Portuguese AI translation SME combining neural machine translation with crowd post-editing for scalable multilingual content.
Unbabel is a Portuguese technology SME that develops AI-powered translation services, combining machine translation with human post-editing to deliver scalable multilingual content. Their core technology bridges statistical and neural machine translation with crowdsourced quality assurance, targeting businesses that need affordable, high-quality localization at scale. Within H2020, they contributed expertise in natural language processing (NLP), neural networks, and distributed crowd workflows for translation tasks.
DeepSPIN focused on structured prediction in NLP with explicit keywords in statistical machine translation and neural networks; the Unbabel SME project centered on scalable content globalization.
The Unbabel SME-1 project specifically addressed distributed crowd-post-editing for seamless content globalization.
INTERACT built an international network on crisis translation, applying Unbabel's translation capabilities to emergency and disaster communication scenarios.
DeepSPIN (2018-2023) explored deep learning methods for structured prediction in natural language processing, representing a shift toward more advanced neural approaches.
Unbabel's H2020 trajectory shows a clear progression from applied commercial translation toward deeper research in AI-driven language processing. Their earliest project (2015) focused on the practical business challenge of scalable crowd-post-editing for globalization. By 2017-2018, they moved into more research-intensive territory — crisis translation networks (INTERACT) and fundamental deep learning research for NLP (DeepSPIN, an ERC Starting Grant). This shift from SME product validation to participation in foundational AI research signals growing scientific ambition alongside their commercial translation platform.
Unbabel is moving from applied translation services toward foundational AI research in neural language processing, making them increasingly relevant for NLP and deep learning collaborations.
Unbabel operates primarily as a participant rather than a consortium leader — they coordinated their own SME instrument project (self-led by nature) but joined INTERACT and DeepSPIN as partners. With only 8 unique partners across 6 countries, their network is modest but geographically diverse. Their participation pattern suggests they contribute specialized NLP and translation technology to research-driven consortia rather than leading large collaborative efforts.
Unbabel has worked with 8 partners across 6 countries, indicating a small but internationally spread network. Their collaborations span both academic research groups (DeepSPIN is an ERC grant) and applied crisis communication networks, giving them connections in both fundamental NLP research and real-world translation applications.
Unbabel sits at the intersection of commercial AI translation services and academic NLP research — a rare combination among SMEs. Their involvement in an ERC Starting Grant (DeepSPIN) signals that leading researchers trust their platform and expertise enough to include them in frontier deep learning work. For consortium builders, they offer both a working translation technology stack and genuine research depth in neural machine translation, not just one or the other.
