Welcome to the Cognate Identification competition organized as part of the FAA course (year 2019-2020), UdL, FIL, Master Informatique
This competition is about predicting whether two word forms (or phonetic realizations of these words) that share the same meaning in two different languages are historically related (in the sense that they come from the same ancestor word). Two words that are historically related are called cognates. For instance, the English "father" and French "père" words are cognates, both derived from Latin "pater", whereas Latin "deus" and Greek "theos", despite their apparent sound similarity, are not cognates as they come from different roots.

From a Machine Learning point of view, we are dealing with binary classification problem, having to decide whether two words are cognates (positive class) or not (negative class). The dataset used for this competition contains around 1.4M word pairs, corresponding to 209 original meanings/concepts realized in over 150 languages. This dataset is split into 1.14M training examples (80%) and .29M test examples (20%).
Acknowledgements
This dataset was extracted from the Indo-European Lexical Cognacy Database (IELex), collected at the Max Planck Institute in the Netherlands: http://ielex.mpi.nl/.
Competition manager
For any information regarding this competition, please contact Pascal Denis (at inria.fr)