Monday, April 29, 2019

Translation Memories


Translation Memories are perhaps the most important and vital technology advancement in the translation industry. They allow us to perform our jobs faster, more accurately, with higher consistency and lower costs than ever before.  TMs however, are NOT that same thing as machine translation, and take significant practice and skill to use effectively. 

At PLG we utilize the commercial software Trados, which is the leading TM program used to store previously completed translations and assist our translators in current projects. As our professional translators work, they enter their translations into a stored database built just for your company. When future material comes up that is similar to something that was previously translated, the software suggests the “memorized” translation to the translator. The translator then reviews the translation, makes any adjustments necessary, and saves the updated translation to the memory once again. This memory applies to content that is repeated not just within a given document, but across documents and projects.

Over time as we work with a client the translation memory grows larger, and our translators have more material to reference. This allows us to provide more accurate, consistent translations both faster and less costly. Two good examples are the TMs we have built for our customers at Walmart and HoMedics in the past years. The TMs combined have over 3 million translation pairs of sentences between a source language and the target!

Machine Translation


The question that every Translation Service Provider answers several times a week. “Why can’t I just use Google Translate?”, and the truth is that you can! Online machine translation tools are a great resource for quickly understanding content, learning how to say a simple phrase while traveling, or doing a trend analysis on large volumes of text. When it comes to professional content being used in a business, legal or medical setting which holds some liability or company image however, that is when it is imperative to seek professional help. 

Machine Translation (MT) has evolved greatly over the years. It started as a basic substitution program, which would replace one word in the source language with the “equivalent” word in the target language. Obviously, this had fairly bad results, as it doesn’t take into consideration changes in syntax, grammatical structure, verb conjugation let alone the issues of context, culture, and other nuanced issues while translating. 

Over the years this system has greatly improved to what is known as Neural machine translation, which is what Google Translate is now. NMT uses a neural network, aka a brain like network, which has the ability to learn grammatical structures and phrasing based on statistical analysis of translations that already exist. Over time the system learns to create better and more natural translations as more and more datasets (i.e. translations) become available to it. 

As amazing as the technology has become, and not doubt will continue to improve, it still has large flaws. For anything beyond basic phrases and sentences, any native speaker can look at a text output from google translate and know that something is off. Even if there are no glaring mistakes, or hilarious gaffes in the translations, they can often simply come across as not quite natural sounding. 

If your goal is what we refer to gist translation, then MT might be perfect for you. No point in paying to have an email translated, if you simply need to know if someone said yes or no to a request. But if your translation is going to be the face of your company in a new country, or hold the risk of physical, legal or financial injury should something be miscommunicated, then Google Translate just isn’t going to cut it. 

Translators spend decades of their lives becoming fully fluent in multiple languages, studying a myriad of topics in both languages, testing and certifying their skills with various organizations, and continuously improving their knowledge sets to perform their jobs well. While there may come a day in the future, where machine translation is flawless and beautiful; conveying everything from technical engineering datasheets, to love poems in perfectly balanced couplets, we are still a long way off. So in the meantime, reach out to your preferred translation vendor for your next multilingual project, leave the languages to us!