Translation LLM now supports full finetuning with LoRA.

The latency and quality of Translation LLM has been improved.

Change

Translation LLM now supports even more languages.
See our documentation for a list of all supported languages.

Translation LLM now supports translation between any of the supported source/target languages without English having to be either source or target.

Glossary now supports Translation-LLM and Adaptive Translation. In GlossaryConfig settings, change contextual_translation_enabled to True in order to apply LLM enhanced glossary.

Change

The translation LLM now supports Polish, Turkish, Indonesian, Dutch, Vietnamese, Thai and Czech. For the full list of supported languages, see the Translate text page.

The translation LLM and adaptive translation now support Arabic, Hindi, and Russian. For the full list of supported languages, see Supported languages.

The translation LLM has been enhanced with the following changes:

• In addition to plain text, you can send HTML as input for text translations.
• When you use the translation LLM for text translations, you can use a glossary with your translation requests.

Change

Adaptive translations with reference sentence pairs support a larger context window, on par with zero-shot translations. For example, when sending a paragraph for translation, Cloud Translation translates the whole paragraph at once instead of translating each sentence one at a time.

For adaptive translations, when you use the API, you can include up to five reference sentence pairs in a request instead of specifying a dataset.

You can request document translations with multi-regional endpoints.

Feature

For AutoML datasets, you can tag segment pairs when importing them through the Google Cloud console.

Adaptive translation is Generally Available and adds Portuguese support, raises the limit for input and output characters, and decreases latency in the API and console.

Adaptive translation, which leverages Google LLMs to tailor translations, is in Preview

Romanization and transliteration are now in Preview.

Creating and managing datasets and custom models is now supported by the Cloud Translation API. All future custom model enhancements will be applied to the Cloud Translation API instead of the AutoML API, which will continue to be maintained.

You can upgrade your existing AutoML resources to use the Cloud Translation API and the Cloud Console without any additional costs. To work with upgraded resources, API users must update existing AutoML API calls to use the Cloud Translation API. For more information, see Upgrade AutoML resources.

Added new fields to the document translation methods to handle translation and parsing issues:

• Shadow text removal for overlapping text (for native PDFs).
• Auto rotate document to the correct orientation (for scanned PDFs).

For document translations, added support for Microsoft DOC, PPT, and XLS files. For more information, see Supported formats.

Glossaries are now supported for the recently added 24 languages.

• Assamese
• Aymara
• Bambara
• Bhojpuri
• Dhivehi
• Dogri
• Ewe
• Guarani
• Ilocano
• Konkani
• Krio
• Kurdish(Sorani)
• Lingala
• Luganda
• Maithili
• Meiteilon(Manipuri)
• Mizo
• Oromo
• Quechua
• Sanskrit
• Sepedi(Pedi)
• Tigrinya
• Tsonga
• Twi (Akan)

For online document translations, you can increase the page limit for native PDF documents to 300 pages.

Support for 24 new languages is Generally Available (GA). Glossaries aren't supported when translating to or from these languages.

• Assamese
• Aymara
• Bambara
• Bhojpuri
• Dhivehi
• Dogri
• Ewe
• Guarani
• Ilocano
• Konkani
• Krio
• Kurdish(Sorani)
• Lingala
• Luganda
• Maithili
• Meiteilon(Manipuri)
• Mizo
• Oromo
• Quechua
• Sanskrit
• Sepedi(Pedi)
• Tigrinya
• Tsonga
• Twi (Akan)

Cloud Translation - Advanced (v3) support for a multi-regional US endpoint is now Generally Available (GA).

For Cloud Translation - Advanced (v3) glossaries, you can now manage glossary entries. This feature is Generally Available (GA).

Cloud Translation - Advanced (v3) support for a multi-regional EU endpoint is now Generally Available (GA).

Document Translation for Cloud Translation - Advanced (v3) is now Generally Available (GA) and includes the following enhancements:

• Right to left language support for PDFs
• Preserves font size, font color, font style, and hyperlinks for native PDFs only
• Batch document translation requests support PDF to DOCX conversions for native PDFs only

Removed the Phrase-Based Machine Translation (PBMT) model. For requests that specify the PBMT model, Cloud Translation uses the Neural Machine Translation (NMT) model instead.

Cloud Translation - Advanced (v3) support for a regional EU endpoint is now in Preview. For more information, see Specify a regional endpoint.

Document Translation for Cloud Translation - Advanced (v3) is now available in Preview. Document Translation supports the DOCX, PPTX, XLSX, and PDF file formats. For more information, see Translate documents.

The Cloud Translation Advanced API is migrating to a new quota system that eliminates the distinction between default and maximum limits.

This change will roll out gradually, starting on March 30, 2020 and continuing through April 30, 2020. Since quotas define the usage limits for a project or a user, the change could impact your billing.

Cloud Translation v3, rebranded as Translation API - Advanced, is now generally available. In addition to the features of Cloud Translation API v3beta1, Translation API - Advanced supports: * User labels * Audit logging * Separate quotas for batch translation and for translation using AutoML models * Client libraries for additional languages PHP, GO, C#, and Ruby

Note: Existing users should pay careful attention to client library versions before updating. Translation API - Basic and Translation API - Advanced client libraries are not backward or forward compatible.

Translation API - Advanced also fixes bugs from the v3beta1 version, notably for the glossary feature.

Change

Translation API - Advanced has a revised pricing structure. Each account receives $10 worth of free usage per month, applied to usage of any of these SKUs:

• Neural Translation Model Predictions (D90A-CFB2-7CCD)
• Neural Translation Model Predictions In Translation V3 (E205-31DB-F1F4)
• Phrase-Based Translation Model Predictions (53BA-5E1D-4314)

This credit replaces the free tier for Text Translation requests using NMT for 0-500k characters on Translation API v3beta1.

Cloud Translation API v3beta1 is now available. This release includes support for AutoML models when translating text. In addition, the release includes the following new features:

• custom glossaries for customer-specific terminology,
• batch translation for asynchronous translation of .txt, .tsv, and .html files saved in Google Cloud Storage.

v3 API offers monthly free tier, and guarded by new Quotas and Limit. Learn more on how to use new features and client library in v3.

The maximum values for Characters per 100 seconds quotas have increased. The new maximum for both Characters per 100 seconds per project and Characters per 100 seconds per project per user is 10,000,000 characters. The default quotas remain unchanged.

The recently updated default Characters per day quota has been adjusted again; it is now 1 billion characters.

See Quotas & Limits.

The change to the default Characters per day quota has taken effect. It is now unlimited. See Quotas & Limits.

On August 15, the default Characters per day quota will be updated to unlimited. For more information, see Quotas & Limits.