Overview of the 
ARPA Human Language Technology Workshop 
Madeleine Bates, Chair, Editor 
BBN Systems & Technologies 
70 Fawcett Street 
Cambridge, MA 02138 
1. PREVIOUS DARPA WORKSHOPS 
For five years, 1988-1992, the Defense Advanced Projects 
Agency sponsored a series of meetings called the DARPA 
Speech and Natural Language Workshops. These workshops 
provided a forum where researchers in speech and natural 
language, particularly as relating to the DARPA programs in 
spoken and written language understanding, could exchange 
information about recent research and technical progress. 
Participants included researchers funded under the DARPA 
programs, other researchers who voluntarily participated in 
these programs or in related evaluations, government re- 
searchers and consumers of these research results, and in- 
vited attendees from inside and outside the US. 
Proceedings of these worksshops were published by Morgan 
Kaufmann. 
2. THE 1993 WORKSHOP: 
EXPANDED SCOPE 
In 1993, the "D" was dropped from DARPA, reflecting a 
change in the organizational purposes and goals. Well be- 
fore this change, however, the committee responsible for 
the annual workshop was directed to significantly broaden 
its focus, and the name of the meeting became The ARPA 
Human Language Technology (HLT) Workshop. 
The I-ILT workshop provides a forum where researchers can 
exchange information about very recent technical progress 
in an informal, highly interactive setting. The scope includes 
not just speech recognition, speech understanding, text un- 
derstanding, and machine translation, but also all spoken 
and written language work (broadly interpreted to include 
ARPA's TIPSTER, MT, MUC, and TREC programs) with 
an emphasis on topics of mutual interest, such as statistical 
language modeling. This workshop no longer focuses on 
spoken language systems evaluation, as another workshop 
fills that need. 
The prime purpose of the newly constituted workshop was 
to facilitate productive technical discussions among key re- 
searchers on topics relating to human language technology 
and of interest to ARPA. This expanded workshop, in par- 
ticular, was designed to facilitate cross-fertilization among 
diverse disciplines, and to introduce researchers to the state- 
of-the-art in areas outside their own. For this reason, the 
papers presented in this proceedings are both more wide- 
reaching and more accessible to non-experts in the field than 
has previously been the case. 
The majority of the workshop participants received funding 
under ARPA's Human Language program. Other partici- 
pants included researchers not funded by ARPA who volun- 
tarily participate in these programs; government researchers 
and consumers of these research results; and, on a rotat- 
ing basis, selected visitors from both inside and outside the 
United States. It is still the intention that the participants 
form a tightly-coupled research community in which results 
and research breakthroughs are evaluated, disseminated, and 
exploited with very short latency. 
Aspects of this meeting included: 
technical presentations of both new research results and 
the ongoing development of large software systems, of- 
ten months before this research is reported elsewhere; 
the presentation of summaries of standardized system 
evaluations; 
discussion of the future direction of the various DARPA 
programs in light of recent progress; and 
much discussion of ongoing work among individual re- 
searchers. 
The format of the meeting was to mix 12-15 minute tech- 
nical presentations (both reviewed and invited) in sessions 
organized around focused topics, with time for informal dis- 
cussion and interaction. In most sessions, the chair presented 
a 15 minute introduction intended to provide a summary of 
the key points in each set of papers, and also a perspec- 
tive of the research context in which the work was done, 
for those not in the primary area addressed by the papers in 
the session. The written version of those introductions will 
serve the same functions in this proceedings. 
Participants selected by the program committee were invited 
to subnfit abstracts in a variety of areas. The committee 
received 131 abstracts (114 for regular presentations; 17 
for der~tonstrafions), 58 presentation and 13 demonstrations 
were accepted. 
The final program sessions were: Spoken Language Sys- 
tems, Invited Overviews of ARPA Program Areas, Continu- 
ous Speech Recognition, Natural Language, Discourse, Ma- 
chine Translation, Demonstrations, Statistical Natural Lan- 
guage, Government Panel, Lexicon, Prosody, Information 
Retrieval, and New Directions. 
Technical highlights included: 
• Machine Translation session, chaired by Alex Waibel of 
CMU. This session featured invited tutorials by Peter 
Brown and Ed Hovy. 
• A demonstration session (chaired by Hy Murveit of SRI 
and organized by Victor Abrrash of SRI) showcased re- 
cent work in a variety of areas, and demonstrations were 
available almost continuously throughout the workshop 
for "hands-on" experience. 
• A lexicon session, chaired by Ralph Grishman (NYU). 
• Invited overviews of each of the areas of interest to 
ARPA in spoken and written language technology: 
MUC (Beth Sundheim), TREC (Donna Harmon), SLS 
(George Doddington), and Tipster (Tom Crystal). 
• A government panel on Government Human Language 
Technology Needs, Funding, In-House Research, and 
Technology Transfer organized by Carol Van Ess- 
Dykema. The panelists were Helen Gigley (NRL), 
Joseph Kielman (FBI), Susan Chipman (ONR), Jesse 
Fussell (DoD), and Y.T. Chien (NSF). 
The workshop attracted 207 attendees, approximately 2/3 
from ARPA sites. The remainder were government repre- 
sentatives (30 individuals from 10 different organizations), 
foreign guests (11 individuals, from 11 organizations), non- 
ARPA attendees from the USA (31 individuals from 27 or- 
ganizations) and young researchers (7 individuals, from 5 
schools). 
3. ACKNOWLEDGEMENTS 
This meeting involved the work of a great many people. The 
Standing Workshop Committee (whose membership rotates 
slowly except for one government member) is responsible 
for the overall seres of workshops. It was chaired this year 
by Patti Price (SRI); other members included Mitch Marcus 
(U. Penn), Madeleine Bates (BBN), Carol Van Ess-Dykema 
(DoD), Cliff Weinstein (Lincoln Lab), and Ralph Grishrnan 
(NYU). The standing committee, and Patti Price in particu- 
lar, provided invaluable direction for the workshop. 
The Program Committee, which is constituted each year by 
adding to the standing committee additional researchers se- 
lected for their expertise in specific areas, was chaired by me, 
and included (in addition to the standing committee mem- 
bers) Stephen Della Pietra, David Lewis, Kathy McKeown, 
Mary Ellen Okurowski, Stephanie Seneff, Beth Sundlieim, 
and Alex Waibel. The program committee was responsible 
for reviewing abstracts to select papers and demonstrations 
for presentation, discussing various policy issues, and orga- 
nizing the overall program. 
Particular thanks are due to Hy Murveit and Victor Abrash, 
who took complete charge of the demonstration sessions and 
handled the myriad of technical problems inherent in ar- 
ranging for a large number of live demos; to Alex Waibel, 
who organized the Machine Translation session (the key- 
stone session of the entire workshop) while commuting from 
his office in Germany to his office in the US, and to Ralph 
Gfishman, who organized the Lexicon session apparently 
effortlessly (which is to say, with extreme competence). 
Carol Van Ess-Dykema, assisted by Mary Ellen Okurowski, 
planned and chaired the government panel. 
Denise Payne served as the workshop administrator, which 
means everything from email archivist to registrar. She han- 
dled all the interactions with the conference center, kept 
track of invitations, abstracts, and attendees. She kept one 
eye on the budget and one hand on the telephone. She pre- 
pared the notebooks of preliminary papers that were given 
to attendees at the workshop, and collected final versions of 
the papers for this proceedings. 
From his new position as Program Manager at ARPA, 
George Doddington provided overall direction and encour- 
agement to the workshop planners. His enthusiasm for his 
new role and his desire to continue the high technical stan- 
dards of previous workshops helped to make this workshop, 
which was the first in a new series, a resounding success. 
4 
