document reference date operation
help

  1. Create new documents by clicking on "CREATE" or insert an example by clicking "EXAMPLE"
  2. Edit your document by using the textarea
  3. Select the model by clicking on the dropdown menu (for "classifier_date": requires a reference date, for "heideltime" models a "mode" is required)
  4. Choose the types (date, set, duration, time) to be processed by toggling them (also filter the types dynamically after tagging)
  5. Start processing by clicking on "TAG"
  6. Reset or download the tagged document by clicking on "RESET" or "DOWNLOAD" / return to document view by clicking on "<BACK"

  1. Create new documents by clicking on "CREATE"
  2. Edit the documents in the editor view by clicking on "OPEN"
  3. Select the documents to be processed by checking the checkbox in the same row
  4. Select the model by clicking on the dropdown menu( "classifier_date": requires a reference date, for "heideltime" models a "mode" is required)
  5. Choose the types (date, set, duration, time) to be exported by toggling them
  6. Start exporting by clicking on "EXPORT", the format of the output files is TIMEML/XML.

  1. Classifiers: transfomer-based tagging models, consisting of three models for English (EN) and one for German (DE). German model (Classifier_DE) is token classifier based on GELECTRA. Classifier_EN is also a token classifier based on BERT, Classifier_CRF is an extension of BERT classifier based with additional CRF layer. The Classifier_DATE additionally takes the reference date of the document as input and appends it to token embeddings.
  2. HeidelTime: we include the python wrapper of HeidelTime to our system. HeidelTime is a rule-based system, automatically extended to many languages. We include the German and English tagger from HeidelTime.
  3. SUTime: is a rule-based temporal tagger built on regular expression patterns and is available as a library for temporal annotations from StandfordNLP.