Understanding AI in Localization: A Phrase Glossary
Artificial Intelligence is transforming the way businesses approach translation and localization, but terms like “AI,” “machine translation,” or “language models” can mean very different things depending on context.
This short glossary is designed to clarify the specific technologies and concepts we refer to across the Phrase platform, and on our blogs, product materials, and communications.
Whether it’s a neural machine translation engine, a large language model, or an automated quality assessment feature, this resource helps ensure a shared understanding of evolving AI terminology, both within Phrase, and across the wider localization industry.
1. Artificial Intelligence (AI)
AI is the development of computer systems that can perform tasks traditionally requiring human intelligence, such as understanding language, recognizing patterns, and making decisions.
In localization, AI plays a key role in automating translation workflows, improving linguistic quality, and adapting content for different audiences. Phrase leverages AI in its translation, localization and quality evaluation solutions to help businesses translate and refine multilingual content at scale.
2. Machine Translation (MT)
MT is the process of automatically translating text from one language to another using AI. It significantly reduces translation costs and speeds up content delivery. In Phrase, multiple MT engines are supported, allowing businesses to choose the best-performing engine for specific use cases.
3. Machine Translation Engine (MT Engine)
An MT engine is the software system that processes and translates text using AI models. Different MT engines have varying strengths based on the languages they support, their training data, and their underlying technology.
Phrase integrates various MT engines, such as Google Translate, DeepL, and Phrase NextMT, to provide flexibility and ensure businesses get the best translation quality based on their needs. MT Engines can also be implemented with Large Language Models such as Phrase NextGenMT or Widn.ai.
4. Neural Machine Translation (NMT)
NMT is a type of MT that uses artificial neural networks to predict the most likely translation of a sentence.
Unlike older methods, NMT considers entire sentence structures and context, resulting in more fluent and natural-sounding translations. For example, Phrase’s NextMT engine applies NMT to improve accuracy and maintain translation consistency for enterprises.
5. Large Language Models (LLMs)
LLMs are deep learning models trained on massive amounts of text data, allowing them to understand, generate, and refine human-like language. LLMs, such as GPT-4, are now being integrated into MT workflows to enhance translation fluency and contextual awareness.
In Phrase, LLMs can be used in translation (with NextGenMT) or for adaption and automated editing of content to ensure optimal quality.
6. Phrase NextMT
Phrase NextMT is a proprietary NMT engine optimized for enterprise-level localization. NextMT allows for custom tuning through CustomAI, meaning businesses can adapt the model using their industry-specific terminology and data.
7. Phrase Next GenMT
Next GenMT is an advanced MT engine built on an LLM (currently OpenAI’s ChatGPT).
Two unique features of Next GenMT are; the use of ‘few-shot’ examples from the customers Translation Memory and Term Base to provide more stylistically faithful translations, and multi-segment batching that enables the system to account for surrounding context and provide more consistent translation.
8. Translation Memory (TM)
A translation memory (TM) is a database that stores previously translated content and retrieves it for reuse in future projects. TM improves consistency and reduces translation costs, as repeated phrases don’t need to be translated from scratch.
In Phrase, TM is integrated with MT workflows, helping enterprises maintain a unified voice across different languages. Phrase also offers partial or ‘fuzzy’ match TM leverage which allows for partially relevant examples to be provided for adjustment through post-editing.
12. Quality Estimation (QE)
Quality estimation is an AI-driven process that assesses the reliability of an MT-generated translation without human intervention. QE can flag translations that need review, helping enterprises prioritize human post-editing where necessary.
In Phrase, we offer the Quality Performance Score (QPS) as a QE system which helps businesses automate quality control and streamline review workflows. QE can also be used as a signal in automated routing workflows in pre-translate steps or Orchestrator workflows.
14. Language Quality Assessment (LQA)
LQA is a systematic evaluation of translation quality based on factors such as fluency, accuracy, and consistency. The Phrase TMS includes an LQA platform that enables customers to conduct assessments following the Multidimensional Quality Metrics (MQM) framework.
15. Auto LQA
Auto LQA is an AI-powered complement to LQA that instantly evaluates translation quality based on the MQM framework. This allows businesses to identify potential translation errors in real time, ensuring high-quality output without extensive manual checks.
16. Retrieval-Augmented Generation (RAG)
RAG is an AI technique that enhances generative models by incorporating external knowledge sources, such as translation memories or glossaries, before generating a response. This improves translation accuracy by ensuring AI-generated translations align with previously approved content.
Phrase uses RAG within NextGenMT in order to leverage TMs that can ensure that MT results remain consistent in terms of style and terminology.
17. Hyperautomation in Localization
Hyperautomation involves using AI and machine learning to automate multiple steps of the localization process, from content ingestion to translation, review, and publishing. In Phrase, hyperautomation allows businesses to handle large-scale localization projects with minimal manual effort.
18. Few-Shot Prompting
Sometimes referred to as In-context learning, this is a technique by which we introduce examples to an LLM with a prompt in order to better guarantee the quality of a result.
Phrase uses few-shot prompting within NextGenMT whereby we provide examples from the customer TM to provide stylistically appropriate output.
19. Term Base
A centralized repository of approved terms and their translations used to ensure consistency and accuracy across all localized content.
20. Glossary
A curated list of specialized terms and definitions that serves as a reference for translators, helping maintain uniformity in terminology across projects. Glossaries can sometimes be used in conjunction with MT to better ensure terminology consistent output.
21. Post-editing
The process of human review and revision of machine-translated text to enhance quality, fluency, and adherence to localization standards.