The Linguist

Chartered Institute of Linguists WINTER 2024 The Linguist 31 SECTION HEADER OPINION & COMMENT Finding the right translation tool for each project ANNA RIOLAND AI translation systems have become a major asset in the toolkit of professional translators, but keeping up with them can be a challenge. From AI-based machine translation (MT) engines such as DeepL to AI-enhanced CAT (computer-assisted translation) platforms like Trados and XTM, the options are vast. However, these tools come with limitations, and their usefulness depends on the context. Among the many available AI-powered MT tools, DeepL enables fast first drafts, which can be a lifesaver for large projects with tight deadlines. Such programs are commonly used to create initial translations, offering a starting point that can be refined by the translator. Several post-editing machine translation (PEMT) tools utilise AI to enhance productivity, speed and accuracy. Trados integrates AI- driven MT engines, allowing users to leverage AI for drafts that can be post-edited. It has also introduced its Trados Copilot – AI Assistant, adding an editing prompt for batch-translating and post-editing, as well as automatically applying terminology. memoQ leverages AI through integration with MT engines and its Linguistic Quality Assurance (LQA), which uses AI to check terminology, grammar and consistency across translations. Adaptive MT ModernMT considers the whole document context, producing more accurate translations based on the surrounding content. Its Human- in-the-Loop product adapts in real time to a translator's changes, enhancing the quality of future suggestions. As you translate, it learns and becomes more accurate in predicting style and terminology, which is particularly beneficial for freelancers who work on varied projects and need an adaptable tool. Lilt combines MT, translation memories and term bases in one adaptive platform that learns in real time from translators' input. Unlike traditional post-editing software, its interactive MT autocompletes whole sentences based on the user's phrasing, enhancing accuracy and saving time. Its MT continuously learns from human corrections to avoid repeating errors across segments, without the need for periodic retraining. Enhanced quality control XTM incorporates advanced MT options such as AI-powered Translation Quality Assurance (TQA), which identifies inconsistencies and errors in translation such as grammar issues and missing terminology. It can even identify non-inclusive or discriminatory language, and suggest improvements. The In-Browser Translator Workbench is fully browser-based, so freelancers don't need to install additional software and can work from any device, which makes it mobile and accessible. Smartcat integrates multiple AI-based MT engines and offers automated quality assurance for consistency and error-checking, while Smartling offers an AI-powered feature called Quality Confidence Score (QSC), which assesses the overall quality of a translation with recommendations for improvement. Smartling's AI Toolkit offers multiple AI- powered features. Its glossary terms insertion function automatically adapts the inflections of glossary terms to the surrounding sentence structure. Its AI Fuzzy Match repair allows users to reuse their existing translations to achieve a high level of consistency, while its AI Formality Adjustments enable users to adjust machine translations from formal to informal register. Where AI falls short AI translation tools often fail when it comes to translating creative content or culturally nuanced text. They tend to produce overly literal translations that can miss the tone of voice or subtleties of the source text, which is particularly problematic in fields like marketing and literature. This, of course, represents a huge opportunity; one of my favourite recent projects was writing a collection of poems and jokes for an AI-powered voice assistant. While customisation is possible, AI systems still struggle with highly specialised and technical terminology, especially in niche fields and marginalised languages. Translators working in legal, medical and scientific fields often find that the machine-generated output lacks precision, requiring thorough post- editing. However, by leveraging translation memories, customising AI models, choosing the right program/platform for each job, and using AI for drafts, we can make the most of these tools while ensuring the final product has that irreplaceable human touch. The AI toolbox Anna Rioland MCIL CL is a localisation expert and multilingual project management professional with a passion for languages and technology. TL © SHUTTERSTOCK | AI-GENERATED IMAGE

• Cover