91黑料

  1. Home
  2. Insights
  3. Blog
  4. A clinical trial linguist’s perspective on machine translation

A clinical trial linguist’s perspective on machine translation

In this blog, we reflect on the adoption of technology in translation and look through the lens of a linguist at some of the transformational changes in clinical trial translations. We asked experienced 91黑料 Language Services linguists for their insights on how machine translation (MT) has transformed some of our workflows and the opportunities this technology brings to our working lives and our customers.

Love it or loathe it, machine translation is here to stay.

The current opinion on MT includes a wide spectrum of views and emotions. On one hand are the technology enthusiasts, keen to adopt the latest and greatest release at the earliest opportunity. On the other hand are the resisters, who push back the increasing advancements of change to well-established working practises. We are all human and have varying change acceptance curves. Where we all eventually agree is that technology, just like any tool, has great potential and great limitations, depending on the skills of the people wielding them. From a linguist’s perspective, we must ensure any adopted tool can capture the rich nuance of one of the pinnacles of evolution: language.

Machine Translation (MT) was not developed and adopted overnight. It took decades of effort to refine and perfect, but the stories surrounding its release often overlooked the extensive development process behind it. MT was introduced as a groundbreaking new concept, without fully acknowledging the history of the work that shaped it into the advanced tool it is today. It was also launched during a time of broader acceptance of transformational change, which helped speed up its adoption. At the same time, a wave of debate emerged — the term "machine translation" created an impression of simplicity in translation, leading to theories suggesting that human linguists might no longer be necessary. This is why myths about MT continue to circulate, and debates about its role persist.

How has MT changed clinical trial translations?

Some of us have been translators for long enough to remember when we relied on printed dictionaries and resources. Back then a rudimentary word processing system limited our ability to store, reuse and adapt our work. In a relatively short space of time, we have witnessed astronomical advances in technology that have transformed our industry, with the introduction of Computer-Assisted Translation (CAT) tools, MT, Neural MT and now Large Language Models (LLM) and generative AI (Gen AI) models. 

On the surface, the translation process remains largely unchanged. Linguists utilise a CAT tool which displays the original text (in the source language) and the translation (in the target language) side-by-side. This leverages phrases and/or entire segments of previously translated text, for the linguist to review, amend and approve for use in the translation. This body of previously translated content, referred to as a Translation Memory (TM), is a valuable resource. The use of legacy wording allows the translator to work efficiently and ensures that language is used consistently. And it is at this stage that MT comes in. 

As explained in a recent 91黑料 blog, MT works alongside Translation Memories, using algorithms and machine learning models to generate an approximate translation in the desired language. The MT engines are regularly “trained”, learning from the corrections made by the linguists at post-editing stage. In this way, MT constantly improves the output quality and evolves the prediction algorithms. Essentially, when we receive a document, part of the content is leveraged from the Translation Memory and the remaining text is machine translated. Linguists are involved at every stage. We may not translate from scratch, but we edit the text provided. We do not use MT as a standalone feature or service in our area of work. It is also important to note that we use machine translation in custom engine setups that protect confidential content and do not share it outside our company or with the public.

How does MT measure up to the clinical trial industry’s high standards?

There are certain text types that are more suited for the use of MT: texts with shorter sentences and repetitions that do not overuse synonyms. Information documents and (patient-facing) clinical material are ideal candidates. Furthermore, with the improved efficiency MT brings, linguists can complete more words in the same amount of time, which means we work on more varied content and build up knowledge on different subjects. As in-house linguists the approach to overall linguistic quality assurance and asset management has changed. While the quality assurance process was traditionally aimed at work created by humans, we now focus on improving data provided by both machines and humans making it easier for the MT engines to learn. 

The introduction of automation to the translation industry spotlights the very thing that attracted many of us to the trade: the fundamental role language plays in determining an individual’s personal and cultural identity. A word may have a given dictionary meaning, but how it is perceived and understood, and the context in which it should be used, can vary significantly. Its interpretation can change from country to country, culture to culture, region to region and even individual to individual. This is one of the things that MT still struggles with. While the dictionary meaning may be strictly correct, the usage of the word may not be appropriate and could cause confusion or offence in another context. 

One of the areas MT does not (yet) handle well is the rendering of nuances of meaning and use of figurative, descriptive language. For example, the title of a patient recruitment text might invite participants by referring to their “Journey With Clinical Research”. This might generate a direct MT output such as “su viaje con investigación clínica” in Spanish. One of the challenges with this output is the direct translation of “journey” as “viaje” (travel). In this context “journey” is not meant as the physical act of travelling from one place to another, but the symbolic and emotional journey of a person living with a disease, encompassing their treatment and how it affects their life. The MT output does not convey any of this. It takes a human, with an understanding of what it means to be healthy or to have a disease, to produce a translation that reflects this. Our task as linguists is to ensure the data used to train the MT engines is of high-quality, so that the engines can start picking up on these subtle nuances eventually. As with any source text intended for translation and/or localisation, the general rule applies: the better the input, the better the translated end result. Clarity of meaning at source (both within a document and project-wide), consistency and grammatical accuracy are of paramount importance for a quality end product, especially when using MT.

Conclusion

Linguists’ working days are very different today from when some of our careers began with a bookshelf of dictionaries by our desks. None of us long to return to those analogue days with clunky word processers and limited possibilities. The combination of customised MT and CAT tools means that we still deliver translations of the highest quality for our customers, but today we do it more efficiently and in greater volumes. A specialised human linguist is essential to skilfully operate those tools, amending and adapting translations and ensuring they meet the rigorous quality standards of the clinical trials industry. For now, at least, machines cannot match the human mind’s ability to differentiate the many subtleties of language which depend on context and audience. From start to finish, linguists are vital to maintain the industry’s high standards by bringing their experiences – as translators and human beings – to the process of conveying important ideas through language.    

 

Contact us to learn more about how we can support your language services needs.

In this section

Connect with us