What is a CAT tool? What does post-editing mean? There are many different technical terms used throughout the language and translation industry. In this blog post, we offer a crash course on what these key elements are, how they’re used, and what you might need to know when working on a translation project.
Computer-assisted translation (CAT) tools use software to aid or assist human translators with translation projects. By using computer processing to automate a number of the tasks involved in language translation, CAT tools can help translators work more efficiently.
How do CAT tools work? Computer-assisted translation often involves translation memories (TM), or databases that store previously translated units of the text(s) in question. Using a CAT tool to store, retrieve, and compare these units of text with new content enables a translator to identify where previous translations can be reused versus where a new translation is needed. CAT tools can be desktop-based or cloud-based.
Other computer-based tools that serve as aids in translation include terminology managers, spell-checkers and grammar-checkers, electronic dictionaries, and more.
CAT tools help translators save time, increase productivity and accuracy, and ensure consistent use of phrasing, especially for long-term or highly technical projects. From a customer’s perspective, the use of CAT tools translates into better cost-effectiveness, improved turnaround times, and easier management—for example, making it simple for editors to publish and maintain the same web content in different languages.
Here are a few examples of leading CAT tools:
- Memsource (cloud-based)
- SDL Trados Studio (desktop-based)
- Smartcat (cloud-based)
- memoQ (desktop-based)
So, are CAT tools the same thing as machine translation? This is a common misconception. In fact, machine translation (MT) uses a computer program to automatically translate text from one language to another without human involvement.
Broadly speaking, machine translation utilizes machine learning and artificial intelligence to receive text input, compare that text to a database, and produce the output of that text in the target language. More specifically, MT can take three kinds of translation approach:
- Rule-based: The machine uses language rules, grammatical standards, and lexicons to parse and translate the text.
- Statistical: The machine uses algorithms to explore a vast dataset of human translation, hunting for the most statistically likely translation of the text.
- Neural: The machine’s artificial intelligence and neural networks “learn” as it translates, predicting and generating the best sequence of words to produce.
Most modern machine translation engines combine one or more of these approaches. Current leading MT engines, all of which use neural translation, include big tech products such as Google Translate, Bing Translate, and Amazon Translate, along with well-established entities such as DeepL and SYSTRAN. At Trusted Translations, we carefully assess each translation project to determine the most suitable engine on a case-by-case basis.
A translator might use machine translation for an initial output (like a first draft) in several situations: for instance, at a live event that requires near-instantaneous results; to speed up processes like subtitling; or to start on extremely large projects. However, machine translation should not be used as a standalone service. Although translation customers may be drawn to its potential for reduced costs, increased speed, and access to a diverse array of world languages, machine translations should ideally always be edited by humans to ensure quality and accuracy.
In post-editing, human translators edit machine translations after they have been generated, with the goal of creating usable final translations. Translators will use CAT tools to compare the machine translation to the source text and assess its quality. Then, they will either amend the machine translation, leave it as is, or retranslate it from scratch.
Different levels of post-editing allow translation companies like Trusted Translations to tailor services to a client’s needs:
- Light and fast post-editing: The editor will make minimal edits focused on critical fixes (such as blatant semantic, grammatical, and spelling errors) and basic clarity. This prioritizes cost and speed, or can be used if the machine translation is already decent.
- Full or ready-to-publish post-editing: Full post-editing brings the text up to the level of a fully human-translated project, making extensive edits to polish stylistic details, consistency, and tone. This level takes longer, but produces the highest quality results.
Regardless of the level chosen, post-editing is an absolutely necessary step in a translation process that employs machine translation (MT). Machine translation often produces clunky phrases, incorrect or inconsistent terminology, and unnatural-sounding outputs that must be reviewed by real translators, especially if tone, emotion, and context are crucial to comprehension. Translation customers should engage reputable human post-editors that can guarantee translations with consistent quality, without machine-generated errors that could be extremely costly, embarrassing, or time-wasting in the long run.
Photo by Miguel Á. Padriñán at pexels.com