NATURAL LANGUAGE PROCESSING

Language is the most natural and most dominant mode of communication and, arguably, one of the main visible signals of higher intelligence. At the same time, language is messy, ambiguous, multimodal and ever-changing so to decipher it you need a good amount of common-sense, contextual and cultural understanding. To fulfil our vision of seamlessly communicating with intelligent devices, existing technology and the methods used to solve natural language processing problems need to be considerably improved.
That is precisely what we do:

• As a European lab of a Korean company, we’re distinctly aware of how real the language barrier can be. We improve the current state-of-the-art in multilingual applications and machine translation, trying to find optimal tradeoffs between efficiency and performance.
• In addition, while natural language generation (NLG) models have recently progressed to a point where they can produce highly fluent texts, they can be deficient on other important levels (producing toxic or socially biased content for instance) so we augment them with explicit controls.
• As far as natural language understanding (NLU) is concerned, we address the challenge of capturing meaning beyond memorising surface patterns and co-occurrences. Our work on this topic applies to document understanding, fine-grained information extraction and spoken language understanding.
• Method-wise, we’re particularly interested in how to combine the power and flexibility of deep neural networks with the rich prior knowledge present in decades of linguistic studies and knowledge of the task at hand. We also investigate how models can continuously and adaptively learn in order to incrementally acquire increasingly more complex skills and knowledge.