The Story of the Characters, the DNA and the Native Language

This paper presents our approach to the 2013 Native Language Identification shared task, which is based on machine learning methods that work at the character level. More precisely, we used several string kernels and a kernel based on Local Rank Distance (LRD). Actually, our best system was a kernel combination of string kernel and LRD. While string kernels have been used before in text analysis tasks, LRD is a distance measure designed to work on DNA sequences. In this work, LRD is applied with success in native language identification. Finally, the Unibuc team ranked third in the closed NLI Shared Task. This result is more impressive if we consider that our approach is language independent and linguistic theory neutral.