Texto4science : Conversions between SMS and normalized French
RALI works at analysing and distributing a corpus of French Canadian SMS, as part of the Texto4Science project, which brings together a number of partners, among which the Observatoire de linguistique Sens-Texte.
In May of 2012, Philippe Langlais, Patrick Drouin and their collaborators published some of their research on the topic in the article Texto4Science: a Quebec French Database of Annotated Short Text Messages, available online.
Annotated and anonymized corpus
The annotated corpus can be downloaded from here [646 kb].
It contains the corpus described in the aforementioned article, and follows the XML format explained by its authors. An XSD validation schema is also included in the archive.
The corpus is anonymized, so that no personal information appears in it. The first names have all been replaced with "Virginie" and "Amélie" for women, with "Florent" et "Alain" for men, and with "Claude" and "Aloïs" when the person's name was ambiguous. The last names all became "Tremblay", the most common in the province of Québec. The addresses and phone numbers have also been obfuscated.
Should you find remaining personally identifiable information in the corpus, we would very much like to know about it. You can notify Philippe Langlais by email.
Demo of the automatic SMS normalizer
This prototype was realised by Fabrizio Gotti and was presented during a Midi Cité talk, described on their website.
Demo of the automatic "texter", from French to SMS
This demo (both the interface and the engine) was realised by Florent Cottin, during a 2 month and a half internship at RALI, during the summer of 2011. He was then preparing a DUT in computer sicence à the IUT d'Orsay, in France. He also greatly helped at anonymising the personal information contained in the SMSs.