For organizations using professional translation services, there is always a debate over which is more accurate: machine translation or human translation. Advances in software and speech recognition have made machines more accurate, faster, and less expensive than human translators, but the technology still lags sufficiently to still raise questions about its efficacy versus a human translator. Several recent tests have pitted one against the other. Here is a summary of the results.
The basics standards for measuring human and machine translation have always been adequacy (it says what the text says) and fluency (grammatically the translation is correct), order (the matched word were in the order they were intended), and sustained accuracy (the length of a sentence and the degree to which its translation was accurate). A popular analysis software geared towards translation evaluation was used as were algorithms are used to assess trends in translation, including error rates and comparative correct strings of words between machine and human translation. Tested was the Google Translation tool and PROMT, a leading translation software provider.
A direct comparison of Google and PROMT found that Google translated more accurately on shorter texts, but both struggled on long strings of text. PROMT did better with grammar because of the RBMT-model of translation, which relies on a sizeable linguistic database and a deeper description of natural languages rules. Because Google is oriented to translating into English, it’s constantly being improved, but in both cases, each was limited to the type of dictionaries used as a reference source. Additionally, with both, post-editing by a human is necessary for the translation end-product to be perfect. Even the best machine translation needs a human editor.
Human translation via a translation service provider is not as flawless as it seems. Much of it depends on the education and knowledge of the human translator. Those limitations can create variances between human translators. For example, one translator might interpret a colloquialism differently than another. In another example, dialect and accents can pose a challenge for both human and machine translations. Then there is the propensity of a human translator to tire over longer periods of time, which increases the rate of errors.
Which Is Better?
It depends on the situation in which that is used. Machine translation still is limited to the size of the dictionary being used, grammatical rules programmed into it and it struggles with idiosyncrasies, such as dialects, accents and colloquialisms. Personnel from a translation agency tire over time and with longer volumes of text due to fatigue. Humans can interpret context, while machines cannot unless programmed to do so.
Overall, a comparison still falls on the side of human translation services if the interaction being translated is live and between humans, say in a business meeting. For machines, longevity seems to be the strong point as it can interpret complex texts accurately, as long as the strings of words are not excessively long.