Leveraging detailed entity descriptions, an encoder-decoder model learns interactions between text and entity candidates to accurately disambiguate entities.
