At ITC Global, we understand the importance of striking the right balance between artificial intelligence and human expertise. We view AI as a useful, complementary tool that helps optimize productivity but not as a replacement for human translators and content writers. Read on to learn more about humAIn®, where artificial intelligence meets human expertise.
humAIn® combines AI-powered technology with human expertise for all your language needs. This service provides optimized quality along with reduced costs and shorter timelines.
Neural Machine Translation (NMT) combines AI with statistical machine translation, using algorithms. The engine reviews the input as a whole, rather than by subsections, so the element to be translated is considered in context.
An NMT engine is AI that learns to translate using an artificial neural network, which is inspired by the human brain’s neural networks.
AI-powered translation always requires review by a human post-editor, except in specific cases (e.g., getting a gist of large volumes of social media posts or patent filings, for example).
The term “engine” in the software field is like the engine in a car. Just like this engine makes the car move, a software engine helps a software system do certain tasks, in this case, translate text from one language to another.
In most cases, clients who use online AI/MT translation, transcription or video tools have free or basic licenses with raw/unedited output, which creates confidentiality, privacy and quality risks. In contrast, we guarantee the security of your data and we only use paid AI-assisted tools provided by companies with stringent confidentiality and data security policies and practices.
For translations, our experienced human post-editors minimize the risks associated with unedited machine translated text including errors and inconsistencies.
When it comes to transcription, subtitling, dubbing, voice-over and other services for video projects, it’s easy to underestimate the resources required to review, format and do post-production on audio and video files. These complex tasks typically require human expertise for polished, professional results.
We test your content to make sure humAIn® is the best option. And we use the best suited engines depending on the project’s domain, content, and language pair.
The human editing step corrects typical machine translation errors for optimized quality. This is typically called post-editing in the translation industry because it involves editing text after it is processed by the machine translation engine.
ITC Global only works with artificial intelligence companies via paid subscription services. This guarantees the content we submit is not permanently stored or used to train their technology or improve the experience of other users. Any content ITC submits is only stored as needed to create translations and, if approved, train your private translation engine. Your content is never shared with third parties. ITC’s machine translation suppliers comply with government regulations like the GDPR and are ISO 27001:2013 certified. We pledge to keep your data secure, protect your privacy and uphold your trust.
Tools like ChatGPT or Copilot are families of large language model tools trained on a wide range of data without regard to copyright. Our AI-powered translation platforms DO NOT use these tools to translate content, so your translations will not infringe on any intellectual property rights. See the “What is Generative Artificial Intelligence” question below for more details.
ITC Global can provide machine translation for most language combinations. We always analyze your project to make sure it’s suitable for machine translation before we recommend this service. As part of this analysis, a trained native translator reviews a sample of machine translated text and grades the output on five criteria including fluency and ease of post-editing.
Data availability
MT systems rely on large amounts of training data to learn patterns and generate accurate translations. For languages with limited resources, collecting sufficient high-quality data is challenging. For instance, despite having 4.2 million speakers across East Africa, the Luo language has zero Wikipedia entries written in the language. This scarcity of data hinders the development of robust MT models for these languages.
Complex linguistic features
Languages like Chinese, Greek, and Arabic have unique linguistic characteristics that make it difficult for MT systems to translate accurately:
• Arabic: Complex grammar, root-based morphology, and varying forms based on context
• Chinese: Logographic writing system (characters represent words or morphemes)
• Greek: Inflectional morphology, complex verb conjugations, and rich vocabulary
Word order and sentence structure
Languages have different word orders (subject-verb-object, subject-object-verb, etc.) and sentence structures. MT systems sometimes struggle to create natural-sounding, consistent and coherent output when translating between languages with diverse structures
While providers offer MT for nearly any language pair, the quality may be insufficient to provide a productivity boost for translators.
Machine translation can work well on content designed to inform the audience, provided the post-editor is specialized in the subject matter and there is enough context. Post-machine-translation human review (post-editing) is crucial to ensure accuracy and appropriateness, particularly in fields where precision is paramount, such as medical or technical instructions.
Some examples of general content that works well with MT include corporate, IT, non-profit, government, retail, legal (agreements, court documents, etc.), medical (patient records, clinical trials), product data, marketing research reports, and technical sections of websites (if terminology isn’t specialized).
Content designed to motivate audiences or create an emotional response has features that often make it unsuitable for MT if you want to recreate the same impact on the target audience. This requires human experts familiar with the source and target culture, as well as the target industry and communication standards. It often requires adapting or completely rewriting the content.
For example, marketing standards, methods and style vary widely, even among countries and regions that use the same language (US and UK, for example). MT may produce a translation that seems acceptable, but that completely misses the mark and fails to achieve the same impact (or has a negative impact) on the target audience.
Examples include ads, slogans, taglines, headlines, subheads, brand brochures, flyers, mailings, email campaigns, website landing pages, and literary works (books, poems, songs, etc.).
Persuasive content like CEO speeches, press releases, blog content, employee newsletters and social media posts may also be unsuitable for MT, depending on your goals. If you prefer to use MT in these cases, we will test the output to see if the results satisfy your requirements.
We research and test the latest AI tools so you don’t have to, then recommend the best solution to meet your goals.
We evaluate your projects and recommend the service that best meets your specifications, so projects have the intended impact on their target audience.
We have an in-house team that manages every step of the humAIn® service, giving you the benefit of AI-assisted services guided by human expertise.
We protect your confidentiality and data security. ITC only works with artificial intelligence companies via paid subscription services. This guarantees the content we submit is not permanently stored or used to train their technology or improve the experience of other users.
Synthetic voices are used to share short messages quickly in multiple languages. It’s essential that terminology and diction match to interpret phone messages, regulatory instructions, voice messages and more. The advantages to this method? It’s faster and more affordable than a traditional recording process with a voice actor.
Our synthetic voices are generated from a written script. This technique is called Text-to-Speech (TTS). The process is simple: you send us your source text and we’ll create your message in multiple languages. TTS software creates phrases based on a language’s graphemes and phonemes. A TTS engine assigns phonetic transcriptions for each word, then a voice synthesizer converts the symbolic linguistic representation into sound.
ITC Global offers a catalog of more than 25 languages produced in realistic synthetic voices. This includes a selection of accents and pronunciations specific to each country, including Portuguese from Portugal and Brazil; English from Great Britain, the United States, Canada, or Australia; and French from France or Quebec.
Voice cloning software analyzes the speaker’s voice on your video and then uses AI voice synthesis techniques to replicate their tone, pitch, and other nuances. We can then use the cloned voice for dubbing or voiceover in your video or audio projects.
Generative Artificial Intelligence (GenAI) is a groundbreaking deep learning technology that enables machines to create original content that mimics human creativity, such as text, audio, video and images.
Large Language Models (LLMs) are one type of GenAI. Open AI’s GPT (Generative Pre-trained Transformer) has brought this technology into the spotlight by making it available to everyday internet users. These models use sophisticated algorithms to process and generate human language, drawing insights from vast datasets of information obtained from the internet and other sources.
There’s lots of hype around GenAI, but progress to achieve the necessary results for translation and localization (performance, cost efficiency) will take years and require significant time and financial investments.
Purpose and Design
LLMs are designed for general-purpose conversation and content generation, handling a wide range of topics. Without rigorous prompt engineering, they are prone to generating blatantly false information while presenting it as true.
Data Handling and Privacy Policies
GenAI tools have more general data handling and privacy policies due to their broad range of applications. Ensuring privacy can be challenging, especially if they are trained on datasets that include sensitive information.
Content Consistency and Reuse
GenAI doesn’t currently offer termbase functionality and translations are often inconsistent. In addition, LLM-based translation tools often add or remove information from translations, deviating from the source text based on the prompt. They have higher latency, working 25 to 100 times slower than neural machine translation.
GenAI introduces new approaches to content creation, translation and more, offering a wide array of benefits.
As your language service provider, we analyze AI innovations and test new tools so you can get the most out of this new technology. We recommend the approach best suited to your language service needs.
Have a project you need to finish quickly? Adding automation to the content production and translation process may be a solution. Working with a tight budget? AI can reduce translation costs and the amount of human intervention required on certain tasks.
While AI is a powerful technology, human translation is still the best choice in certain situations. We always consider factors like the language combination, the technology used, the content, and the area of specialization, as well as our clients’ end goals, when recommending AI or traditional methods.
Step into the future of language services with humAIn® by ITC Global. Contact us today to take advantage of AI-powered translation and optimize your translation process.
Speak the language of your customers, prospects, partners, and employees around the world with ITC Global’s full suite of solutions powered by our unique blend of talent and technology. Every language solution you need, from translation to AI technology.
Tailored to you. All in one place.
Curious about our machine translation technology and processes? We have lots of details for you in our resource center. Here are some suggestions to get you started.
