, in the
tracer_config.xml file, users add the paths to the
.lemma files generated by TRACER.
Finally, users can now execute TRACER.