Entity Linking is crucial for numerous downstream tasks, such as question answering, knowledge graph population, and general knowledge extraction. A frequently overlooked aspect of entity linking is the potential encounter with entities not yet present in a target knowledge graph. Although some recent studies have addressed this issue, they primarily utilize full-text knowledge bases or depend on external information such as crawled webpages. Full-text knowledge bases are not available in all domains and using external information is connected to increased effort. However, these resources are not available in most use cases. In this work, we solely rely on the information within a knowledge graph and assume no external information is accessible.

To investigate the challenge of identifying and disambiguating entities absent from the knowledge graph, we introduce a comprehensive silver-standard benchmark dataset that covers texts from 1999 to 2022. Based on our novel dataset, we develop an approach using pre-trained language models and knowledge graph embeddings without the need for a parallel full-text corpus. Moreover, by assessing the influence of knowledge graph embeddings on the given task, we show that implementing a sequential entity linking approach, which considers the whole sentence, can outperform clustering techniques that handle each mention separately in specific instances.