Machine translation and artificial intelligence: two technologies transforming the world, simultaneously arousing fascination and inciting panic across the language industry. As an SEO specialist with decades of experience working as a linguist, I often find myself disentangling and correcting unintelligible texts generated by machines.
Editing texts produced by machine translation
Linguists around the world recognize just how impressive current machine technology is. These machines and programs have the capacity to sort through and produce enormous volumes of text in short periods of time.
And, neural networks, deep learning, and artificial intelligence can often quickly generate astoundingly accurate texts.
What has been clear to editors and my fellow linguists for years, as we’ve been correcting the errors in machine-translated texts, is that while the speed and volume are impressive, the quality and depth are often lacking.
Many times as I’ve finished reviewing a few astonishingly good paragraphs and had almost nothing to correct, I’ll happen upon hollow phrases that don’t properly convey the original text’s nuance, or bizarre turns of phrases that could plausibly be the work of surrealist poets where the machine tripped up on simple wordplay or missed cultural context.
Editing texts generated by AI
With AI’s capacity to generate paragraphs in seconds, at first glance, it can be easy to fear that AI will render human brains obsolete. The first time I read about ChatGPT and then used it to produce an original text, I felt a sense of doom.
But, upon additional AI-generated texts, I became aware of a standard style that was being applied over and over again, regardless of the type of text or required tone. And, for now, it seems that despite multiple rounds of adjusting and optimizing prompts, AI still produces texts that can be soulless and lack nuance.
So, even though AI’s capacity to generate text is incredibly impressive, the final product often falls short of achieving compelling texts that encourage target readers to think differently about a subject or inspire readers to do something or buy something. When there are many layers of nuance and cross-cultural contexts to consider, professional human writers still have the capacity to create superior written materials.
Editing machine-generated materials
It would be reckless to publish translations and texts generated by machine and AI programs without first having a human proofread and edit them.
The table, below, provides an overview of the different criteria for humans editing translations generated by machines (machine translation post-editing or MTPE) and humans editing texts generated by AI programs.
Criteria | Editing texts translated by machines (MTPE) | Editing texts generated by AI |
Goal | Improve translation quality and ensure texts contextually make sense. | Improve texts’ quality, coherence, and relevance. |
Typical challenges | Incorrect terminology, lack of nuance, grammatical errors, style issues, and inconsistent tone. | Lack of precision, redundancies, logical inconsistencies, style issues, and inconsistent tone. |
Degree of automation | High: Machine translation programs can manage huge amounts of text and do so with great speed. | Varied: The amount of post-machine editing required depends on the output’s quality can greatly vary. |
Editor’s required qualifications | Knowledge of both languages, training in translation and editing, and cross-cultural understanding. | Experienced editor , knowledge of the topic, and a thorough understanding of the target audience. |
Human involvement | Required to correct errors and sharpen-up the text. | Required to proofread, be aware of nuances, and improve texts’ precision. |
Uses | Translation of documents, websites, software, and technical instructions etc. | Production of content for blogs, articles, product descriptions, social media posts, corporate emails, etc. |
Level of contextual understanding | High: Context is essential for accurate and coherent translations. | High: Context is essential for accurate and coherent content. |
While using a machine to translate or AI program to generate text can initially save time and money, the quality and accuracy of the work can be more prone to inconsistencies.
That said, as the technologies continue to advance, such as with neural networks for machine translations and GPT-4 for text generation, the quality of the output will improve.
Professionals working with AI
Do I think that AI will relegate humans to the role of editor? No. As mentioned above, AI is not currently able to effectively create certain types of text, such as those that require subtle metaphors or nuance, among other things. AI is greatly deficient when it comes to creating texts that employ humor, especially if the text is to be effective across different cultures.
Technological advancements will continue to transformation the language industry along with thousands of other industries. Innovation is, overall, a positive force for the world, industry, and people. For centuries, humans have adapted and benefitted from technological developments from the printing press to the combine to automated factories to personal computers and hand-held devices etc.
Just as professionals in other industries have done, the linguists and writers of Leinhäuser Language Services will continue to adapt to new technologies so we can harness the beneficial efficiencies they offer, while continuing to imbue our work with our expertise and myriad insights to deliver flawless products to clients.
If you’d like to learn how AI efficiencies can be used to benefit your translation, transcreation/cross-cultural translation, and copywriting projects, contact us at info@leinhaeuser.com. We look forward to talking with you.
About the author: Udo Leinhäuser was managing partner of Leinhäuser Language Services until the end of 2012. Since then, he has been intensively involved in SEO, with a focus on international search engine optimization. E-mail contact: udo@i-seo.works