Machine Translation Risks: Why LSP's Still Have the Advantage

07 01 2025

It's impossible to ignore the power, speed and convenience that machine translation (MT) now offers global businesses. The ability to translate massive amounts of text in seconds has revolutionised how companies and individuals communicate across language barriers. Yet, as with all new technologies, the potential risks surrounding MT should not be overlooked.

While MT technology has been steadily evolving since the 1950s, with significant advances in recent years, the dangers of relying on it too heavily are now clear. From major security breaches to embarrassing mistranslations, these automated tools come with pitfalls that can have serious consequences.

As more companies turn to MT for its efficiency and convenience, it's crucial to recognise its limitations. Join us as we explore how exactly MT falls short before outlining how businesses can clearly manage its risks while still utilising this game-changing technology.

Security

As with any industry or sector, data security is paramount when it comes to translation. Whether it's legal documents, corporate communications, or personal records, any breach in security can have severe consequences.

The convenience of free machine translation services often comes with data security concerns of their own. Most free MT tools and apps use your data to train their machine learning models, which can lead to significant security risks. The terms and conditions for these services often obscure the full extent of data usage, leading users to unknowingly expose sensitive information.

A striking example of this risk is the privacy breach involving Translate.com. In 2017, it was revealed that sensitive documents, including resignation letters, employment contracts, and details of corporate mergers, were accessible online through Translate.com's free service. This breach occurred because the platform stored translations in a way that allowed them to be indexed by Google, making confidential information publicly available. The breach was so severe that organisations, including the Oslo Stock Exchange, reacted by blocking access to Translate.com.

Beyond data exposure, MT systems are also susceptible to backdoor attacks. A study found that injecting less than 0.01% of poisoned data into a low-resource language pair can achieve a 20% success rate in attacking high-resource language pairs. This vulnerability could lead to malicious translations, putting both your data and the accuracy of the translation at risk.

Mitigating these risks requires a comprehensive approach. First and foremost, businesses should implement strict guidelines on how employees use publicly available translation apps and platforms. These guidelines should include a careful reading of terms and conditions, opting for paid versions when necessary, and being mindful of the permissions granted to the software.

However, the most effective strategy is partnering with a trusted language service provider. These providers are experts in machine translation, ensuring both data security and high-quality translation outputs. By using secure, professional services, businesses can safeguard their information while still benefiting from the speed and efficiency of machine translation.

Quality Matters

One of the main attractions of machine translation is its speed. A professional human translator may translate around 3,000 words per day, whereas MT tools can process vast amounts of text in just minutes. However, this incredible speed does not necessarily translate to quality.

The quality of MT output varies significantly depending on how well the engine is trained and how customised it is for specific clients. Generic MT tools, while easily accessible, do not guarantee accuracy. Mistranslations can lead to serious errors, especially in low-resource languages where the risk of distorted meaning is higher. For example, legal or medical documents require precision, and any error can have severe consequences.

MTs are excellent for getting a rough understanding of a text or assisting in translating languages you or your employees already have a strong understanding of. However, when it comes to detailed understanding, published content or legally binding documents, the risks must be managed carefully. Several strategies can help reduce these risks:

● Choose your MT tool wisely: Different tools offer varying levels of quality depending on the language combination and the domain. Not all MT tools are created equal, and selecting the right one is crucial.

● Train your MT engine with your bilingual data: Customising the MT engine with your own data improves its accuracy and relevance to your specific needs.

● Enhance MT performance by adding termbases: Termbases ensure that specific terminology is translated consistently and correctly.

● Use raw machine translation only for texts you already have a strong understanding of: When you only need a basic grasp of a document or its less critical details, raw MT might suffice.

● Conduct careful post-editing of the MT output: Ensure a skilled translator reviews the machine-generated text to correct errors and improve accuracy, producing a polished final version.

● Collaborate with language professionals: Experts can advise when human translation is necessary and when MT can be effectively used to streamline processes.

For businesses and individuals who prioritise accuracy, context and quality, human translators remain the superior choice. While MT can save time and resources, it should be used as a tool to assist human translators rather than replace them. Human expertise is essential to ensure the final output is accurate, culturally appropriate and of high quality.

Managing Language Coverage Risks

While many languages are now supported by top-tier MTs, the extent and quality of that support can vary significantly, especially for less widely spoken languages. The increasing availability of MT has provided unprecedented access to languages that might otherwise be inaccessible. For instance, Google Translate recently announced the addition of 110 new languages, expanding their total supported count to over 250.

However, the level of support for these languages varies, and the gap between high and low-resource languages remains wide. For lower-resource languages, the risk of inaccurate MT output is higher. These languages often lack sufficient training data, leading to frequent errors and less reliable translations.

MT can be a useful tool for an initial understanding of a text. However, users should not expect high accuracy, especially in low-resource language pairs, where relay translation (translating through an intermediate language) is often used. Relay translation can introduce additional errors and distort the meaning further. Once the relevance of the text is established, investing in human translation or thorough post-editing is advisable to ensure the text's accuracy.

To manage these risks, it's important for users to understand the limitations of MT and use it appropriately. This might involve combining MT with human expertise, especially when working with low-resource languages. It's also crucial to stay informed about the specific capabilities and limitations of the MT tools being used and to apply caution when dealing with sensitive or critical information that requires high accuracy. Leveraging the expertise of professional language service providers can help ensure that even in less common languages, the translation meets the necessary quality standards.

Communication Mishaps

The development of AI tools has improved global business communication in many ways, but it has also introduced new challenges. MT and other AI-powered translation tools can be useful for understanding basic information and facilitating initial communication. However, they often fall short when it comes to critical tasks like translating legal or regulatory documents and launching high-impact marketing campaigns.

A prime example is Amazon's launch of its Swedish website in 2020. Many translation errors marred the site, leading to confusion and embarrassment. For instance, a collection of Second World War-era Russian infantry figurines was translated as "Russian toddlers". Such mistakes can damage a company's reputation and, in some cases, lead to financial loss or legal complications.

This issue is not exclusive to MT - using low-quality language service providers can also result in costly errors. A famous example is HSBC's rebranding in 2009, where their tagline "Assume Nothing" was translated into other languages as "Do Nothing," costing the company $10 million to fix. This demonstrates that even with human translation, quality matters, and choosing the right provider is absolutely crucial.

High-impact translation tasks should be entrusted to established language service providers with a proven track record of delivering quality results. These providers have the experience, expertise and resources to handle complex projects, making sure that all translations are accurate, culturally appropriate and aligned with the client's goals.

Conclusion

Machine translation offers significant advantages in speed and accessibility, but these come with considerable risks, particularly in areas of security, quality and language coverage. The potential for communication mishaps is high, especially when dealing with sensitive or critical information. To avoid these risks, it's essential to understand the limitations of MT and to combine it with human expertise where necessary.

Partnering with a reliable translation service provider is the best way to ensure that your translations are secure, accurate, and of the highest quality. As experts in the language services industry with a long history of satisfied clients, Diskusija is here to help you navigate these challenges with confidence.

Back to feed