BehindTheWords

Many language technology applications require the ability to model the meaning of language statements, regardless of their exact wording. The sentences “Can you corroborate that?” and “I need proofs.” have no common words, yet their meaning is almost the same. Being able to model this relation would enable considerable improvements in document search, clustering and analysis, language generation, and other applications which deal with the meaning, rather than the surface form. While easy for a human, the computational models of paraphrase are still weak, and mostly unavailable for Finnish, preventing many advanced language technology applications in the Finnish industry. We aim to address this gap and develop the deep learning models and, in collaboration with the industry, pilot several applications. To achieve this goal, the project will also develop a unique paraphrase dataset. The project is a collaboration between the natural language processing groups in Turku and Helsinki.

2021

2024