Found in translation

Abstract

We describe translation effects that have been studied in the automated text classification literature. We expand on a point within this research space, quality effects, with our own work in this area. We present an efficient method for evaluating text quality on the basis of reference texts. The method, which is general to text classification problems more widely construed, is related to the background literature and argued to be effective on the strength of the fact that it enables quality checking of amounts of text that exceed what is humanly feasible to verify. The method partially automates the process: in processing the entirety of a translated corpus being probed, it ranks items for stylistic conformity with a reference corpus, and the least conforming ranks are indicated as the items most likely to require human intervention.