ajfisch / QA-Turk

Intuitive Annotation Tool for Extractive Question Answering (see https://github.com/Varal7/ieturk)

QA-Turk: Annotation tool for Extractive Question Answering

Adapted from Victor Quach's great ieturk repo.

Input Files

Requires a CSV input file with description,question,spans input. Spans are start,end token tuples that will be prefilled in the interface.

Notes:

• Multiple spans are specified as start_1,end_1,start_2,end_2,....
• No spans are specified as "".
• Use a standard CSV reader/writer to make sure everything is escaped properly.

Annotate

Annotate using Mechanical Turk

• Create an new project in Amazon Mturk interface.
• Paste the content of annotate.html in the second tab Design layout.
• Insert the scripts config.js, annotate.js and the CSS file style.css into the document as well
• Prepare a tokenized version of the entry, separating tokens with spaces.
• Submit batch

Annotate locally

Requires localturk. Install using

npm install -g localturk

Modify config.js with the name of the fields of interest. Then simply run it using the same tokenized csv file as for Mechanical Turk.

localturk annotation.html input.csv output.csv

Visualize

Simply open visualize.html with any modern browser. Then choose the .csv that came from either localturk or Mechanical Turk.