LETTER TO ASSISTANT SECRETARY OF STATE FOR OCEANS AND ENVIRONMENTAL AND SCIENTIFIC AFFAIRS REQUESTING HIS APPROVAL FOR A MEMBER OF HIS STAFF TO PARTICIPATE IN A COMMUNITY WORKING GROUP

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP85M00158R000600010013-3
• Release Decision: RIPPUB
• Original Classification: K
• Document Page Count: 14
• Document Creation Date: December 22, 2016
• Document Release Date: June 22, 2009
• Sequence Number: 13
• Publication Date: June 6, 1983
• Content Type: MEMO

Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
TRANSMITTAL SLIP DATE 6'_3
TO: /cl
ROOM NO. B DING
REMARKS:
FROM:
ROOM NO. BUILDING
EXTENSION
OOR
1 FEB 55 241 WHICHCMAYB BE M USED.
(47)
Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
FLTC-83-015
S JUG!
MEMORANDUM FOR: Deputy Director, Intelligence Community Staff
Executive Secretary, LTC
SUBJECT: Letter to Assistant Secretary of State for
Oceans and Environmental and Scientific
Affairs Requesting His Approval for a Member
of His Staff to Participate in a Community
Working Group
1. Action Requested: That you sign the attached letter to Ambassador
Malone.
2. Background: Mr. Arthur Corte, a State Department officer on
Ambassador Malone's staff, is one of the most knowledgeable U.S. Government
officials on developments within the American and foreign business communities
on Machine Assisted Translation (MAT). At the recommendation of ORD, he
participated in the 20 May meeting to discuss the application of MAT to
overtly acquires He would be most willing to be a
member of the working group to investigate this matter further but requires
the concurrence of his Bureau Chief, Ambassador Malone. The attached letter
from you provides Ambassador Malone with the background of this project and
requests his approval for Mr. Corte to continue to participate in it.
Attachment:
Ltr to Ambassador Malone
STAT
STAT
STAT
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
Distribution
FLTC-83-015
Original -
DD/ICS
1 -
FLTC Chrono
1 -
FLTC Subject
1
DD/ICS
IC Registry
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
The Director of Central Intelligence
Washington, D.C. 20505
Intelligence Community Staff
FLTC-83-015/1
7 June 1983
The Honorable James L. Malone
Assistant Secretary of State for Oceans and
International Environmental and Scientific Affairs
Room 7831 New State
Washington, D.C. 20520
Dear Ambassador Malone:
The Director of Central Intelligence's Foreign'Language Training
Committee (FLTC) has been charged with investigating the applicability of
Machine Assisted Translation (MAT) techniques to the translation of overtly
acquired Japanese scientific and technical documents collected, as a service
of common concern, by the Foreign Broadcast Information Service (FBIS).
Several components of the defense, foreign affairs, and intelligence
community, among them the School of Language Studies of the Foreign Service
Institute (whose Dean is the State Department's representative on the FLTC)
have long been interested in the potential of MAT. A meeting of
representatives of those organizations concerned with MAT was held on 20 May
at which it was decided to form a Working Group to explore further the
feasibility of applying MAT to these Among the
participants was Mr. Arthur Corte of your staff.
STAT
The Working Group, which will be chaired by of CIA's STAT
Office of Research and Development, will-prepare a report summarizing the work
presently being done in this field by U.S. Government departments and
agencies. It will also ensure that these findings are made known to an
independent study group under contract to FBIS and that, to the greatest
degree possible, the study group's research serves Community as well as FBIS
interests.
Because of his background and knowledge, Mr. Corte would be an invaluable
member of this Working Group. I should be most grateful if you could arrange
for Mr. Corte to participate in this important undertaking.
Sincerely,
Enclosures
Eloise R. -age
Deputy Director
STAT
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00l58R000600010013-3
FLTC-83-015/1
Ltr to J.
Malone
Original -
Addressee
_1 -
DD/ICS
IC Registry
1 -
FLTC Chrono
1 -
FLTC Subject
Approved For Release 2009/06/22 : CIA-RDP85M00l58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
Meeting of Intelligence Community Representatives
to Discuss the Feasibility of Utilizing Machine
Assisted Translations for
1. A meeting of Intelligence Community representatives concerned with
and knowledgeable in the developments of MAT met under the sponsorship of the
DCI's Foreign Language Training Committee (FLTC) on Friday, 20 May at the
o discuss the formation of a working group to
assess the feasibility of applying MAT to overtly acquire
documents. A list of those attending is appended. In his opening remarks,
C/FLTC, advised the participants that the Committee had been
charged by the Director of the Intelligence Community Staff with coordinating
the responses to the DCI Guidance (FY 1985-89) "for the CIAP and the GDIP" to
"submit coordinated program initiatives designed to improve the Community's
capability to produce timely and accurate translations of
material through the ... utilization of ... machine-assisted translation
technology". Moreover, the guidance also requested the CCP to accompany its
1985 budget submission with a report on the applicability of this technology
to NSA's requirements. The FLTC had attempted, at this meeting, to assemble
representatives of all elements of the Community interested in this problem
and to constitute a working group to prepare a report to the DDCI with
recommendations for possible action and funding to resolve this problem.
2 . I I then asked
Branch, to brief the group on the problem of these
Chief of FBIS' Asia
documents from
the point of view of his organization which is responsible for their
collection and translation.
stated that FBIS had been tasked in
1980 to investigate the problem of the acquisition and translation of
STAT
STAT
I
STAT
STAT
STAT
STAT
STAT
1
STAT
STAT
STAT
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
. There are at least 9000 such publications ij and FBIS
has subscriptions to some 300 of them. However, it is faced with a scarcity
of competent translators
who must be at the "5" or native fluency
level to meet FBIS's rigid requirements for finished, literate translations.
An effort is now being made to supplement the translators available to FBIS in
Washington, with people recruited inF---)ho will be working under the
direction of an FBIS officer there. Although the number of translators hired
thus far may quadruple current output, a massive backlog of documents awaiting
screening and/or translating remains. There is, perhaps, hope in a
technological breakthrough represented by fifth generation computers,
but this technology is not yet available. Even if technological help should
be forthcoming, FBIS is faced with a most serious problem of maintaining the
morale of its human translators who fear that should FBIS move to Machine or
Machine-Assisted Translations, they would be relegated to the most boring and
menial work of editing, etc. In concluding his presentation,
emphasized that for any form of MAT suitable for his needs, machine readable
input is essential. Other participants agreed.
3. Chairman of the Information Handling Committee
(IHC), stated that the importance of to the United
States will grow "by leaps and bounds". The IHC has a charter to facilitate
the automation of all aspects of intelligence processing and will lend
whatever assistance it can to the problem of Even
though we are successful in adapting some form of machine-assisted translation
to these documents, human assets will still be needed.
4. reported that Dr. Thoma, president of the LATSEC
Corporation which developed the SYSTRAN system of Machine Translation utilized
by FTD, is working on a system suitable for Mr. Corte
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
(State), who is familiar with industry's needs and experimentation with MAT
pointed out that American. business has access to some
10,000 ~ publications and that one must explore the technology which
could be useful in Machine assisted translation. Mr. Crawley, NISC, has been
working this problem, not only fo but for Korean and Chinese as
well.
5. Dr. Allen Weinstein (FSI) said one must make a distinction between
"translations and information capture". The latter might be well served by
MAT while the former might require, in addition, the services of skilled human
in screening
ORD) agreed that
machines could probably be
most helpful
juncture, adequate translation would be possible.
(NSA)
was a bit more optomistic. Then followed a brief discussion of various
systems for machine translations being evaluated or actually employed in
government and industry. Mr. Corte (State) who, through his work at MIT, is
familiar with industry's efforts to cope with pointed
out that the Europeans are faced with essentially the same problem as are the
Americans and that it would'be useful to learn what progress they have made in
this field.
6. To the surprise of most, and the delight of all participants,
reported that the CIA's Directorate of ,Science and Technology
(DDS&T) has contracted with the JASONS for a study on the applicability of MAT
to speaking for the group, welcomed
this development, expressed the hope that CIA would make known to the JASONS
the work presently being done in this field by the Intelligence Community, and
that the JASONS would, when addressing the problem of the FBIS, take into
account as well, the requirements of other elements of the Defense, Foreign
but doubted that, at this
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
Affairs, and Intelligence Communities. He advised participants that ORD has
already agreed that ould be made available to chair a
working group to address this problem and asked those present to advise
their willingness to participate in such a group.
said that he would prepare a Memorandum for the Record of the meeting for
review by the particpants and that he would personally contact them after
meeting with
work to be done.
(who was away on official travel) concerning further
thanking members for their attendance,
expressed the common view that the meeting had served the most useful purpose
of bringing together members of the Community who have independently been
wrestling with a common problem. He trusted this would be but the beginning
of a fruitful, joint cooperation.
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
ATTENDEES
C/FLTC
ES/FLTC
C/IHC
ICS/PBS
ICS/PBS/CIAP Monitor
ICS/PBS/GDIP Monitor
HUMINT Committee
ICS/PPS
CIA/ORD/Information Systems Research
Division
CIA/FBIS/C/Asia Branch
CIA/FBIS/C/Systems Development Staff
CIA/FBIS/Systems Development Staff
DIA/DST/3
NISC Translation Division,
Information Management Department
NISC Translation Division,
Information Management Department
Department of State, Foreign Service
Institute, School of Language Studies
Department of State-OES/SAT
Approved For Release 2009/06/22 : CIA-RDP85MOOl58R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
CIA/OTE/Language School
NSA
NSA
USAF/Foreign Technology Division (FTD)
Army ACSI
Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
1983 JASON STUDY FOR FBIS
.General Statement of Task
Currently, available Machine Translation (NIT) systems cannot produce
acceptable English translations without extensive pre-editing and/or post-
editing by a human. We need to assess whether ongoing R&D activity will
produce a next generation MT system which will satisfy our requirements or
whether we should support some effort in a direction not now being pursued.
This study will therefore consist of a survey and analysis of R&D activity
(in the United States, Europe, and Japan) in artificial intelligence/natural
language processing. The result of this effort will be a report and bricling
identifying centers of this activity, the barriers to and shortfalls in the
development of an efficient, accurate MT system, recommendations for FBIS-
supported development directed toward overcoming these barriers, and the
prospects of achieving such a system.
Background
FBIS currently does not have enough translators who are skilled in
both the languages and the subjects of the information we collect.
While we are increasing our efforts to hire and train new translators,
we do not anticipate a significant increase in our translation capacity.
Meanwhile, the number of foreign sources and the amount of information
in foreign languages (both printed and spoken) of interest to the
Intelligence Community continue to increase. Our improved collection
capability must be matched by improvements in our ability to process
(select, translate, and edit) the collected information.
Automation of these processing functions is a key goal of the FBIS
Modernization Program. In particular, automation of the translation
function would have the greatest payoff (in terms of processing through-
put) if it could be achieved. At present, there are a variety of MT
systems available in use which "automate," with human assistance, tome
translation of text from a source language to a target language.
of these systems are simply target language word processors used by
human translators. Others are claimed to be "fully automatic MT systems"
because the human involvement is "pre-editing" the source text and "post-
editing" the target text. The actual translation is performed by the
system based on store vocabulary and rules of grammar. All of these
systems suffer from performance that is unacceptable to FBIS. That is,
they produces target language products of unacceptable quality. The
users of these systems must either train themselves to read and accept
the MT output or they must employ a human translator in the post-edit
clean-up fldll\.1 loll. Most skilled translators reject this task, claiming
that is is easier to translate from the source text without the system.
Consumers of FBIS products require and expect English text whose meaning
and grammar are comparable to that of the source text.
Attachment 2
Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
Plans to improve the existing MT systems by enlarging the dictionaries
and refining the rules of grammar do not seem to us to be viable solutions
to the problems of accurately and efficiently translating a natural (human)
source language. Natural language is too ambiguous, too imprecise, too
ill-defined for any dictionary/grammar based MT system to cope adequately
with accurate, idiomatic translation of source language in context. A
totally new approach seems required.
Assumption
We have become aware of research in Artificial Intelligence/Natural
Language Processing (AI/NLP). The goal of AI/NLP (handling and under-
standing. natural language) matches what we perceive as the significant
problem in today's MT systems. We assume, therefore, that if AI/NLP
techniques were applied to the translation function, the translated
product would have improved quality.
Requirements
FBIS is interested solely in the. translation of other languages into
English. Japanese is currently the source language of greatest interest,
but an MT capability in Russian, Arabic, Chinese, and other languages of
interest would be beneficial.
Our goal is to eliminate the need for pre-editing but not necessarily
for post-editing. Our approach is to increase the productivity (through-
put rate) of the human translator by providing him with an MI' system that
can learn from him and require him to clean up only the unusual and
"challenging" passages. A productivity increase of at least a factor
of two (compared to the throughput rate of the human translator alone)
does not seem unreasonable.
Our schedule permits incremental improvements in productivity.
Depending on cost, we could use today an MT/human post-editor system
that offered a'33% increase in productivity in translating Japanese S&T
material. In two to four years, we will need productivity increases of
66% to 100%. By the end of the 80's, we will need even greater improve-
ments for a variety of source languages.
Ancillary Issues
Although not the primary goal of this study, there are two related
issues in which we may derive benefit. The first involves getting the
source-language text into machine readable form. We collect foreign-
language material in both printed and spoken forms. Converting this to
machine readable form will be done most efficiently by foreign language
optical character readers and foreign language voice detection/analysis
systems.
Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
 Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3
The second related issue involves the selection function: of all
the material collected, which items should be translated? If AI/NLP
systems can understand the content of the source language text well
enough to translate it, they should certainly be able to compare against
stored criteria to perform the selection function. With a human, of
course, hovering about to resolve issues of indecision.
These two issues should be kept in mind during the survey of AI/NLP
research activity. Any intelligence derived, which is applicable to
them, should be reported.
Questions to be Addressed
Where is research taking place in AI/NLP?
How has the field been structured?
What are the key attributes of this research?
Which attributes are critical to the development of new MT techniques?
What are the plans to develop new MT systems based on AI/NLP?
Are there existing MT systems which incorporate AI/NI.P?
V,hat are the plans to improve these systems?
1that approaches should be pursued to achieve a significant improvement
in the quality of current MT output?
Should FBIS support the development of an MT system employing AI/NLP?
Should FBIS acquire any existing MT system?
Approved For Release 2009/06/22 : CIA-RDP85M00158R000600010013-3