memoQ Tips
Translation AI 'Lara' Now Available in memoQ 12.4.36
memoQ 12.4.36 was released on 16 July 2026. This release introduces a plugin for Lara, an AI-powered translation engine.
At present, however, Lara does not appear in memoQ’s standard list of MT plugins. This is because the Lara plugin has not yet been signed and published in the memoQ Marketplace: to use it, you need to obtain the installer from Lara (Translated) and install it manually. During installation, memoQ shows an “unsigned plugin” warning, which is normal for plugins distributed outside the memoQ Marketplace; you select “Yes” to proceed. Once installed, “Lara Translate” appears in the list of MT plugins. A signed version for the Marketplace is expected to follow.
memoQ already supports a wide range of machine translation engines. Lara is a new translation engine developed by the Italian company Translated as the successor to ModernMT, one of the engines already available in memoQ. Translated describes Lara as a large language model designed specifically for translation. Its key features include the ability to translate in the context of the entire document and to adapt in real time based on users’ translation resources and confirmed translations.
Adaptive and domain-specific AI and machine translation engines are not new. ModernMT also took account of document-level context and incorporated translators’ revisions in real time. Lara can be seen as combining these established adaptation technologies with an LLM designed specifically for translation.
Translation work in memoQ is performed segment by segment, but Lara processes the entire source document as context before generating a translation. In addition, when Self-Learning MT is enabled and a Lara translation memory is selected for updating, segments confirmed in memoQ are sent to Lara and immediately added to the connected translation memory. Subsequent translations can then take those segments into account.
This is an attractive system. However, depending on the mode selected, translation content may be stored by an external service and used to improve its models. According to Lara’s documentation, its standard Learning Mode stores source text, translations and contextual information, which may then be used to improve Lara’s models.
Translated also explains that translation memories and termbases uploaded by users are not themselves used directly as a separate source of training data for the shared model. However, translations generated in Learning Mode may reflect content from these resources, and the resulting translations may in turn be used for model improvement.
Data handling requires particular care in patent translation, which is our main area of business. Most of the work we handle relates to PCT applications. By the time we translate these applications, they have usually already been published internationally, meaning that the substance of the inventions being translated is generally already public.
Even for published applications, however, the existence of the engagement itself, client information, the scope of the material being translated, and any unpublished amendments or additions do not necessarily constitute public information. Confidentiality agreements and client-specific security requirements also apply independently of whether the application has been published.
Some work, including direct national filings, involves material that has not yet been published at the time of translation. In such cases, it is important to review the relevant client agreements and security requirements and carefully determine whether an external AI or machine translation service that may retain or learn from the translation content can be used.
Lara also offers an Incognito Mode in which translation content is neither retained nor used for model training. Existing translation memories and termbases can still be used as reference material in this mode. However, the translation request itself is not used to update the translation memory or improve the model, so confirmed translations produced during the project cannot contribute to ongoing adaptation.
When using Lara through the memoQ plugin, it is also important to confirm in advance under which settings or contractual arrangements Incognito Mode applies.
For those of us engaged in day-to-day translation work, the growing range of AI and machine translation options is a welcome development. At the same time, it is increasingly important to understand how each service works and how it handles the data we submit before deciding whether to use it.