Wiki

Clone wiki

Questimate / Home

Questimate Wiki

Questimate is a tool primarily meant for machine translation quality estimation, but it is possible to use it for many other purposes.

What It Does

Questimate be used for the following main purposes:

  • Extraction of features from data meant for (or the output of) machine translation systems. These feature are primarily meant for quality estimation (or 'confidence estimation') of machine translation output, but they can be used for various other purposes, wherever suitable.

  • Viewing and manipulating (deleting, merging) features in the feature file created by Questimate or through some other means. This part also uses the Weka API.

  • Looking at the attributes and their relations with help of some common kinds of statistical plots. This part currently uses the JFreechart API.

  • Training the quality estimation system and predicting the quality, based on the features extracted and made available to the quality estimation system. This system is based on Weka and can use all the algorithms available in Weka.

  • Reading word/phrase graphs/lattices and generating WFST files in the OpenFst format. The n-gram mapping FSTs for these lattices can also be generated and they can be composed with the lattices to give n-gram FSTs.

  • These n-gram FSTs can then be used for computing word and n-gram posterior probabilities, which can again be used as features for the quality estimation system, or for any other purpose where they can be useful.

Installation

Since the tool is implemented in Java, no special installation is required, except that you should have JDK-1.6 installed on your system.

If you download a tar or a zip file, you can uncompress it and use it directly.

If you start with the source code, you can build the system easily if you have Ant installed on your system:

$ ant

You can get the complete source code by this command:

$ git clone https://anilsingh@bitbucket.org/anilsingh/questimate.git

However, there is some preparatory work that needs to be done, which will be detailed in a section about using the tool.

Getting Started

(More details soon to be added.)

How to Use it

To be added

Updated