CALL FOR PAPERS

	Analyzing Conversations in Text and Speech  (ACTS) 

		       HLT-NAACL 2006 Conference Workshop

			   New York, NY
			     June 8, 2006

	     http://www.isi.edu/~liangz/WORKSHOP/acts.html

	     *** Submission Deadline:  March 17, 2006 ***

-----------------------------------------------------------------
In recent years we have observed indications of convergence between 
the Natural Language Processing (NLP) and Speech Analysis research 
communities on the topic of conversation/dialogue analysis.  This 
proposed workshop will focus on the challenges posed in multi-party 
exchanges, shared by researchers in both areas.



Modeling and analyzing human-to-human multi-party conversations has 
been investigated by the Speech community in the context of 
telephone conversions and meetings.  As very large collections of 
high-quality audio and video recordings become available, we are 
provided with a new opportunity to perform deep analyses of 
multi-party exchanges which are becoming more and more useful.  For 
instance, one might wish to be able to extract the main theme from a 
talk show, the headlines from a live news broadcast, and the main 
topics from a meeting, from any desired point of view.

New types of conversational data recorded in texts on the Internet 
also exhibit multi-party responsive structures.  These naturally 
exist in various forms such as electronic mail, mailing lists, 
Instant Messaging (IM), Short Message Service (SMS) notes, chat 
rooms, Usenet newsgroups, threaded Web-based discussion lists, and 
massive multi-player online role-playing games (MMORPGs). Direct 
applications of NLP techniques, developed primarily for 
non-conversational data, are not adequate and often return poor 
results for these types of data.  The analysis of conversational 
texts poses significant challenges in discourse analysis, 
cross-document structure (CST), topic detection and tracking (TDT), 
summarization, etc.  In addition, problems such as boundary 
detection and speaker identification that used to be solely 
associated with traditional speech analysis are now coming under the 
focus of the NLP community. The new data types are allowing us to 
combine and apply traditional NLP and speech techniques to achieve a 
shared goal - understanding conversations from the input. 



The goal of this workshop is to highlight the set of core problems 
shared by research in speech and NLP with respect to multi-party 
dialogues. Authors from both communities are encouraged to submit. 
We hope to foster discussions about, but not limited to (the 
following topics are equally important; detailed descriptions for 
some are provided only for clarification purposes): 

-content identification and segmentation: techniques for identifying 
and subdividing conversational input (both speech and text) into 
passages or segments that are relevant to specific subtopics. 

-speaker-turn identification: in speech analysis, speakers take turns 
in contributing to a conversation. It is more subtle in conversational 
data recorded in texts. For example, sorting messages posted by a 
particular newsgroup by timeline does not produce a naturally flowing 
conversation. Individual messages can refer to any of the previously 
posted messages. Identifying speaker turns becomes a part of the 
conversation extraction task. What are the techniques that we can use 
from speech analysis to solve this problem? And how do we apply them
to the new data types?

-social network analysis: identify the conversation participants and 
how they (or the messages they send) are related. 

-expert search: identify the authority figures (recognized by other 
participants through exchanges) in a conversation.

-cross-document relations: multiple messages or segments may
 contribute 
to a conversation. What is the function of each and how are they 
related to one another?

-dialogue structure

-multi-party discourse analysis

-analysis of task structure

-topic detection and tracking

-summarization

-evaluation


SUBMISSION:

Full-paper submissions should be a maximum of 8 pages, short-paper 
submissions should be a maximum of 4 pages, and position papers should 
be of 2 pages. Please use the HLT-NAACL formats. The reviewing process 
will be blind, so authors' names, affiliations, and all
self-references 
should not be included in the paper. 

Submissions should be sent through the HLT submission page: 
http://www.softconf.com/start/HLT-WS06-ACTS/submit.html

Submission Deadline: March 17th, 2006. 


ORGANIZING COMMITTEE:


Eduard Hovy (USC-ISI)
Klaus Zechner (ETS)
Liang Zhou (USC-ISI) 


PROGRAM COMMITTEE:

Regina Barzilay (MIT)
Bob Frederking (CMU)
Graeme Hirst (U. of Toronto)
Miralla Lapata (U. of Edinburgh)
Anton Leuski (USC-ICT)
Chin-Yew Lin (USC-ISI)
Kathy McKeown (Columbia U.)
Doug Oard (U. of Maryland)
Mari Ostendorf (U. of Washington)
Gerald Penn (U. of Toronto)
Dragomir Radev (U. of Michigan)
Alex Rudnicky (CMU)
Karen Sparck Jones (U. of Cambridge)
Simone Teufel (U. of Cambridge)
David Traum (USC-ICT)
Klaus Zechner (ETS)
Ming Zhou (Microsoft Research Asia)