Most problems have a curve whereby the results improve as data are added but level off at some point.
Are there research papers or industry results that discuss the correlation between data set size and prediction accuracy for natural language identification?
LIDE: Language Identification from Text Documents benchmarked the difference between using character and word n-gram information for natural language identification.
Word-length algorithm for language identification of under-resourced languages proposed new algorithms for languages with limited available data.
