User:Taeyasu/sandbox
Translated
[edit]Translated is a provider of translation, localization, and language services headquartered in Rome, with offices in the United States, France, and Germany. Founded in 1999, it was one of first online translation agencies and one of the first to use artificial intelligence (AI) to support professional translators. The company has been recognized as a leading provider of AI-enabled translation services[1] and a global leader in enterprise translation AI.[2][3]
Translated offers services in 201 languages and 40 areas of expertise, serving a client base that includes multinational corporations and small enterprises. In 2021, the company signed a significant deal with Airbnb, described as the largest-ever private deal in the translation industry.[4]
History
[edit]Translated was founded in 1999 by linguist Isabelle Andrieu and computer scientist Marco Trombetti with the ambition to allow everyone to understand one another in their native language by combining professional translators and machine translation (MT).
CEO Trombetti spearheaded investments in research and development, focusing on tools to foster a symbiotic relationship between professional translators and AI. The most significant projects coming out of this were Matecat, which led to the release of the first computer-assisted translation (CAT) tool that integrated adaptive machine translation[5], and one of the first neural adaptive machine translation systems, ModernMT.[6] In 2018, ModernMT was one the first commercial applications using the Transformer deep learning architecture[7], which was introduced to the public few months earlier via the landmark research paper, “Attention is all you need,” and later became foundational in generative AI.
In 2021, Translated secured a $30 million investment led by Ardian, with contributions from an undisclosed number of business angels.[8]
In 2024, the company, leveraging the technologies behind Matecat and ModernMT, developed Lara, claimed to be the first AI-powered translation architecture capable of evaluating options and providing the reasoning behind its translation choices.[9] The company state on the tool website that the translation model was trained with the corrective feedback provided by professional translators working with ModernMT in Matecat plus the discussions between translators and reviewers when disagreements arose during the revision process.
Research projects
[edit]Translated also frequently collaborates with universities and research centers to develop innovative AI solutions in the field of translation. The company has participated in several research projects funded by the EU Commission, including Matecat, ModernMT[10], and Meetween.[11]
Translated is continually measuring the progress towards singularity in translation – when machine will be better than humans at translating – quantifying the progress using Matecat data.[12]
In many cases, such as with Matecat, Adaptive MT, and ModernMT, Translated has successfully brought research to market by releasing commercial products based on its research.
Tools for symbiosis between translators and AI
[edit]Matecat
[edit]Matecat is a web-based, open-source CAT tool designed to enhance translation speed through real-time learning, contextual text recognition, and access to a database of shared machine translations and glossaries that were revised by translators. It was the first CAT tool to integrate Adaptive MT.[13] In 2015, the European Commission recognized MateCat as one of the top AI research projects of the preceding seven years.[14]
Matesub
[edit]Matesub is a web-based, AI-powered subtitling tool designed to streamline the subtitle creation process by automating tasks such as subtitles positioning and integrating machine translation in over 200 languages. The tool allows users to upload videos and automatically generates subtitles, which can then be edited for accuracy. The system is designed to improve its performance over time by learning from user edits.
The tool leverages machine learning and data analysis to automate processes such as project creation, tag positioning, and post-editing, aiming to enhance efficiency and reduce manual effort.
Matedub
[edit]Matedub is an AI-powered dubbing software designed to facilitate the creation of AI-generated voice-over projects. It enables users to edit AI-generated translations and produce voice-overs using AI-generated voices. The tool was released in a private beta version in 2020. It supports multiple languages and is used for projects like film and video game localization. The software integrates machine learning algorithms to improve the quality of its translations and voice synthesis over time, adapting to the specific needs of each project.
MyMemory
[edit]MyMemory is a collaborative translation memory project lead by Translated, claimed to be the world’s largest free translation memory. It contains over 8 billion translated segments[15] sourced from professional translators, language service providers, customers, and multilingual web content.
The tool provides both machine and human translations and uses an algorithm to deliver relevant matches for a given source text. Translators can download translation memories for use with various computer-assisted translation (CAT) tools and contribute to the project by sharing their own translation memories or editing existing entries.
Translated states that MyMemory was created by aligning multilingual texts from sources such as the European Union and United Nations, as well as domain-specific websites.
Translation technology
[edit]ModernMT
[edit]In early 2018, Translated launched ModernMT[16], an adaptive neural machine translation system developed to provide context-aware translations for enterprises and professional translators. The platform employs a neural adaptive approach, allowing the system to learn continuously from user inputs and align translations with specific terminology and contextual nuances. It supports over 200 languages and is compatible with various computer-assisted translation (CAT) tools.
ModernMT was created by Translated and Fondazione Bruno Kessler in 2017 and merged into Translated in 2022. The initial research was led by Translated, Fondazione Bruno Kessler, and a consortium that included The University of Edinburgh and Taus. After the research project ended, Translated and Fondazione Bruno Kessler joined together to establish a company to spread the commercial use of the technology.
In 2019, Gartner named ModernMT a "Cool Vendor" in Speech and Natural Language. In 2023, CSA Research recognized it as "the most advanced implementation of responsive machine translation".[17][18]
Lara
[edit]Lara is a translator based on the Transformer architecture and specifically trained for translation. Created by Translated, it debuted in November 2024. Lara was trained on a collection of documents translated by professional translators with the initial support of machine translation, and revised by professional translators selected for the specific task. Lara includes a reasoning system to evaluate translation options and explain the reasoning behind its choices.
References
[edit]- ^ "Gartner: Translated Leading Advances in AI in the Translation Industry | Business Wire". web.archive.org. 2022-07-15. Retrieved 2024-12-31.
- ^ https://www.idc.com/getdoc.jsp?containerId=US48325622.
{{cite web}}: Missing or empty|title=(help) - ^ https://translated.idc-custom.com/marketscape/us48325622/.
{{cite web}}: Missing or empty|title=(help) - ^ "Airbnb awards largest contract in language industry | MultiLingual". multilingual.com. 2021-11-09. Retrieved 2024-12-31.
- ^ https://aclanthology.org/www.mt-archive.info/10/AMTA-2014-Federico.pdf.
{{cite web}}: Missing or empty|title=(help) - ^ modernmt/modernmt, ModernMT, 2024-10-31, retrieved 2024-12-31
- ^ "Adding Transformer/CNN model · Issue #377 · modernmt/modernmt". GitHub. Retrieved 2024-12-31.
- ^ "Translated Receives $25M Investment From Ardian | Business Wire". web.archive.org. 2021-06-30. Retrieved 2024-12-31.
- ^ Abirascid, Emil (2024-11-05). "Meet Lara, Translated's artificial intelligence for translations". Startupbusiness.it. Retrieved 2024-12-31.
- ^ "MMT will deliver a language independent commercial online translation service based on a new open-source machine translation distributed architecture | MMT Project | Fact Sheet | H2020". CORDIS | European Commission. Retrieved 2024-12-31.
- ^ "My Personal AI Mediator for Virtual MEETtings BetWEEN People | Meetween Project | Fact Sheet | HORIZON". CORDIS | European Commission. Retrieved 2024-12-31.
- ^ "Humanity May Reach Singularity Within Just 6 Years, Trend Shows". Popular Mechanics. 2024-11-30. Retrieved 2025-01-01.
- ^ "Translated Introduces Free Adaptive MT in Matecat to Increase Linguists' Productivity | MultiLingual". multilingual.com. 2024-05-07. Retrieved 2025-01-01.
- ^ "EUROPEAN PARLIAMENT: THE ARTIFICIAL INTELLIGENCE OF TRANSLATED, FBK AND PERVOICE FOR REAL-TIME TRANSLATION OF DEBATES". FBK Magazine. Retrieved 2025-01-01.
- ^ "About MyMemory - A collaborative language resource". mymemory.translated.net. Retrieved 2025-01-01.
- ^ https://aclanthology.org/2018.eamt-main.46.pdf.
{{cite web}}: Missing or empty|title=(help) - ^ https://csa-research.com/ModernMT.
{{cite web}}: Missing or empty|title=(help) - ^ Translated (2023-02-03). "Translated's MT Recognized by CSA Research as the Most Advanced Responsive MT". Slator. Retrieved 2025-01-01.