Introducing Neural Fragment Recall: A Smarter Way to Work
3 mins read
At Trados, we’re committed to driving innovation that empowers translators and reviewers to work easier, faster and smarter. With that goal in mind, we’re thrilled to introduce Neural Fragment Recall, the new AI-enabled generation of our existing upLIFT Fragment Recall feature. Powered by AI-based neural alignment, this new capability elevates translation efficiency by delivering more accurate and consistent fragment suggestions in real-time. Neural Fragment Recall is now uniquely available in both our desktop and online editors when working with cloud translation engines.
The power and limitations of upLIFT Fragment Recall
For the last decade, upLIFT Fragment Recall has been an essential tool in our editing environments, making translation memories smarter by detecting and aligning parts of source and target sentences, then suggesting them as fragments during translation. Without this feature, linguists would have to manually search for and copy previously translated phrases into the editor - a time-consuming process.
upLIFT Fragment Recall is a powerful tool that uses a statistical approach for word and phrase alignment. However, this technique lacks a deeper understanding of how words relate to each other within a shared semantic context, which can result in missed connections between source and target words. Additionally, upLIFT Fragment Recall requires relatively large translation memories - typically between 1,000 and 5,000 translation units (TUs) - to deliver accurate and reliable results.
Enhance efficiency with Neural Fragment Recall
The new AI-powered Neural Fragment Recall improves on the original approach by leveraging a multilingual model pretrained on vast amounts of text. By analyzing text at the subword level, it can effectively process new documents even if they contain words the model has never encountered before.
Neural Fragment Recall automatically aligns source and target words or phrases when translation units are added to a TM. The alignment data is then stored, enabling fast, consistent fragment retrieval and ensuring translations stay current with TM updates.
Here’s how it can improve your translation processes:
- Real-time neural alignment: It automatically aligns words and phrases saved to your translation memory, ensuring that newly added fragments are quickly available for recall.
- No minimum TM size required: Unlike its predecessor, this feature doesn’t depend on translation memories having a specific number of translation units to get started. Users can get up and running with Neural Fragment recall immediately, unlocking the full potential of their TMs from the very first segment they translate.
- Improved consistency: By relying on neural alignment instead of traditional statistical matching, it delivers more accurate and consistent results every time.
- Ideal for less-repetitive content: Particularly useful for legal and public sector content, where maintaining consistency is critical, even when segment-level repetition is low.
Image of Neural Fragment Recall in the online editor. The TM attached to the project contains two saved translation units (segments 6 and 7). Previously, upLIFT Fragment Recall would provide no suggestions due to an insufficient number of TUs available. With Neural Fragment Recall, only one TU is needed to start aligning fragments and generating suggestions.
Currently, Neural Fragment Recall is available for three language pairs - English <> German, English <> French, and English <> Spanish - with additional languages to be added based on user feedback and demand.
With fragment matches delivered via AutoSuggest in both the desktop and online editor, translators can quickly leverage results as they work, boosting efficiency and reducing project turnaround times.
If you’d like to discover what else is new in Trados, read our blogs on the RWS Community.
Nicole Loney
All from Nicole LoneyRelated Articles
