Hybrid (AI machine + human) translation

Our customers need high-quality translations as soon as possible and at a reasonable price. But these three common criteria are hard to combine. To meet this demand, InText employs a wide range of software tools including ones that perform machine translation (MT).

Machine translation is the process of automatically generating text in a target language based on text in a source language. In recent years, machine translation technology has made significant progress due to the application of neural networks and artificial intelligence (AI). Neural machine translation (NMT) is the latest MT technology, providing acceptable results for various business cases.

Machine translation offers two main advantages: high speed and low cost.

Hundreds or even thousands of pages can be machine translated in a single day for a fraction of the cost of a full human translation. But this technology, like any other, should be used with knowledge and caution. Even the most sophisticated algorithms do not guarantee 100% accuracy in the target language, so qualified linguists are vital for achieving the desired quality.

The feasibility of using machine translation depends on three major factors:

• Nature of the source text. Easy-to-read and well-structured descriptive texts are best suited to machine translation.

• Fluency requirements for the translated text. When the speed of communication is more important than style, machine translation can often provide a good result.

• Possible risks of miscommunication. Machine translation is suitable for situations where the possible consequences of a misunderstanding are negligible. But when information is crucial — such as for business negotiations or healthcare decisions — you need to perform a careful quality check of machine translated texts or opt for full human translation.

At InText, we adhere to ISO 18587:2017 standards for post-editing of machine translations. ISO 18587:2017 applies only to content processed by MT systems, providing requirements for the process of full human post-editing of machine translation output and for editors’ competences.

MT-related services

Our range of machine translation services includes:

• Assessing your translation project. We always strive to provide our clients with optimal solutions. To do this, we analyze your source files and requirements and recommend the most efficient translation workflow.

• Preparing source files for machine translation. It’s not always possible to simply upload files to a machine translation system. For example, scanned copies or files with sophisticated layouts may be challenging. We can preprocess your files to remove stumbling blocks for MT software.

• Machine translation with full post-editing. Comprehensive checks and corrections of machine-generated output will give you quality comparable to a human translation.

• Machine translation with premium post-editing. Depending on your requirements, a quality check of machine-generated text by a human linguist and revision by a second linguist will be the best choice to ensure your message is clearly delivered to your audience. In this case, the quality of the translated text will equal the quality of a human translation.

Large Language Models (LLMs)

Large language models (LLMs) such as GPT-3, GPT-3.5, and GPT-4 use deep learning algorithms to perform a variety of natural language processing (NLP) tasks. LLMs use transformer models and are trained using massive datasets, enabling them to comprehend, translate, predict, and generate text or other content. ChatGPT, Claude, Gemini, and Llama are considered some of today’s largest language models.

LLMs are transforming the localization industry, helping to significantly improve the translation and adaptation of content for diverse markets. Here are some of the key areas where LLMs are actively used:

• Post-editing. Our editors use LLMs as a tool to refine human-translated texts by checking for accuracy, cultural adaptation, and stylistic consistency. This speeds up post-editing and reduces costs. (Note that InText only implements LLMs with client approval.)
• Content localization. Large language models can consider cultural and linguistic nuances, drawing from multiple data sources to understand the context of a localization project and offer an accurate and culturally appropriate translation in the target language.
• Automation of routine tasks. LLMs help to automate tasks such as segmenting text, searching for similar phrases, and extracting key concepts, which speeds up the work of translators.
• Localization testing. LLMs can be used to test the quality of localizations, helping to identify errors, inconsistencies, or phrases that do not sound appropriate in the target language.

While InText is currently using LLMs with human assistance, we are exploring ways to automate their use. It’s important to note that LLMs are not yet fully used for translation, as they underperform compared to Neural Machine Translation (NMT) models.

LLMs significantly speed up localization workflows, improve translation quality, and help to reduce costs. However, human input remains essential, especially when deep understanding of cultural and linguistic nuances is required. InText is committed to striking the right balance, using LLMs while ensuring that the final product is approved by professional human translators.

A more detailed description of our services is provided in our machine translation menu, which is available here:

InText_Machine_Translation_Menu.pdf Download