Unknown language rejection in language identification system

The number of languages in the world is much larger than the number of target languages that current language identication systems can handle. Therefore, we propose here the use of a multilayer perceptron neural network as a means to prevent those unknown language inputs from being misidenti ed as one of the target languages. We consider not only the target language identi cation rate but also the unknown language rejection rate. Results reveal that with the use of phonemic unigram as the input features to the neural network, a target language identi cation rate of 93.5% can be achieved for 3 languages. By varying the thresholds of the outputs, good unknown language rejection rate can also be obtained at the expense of lower identi cation rate.