Return Translation:Measuring What’s Lost In Translation

The purpose of this research is to find out what is lost in translation through the analysis of return translation accuracy to its original. A return translation, or back translation, is when a work is taken from a source language, translated to a target language, and then translated back to the source language. Furthermore, we would like to find a method that can give us some degree of loss comparable to the overall work, in this case a percentage. To be more specific about the ability of what can and should be measured, we specifically only focus on semantics and not on other factors like tone, so long as they do not have a direct impact on the meaning. To check if our method is working or is accurate, we use two pairs of two translators to process the same data into Chinese and back as an attempt to account for flukes and irregularities that could potentially affect the data. Moreover,we have used two separate methods for measuring this semantic change. One is sentence based and the other is word segment based. To come up with these methods we consider what makes up an individual part in spoken and written language. Our first thought was that sentences would possibly be representative of this, at least on some level and for that reason this is how method one is measured. But the problem with the sentence method is that the length of each sentence varies a lot. And so in order to get an accurate percentage, each part measured should be equal to every other part. Next,we think that the smallest part of each sentence that has equal weight against each other and that is a complete idea could be phrasal parts. Since each phrasal part carries about the same weight, it would be ideal to calculate our percentages, which is what we do in method two.In the research we were able to make a measurement that was stable between individual graders working individually. We first looked into the two methods, but the data suggests thatthe first method is inaccurate, while the second method showed promising more accurate results most likely because of the higher degree of depth of measurement. We determine this by looking at the relative amount measured between each work, meaning that although two evaluators" grades are not exactly the same, two evaluators measuring the same three works, measure very closely to the same difference between those same works. Later we analyze the areas where both translators agree there is loss in meaning against the occurrences of words not found in the basic Dale-Chall word readability list. The results suggest at least slight correlation, even with a limited data set. This correlation suggests that there is an element of predictability in measuring the loss.