This work package will be led by Saltlux. The consortium partners will support the WP leader in developing loosely coupled, REST-based modules for query formulation via text input and speech.
The text input module will be able to suggest search query completion via ontology and log-message-based machine learning algorithms and gazetteers. As soon as a user starts typing the already existing query segments will be used to suggest possible sensible search queries to facilitate a faster search and presenting the user with queries that the QA engine might understand. This bidirectional feedback will showcase the abilities of the QA engine to the user while the input module will learn the type of queries that are of interest to the user. Furthermore, the module will comprise an autocorrection ability to fix typos and misspellings.
Speech input is the natural evolution for search input and will be provided by existing off-the-shelf technologies that will adapted to the needs and use cases of consortium partners. Especially, an in-depth evaluation for choosing the winning speech recognition solution will be conducted. The winning module will be extended subsequently to reliably capture questions from a wide variety of users.
Furthermore, the resulting modules will reuse knowledge from the underlying structured and unstructured knowledge bases to facilitate faster recognition and supporting of the user’s input. All developed modules will be multimodal, i.e., able to work on hybrid datasets stemming from structured, unstructured and semistructured sources as well as be multilingual for three European languages (Russian, English, German) as well as Korean.