Event
Extraction and Synthesis 
A workshop in conjunction with AAAI-06, July 16th 2006, Boston, MA
Description
The focus of this workshop is on the extraction and synthesis of events from raw information in text or other modalities. Event extraction from raw input is of significant interest for a variety of applications such as situational awareness systems, intelligence analysis etc. Consider the following sentence from a Voice of America news report: “The United Nations says Somali gunmen who hijacked a U.N.-chartered vessel carrying food aid for tsunami victims have released the ship after holding it for more than two months.” This sentence expresses multiple events--a focal event (the release of a hijacked ship), and several contextual events providing a temporal and semantic framework for it. Not only is an adequate accounting of this structure challenging, but the information contained in the event is often partial, subject to evolution over time, conflicting reports from other sources, and the question of trust in the source of the information. Solving a task of this complexity appears to call for ideas from multiple disciplines, such as machine learning, natural language understanding, knowledge representation, data management, and linguistic theory related to language semantics. It is ultimately not enough to label text spans; accounting for uncertainty, both that attached to a whole report, and any expressed in the report itself, requires measures of certainty and techniques for manipulating them. Such measures are a common subject in related disciplines (e.g., "information quality" in the context of multi-agent systems). Also, while work on event extraction has mostly focused on text, events are certainly embedded in information streams in other modalities as well, such as audio and video. The problem of corroborating uncertain information argues for approaches that span modalities.
The goals of this workshop are the following:
1. Obtain a clear understanding of the new challenges posed by event-oriented information extraction vs. work done earlier in relation or entity extraction.
2. Discuss approaches and techniques, including combinations of techniques from different disciplines to perform efficient event extraction.
3. Identify unifying concepts valid for multiple modalities.
4. Grapple with questions of uncertainty and reliability, in an attempt to promote the use of uncertainty measures in extraction systems.
Topics of
Interest
We are particularly interested in position, vision, or research papers, and system demonstrations outlining or addressing challenges in the extraction of event oriented information. Topics of Interest include, but are not limited to:
- Event definition, modeling, and representation
- The role of semantics in event extraction
- Combining multiple techniques (machine learning, ontologies, NLP, structured machine learning, etc.,) for event extraction
- Audio, visual, and audio-visual event extraction
- Temporal aspects and evolution of extracted information
- Extracted information quality and pedigree (trustworthiness, uncertainty)
- Event identity and disambiguation, or event co-reference
- Evaluating event extraction systems
Workshop Format
We anticipate a day long workshop that will comprise of paper presentations, and a systems demonstration and poster session.
Schedule and
Deadlines
Paper submission: March 31, 2006
Notification: April 24, 2006
Papers should be no more than6 pages in length and should conform to the format
Please
email inquiries andsubmissions to events2006[at]ics [dot] uci
[dot]edu
Invited Speaker
Dr. Ralph Weischedel, BBN Technologies
Topic: TBA
Organizers
Doug Appelt SRI
Naveen Ashish UC Irvine
Dayne Freitag Fair Isaac
Dmitry Zelenko SRA
Demonstrations Chair
Fabio
Ciravegna
Program Committee
Eugene Agichtein, Microsoft Research
Chinatsu Aone, SRA
George Doddington, NIST
Kamal Nigam, Intelliseek
Steffen Staab, University of Koblenz-Landau
Utz Westermann, UC Irvine
Min-Yen