SOVIET DOCUMENTS ON ELECTRONIC COMPUTERS: DIGITAL COMPUTERS AND TRANSLATING MACHINES

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP80T00246A031800500001-3
• Release Decision: RIPPUB
• Original Classification: S
• Document Page Count: 31
• Document Creation Date: December 22, 2016
• Document Release Date: February 23, 2010
• Sequence Number: 1
• Publication Date: February 28, 1957
• Content Type: REPORT

Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
pcI ESSING COPY
CENTRAL INTELLIGENCE AGENCY
This material contains information affecting the National Defense of the United States within the meaning of the Espionage Laws, Title
18, U.S.C. Secs. 793 and 794, the transmission or revelation of which in any manner to an unauthorized person is prohibited by law.
S-E-C-R-E-T
DATE OF
INFO.
PLACE &
DATE AC Q.
NO.
28 February 1957
1
RD
documents publU.shed by the USSR Academy of Scienceg Moscow:
High-Speed Electronic Computer of the Academy of Sciences of the USSR,
by S.A.Lebedev, 1955.
The High-Speed Electronic Computer of the USSR Academy of SciencespB
It F Reliability and Methods of Checking, by V.A. LL_elnikov, 1956.
Les Experiments ...e la Traduction Automatigue de L'Angla.ia_ en
Russe a T?'Aide de la C .lculatrice BESM by Korolev, S. Rasoumovskiy,
and G.Zele kevich, 195 .
Attachments arc not classified when detached from this report.
STATE X ARMY
Soviet Documents on Electronic DATE DISTR.
Computers '
~y~~ ~ NO. PAGES
~'-6 etc REQUIREMENT
X NAVY IX AIR IX FBI
AEC
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
STAT
_d-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
trol are carried out in 77 microsecs. Addition and subtrac-
tion is carried out in 77 to 182 microsecs., depending on
the necessity of equalizing characteristics on normalizing
results. Multiplication takes 270 microsecs. and division
288 microsecs. When solving complex problems on the
machine, the average operation speed k from 7,000 to
8,000 three-address operations per second, including
reference to the magnetic drum and the magnetic tapes.
The machine operates 24 hours per day, part of the time
being spent on checking. The useful operating time of
the machine is 72 per cent, the time spent on checking
is 20 per cent, and error losses total. 8 per cent (these loss-
es including not only the time required to find the source
of trouble in the machine, but that required to repeat.
the calculations as well).
The machine is checked by means of test programmes
with the electron tubes working under worsened' conditions
(marginal checking). At first special circuitswere provided .
in the machine for marginal checking. Afterwards condi?
tions for marginal checking were secured by changing
the incandescence of the tubes. For marginal checking
of the memory the displacement in the automatic ampli-
fication control units is altered.
The machine is made up of small standard plug-in
units. A general view of the ' machine is shown in the
photographs.
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
ACADEMY OF SCIENCES OF THE USSR
INSTITUTE OF EXACT MECHANICS AND
COMPUTING TECHNIQUE
S. A. L1:BB1:DEV
Iq
COMPUTER OF THE ACADEMY
OF SCIENCES OF THE USSR
HIGH-SPEED ELECTRONIC
PUBLISHING HOUSE OF USSR ACADEMY OF SCIENCRS
1o04COw 1063
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
ACADEMY OF SCIENCES OF THE USSR
INSTITUTE OF EXACT MECHANICS AND
COMPUTING TECHIQUE
HIGH-SPEED ELECTRONIC
COMPUTER OF THE ACADEMY
OF SCIENCES OF THE USSR
THE PAPER
FOR THE INTERNATIONAL CONFERENCE
ON ELECTRONIC COMPUTERS
IN DARMSHTADT
OCTOBER 1955
PUBLISHING HOUSE OF USSR ACADEMT OF SCIENCE Si
Moscow +ftb
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
The USSR Academy of Sciences high-speed computer
(BESM) is a digital electronic machine for the solution
of laborious problems in physics, mechanics, astronomy,
engineering. etc. In designing the machine emphasis was
laid on convenience of programming and simplicity of
operation.
A binary system wi`h a floating decimal point was
selected. Although the floating decimal point somewhat
complicates the logic diagram of the machine, it greatly
simplifies programming and mathematical operation. Cal-
culations are made, as a rule, with normalized numbers,
If the result of any arithmetical operation, say, subtrac-
tion, is unnormal, it is automatically normalized. Be-
sides, provision is made in the machine for carrying out.
calculations with unnormalized numbers and with Fe'
suit normalization interlocking. The mantissa of the
number is represented by 32 binary positions; then,, there
Is. one position for the sign of the number, 5 positions,
for the characteristic of the number and one position for
the sign of the characteristic.
Thus, the machine can represent numbers from 2-"
to 2+'1. The number. of digits selecte i ensures the accu-
racy required for most problems. In some cases, for in-
stance, when solving certain astronomical problem;,
the calculations _may be carried out on the machine with
double the number of digits. Conversion from the decinl
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
to the binary system and vice versa is effected directly on
the machine by means of suitable subprogrammes.
The machine has a three-address system. The code
of each address consists of 11 bits, the operation part -
of 5 bits; the sixth operation position is for result normal-
ization interlocking. Thus, provision has been made in
the machine for 31 instructions. Besides the four arithmet-
ical operations, the results of which are rounded to re-
duce accumulating errors, there are instructions for multi-
plication with presentation of the product in double the num.
her of bits and for division with presentation of the remainder
for the sake )i simplifying calculations employing double
the number of positions; several' instructions. facilitating
operations on the characteristics of numbers; instructions
for exchange of codes between the separate memory units;
conditional and unconditional transfer Instructions;
number of logical instructions.
The instructions are selected from. the: memory by.
two independent instruction control systems.- The-. set of
instructions includes control frensfer operations -from
one system to the other both with and without changing
the number to the instruction. Double instruction con-
trol facilitates programming transitions to subprogrammes
and returns to the main programme.
All operations are carried out by a single universal
parallel action arithmetical unit based on trigger cells.
The arithmetical unit consists of two receiving Misters
and an adder. A special circuit has been employed in
the adder to eliminate cascade carry. Code addition
is effected in less than 3 microsecs. In- takes less than
2 microsecs for the codes to shift to the next position.
When adding or substracting numbers with. different
characteristics, the latter are equalized by shifting the
smaller number, prior to the actual addition of the codes.
If an unnormal result is obtained, it is normalized by
shifting the code to the left and respectively changing
its characteristic. Multiplication and division are carried
out by addition or subtraction of the code in the adder
and shifting.
Special cathode-ray tubes are employed for the mem-
ory. There is one tube for every binary position. The
capacity of the memory is 1023 numbers. Number selec-
tion from the memory and result recording are carried
out parallelly. The level of the reading signal at the en-
trance to the amplifier is 10 mV. A bridge circuit Is em-
ployed to reduce overloading of the reading amplifier.
This has made it possible to reduce the false signal at
the entrance to the amplifier when recording, to the level
of the reading signal. Moreover, the balance does not
depend on the working conditions of the tubes and the
supply voltages. The level of the reading signal is adjusted
automatically by referring to the control point outside
the raster, in which reading and recording are done after
every 32 regenerations.
Recording (including preliminary reading) takes 6.5
microsecs., and reading and regeneration-5 microsecs.
The full time of reference to the cathode-ray tubes, in-
cluding the adjustment of the ray and reading, or record-
ing, is 12 microsecs. The memory Is provided with auto-
matic internal control, vermitting simultaneous veri-
fication of all the positions and emission of the corres-
ponding signals.
Besides the cathode-ray tube memory there is a setup
unit based on germanium diodes, having a total capacity
of 376 numbers. The desired numbers or Instructions can
be selected from a memory of this type, but results cannot
be reoorded in It. The numbers 6f Instructions are set bp
by punching codes on punchcards. The punehcards ails
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
placed in special arrangements called 'books". When
the books" are closed contacts separated from the other
circuits by the diode make through the punched holes.
Besides the punchcards there is provision for setting up
the numbers on a plugboard. The corresponding positions
in the "books" and on the plugboard are in parallel.
The diode set-up device has also a number of removable
units with typical subprogrammes (subprogrammes of
trigonometric functions, logarithms, exponent functions,
transition from one numerical system to another, etc.)
permanently set up (by soldering) in the form of diode
matrices. The diode set-up device is employed mainly
for typical subprogrammes, for setting coefficients which
change frc,m one type of calculation to another, for pre-
determining the error of coincidence of iterative pro-
cesses, fr,r manual control of calculations, etc.
In order to extend the sphere of solvable problems
to include such as require large storage capacities, the
machine. Is. provided with magnetic drum and a magnetic
tape storage device.
The magnetic drum had a capacity of 5120 numbers
(five groups of 1023 numbers each). Any group of a pre-
determined number of codes can be transferred to the
memory, and vice versa, groups of numbers from the
memory can be recorded on the drum. In order to reduce
the amount of equipment, a series system of reading and
recording on the magnetic drum has been provided.
Conversion of the series to the parallel system and vice
versa, necessary in exchanging codes between the magnetic
drum and the memory, is effected in the arithmetical
unit.
There are 84 reading and recording heads mounted
on the magnetic drum, 80 of which are for codes, one for
clock pulses, one for registering the start pulse and 2
spares.: The clearance between the magnetic dh=1 k'}(
the heads is of the order of 35 microns. Each.ta
drum holds 64 numbers. The recording density
3 pulses per millimeter. The drum does 750
per minute. Hence, the average time spent An Valth>g,
for the required code to come under the magnet. had fe,
40 microsecs., the subsequent selection 'or recordiuf_
taking place at a rate of 800 numbers per, second. The pulse
frequency of the magnetic drum 'amounts to - about
35 kilocycles. The amplitude of a reading pulse at the mag-
netic head constitutes from b0 to 60 O:V, Reading and
recording amplifiers are provided for egch group (5' ito-
gether). They are switched from track tb,ick *fthtn
each group automatically by means of a cal #W
k ~7
diode circuit.
The magnetic tape storage consists of 4. , ee action
magnetophones with tape 6.5 mm wide. 'tote 'tape ';has
two, tracks One of them is for recording =t k pulses,
and the other for code recording. The group ' fi~fimber' .1$
also recorded on the code track. Like the magnetic'drwii,
the magnetic tapes are for exchanging codes with the
memory. The series codes are transformed to parallel
ones In the arithmetical unit.
The magnetic tapes can move both, forwards = and
backwards. Only the forward movement is the working
stroke. It Is during this stroke that any subsequent re-
corded block is read out, according to a pre-determined
number, or that mcording Is done. The reverse movement
Is for automatically conveying any block that has passed
to the magnetic head. This makes it possible to use the mag-
netic tape operatively for reading and recording within
the limits of one or several integral blocks.
The length of the magnetic tape on each magna ?
ph,ine is 200 meters. Eight pulses can be recorded on
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
millimeter. One tape can hold an order of 30,000 numbers
(with allowance for the intervals between groups), making
a total of about 120,000 numbers for the 4 magnetophones.
The tape reels can be replaced in a very short time, and
this is usually done without stopping the machine. The
heads are mounted on the magnetophones with high
enough precision to permit transposing recorded tape
reels from one magnetophone to another. The tape speed
is 2 meters per sec. The pulse frequency is 16 kilocycles,
and the rate of selection or recording is 400 numbers per
second. The amplitude of a reading pulse at the magnetic
head is 15 to 20 mV. Each magnetophone is provided
with reading and - recording amplifiers.
The numbers and instructions are put into the machine
from a punched tape in the form of a series code. The
punched tape has two tracks, one for clock pulses, and
the other for code pulses. The series code is tranformed
into a parallel one In the arithmetical unit. The punched
codes are read-out by means of photoelectric cells. The
reading rate is 20 numbers per second. .
The punched tapes are prepared independently on
special punchers. For verification purposes two identical
tapes are punched independently on different punchers.
The punched, tapes produced are then compared automa-
tically on a special verifier.
When the data to be put in are the results of calcula-
tions carried out on the machine they are recorded oh
a magnetic tape, and are put in from this tape. In case
of an interruption in the calculation of a large problem
the used programmes or necessary data are likewise recorded
on a magnetic tape. The subsequent input is also. from
this magnetic tape.
The results of the calculations are withdrawn from
the machine by recording on magnetic tape and their
printed on a motion picture film independently on a
special printing device. The series code passes from the
magnetic tape to a shift register, on which it is trans-
formed into a parallel code. Suitable decorders for each
digit control pinpoint light sources, which project the
image of the digit on a motion picture film. The printing
device work's at a rate of 200 numbers per second. The
film is developed in a developing machine. Paper copies
are likewise made in a special machine.
Besides the photographic printing unit, there is an
electromechanical printing unit, controlled directly by
the machine. The printing rate is 1.5 numbers per second.
This printing unit is used when the material to be with-
drawn is small in volume compared to the calculations,
and to print control values for checking the progress of
the calculations.
The controls of the machine are centralized for most
operations that fit into the time of the common cycle.
A number of elementary operations which are individual
for certain instructions and which would substantially
slow down the fulfilment of the other instructions if
centrally controlled, are carried out. by local controls.
In carrying out the individual instructions the transition
from the central to the local controls and vice versa takes
place automatically at the correct moments. The ele-
mentary operations under local control include: multi-
plication, division and number shifts, as well as equali-
zation of the characteristics of numbers and result nor-
malization for addition and subtraction instructions.
The standard cycle of the machine includes selection
of two numbers from the memory carrying out the pre-
determined operation with these numbers, sending the
result to the memory and selection of a new instruction
from the memory. Instructions not requiring local con-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
 'A Sr'd
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
STAT
_d-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
ACADEMY OF SCIENCES OF THE USSR
V. A. Melnikov
THE HIGH-SPEED ELECTRONIC COMPUTER
OF THE USSR ACADEMY OF SCIENCES
B E S M,
ITS RELIABIL'.TY AND METHODS OF CHECKING
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
The high-speed electronic computer of the USSR Aca-
demy of Sciences (BESM) was developed at the Institute
of Precise Mechanics and Computing Technique by Aca-
demician S. A. Lebedev and his associates.
The BESM is a digital electronic machine for the solution
of laborious problems in physics, mechanics, astronomy,
engineering, etc. In designing the machine emphasis was
laid on convenience of programming and simplicity of ope-
ration.
A binary system with a floating decimal point was se-
lected. Calculations are made, as a rule, with normalized
numbers. If the result of any arithmetical operation, say,
subtraction, is unnormal, it is automatically normalized.
Besides, provision is made in the machine for carrying
out calculations with unnormalized numbers and with
result normalization interlocking. The mantissa of the
number is represented by 32 binary positions; then, there
is one position for the sign of the number, 5 positions for
the characteristic of the number and one position for
the sign of the characteristic.
The machine has a three-address system. The code of
each address consistes of I I bits, the operation part --
of 5 bits the sixth operation position is for result nor-
malization interlocking. Thus, provision has been made
in the machine for 31 instructions.
All operations are carried out by a single universal
parallel action arithmetical unit. See fig. 1.
The arithmetical unit consists of two receiving registers
and an adder. A special circuit has been employed in
the adder to eliminate cascade carry. Code addition is ef-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
fected in less than 3 microsecs. In takes less than 2 mi-
crosecs for the codes to shift to the next position.
The standard cycle of the machine includes selection
of two numbers from the memory carrying out the prede-
Addition and subtraction carried out in 77 to 182 mi?
crosecs., depending on the necessity of equalizing charac-
teristics on normalizing results. Multiplication takes
270 microsecs. and division 288 microsecs.
Fig. I. The arithmetical unit.
termined operation with these numbers, sending the result
to the memory and selection of a new instruction from
the memory. This cycle takes 77 microsecs.
Fig. 2. A ganemal view of the cathode-ray tube memory.
Special cathode-ray tubes are employed for the me-
mory. There Is one tube for every binary position. The
capacity of the memory is 1023 numbers. See fig. 2.
5 -
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
Number selection from the memory and result recording
are carried out parallelly.
Recording (including preliminary reading) takes 6.5
microsecs., and reading and regeneration - 5 microsecs.
The full time of reference to the cathode-ray tubes, in-
cluding the adjustment of the ray and reading, or record-
ing, is 12 micresecs. The memory is provided with auto-
matic internal control, permitting simultaneous verification
of all the positions and emission of the corresponding
signals.
Besides the cathode-ray tube memory there is a set-up
unit based on germanium diodes, having a total capacity
of 376 numbers. The desired numbers or instructions can
be selected from a memory of this type, but results cannot
be recorded in it.The diode set-up device is employed ma-
inly for typical subprogrammes, for setting coefficients
which change from one type of calculation to another,
for predetermining the error of coincidence of iterative
process, for manual control of calculations, etc.
In order to extend the sphere of solvable problems to
Include such as require large storage capacities, the ma-
chine is provided with magnetic drum and a magnetic
tape storage device. See fig. 3 acid 4.
ME
Fig. a The magnetic tape storage device.
Fig. 4. The ma`netle data.
The magnetic drum had a ca acity of 5120 numbers
(five groups of 1023 numbers each). Any group of a pre-
determined number of codes can be transferred to the
memory, and vice versa, groups of numbers from the
memory can be recorded on the drum. In grder to reduce
the amount of equipment, a series system of reading and
recording on the magnetic drum has been provided. Con-
version of the series to the parallel system and vice versa.
necessary in exchangin codes between the magnetic dr-im
and the memory, Is effected in the arithmetical unit.
There are 84 reading and recording heads mounted
on the magnetic drum, 80 of which are for codes, one for
clock pulses, one for registering the start pulse and 2
spares. The clearance between the magnetic drum and
the heads is of the order of 35 microns. Each track of
the drum holds 64 numbers. The recording density is about
3 -pulses per millimeter. The drum does 750 revolutions
per minute. Hence the average time spent in waiting for
the required code to come under the magnet head Is 40 mi-
crosecs., the subsequent selection or recording taking
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
pace at a rate of 800 numbers per second. The pulse fre-
quency of the magnetic drum amounts to about 35 kilo-
cycles. The amplitude of a reading pulse at the magnetic
head constitutes from 50 to 60 mV. Reading and recording
amplifier are provided for each group (5 altogether). They
are switched from track to track within each group auto-
matically by means of a special tubediode circuit.
Th magnetic tape storage consists of 4 series action
magnetophones with tape 6.5 mm wide. The tape has two
tracks. One of them is for recording clock pulses, and
the other for code recording. The group number is also
recorded on the code track. Like the magnetic drum,
the magnetic tapes are for exchanging codes with the me-
mory. The series codes are transformed to parallel ones
in the arithmetical unit.
The magnetic tapes can move both forwards and back-
wards. Only the forward movement is the working sroke.
It *s during this stroke that any subsequent recorded
block is read out, according to a pre-determined number,
or that recording is done. The reverse movement is for
automatically conveying any block that has passed to
the magnetic head. This makes it possible to use the mag-
netic tape operatively for reading and recording within
the limits of one or several integral blocks
The length of the magnetic tape on each magnetophone
is 200 meters. Eight pulses can be recorded on each mil-
limeter. One tape can hold an order of 30,000 numbers
(with allowance for the intervals between groups), making
a total of about 120,000 numbers for the 4 magnetophones
The tape reels can be replaced in a very short time, and
this is usually done without stopping the machine. The
heads are mounted on the magnetophones with high
enough precision to permit transposing recorded tape
reels from one magnetophone to another. The tape speed
is 2 meters per sec. The pulse frequency is 16 kiolcycles,
and the rate of selection or recording is 400 numbers per
second. The amplitude of a reading pulse at the magnetic
head is 15 to 20 mV Each magnetophone is provided with
reading and recording amplifiers.
The numbers and instructions are put into the machine
from a punched tape in the form of a series code. The
punched tape has two tracks, one for clock pulses, and
the other for code pulses. See fig. 5 and 6.
The punched codes are read-out by means of photo-
electric cells. The reading rate is 20 numbers per second.
The punched tapes are prepared independently on spe
cial punchers.
Fig. 5. The input unit.
Fig. & A punched tape.
The results of the calculations are withdrawn from
the machine by recording on magnetic tape and then
printed on a motion picture film independently on a special
printing device. The series code passes from the magne-
tic tape to a shift register, on which it is transformed Into
a parallel code. Suitable decorders for each digit control
pinpoint light sources, which project the image of the digit
on a motion picture pilm. The printing device works at
a rate of 200 numbers per second. The film is developed
in a developing machine. Paper copies are likewise made
in a special machine. See fig. 7 and 8.
Besides the photographic printing unit, there is an
electromechanical printing unit, controlled directly by
the machine. The printing rate is 1.5 numbers per second.
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
Fig. & A tai a( We pbob hft i. - ~oa~adk tem
~. - deoodw, 3~ - flaw; t -eon IaaR; A - leas.
This printing unit is used when the material to be with-
drawn is small In volume compared to the calculations,
and to print control values for checking the progress of
the calculations. See fig. 9.
Electronic computers are characterized by their sta-
bility and reliability in operation as well as by their
storage capacity and access times. It is general practice to
have the machine inspected regulary in order to keep it
operating properly.
Reliability of the machine is guaranteed if special
checking procedure has been carried out, in the course of
which every element of the machine was given marginal-
checking and those of them that proved unreliable were
replaced by reliable ones.
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
Fig. 9. A general view of the electromechanical printing unit.
The BESM machine of the USSR Academy of Sciences
has been in operation since 1952. During this period spe-
cial methods of checking procedure have been worked out
with special test programmes which make it possible quickly
to eliminate trouble in the machine which may lead to
errors in doing problems.
- 12 -
The BESM is basically made up of separate strandard
plug-in-units assembled on two-tube or four .tube chassis
These units are mounted on racks between which the Inter-
unit connections are made. The control unit and the arith-
metical unit take up a separate rack. Both operative me-
mory and the external magnetic drum and magnetic tape
storage devices are also mounted on separate racks. This
individual location of the separate units greatly facilitates
technical maintenance by making it possible to ca out
prophylactic 'preparations and to check the machineyrack
by rack, independently of one another.
There are 8 types of standadt plug-in-units;
I. Trigger;
2. Gate;
3. Shaper;
4. Cathode follower;
5. Amplifier with electromagnetic delay line;
6. Inverter;
7. Diode unit;
& Ripple throughs carry unit.
These are the types of plug-in units the control and
arithmetical unit circuits as well as those of the automatic
control are made of. Thus it becomes clear that the reliability
of the machine depends to a considerable extent on the
reliability of the standard plug-in-units.
The most important factors influencing stability of
operation are the following:
a) time variation of the parameters of the electronic
tubes;
b) appearance of p o o r contacts in some of
the circuits, due to inevitable vibrations of the machine,
contamination and oxidation of the contact surfaces;
c) time variation of the resistance parameters. It is
obvious, that tubes and circuits of low reliability should
be detected beforehand to guarantee the reliability of the
machine at work. In designing the BESM methods of check-
ing procedure of the standard plug-in-units were devised,
based on alterations in some of the feed voltages. Tumbler
switches were provided so that different groups of units
could be tried under marginal conditions. However, expe-
rience in operation showed that the methods provided for
the marginal checking did not prove sufficiently effective
(with very few exceptions).
-13-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
While running the machine we have learnt, that the best
method ci checking the tubes in the BESM for emission loss
is that of decreasing the incandescence.
This method proved efficient for all the types of stan-
dard plug-in- units. The only exception is the trigger. In
the case of the trigger, besides lowering the incandescence.
another method should be used, which includes changing
the divider bias 3 to 4 per cent, thus disturbing the sym-
metry of the trigger. By using this method both the tube
and the stability of the divider resistances are checked,
the latter being of great influence on the stability of ope-
ration of the trigger.
The possibility of checking the units by lowering the in-
candescence is due to the fact that the cathode current
of a tube which has lost its emission depends greatly
on the incandescence, while that of a new tube remains
practically constant with incandescence fluctuations within
t 10 per cent.This property Is utilized in the checking pro-
cedure, of the standard plug-in-units. The presence of false
signals is revealed by the method of increasing-the incan-
descence.
Malfunctions sometimes occur and their frequency and
their nature does not change when the incandescence, current
frequency of feed voltages are varied. Malfunctions of this
kind are usually due-to unreliable contacts or the plugdn-
units being insufficiently shockproof.
This is the reason why regular checking must be carri-
ed out once or twice a month for shockproofnees. For this
purpose a test programme is put into the machine while
panels and separate units are tapped. In most cases
either poor contacts are found on the tube panels
or poor solder joints in the asembly. Poor contacts in
the plugs and sockets by which the unit is connected to
its rack practically do not occur.
The same methods of checking procedure are used for
the non-standard units in the storage devices. An addi-
tional position checking is carried out in the operative
cathode-ray tube memory. This is done by checking the
rewriting of code el* into code c0* and vice versa. in all
the positions of the storage unit with a corresponding bias
change on the read-out gates.
The following test programmes were compiled for
complete checking of the machine:
1. A programme for checking the main operations.
2. A programme for checking the adder with a vari-
able code.
3. A programme for checking the special plug-in-units.
4. A programme for checking the internal and exter-
nal storage devices.
5. A programme for checking the operative memory
for erewrltings.
6. A programme for checking the operative memory
for multiple reference.
g problems king Is similar to the operation
duty prooin
gr
All the test programmes are recorded in one of the
groups of the magnetic drum, and are kept there all
the tim . When cbedking Is necessary, reading out the pro-a few soconds, The time for
grammes
single fulfilment of al the 'r6e r+o~r Is 1 minute
Any of the test prograrrmea can adrsded by
manipulations at the central control desk In can rapid
checking of the machine Is required. Tumbler swit es
on the control daft can be used to set up automatic
successive fulfilment of the test pragrassntes or cyclic
fulfillment of one of tem. N all the tat
are fulfilled satiisfactorily under the ma conditions.
the machine may be considered ready for computations.
The statistics collected shows that the work carried
out with a view to improving certain units of the machine
as well as continuous Improvement of the checking pro.
cedure methods, have made It possible to Improve con-
siderably the stability and reliability of operation of
the machine.
When solving complex problems on the machine, the
average operation speed Is from 7.000 to 8,000 three-ad-
dres operations per second, Including reference to the mag-
netic drum and the magnetic tapes: The machine operates
24 hours per day, part of the time being spent on checking.
The useful operating time of the machine is 72 per cent,
the time spent on checking is 20 per cent, and error losses
total 8 per cent (these losses Including not only the time
required to find source of trouble In the machine, but that
required to repeat the calculations as well). See fig. 10.
The time of operation without malfunctions for periods
of ten consecutive hours and more constitutes about 70 per
- 15-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
riot /at per waR due tc
sESM *fiWVtkV ieANrb 124 h. d A'ri,J with MWM"y AWd
A+ ~tia/s~or?-type eat4ede-my tabs:. Averog-
7,A>b to AV p&wtiau Arp std
2$U tZW 40 W E 2$j 2a .9,*
Z #10A W:
4br/rt wwwao f ii-
~-" i/ Id
aaas rim Wiew AN to AVwe't au y
--- - now y,+V^e, rwt.a
Fig. 10. The BESM operating shedule
"55
(24 h. p. d.).
cent of the useful operating time. The maximum time of
operation registered without a single malfunction was
42 hours. Table I shows the percentage of mulfunctions
traced to various units. TABLE 1
3
~
e,
11
a
all
to 11
a>
-all
g 8
3'11
13 p, c.
3 2 p. c.
20 p. C.
4 p. c. 1 9 p. C.
10 p. c.1
13 p. c.
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
STAT
_d-
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
L. Korolev, S. Rasoumovski,
G. Zelenkevitch
LES EXPERIMENTS DE LA TRADUCTION
AUTOM,\TIQUE DE L'ANGLAIS EN RUSSE
A L'AIDE DE LA CALCULATRICE b )CM
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
ACADEMIE DES SCIENCES DE L'URSS
L. Korolev, S. Rasoumovski,
G. Zclenkevitch
LES EXPERIMENTS
DE LA TRADUCTION AUTOMATIQUE
DE L'ANGLALS_EN RUSSE A L'AIDE
DE LA CALCULATRICE 53CM
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
Le travail de la traduction automatique de l'angiais
en russe a ete commence spar I'Institut de Mecanique
Exacte et de Technique de Calcul et par l'Institut d'Infor-
mation Scientifique de I'Academie des Sciences de W.R.S.S.
Les collaborateurs scientifiques suivants ont pris part
a, ce travail: D. Yu. Panov, 1. S. Mouchin, I. K. Belskaia,
S. N. Rasoumovski, L. N. Korolev, G. P. Zelenkevitch.
Les experiments de la traduction ont ete realises par
la calculatrice 6lectronique, 6.3.C.M. Cette machine a etc::
construite par I'academicien Lebedev.
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246AO31800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
Les plus in,portants progres de la technique de la der-
ricrc dizairie d'annees se sont manifestos dans la construc-
tion des calculatrices electroniques sous le controle de
pro~rafnrne.
Les calculatrices universelles electroniques modernes
cilectuent plusieurs milliers d'operations arithmetiques
daps one seconde et elles remplacent le travail de plu-
sicurs dizaines de milliers d'hommes-calculateurs.
II est patent que la grande vitesse avec laquelle les
operations arithmetiques s'effectuent, n'aurait aucun lens
si I'on ne pouvait pas completement automatiser le proces
de calcul qui contient I'ordre etabli d'execution des opera-
tions et qui conserve les resultats intermediaires pour
leur utilisation soivante etc.
II faut marquer a part les commandes du passage
conditionnel qui permettent a la machine, selon les re-
sultats de calcul, de passer automatiquement de remplir
Tune ou I'autre sequence des commandes, c'est-a-dire de
choisir la vole des calculs suivants. On peut done questi-
onner la machine; elle repondra selon les circonstances
aoui), ou (non).
On appelle les commandes du passage conditionnel
'g les points de la bifurcation de programme*. Les pro-
grammes affectes a effectuer les problemes logiques ont
beaucoup de points de la bifurcation, tandis que les pro-
grammes des operations de calcul ont peu de points
pareils.
Le nombre de points de la bifurcation determine la
complexite de la structure de programme quand on le
compose.
La possibilite de modifier les commandes du program-
me, au cours du travail, et de choisir la voie suivante de
calculs selon les resultats obtenus, permet d'utiliser les
calculatrices electroniques pour executer une vaste classe
d'operations logiques.
L'operation de la traduction automatique d'une langue
en autre se refere a la mem@ categorie d'operations.
Depuis le janvier 1955, le groupe des collaborateurs
scientifiques de I'Institut de Mecanique Exacte et de
Technique de Calcul et les collaborateurs scientifiques de
I'Institut d'Information Scientifique de I'Academie des
Sciences de I'URSS commenga a developper les problemes
de la traduction automatique de I'anglais en russe.
Au bout de 1955 on a requ les premieres traductions
experimentales effectuees par la calculatrice electronique
63CM (la construction de I'academicien Lebedev).
Ces resultats experimentaux ont montre qu'il est
possible de creer le systeme de I'analyse de proposition
qui permet de fixer le sens de tous les mots de cette pro-
position en se basant sur les connaissances de la grammai-
re et du lexique de la langue. Le systeme pareil ne depend
pas pratiquement du vocabulaire et convient a la traduc-
'tion non seulement du texte specialement choisi mais it
convient a tous les, textes aux elements techniques et sci-
entifiques du profit choisi.
11 est vrai que la langue est le moyen des relations
entre les gens et qu'elle represente le systeme determine
dont le sens des mots et de leurs modifications peut @tre
exprime par les moyens lexicaux et grammaticaux. Ces
moyens s'enoncent comme les regles definies du lexique
et de la grammaire, comme les lois de la construction de
la phrase. En consequence, it est possible d'elaborer les
regles de la traduction qui prendront en consideration
toutes les particularites de la langue et permettront d'etab-
lir he sens des mots et les rapports entre les mots dans
la proposition. En d'autres termes, ii est possible d'etablir
les regles de la traduction automatique qui permettront
d'effectuer, a condition de leur satisfaction, la traduction
sans travailler prealablement et sans reviser ensuite des
phrases et des textes differents.
Il est necessaire d'avoir le dictionnaire pendant la
traduction ordinaire et pendant la traduction a I'aide de
la machine.
La calculatrice electronique opere sur les nombres,
c'est pourquoi les mots dans he dictionnaire doivent We
representes sous une forme de nombres, c'est-a-dire its
doivent ktre codes. En mettant, par exemple, les hombres
du code connu Bodo (fig. 1) en conformite avec chaque
lettre de I'alphabet latin et de I'alphahet russe, nous aurons
la possibilite d'inscrire chaque mot anglais et sa traductioni
russe par he nombre definitif, qui est l'unique pour ce mot.
Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
 Sanitized Copy Approved for Release 2010/02/23: CIA-RDP80T00246A031800500001-3
a-16 I -14 k -19 p-24 u -20
b-06 g- 10 I -27 q-23 v -29
c-22 h-26 rn - II r -07 w-13
d-30 i -12 r -15 s-05 x -09
e-08 j 18 o -28 t -21 y -04
Flg. I. Le codage de ('alphabet latin.
Par exemple, Ies mots:
There, short, into, numerical, methods
on inscrira sous !'aspect de nombres.
2126080708, 0526280721, 12152128, 152011160712221627,
11082126283005.
Chaque mot a son code et son numero d'ordre (de voca-
bulaire) definitif.
Le vocabulaire affecte a la traduction automatique se
distingue du vocabulaire ordinaire, car it a une serie de
renseignements supplementaires (des indices) qui se rap-
portent a la grammaire du mot russe, sauf le mot russe
qui correspond au mot anglais donne. Pour le nom ce
sont des renseignements suivants: genre, declinaison,
theme mouille ou dur presence ou absence du chuintant
dans le theme, quel objet signifie le mot: anime ou non
etc.
Pour le verbe ce sont: conjugaison, aspect etc.
Pour l'adjectif - theme mouille ou dur, degres de com-
paraison etc.
On code les principes de grammaire selon les regles
etablies. Ces principes occupent 39 digits binaires, se
representant une cellule de memoire. Les codes des mots
anglais et les codes de leurs indices se representent la
partie aanglaise3~ du vocabulaire.
La seconde partie du vocabulaire est Krusse,, elle com-
porte des mots russes inscrits selon 1'ordre de leurs nu-
meros indiques dans la partie anglaise du vocabulaire.
On avait prepare le vocabulaire compose de 952 mots
anglais et de 1073 mots russes pour les experiments, rea-
lises daps la machine 53CM. Si I'orthographe du mot du
texte correspond exactement a l'orthographe du mot du
vocabulaire c'-est-a-dire si Ies deux mots sont representes par
les rntmes nombres, on peut le constater facilement a !'aide
de !'operation de comparaison. Le choix des mots dans
le vocabulaire est base sur ce principe.
Pour choisir les mots qui prendront quelques term't-
J aisdTts (en anglais ce sont -s, -ing, -ed, -er, -est, -e, -y,
etc), it faut supprimer s terminaisons et recommencer
les recherches du mot sans la terminaison dans le voca-
bulaire.
C'est tres facile de trouver les mots monosemantiques dans
le vocabulaire. Mais c'est plus difficile de determiner le sens
du mot dont on a besoin si le mot est polysemantique.
Pour determiner le sens du ' mot polysemantique qui con-
vient a la pr' se donnee it faut faire !'analyse des mots
voisins, examiner leurs significations et leurs caracte-
ristiques grammaticales. Les regles pour determiner la
signification necessaire du mot polysemantique sont
basees sur !'analyse du grand materiel et elles sont com-
posees en schema qu'on appelle csupplement,.
Cette schema contient de meme les expressions idio-
matiques. On peut ordinairement mettre en relief tie mot
typique, dans chaque expression de ce genre, le mot qui
se rencontre dans toutes les variantes de cette expression
idiomatique.
On a donne a ce mot l'indice gpolysemantique, dans
le vocabulaire. Le supplement comprend le contr6le des
mots voisins pour constater que le mot ne fait pas partie
de !'expression idiomatique.
Si la reponse est positive, toute 1'espression idlomati-
que se traduit de !'expression equivalente. Par exemple,
le mot anglais cables on traduit ordinairement comme
l'adjectif tCnoco6HblI1,, mais en combinalson avec les
formed verbales tbe, (is, are, were, been, being), c'est
mfeux de le traduire comme le verbe