Context Token
Context tokens are supplemental information added to model inputs to improve performance in various natural language processing (NLP) tasks, such as machine translation and code vulnerability detection. Current research focuses on optimizing the design and integration of these tokens, exploring their impact on model architectures like transformers and neural transducers, and investigating methods for effectively encoding extra-sentential information (e.g., speaker turn, scene type). This work aims to enhance model capabilities by leveraging contextual information, leading to improved accuracy and efficiency in diverse applications, including code analysis, dialogue translation, and long-form speech recognition.
Papers
September 11, 2024
November 20, 2023
November 9, 2023
November 17, 2022
October 13, 2022