The article Automatic Bilingual Language Identification in Short, Noisy Texts presents a language identification solution designed to deal with noisy and short texts. The identifier is implemented as part of Meedan linguistic server Alegre and it is affectionately known as
It works identifying one language or pairs of languages. In case of pairs, one of the languages is assumed to be English. The presented solution combines TF-IDF VSM similarity of character 3-grams with heuristics for single language identification and uses words 3-grams windows for bilingual identification.
Follow Clarissa on GitHub.