Human Guidance for Machine Translation

May 24, 2014 |

The rook hovers stealthily from his corner towards the center of the playing field displayed in the blue light of my computer screen. I feel like the next Gary Kasparov as I attempt to counter the moves of the artificial chess player I am entrenched against. My strategy is to not make any expected moves so as to confuse the machine, or risk the “you lose” animation sooner than later. Indeed, artificial intelligence functions through patterns and calculated outcomes, new schemes tending to reduce efficacy. This is the case especially for “machine translation”, the 21st century solution to language barriers, that increases its abilities every day (since 1970s and rules-based machines). Different syntactical patterns, the slightest changes in tone or register are enough for the machine translation to falter. Yet translation is not like chess, and both man and machine are on the same team. Machines have to be “taught” a specific translation pattern and guided from start to finish, suggesting the need for a LSP (language service provider) to, at the very least, mark-up raw translations.

Human involvement is more than just the typical “Add to Dictionary” you might see in Microsoft Word, yet completing glossaries remains critical in the customization process. While full dictionaries accessed by the machine are sufficient for many technical terms, a human team can point the machine in the right direction. A run-time glossary (GLO) has a list of words translated a specific way which influences the machine holistically. A list must also be compiled of non-translatable terms (NTT) for brand names or intended regional demarcation. An example of NTT are references to specific laws which would be a large component of the “Community acquis”, the European Union’s compilation of laws for each country. Legal, political and economic translation in Europe could use this compilation database effectively. Another useful source is Bitextor, an open source engine which compiles bilingual glossaries from client’s websites. Finally, Microsoft Translator Hub, a software specialized in preparing documents before their machine translation, leads the field in teaching machines the accurate terminology with the best handling of colloquialisms. Indeed, register and cultural format are aspects that differ depending on the projects and are aspects that machines cannot anticipate. And even with two software programs working together, humans still have to organize the process and eventually ensure the post-translation, which is time consuming. In fact, man is irreplaceable in the tasks of cultural reference interpreting, data cleaning, data preparation, training, diagnostics, fine-tuning, and quality assurance.

This last task, quality assurance, is always important, but crucial for rules-based translation which doesn’t have the statistical edge allowing the machine to know what has been written previously. And even with a statistical approach there is always brand new content to translate and different bridges to gap. Additionally, a human can help the machine tweak errors to perform better for subsequent translations. However, post-editors must be careful because working with machines is not the same as with human translators, whose error patterns differ greatly. Instead of misspellings, the natural flow of the text will likely be the most affected.

With this reasoning, it makes sense for businesses to work with LSPs rather than relying on machine translation. Whether this method was chosen due to a restricted time limit or a constraint on funds, machine translation is not a feasible alternative for quality translation work. However complex, the human component is still the best move compared with machine translation and will continue to be. Arguably, better technology in this field requires more supervision and skepticism so that people can keep track of changes to a document and understand the process! For a company with an international exposure, it is best to leave the important documents to the human translators. A company who has explored the positives and negatives of machine translation and has decided to partner with a reliable translation vendor is already a step ahead of their opponents. And in chess that warrants a “Check-mate”!

For an overview of our translation expertise, visit our editing services page.

 

images-1

Tags: , , ,

Category: Translation Tools

Skip to content