LEVERAGING COGNATES TO ENHANCE LOW-DENSITY NLP

The standard approach to many Natural Language Processing tasks is a statistical one, which depends upon large tagged or aligned corpora. Yet the vast majority of languages are “low-density languages,” in that they lack these, and other, linguistic resources. There are several approaches to performing NLP on these low-density languages. Two interesting approaches are: using tools of high-density languages on genetically close low-density ones; and projecting linguistic resources from high-density to low-density languages. Cognation is the detection, or generation, of words in one language which are the parallels of words in a genetically related language. The field of cognation is related to that of machine transliteration, in which the parallel words are in a different alphabet, and are most often “borrowings” (rather than cognates) from a genetically unrelated language. Cognation can assist in various NLP tasks, such as improving word-alignment for the purpose of machine translation. For low-density languages which are genetically close to high-density languages, cognation can assist in the two aforementioned low-density NLP approaches.