Skip to content

Artificial Intelligence and Machine Translation

Artificial Intelligence and Machine Translation

Leading organizations around the world, and from all major industries, are beginning to see the value in state-of-the-art neural machine translation (NMT). GPI NMT solutions, within the right context, can allow companies to complete translation projects in less time with lower costs. Our high-quality NMT solution can be an effective tool for rapid translations.


Artificial Intelligence and Machine Translation


GPI offers AI-based machine translation options where appropriate, combining neural machine translation (NMT) and using certain generative AI technologies for post-editing and additional quality control. It is important to remember, that NMT is a very specific linguistic application of AI which automates to a degree language translation. AI in general, includes a much broader spectrum of approaches and technologies with the aim of simulating human intelligence in a range of domains beyond translation such as robotics, computer vision, and speech recognition, to name a few.

Advances in machine translations have evolved quickly and the language industry has used three basic machine translation technologies in the last few decades.

“Rule-based MT” which is an early form of machine translation that required the manual addition of languages and human post-editing with predefined sets of rules to help determine and transfer meaning between languages. “Statistical-based MT” which creates a statistical model of relationships between words and phrases in the text and then applies it to another language than the source.  And “Neural-based MT” that mimics the neural networks of the brain and encodes and decodes source text for translations.


What is Neural Machine Translation (NMT)?

NMT is the latest approach in a long line of machine translation processes dating back to the 1990s. The goal of NMT is to replicate the human brain’s ability to learn through trial and error as well as to adapt to new experiences. NMT uses a “deep learning” process where artificial neural networks predict the probability of a sequence of words. Building upon previous iterations of machine translation (statistical and rule-based), NMT provides a whole new level of quality. Coupled with a human review, and the proper front-end preparation, our NMT solution produces improved translation


What is Neural Machine Translation (NMT)?



The GPI NMT Methodology

GPI’s primary approach to Neural Machine Translation (NMT) is first and foremost, to ensure that it is a good fit for our client’s content. Based on decades of MT, NMT, and newer AI language translation applications, GPI clientele can rest assured our NMT solutions will introduce additional savings in time and money and deliver acceptable quality.


Basic Steps in Neural Machine Translation Projects:

  1. Conduct an Initial Quality Assessment (IQA) of your content to determine the suitability for using an NMT solution and the amount of human post-editing required
    • While the quality of NMT has greatly improved, it is still not for every use case, which is something we will determine early on. Internal communications might be an ideal fit, but client-facing ad copy or highly technical content likely is not
  1. As part of the testing phase, we will build a custom NMT engine per language pairs required
    • These engines need training with your content and domain
  1. To achieve the best results, it is ideal for us to train the engine with any previous translation memory assets or multilingual glossaries that are available
    • These steps will only increase the eventual output quality. If previous materials are not available, we test with generic NMT engines
  1. After engines are built, we will run your content through the machine translation tool and then re-evaluate the larger set of content (greater than the IQA) and estimate the amount of human post-editing that will be required for predefined quality levels
    • This process gives valuable information that will help to plan translation projects using NMT technology. Quality and feasibility levels can differ between different types of content and language combinations – no singular solution exists
  1. After your content is input through the machine translation tool, it produces a raw translation output. Then our qualified human translation teams will complete Machine Translation Post-Editing (MTPE) to post-edit the translation
    • MTPE is a requirement and may involve retranslation, infusion of SEO keywords, grammar checks, editing for appropriate style as well as cultural correctness review, to name a few of the linguistic tasks involved


GPI has used a range of NMT solutions including but not limited to:

  • Google Translate
  • Microsoft Translator
  • Google Cloud Translation API
  • Unbabel
  • Azure Translator Text API
  • Amazon Translate
  • Language Weaver
  • DeepL
  • IBM Watson NLP
  • Bing Translator
  • Katan MT


Reach out to GPI’s team of language and technology professionals today to see if an NMT solution is right for you and your content!


To learn more, request a demo of any of GPI's translation tools.

Request a demo