Météo: translation of weather reports and warnings

Since 2004, RALI has been investigating how well current SMT approaches deal with a real-world task. We have reconstructed translation systems for dealing with weather bulletins and warnings issued by the Canadian Meteorological Centre of Environment Canada.

This work has been performed in two main steps:

  • 2004-2005 : feasibility demonstration
  • 2008-         : development of an operational warning translator 

2004-2005 : Feasibility demonstration

We showed how a combination of a sentence-based memory approach, a phrase-based statistical engine and a neural-network rescorer can give results comparable to those of the current system. We also explored another possible prospect for MT technology: the translation of weather alerts.

The methodology, the experiments and the results of this project are fully described in

Philippe Langlais, Simona Gandrabur, Thomas Leplus and Guy Lapalme, The Long-Term Forecast for Weather Bulletin Translation, Machine Translation, Vol 19, no 1, 83-112, Kluwer Academic Press, 2005.

Other publications about this project.

Examples of translation

You can have a look at the translations, with the source sentence and the reference translation, we obtained by the best combination of statistical translation techniques

Important : Translations are encoded in XML, so

  • the referenced stylesheet in the source XML should allow your browser to visualize these files in a friendly way
  • If you see that it does not, change your browser ;-) or try the html format
  • To download the file on your machine, right-click on the link
    Pairs for learning# of distinct translationsTable in paper
Reports [html] 4 000 000 5994 VI
Omnibus bulletins WWCN [html] 78 000 1532 X
Violent storm warnings WUCN [html] 21 000 653 X

The XML file are valid according to the following XMLSchema. The display of the files is done using the following XSL Stylesheet.

Meteo bitexts

One of the results of this project was the development of French and English bitexts which we make available to the community.

For more details: Philippe Langlais.

2008- : Development of an operational warning translator

In the context of the Multi-format Environmental Information Dissemination project, RALI developed WATT (Warning-Avertissement Translator Traducteur), a prototype weather warning translator which can be tried on line.  

This work is the result of many years of work is organized around pipeline of linguistic tools that can go from the French or English source as produced by meteorologists (i.e. all upper-case) at Environment Canada to the final warning in the other official language of Canada in a publishable format (i.e. mixed-case). 

The following schema shows the many steps involved in the production of a translation.

WATT pipeline

The prototype is described in detail in this publication and it can be tried online

For more details: Guy Lapalme