On Achieving and Evaluating Language-Independence in NLP

Abstract

Language independence is commonly presented as one of the advantages of modern, machine-learning approaches to NLP, and it is an important type of scalability.

In this position paper, I critically review the widespread approaches to achieving and evaluating language independence in the field of com- putational linguistics and argue that, on the one hand, we are not truly evaluating language independence with any systematicity and on the other hand, that truly language-independent technology requires more linguistic sophistication than is the norm.