TECH
Generative AI
The future of translation expertise
By Martinien Afangbedji
Introduction
In the past few years, the translation industry has undergone substantial changes, primarily driven by the progress made in artificial intelligence (AI) and machine learning. A noteworthy domain that is reshaping the industry is generative AI, a branch of AI that uses advanced deep learning models to create fresh content like essays, images, and even voices. With expertise in generative AI prompt design and engineering, translators can augment their work by leveraging these innovative AI technologies, improving efficiency and accuracy in their many translation tasks.
In this article, we’ll take a look at how knowledge and experience in generative AI prompt design and engineering can reshape the future of expertise in the translation field. We will discuss practical examples from using the OpenAI Translator API integrated with Trados Studio. By applying the best practices of prompt design, we can enhance the contextual accuracy of translation tasks and unlock new possibilities for the industry.
Understanding generative AI and its potential impact on translation
As defined in a recent article from Forbes, generative AI is the subset of AI that utilizes deep learning algorithms to produce new content, including images, texts, or sounds. These algorithms, commonly built upon neural networks, possess the ability to discern patterns and structures from extensive datasets and subsequently generate novel outputs that closely resemble the input data. This implies that, in the context of translation, generative AI can be used to generate translations that were not previously seen in the training data. It proves valuable in generating precise translations of texts, taking into account the context, style, and tone of the original content.
Here are examples of how generative AI is being used today:
- Generating text: Generative AI can be used to generate text, such as news articles, blog posts, and marketing copy. This may help organizations save time and money while also enabling them to produce more interesting and individualized content. Some examples of text generative AI include GPT-4, Bard, Jasper, and Bloom.
- Generating images: Generative AI can be used to generate images, such as product photos, marketing materials, and even works of art. This can help businesses create more visually appealing and impactful content. Examples of image generative AI: DALL-E 2, Midjourney, Artbreeder, Dream by WOMBO, Deep Dream Generator, and many more.
- Generating music: Generative AI can be used to generate music, such as soundtracks, jingles, and even entire albums. This can help businesses create more original and engaging music for their products and services. Some examples of music generative AI are Amper Music, AIVA, Google Magenta, Jukebox, and others.
Figure 1: Image source: Lionbridge
The role of generative AI in translation
Generative AI has the capacity to revolutionize the translation industry through its ability to deliver remarkable speed, accuracy, and efficiency. In a comparative evaluation of generative AI and machine translation (MT) tools, researchers at Lionbridge discovered that generative AI-based translation systems exhibited superior performance when compared to neural machine translation (NMT), despite NMT being considered a more advanced approach than traditional rule-based and statistical machine translation systems.
The evaluation highlighted the remarkable gains achieved by generative AI in terms of translation quality, speed, and efficiency. Specifically, the study found that generative AI-based systems produced translations that were more fluent, accurate, and natural sounding, while also requiring less post-editing effort from human translators.
The image to the upper right illustrates the similarity in reverse edit distance between the NMT engines and LLMs, indicating comparable performance. However, GPT-4 demonstrated slightly superior translation quality compared to Yandex NMT for the English-to-Chinese language pair. Edit distance is a measure of the number of edits required to align the MT output with multiple human translations, representing translation quality. The study involved comparing the raw MT output against ten distinct human translations. A higher inverse edit distance signifies better translation quality.
By leveraging the capabilities of these AI models, translators can surpass the constraints imposed by conventional translation approaches like rule-based or statistical machine translation. Generative AI empowers translators to produce translations that not only maintain grammatical correctness but also capture the contextual nuances and stylistic coherence of the source text.
OpenAI Translator API in Trados Studio
At the 2023 ELEVATE Trados online conference, RWS Group made an exciting announcement regarding the integration of OpenAI Translator, powered GPT-4, GPT-3.5-turbo, and text-davinci-003 large language models (LLMs),with Trados Studio 2022 and above. This integration allows users to access the power of generative AI-based translation directly within Trados Studio, improving translation quality and efficiency.
After participating in the conference and testing out the plugin myself, I felt I had to share it with my colleagues. “I quickly downloaded the plugin and put it to the test within my trusted Trados application. And let me tell you, the results are simply astonishing! With the OpenAI Translator plugin, I’ve witnessed a remarkable boost in my productivity as a translator. The quality of machine translations has soared, and I believe it’s going to make a world of difference for my team,” reads a LinkedIn post that I excitedly typed up afterward.
Figure 2: A look at the interface of the Trados Studio 2022 OpenAI Translator plugin
This integration allows translators to access the cutting-edge AI translation technology directly within their Trados application, significantly boosting productivity and translation quality.
Benefits of the OpenAI Translator integration
With the OpenAI Translator plugin in Trados Studio, translators can experience various benefits, including:
- Assistance in translation projects: The plugin is designed to assist users in their translation projects by leveraging OpenAI’s language models. It provides support and resources to enhance the translation process.
- Alternative translations based on prompts: The plugin offers alternative translations based on user-defined prompts. Users may use this tool to go over many translation possibilities and select the best one for their particular requirements.
- Additional information about source/target content: The plugin provides additional information about the source and target content being translated. Users’ comprehension of the context will likely increase as a result, and their translations will likely be more accurate and of higher quality.
- Automatic or manual translation triggering: Users have the flexibility to choose between automatic or manual translation triggering. This means they can either let the plugin automatically generate translations or manually control when the translation occurs, based on their preferences and work requirements.
- Support for different language models: The plugin supports various language models, including GPT-4, GPT-3.5-turbo, and text-davinci-003. Each of these models has distinct capabilities and performance levels, providing users with options to choose from based on their specific translation needs and goals.
- Custom prompts: Users can create custom prompts to guide the AI. This feature allows users to tailor the translation process and provide specific instructions or guidelines to the AI model, further improving the accuracy and relevance of the generated translations.
- Easy installation: The OpenAI Translator plugin can be easily installed via the RWS AppStore or directly within Trados Studio 2022. This straightforward installation process ensures convenient access to the plugin and its functionalities.
The images below depict the application of multiple translations generated from carefully crafted prompts using OpenAI Translator in Trados Studio 2022.
Figure 3: Prompt design/engineering window of the OpenAI Translator plugin in Trados Studio 2022
The importance of prompt design in generative AI
Prompt design refers to the process of crafting effective prompts or input sequences for generative AI models to produce desired outputs. In the context of language models like OpenAI’s GPT, prompt design plays a crucial role in guiding the model’s behavior and generating accurate and relevant responses. A well-designed prompt can significantly influence the quality and relevance of the AI-generated content. In translation, prompt design involves formulating input text in a way that encourages the AI model to generate accurate, contextually appropriate translations.
Here are some practical examples to illustrate prompt design:
Translation prompt for a technical document
- Prompt: “Translate the following technical document from English to French. Maintain the technical terminology and ensure the accuracy and clarity of the translated content.”
- Desired response: The AI model generates a translation of the technical document into French while preserving technical terminology and ensuring accuracy and clarity.
In this example, the prompt clearly instructs the AI model to translate a technical document from English to French. It emphasizes the importance of maintaining technical terminology, accuracy, and clarity, guiding the AI model to prioritize these aspects during the translation process.
Contextual prompt for marketing content
- Prompt: “Translate the marketing copy for a new fashion brand into Spanish. Adapt the tone and style to resonate with the target audience, which consists of young, fashion-conscious individuals.”
- Desired response: The AI model generates a culturally appropriate translation of the marketing copy for the new fashion brand into Spanish, ensuring it resonates with the target audience of young, fashion-conscious individuals.
In this example, the prompt provides context and specifies the target audience, enabling the AI model to produce a translation that aligns with the cultural preferences and stylistic expectations of the Spanish-speaking audience. It guides the AI model to adapt the tone and style accordingly.
Localization prompt for a mobile application
- Prompt: “Localize this user interface (UI) string of a mobile application from English to Eʋe. Ensure the length of the translated UI elements corresponds with that of the original elements in order for it to fit within the designated space and maintain consistency with the app’s overall design.”
- Desired response: The AI model generates a localized version of the mobile application’s UI strings in Eʋe, considering space limitations, and maintaining consistency with the app’s design.
In this example, the prompt focuses on localization, which involves adapting the UI strings of a mobile application to the Eʋe language and culture. It highlights the importance of considering space limitations and maintaining consistency with the app’s design, guiding the AI model to produce a seamless and user-friendly localized version.
Legal translation prompt for a contract
- Prompt: “Translate this content of a legal contract from English to French. Pay close attention to legal terminology, and preserve the legal intent of the original text.”
- Desired response: The AI model generates an accurate and legally sound translation of the contract from English to French, demonstrating a deep understanding of legal terminology, precise interpretation of clauses, and preservation of the original document’s legal intent.
In this example, the prompt addresses the specific requirements of legal translation. It emphasizes the need for a meticulous approach, focusing on legal terminology, accurate interpretation of clauses, and maintaining the legal intent. The prompt guides the model to prioritize precision and legal accuracy in the translated contract.
Best practices for prompt design
To maximize the effectiveness of generative AI models in translation tasks, it is essential to adhere to the best practices of prompt design, such as:
- Be concise: Keep the prompt short and to the point, making it easier for the AI model to understand the task.Be specific: Clearly define the desired output, reducing the chances of ambiguous or irrelevant translations.
- Ask one task at a time: Break down complex tasks into smaller, more manageable prompts to ensure better results.
- Ask to classify instead of generate: Request the model to categorize or classify information, rather than generating new content, to improve the translation’s accuracy.
- Include examples: Provide examples of the desired output, giving the model a clearer understanding of what is expected.
The future of expertise in translation
As generative AI becomes more prevalent in the translation industry, professionals will need to adapt to new roles and acquire new skills. Expertise in generative AI prompt design and engineering will become increasingly valuable, enabling translators to stay ahead of the curve and offer their clients the best possible services.
Here are a couple of the new skills and roles that are likely to be in high demand:
- Generative AI prompt designer: The role of a prompt designer in the context of generative AI involves creating effective and well-crafted prompts that guide the AI models to generate accurate and contextually appropriate translations. The prompt designer should have a deep understanding of the intricacies of the source and target languages, as well as the specific domain or subject matter being translated. They will need to carefully craft prompts that provide sufficient context, clarify ambiguous phrases, and capture the desired tone and style of the translation.
- Human reviewer and AI ethicist: With the growing influence of AI in translation, the role of a human reviewer and AI ethicist gains prominence. These professionals will be responsible for reviewing translations generated by generative AI models, evaluating, and addressing ethical considerations related to the use of generative AI in translation. They will assess potential biases, cultural sensitivities, and privacy concerns, ensuring responsible and inclusive AI practices.
Revolutionizing the translation industry
The integration of generative AI into translation tools like Trados Studio is just the beginning of a new era in the industry. As AI technologies continue to advance and improve, we can expect even more sophisticated translation solutions that further enhance the speed, accuracy, and efficiency of multiple translation tasks.
These advancements will not replace human translators but rather empower them to deliver higher-quality translations in less time. With the assistance of generative AI, translators can focus on the creative and nuanced aspects of their work, while AI handles the more routine and repetitive tasks. The collaboration between human expertise and AI capabilities will revolutionize the translation industry, opening up new possibilities and opportunities for professionals in the field.
Translators who embrace and adapt to these advancements will be well-positioned to thrive in the future of translation.
The future of expertise in translation lies in harnessing the power of generative AI. Translators can significantly enhance the quality and efficiency of their work by applying advanced technology and adhering to industry best practices. This transformative technology has the potential to revolutionize the entire industry. To stay relevant and succeed in this dynamic field, professionals must continuously stay updated and adapt to the evolving landscape.
Martinian Afangbedji is the chief technology officer at Pairaphrase. is a language services professional based in Ghana.
RELATED ARTICLES
The Limits of AI with Language
Large language models (LLMs) are all the rage nowadays and it is almost impossible to get away from the news frenzy around ChatGPT, BingGPT, and…
→ Continue Reading
Spence Green An AI Vision For Localization
Supported By Lilt Spence Green explains how his experience as a research scientist at Google Translate over a decade ago sparked a grand idea: use…
→ Continue Reading
Best practices for securing enterprise machine translation
Organizations today find themselves dealing with an ever-growing volume, velocity, and variety of multilingual content and data. Digital transformation, including the recent rapid advancements in…
→ Continue ReadingWEEKLY DIGEST
Subscribe to stay updated between magazine issues.
