With advances in technology, it is impossible to ignore the advances in Machine Translation or MT. With the power of user contributions and neural engines, machine translation is improving, albeit at a plodding pace. It still is not at the level of a human linguist, but with the proper attention and monitoring, it may get there one day.
It's interesting to note all the times we in the translation industry have heard, “Machine translation will make your job obsolete.” While that may be the case (very far) down the road, it's important to acknowledge that the quality of a machine translation, even with fully-trained AI models, still cannot entirely match the expertise of a human translator. In this post, we will examine some of the pros and cons of standard machine translation (often referred to as unverified machine translation), such as tools like Google Translate, as well as some alternatives that might be better options.
Standard machine translation (MT) can be great for getting the “gist” or a general understanding of a file. It won’t sound natural to a native speaker, but they will be able to glean the meaning. It will be at or below roughly a 3rd-grade reading level. It will sound clunky and disjointed. There will likely be some terminology mismatches or incorrect translations of specific terms for a given context.
Here is a good use case for standard MT that we have seen quite often: a company will receive a rather large document in a language that they do not understand. They don’t have the budget to translate with a human editor, and they want to know if the document applies to their business. An excellent example of this would be if an American firm received an RFP in Dutch. They would have no idea what it says but would need to know if they should respond to the RFP. In that case, a standard MT pass could be a good start to see if the RFP applies to their business. If so, then they can identify the critical sections, and do a human translation pass of that content.
Standard MT is the most cost-effective solution on the market. In some cases, it is free, and in other cases, it costs very little. Quite simply, it is the cheapest solution on the market. We’ll dive more into this in the cons section.
Standard MT can happen at the snap of a finger. It can be output very quickly and requires little to no QA. This process is, of course, referring to machine translation with no editing pass.
As mentioned above, the cost can be an advantage, but it can also be a negative factor. There needs to be a reasonable expectation of quality on a free/cheap option. There are not many services that are both high quality and offered for free. I like to use the analogy of the car market in the US. There are luxury cars (translator and independent editor), electric cars (translator only), economy cars (machine translation with post-editing), and go-karts (standard machine translation). All are cars/locomotives, but not all accomplish the same thing. I use this analogy in terms of levels of service in translation projects (in parenthesis by each option above). There is a reasonable expectation that a luxury car will outperform a go-kart.
Similar to the car analogy, if something is completed very quickly, there is generally a reasonable expectation that it will not be of high quality. There are exceptions, of course, but it stands to reason that quality work takes more time, care, and attention.
One area where standard MT struggles is context. The MT process can take the same term, when it appears in different sections of a document, and translate that term differently. On the contrary, a human translator and editor are employed to ensure that terminology is consistent throughout a project. This attribute is crucial, so you do not confuse your reader when referring to the same thing.
Our suggestion is to never use standard MT for anything that is customer-facing. We have a story of a client using standard MT to save a few bucks on the front end of their project. They used MT on a package for a product sold by a major retailer. The customer received negative feedback from the retailer. The translations were of low quality, and multiple customers complained. As a result, they had to recall $50,000 worth of merchandise. The product had to be translated by humans and repackaged. The cost to have the content correctly translated upfront would be nothing compared to the price of the recalled products.
Another area where standard MT struggles is with subjective language. There are so many things that are difficult to translate, like idioms, plays on words, and expressions that have no direct equivalent. Running those types of content through a standard MT engine would produce subpar results.
One thing that a buyer of standard MT may not consider is the condition of the source file. If there is complex formatting, breaks, and hard returns, that will pose a serious issue for standard MT. It will segment text in the middle of sentences, which would make the MT have no context.
An alternative to pure standard MT would be to have a linguist edit the results of a translation, referred to as machine translation with post-editing (MT + PE). However, there are some things of which to be aware. Some linguists refuse to edit standard MT simply because of the typically poor quality. If they do edit MT, there is always the chance for fatigue, and the results will be “good enough” rather than great.
This approach offers a strong alternative, as it comes at a similar cost to MT + PE. The linguist has the freedom to translate as they see fit, potentially resulting in higher quality compared to MT + PE. This method relies on a professional translator and presents a viable alternative to standard MT.
There are times when standard MT can be a suitable option; however, beware of the price you pay and have reasonable quality expectations with the price. If you are a shopping provider, be sure to know what you are paying for, and what level of service you are receiving.