How to Create Authentic Translated Content in the AI Era

With recent advancements in artificial intelligence (AI), many are tempted to rely solely on machine translation (MT) to save time and reduce costs. AI can be an opportunity for small companies to generate authentic translated content at a reduced cost. However, its uncontrolled use can be a double-edged sword: many companies have experienced a dramatic drop in their Google rankings after using content generated exclusively by AI.

Unedited machine translated content is proliferating across the web, often doing more harm than good to a company’s brand image. According to Mark Heitmann, Professor of Marketing & Customer Insight at the University of Hamburg and Adjunct Professor of Marketing at NOVA School of Business and Economics in Lisbon, “the productivity frontier of content quantity and quality continues to expand.” However, in an article published in Multilingua, the author writes that “high quality translation is a complex cognitive process that requires skills AI currently lacks, including authenticity.” In this article, we will explore what MT is, as well as its strengths, its limitations, and its acceptable uses.

What is Machine Translation, and How Mature is This Technology?

Machine Translation (MT) is the automated process of translating text or speech from one language to another using computer algorithms and AI. Since the quality of MT output is quite variable, ongoing research and development is needed to improve its effectiveness. Furthermore, although MT is available for many languages, translation accuracy can vary between language pairs, which militates in favor of a human-in-the-loop approach. 

The History of Machine Translation: From Statistical Machine Translation to Neural Machine Translation

Machine translation began with rule-based systems that relied on predefined linguistic rules. “Machine translation was long characterized as a complex, expensive technology requiring a lot of computing horsepower. In 1956, if you wanted MT, you had to go to the United States Navy or to the KGB to get it. Today, the options for using MT ranged from desktop to in-house servers to free online access,” explains DePalma.

The 1990s saw a shift to statistical machine translation (SMT) using bilingual text corpora for predictions, while the 2010s introduced neural machine translation (NMT) with artificial neural networks for improved accuracy and fluency. Hybrid models combining rule-based and statistical methods emerged, leading to more robust systems. Rapid advancements in computational linguistics and AI culminated in technologies like OpenAI’s ChatGPT. In 2023, ChatGPT, built on the GPT-3.5 architecture, revolutionized translation with highly accurate, contextually relevant translations across multiple languages, showcasing the potential for seamless, high-quality automated translation solutions. 

What is the Difference between AI Translation and Machine Translation?

AI significantly enhances language processing through the use of machine learning models and neural networks. AI tools can generate human-like text and translations across many languages. They certainly showcase advancements over statistical machine translation. AI capabilities extend beyond language conversion, revolutionizing language processing, enhancing communication, and bridging linguistic gaps. Tools like OpenAI’s ChatGPT, DeepL, and Google Translate exemplify these advancements.

When and How to Properly Use Machine Translation

Machine Translation (MT) is effective for translating large volumes of straightforward documents, but struggles with specialized texts requiring creativity and precision, such as legal and literary translations. According to an article published in ScienceDirect, the use of MT has shown great potential in the medical field, but there are ethical implications, and “incorporating both human expertise and machine translation technology can yield more precise and practical translations.” To ensure quality, MT must always be used as a tool alongside human verification. This process is known as machine translation post-editing (MTPE). Light post-editing (PE) corrects minor errors, while Full PE involves extensive rewriting of machine translation output to achieve a more natural flow.

Citing research conducted by CSA, the Intelligent Information Blog explains that, in the traditional post-editing process, language service providers (LSPs) send linguists raw machine translation output, which often requires extensive post-editing for a fraction of the price of a human translation. This undervalues linguists’ skills and typically offers no improvement for future cycles, leading to dissatisfaction among linguists. Conversely, CSA Research advocates for “augmented translation,” where artificial intelligence aids rather than replaces human translators. In this model, human translators remain central to the process, and all translation is ultimately human-driven, enhancing efficiency and decision-making.

How to Optimize Machine Translation Output

To enhance MT output, integrate MT engines (e.g., DeepL), CAT tools, translation memory systems, glossaries, and online translation services. This approach ensures heightened accuracy and consistency. Neural Machine Translation (NMT) engines can be trained with clients’ glossaries and industry-specific terminology.  

Creating authentic translated content in the machine translation and AI era involves combining the strengths of advanced AI tools with human expertise. Begin by using reliable AI translation tools such as DeepL, Google Translate, or OpenAI’s ChatGPT to generate a preliminary translation. Next, engage professional translators to review and refine the content to ensure that cultural nuances, context, and idiomatic expressions are accurately conveyed. This hybrid approach leverages the efficiency of AI, while maintaining the authenticity and quality that only human insight can provide, resulting in authentic translated content that resonates well with the target audience. 

What are the Limitations of Machine Translation?

CSA Research states clearly that “MT and large language models cannot be trusted to produce high-grade translations.” 

Utilizing free machine translation engines like Google Translate or DeepL can raise privacy concerns, as input data might be repurposed to train neural machine translation (NMT) engines. While machine translation (MT) has made significant strides, it still faces challenges with technical terminology and idiomatic expressions, which require human expertise for accurate interpretation. Automated translation systems often overlook subtle language nuances, leading to issues like gender bias and potential inaccuracies or omissions in the translated text. Despite these limitations, MT remains valuable for specific tasks, especially when complemented with human post-editing to refine and enhance translation quality.

What are the Common Errors of Neural Machine Translation?

Neural Machine Translation (NMT) often struggles with interpreting context and cultural nuances. As such, although it produces translations that may be accurate word-for-word, it often fails to convey the intended meaning, particularly when processing ambiguous language or long sentences. This leads to inconsistent terminology, incorrect handling of words that shouldn’t be translated (DNT), and errors with acronyms, formatting tags, spacing, capitalization, and hyphenation. Consequently, NMT can result in awkward, non-idiomatic translations that deviate from clients’ style guides and project specifications, necessitating extensive rework by human translators to ensure quality and accuracy.

Is AI Sexist?

Gender bias in AI translation highlights the need to address systemic biases in training data. The reason is that AI models often unintentionally reinforce societal gender stereotypes learned from extensive datasets. This issue is particularly significant in languages with gendered grammatical structures, which can perpetuate stereotypical associations. The lack of diverse voices in training data exacerbates these biases. This results in translations that reflect narrow perspectives, especially in gender-specific roles or professions. To combat this, it is crucial to diversify training datasets, implement bias-detection mechanisms, and ensure transparency in AI development processes. Prioritizing diversity and inclusivity is crucial. It will help create translation systems that accurately reflect the diversity of human experience and uphold ethical standards.

Tips and Reminders about MT 

• MT should always be a means to an end, not an end itself. Machine Translation Post-Editing (MTPE) by a professional translator is crucial. 
• Be cautious with free generic machine translation services, as using them may raise privacy concerns. 
• MT is valuable for technical translations, but less effective for texts requiring creativity or cultural knowledge.

Our Experience

ArtLingua considers MT as an additional tool in the translation process. We will use it when requested or when it is deemed appropriate for a given project. However, we always prioritize the human aspect, and place great emphasis on:

terminology research, project specifications, and professional translators to produce idiomatic translations and create authentic content. 

With 25 years’ experience in translating and localizing multilingual content, ArtLingua can assist you in effectively conveying your brand message and establishing a strong presence in international markets. Specializing in European languages, our expert translators produce high-quality translations that resonate culturally and linguistically. Our extensive experience in post-editing machine translation unlocks new opportunities for growth and success.  

Do you need translation services that leverage technologies, but still put a premium on the skills of language professionals? 

Feel free to reach out to us for a free consultancy.