The Automatic Content Extraction (ACE) Program Tasks, Data, and Evaluation

The Automatic Content Extraction (ACE) Program Tasks, Data, and Evaluation

September 2004 | George Doddington@NIST, Alexis Mitchell@LDC, Mark Przybocki@NIST, Lance Ramshaw@BBN, Stephanie Strassel@LDC, Ralph Weischedel@BBN
The Automatic Content Extraction (ACE) program aims to develop technology for automatically inferring entities, relations, and events from human language data, including text, audio, and image sources, in multiple languages such as English, Arabic, and Chinese. The program began in 1999 with a pilot study and is scheduled for its next evaluation in September 2004. The core tasks include entity detection and tracking (EDT), relation detection and characterization (RDC), and event detection and characterization (VDC). Annotators at the Linguistic Data Consortium (LDC) at the University of Pennsylvania develop annotation guidelines and corpora to support these tasks. The evaluation process involves scoring entities, relations, and events based on their detection and recognition accuracy, with a focus on maximizing the value of correctly identified objects and minimizing the value of errors. The scoring methods are detailed, considering factors such as entity type, relation type, and event type, and accounting for cross-document consistency.The Automatic Content Extraction (ACE) program aims to develop technology for automatically inferring entities, relations, and events from human language data, including text, audio, and image sources, in multiple languages such as English, Arabic, and Chinese. The program began in 1999 with a pilot study and is scheduled for its next evaluation in September 2004. The core tasks include entity detection and tracking (EDT), relation detection and characterization (RDC), and event detection and characterization (VDC). Annotators at the Linguistic Data Consortium (LDC) at the University of Pennsylvania develop annotation guidelines and corpora to support these tasks. The evaluation process involves scoring entities, relations, and events based on their detection and recognition accuracy, with a focus on maximizing the value of correctly identified objects and minimizing the value of errors. The scoring methods are detailed, considering factors such as entity type, relation type, and event type, and accounting for cross-document consistency.
Reach us at info@study.space