DISTINGUISHING FACT FROM OiqNlON AND EVENTS FROM META-EVENTS 
Chrlltlne ,4. Mor~gomery 
Ol~r~lng ~/zrtems Division 
I.~Icon, Inc. 
Woodlen~ NIIIw, California, U,9~d. 
ABSTRACT 
A major problem In automatically anaiy~dng the text of 
m~lltary messaq;les In m~lar to synthesize data b~le ele- 
ments is separating fact from opinion, I.e., Ident~n9 
factual statements versus evaluative commentary such 
4,; degree Of belief or confidence In the parameters of s 
particular event, comparative evaiual~ons of entitles and 
events, predictions of future events, etc. This paper 
desmlbes a m~iel which accmmta for the range of fac- 
tual to evaluative Information In the message traffic, and 
dlocusses a means for repreaenUng such Ioformation In 
the context of an oxperlmentsJ system for automated 
data bale generstlo~ 
1.0 EIACI~GFIOUND 
For several years, with some Inl;erroptlone, a research 
end development effort hal been In procmm In the 
Operating Systems Division of Logicon (and its prede- 
came' company, OpereUng Systems, Inc.) aimed at the 
automated creation of data base elements from the text 
~f electronically transmitted military messages. 1 \[Kuhna 
and Montgomery 1973, Kuhns 1Q74, Kuhns ot ai 1975, 
SIlva end Montgomery 1977, Silva et el 1979e, 19T9b, 
Owigglns and Silva 1981.\] The objective of this 
research has been to provide an automated capability to 
supplement the presently largely manuel, isbor-intermive 
task of meintainlng the currency of data bases which 
derive their InformaUon elements from the text of mili- 
tary rues=ages. Although some effort has been devoted 
to primarily Interactive approaches to the problem, and 
to mwmages which have highly predictable cniumnar 
summary formats, the majority of the research and 
development work has concentrated on the more difficult 
task of analyzing unformatted narrative l:axt with user 
Interaction limited to occasional sssistmlca to the 
automated system. 
A teethed system called MATRES has been constructed 
In Prolog to run under the UNiX operating system on the 
POP 11/70. MATRES Is a knowledge based system for 
undorotanding the natural language text of event- 
oriented massages tn the domains of air activities and 
space/missile (S&M) activities. The knowledge struc- 
tures In MATRES, called "templates a, are essentially 
frames or scripts describing entities and events, which 
answer the military u~er's bsolc questions about these 
1. This work has been carried out under the 
sponsorship of the Rome Air Development Center 
(RADC), U. S. Air Force Systems Command, Grlffiss Air 
:'~,-~e 8aBe, New York. 
MlenmmmL as Illustrated In the simplified view of an 
event template prlmentod In Figure 1. 
The templatee are hierarchically organized; lower level 
templates deal with ot)Jects or times, mid level with 
evarlts containing objects and times, higher level with 
activities composed of events. The slots in the tam- 
plates contadn procedures which operate upon the out- 
put of the Definite Clause Grammar (DCG) to Instantlata 
the templatlm. 
We are currently using a corpus of spprcximstaF/ 125 
messages In the S&M domain s~ a basis for developing a 
scenario for evaluation of the extended MATRES 
teatl~bed, ms well ms a teethed for a related knowledge 
baaed ayltem, the Active/Introspective information Sys- 
tem \[desCribed In Montgomery and Ruapini 1981, and 
Ruapinl 1982\] for which MATRES serves an a front end. 
The scanmlo Involves two simulated nations, the Delta 
Col~edaratton of the Atlantic States and the Epsilon 
Repobllk:. Both netlone have space programs, and each 
Is interested in monitoring the technological progress of 
the other, using their own sataillte and sensor resources 
mid those of other friendly neUon8. The sot of fifes- 
sages to be analyzed by MATRES are mainly reports of 
space and satellite launches and orbital activities of the 
Delta Confederation, which are being monitored and 
evaluated by the Epsllon RepubUc. The text o1' mes- 
sages used In the scenario has the structure and format 
of actual messages reporting on S&M activities, although 
the lexicon is substantially different. 
All discumled In several previous technical reports 
prepared under earlier contracts with RAOC (~Kuhns and 
Mantgamary 1973\], \[Silva at el 19790\], \[Silva et sl 
1979b\]), the subset ot' the English language on which 
the text Of Intelligence messages 18 based Is essentislly 
a specialized language for reporting events. Intermixed 
with factual statements reporting on entitles and 
events, however, Is much evaluative commentary. More- 
over, proms announcements of the Oeita Confederation 
are Included In the reports, and evaluative comments are 
made both about the events reported In the press 
announcements and the announcements themselves. In 
synthesizing data base elements from these messages, 
it is crucial to sort out these different levels of informa- 
tion. 
This paper defines an approach to identifying and label- 
Ing those types of Information so that they can be 
exploited In the context of MATRES and the data base 
systems which It serves. 
55 
QUESTION 
WItAT 
WHO 
WHEN 
WHERE 
TO WHOM 
~n4Y 
INFORMATION PARAMETER8 OF A PROTOTYPE TEMPt.ATE 
PROTOTYPE EVENT 
EVENT TYPE 
AGENT (OR OBJECT PLUS 
OWNER) 
TIME OF THE 
EVENT 
LOCATION AT WHICH 
EVENT TOOK PLACE 
PATIENT, OR ENTITY AF- 
FECTED BY THE EVENT 
INTERPRETATION OF THE 
EVENT 
PROTOTYPE OF 
SPECIFIC EVENT 
AIRSPACE VIOLATION 
AIRCRAFT OWNED 
BY COUNTRY C 
TIME AT WHICH VIO" 
LATION OCCURRED 
LOCATION AT WHICH 
VIOLATION OCCURRED 
OWNER OF VIO- 
LATED AIRSPACE 
PROBABLE REASON FOR 
VIOLATION EVENT 
EXAMPLE 
AIRSPACE VIOLATION 
A UGANOAN FIGHTER 
AIRCRAFT 
AT AB(~T 0200 HOURS 
ON 25 APRIL 1978 
8 MILES FROM THE KENYA 
80ROER NEAR SUAM 
KENYA 
PROBABLE RECONNAISSANCE 
MiSSiON 
Flguro 1, Four Aspects of an Event Template. 
2.0 THE RESEARCH PROBLEM 
/Ul noted above, the subiengusge of the message car- 
pore for the air sotivitJes and S&M domains Is essentially 
a vehicle for ~ng events. Kuhns and Montgomery 
\[1978\] presented • detailed .:ethodology for classifying 
the various types of events described In the .:essages, 
which Is shown In Figure 2, and summmtzed below. 
Before describing the event clmlflcatlon, however, It Is 
enlightening to review briefly soma example .:esaages In 
order to understand the motivation for this rather com- 
plex model 
Some mmsgas w for example, those encountered In 
our previotm research on the air activities domain -- may 
report only primitive events. However, as noted shove 
and IlluatTerted in Figure 3, s message may in fact be a 
report of • report-- that Is, It may Include e report of an 
event by so.:e other source than the originator of the 
message. The "announcement m Is thus a row0rt of • 
• launch m event, which Is the basic or prlmiUve event 
being reported. The mmnrmunoementm \],., an event, but It 
Is clearly not on the same level as the primitive event. 
Rather, it is s report about the launch, s .:eta-event 
that Incidentally introduces s new Infatuation source of 
different credibility than the originator of the .:esaege. 
However, this dlsUnotJon alone Is not sufficient to 
account for the difference between the Initial two sen- 
tencea of the sxa.:pia message end the third sentence, 
whloh contains an evaluation of the announce.cent, stat- 
Ing that It wu characterized by "routine" wording. It Is 
thus an evalueUve com.:entary on the press snrmunce- 
,:ant of the launch event Since the announcement has 
been defined am s .:eta-event, the comment represents 
another mete-leveL In fact, In reviewing additional 
example8 of the .:emsege traffic in this scenario, it Is 
clear that, In order to accurately distill and represent 
Information contained in the text of these messages, the 
enalyl:lcal .:ethodo~ogy must Identify end uniquely label 
the following types of information: 
Mate-events 
llon.-meta-evm~bn 
OIbsel, vational events 
Primitive events 
AtUtbutlva events 
Relational events 
World point events (location events) 
World point qualification events 
(location event qualifications) 
N(m-world point events, or events involving 
or more objects or locations 
Figure 2. Classification of Events for the Message Sublanguaga 
56 
NYI" ANNOUNCES LAUNCH OF TERREX 588 AND TERREX 580. 
AT 1087Z, NYT-THE DELTA NEI$ AGENCY-ANNOUNCED THE LAUNCH OF 
TERREX 588 ANO TERREX 589, THE MAN RELATED SPACECRAFT LAUNCHED FROM 
THE BOGOTA MI881LE TEST RANGE AT OSOOZ ON 17 NOVEMBER. THE ANNOUNCE- 
MENT CONTAINED ROUTINE WORDING ANO STATED THE ES1/'8 WERE PLACED iN 
ORBIT BY A SINGLE CARRIER ROCKET. ORBITAL PARAmI= i ~R8 CONTAINED IN 
THE ANNOUNCEMENT ARE, 
APOGEE KILOM= u =AS 
PERIGEE KILOMETER3 
INCUNATION . DEGREES 
I~ 1. Ixepio SiN lq~iqo. 
• -factuw stetemon~l a~ events made by Epai- 
Republic reporters, 
• dlgfaen of bailer In ~ ststamento, 
• evaluations of events and predlctl~ta of future 
events, 
• do~'eea of belief In those statemont~ 
• factual statements about events made by the 
Delta Presu, 
• degree of bailer of the Epaikm Republic rel~urt- 
ors In such statements, 
• prediction8 of future events In the press 
m~nouncomen~ 
• evaJuatlona of evento reported and predlctlmls 
made by the EpWIon RepubUc reporters. 
a.o AMALYTICAL MLrrHoDQLOGY 
~mdyticaJ methodology IS baaed on • model of the 
mW world situeth=n which underlies the scenmto. Part of 
the modet derives from the event clmificaUon schema 
mentlmled shove, and Illustrated in Figure 2, which was 
deveiooed to account for the levels of content occurlng 
in the event-oriented message disooume, in this clan- 
slficetion, there are two major types of events, mete 
meats and n~t-mete event I. Of the letter, events may 
bo al~lorvartkNlei, Or prlmltive. An obaervetlonaJ evenf~ 
la a direct perception of an event, which may be a visual 
percopUon (e.g., "observe", "sightS), Or In Ute cane of a 
sensor, an electronic measurement of the emitted 
energy charanterizlng the event. A primlt|ve event is 
thus a physical event of some kind which does not 
Involve an obaervaUon Or per~epUQn. Primitive evento 
may be attributive Or relatloeai. An attribuUve event 
doacriben a situation In which a particular entity hag • 
partiooiar attribute at a certain time or during a particu- 
lar time Interval (other than the attribute location, which 
Is coveted under relational events), for example: WTer- 
rex 534 operates In the high density mode". A relational 
event Involves entitles which stand In an n-sty relation 
with oath other st a certain Urea or during s fixed time 
period. The Importance of the subclasses of world point 
and world point qualification event8 la in defining the 
world lira of sn entity, say the track cf s ship or subma- 
rine. Of these distinctions, the moat relevant for this 
discussion ere those Involving mete-events and me- 
mete events, and of the latter, primiUve vef=us obaer- 
vetkmad events. 
In torero of the scenm'io described aDQve, a primitive 
ovwrt many occur, say, • aatelllte launch by tllo Delta 
Confederation, am Illustrated In Figure 4. This event, like 
any other event, Involves the emission of energy. Such 
in omission 18 perceived by a sensing device of the 
El~ilml RopubllG. The device genefetoe (down arrow) a 
report of the giv|m event, In terms of the partlcuisr 
attributes of the event It la designed to measure. This 
sefmer report Is an obaervetlonaJ event, entailing an 
obaervetlml of a primitive evonL An S&M analyst for the 
Repubilk: ancemles (up arrow) this report, which 
contains digitized Information generated by the sensor, 
Ilttefll~to this Information ~1 • launch event, and Issues 
his own report about that event. 
HI8 rel~rt, which la an IntsrpretaUon of the primitive 
event baned on the obaervetionW event, Is s zeroth 
m~lm. mete-event= the common denominator of the rues- 
mlge traffic. At the same time, the Deltas may release 
an Internal report about the launch, which would also 
(mnstltlxte s zeroth order mote-event. Based on that 
aport, the Delta prim agency, NYT, may Issue an 
snnounomaent of the primitive event, the snnmmoemsnt 
thus e~nwUtutea s f~der mete-event. An Epsllon 
Republk: rq~rter may then make an Interpretation of. 
that ennota~enent~ In the form of a report, which 
being a report of 4 first-order mete-event -- Is there- 
fore m aecmtd ~ mete-event. Corrections or other 
changes made by El=~lllon reporters to these messages 
constitute a third mete level of reporting event, since 
they may reference reports of reports of events. 
The m(xlet thul fair accounts for the event reporting 
structure which undmtioa the Delta/Epsilon scenario, but 
we must slam acm~alt for the repo~ar's comments about 
the event -- Le., his interpretation or evaluation of the 
event -- which can occur at any of these levels. 
The moortor~s goaJ Is to Identify and denoribe all the 
relevant parameters of an event (exemplified by the 
~ots In the template for a launch event, shown in the 
center of Figure 5) based on the observational report 
produced by the sensor and any other information he 
may hive (e.g., knowledge that a replacement of s non- 
functioning communications satellite is likely within a 
given time frame). However, if the reportor's Information 
Is Incomplete or Imprecise, he cannot exactJy describe 
the parameters of an event, but will give his best 
Interpretation of the event baaed on whet he knows. 
Thus he may report a launch of "an unidentified satel- 
lite", "a probable television support satellite", "s possi- 
ble CE satellite". In some cases, ho may have enough 
Information to make a comperaUve evaluation with launch 
events which have occurTed In the pest: "a new ESV", 
"the second CE satellite to be successfully orbited by 
the Deitim this year". SUil another type of mete Infer- 
57 
OELTA CONIFEDERATIQN EPSlLON REPUBUC 
I 
II f"'--'°~ ~ olr Obe. E~Int 
I ~ Ol~mrvaIJonal & 
I 
I 
Ideta Event 
Prmm Annauncement I I 
of Prlmtlvt Event i~~1 ~ 
I 
I 
t 
I 
I 
Immq~rg1~m1~a~' 
P~ Announcement 
Om Order Idet~ Event 
k L 
2rid O~llr Meta Event 
,,,,u 
3rd Order I ~ Svent~ 
ChmnQes. 
UpdalLlm to Mite Evenl~ 
RgUrO 4. Modet for Orders of Metro Evenl~l 
58 
sauce 18 exemplified by the last sentence In the sample 
memmge presented above: the assignment of an Iden- 
UflcaUon number to s space object. 
If a reporter'a InformaUon is good, I.e., complete and pre- 
cise, the fo4iowing type of launch report i8 produced: 
the s~Jrce 18 the actual originator of the message. 
Thus, In the cane of the "designate: mete template, the 
mlnfoamm=e" of the designation Information (i.e., that the 
particular satellite launched from that site st that date 
end time hen been designated s space object called 
Meg. g4-OOg 
VARIABLE DENSITY CE AGSAT LAUNCH, 20 SEPTEMBER 1983 
A VARIABLE DENSITY CROP ENHANCEMENT AGRICULTURAl. SATELLITE 
WAS LAUNCHED FROM THE HARRI88URG MJS.51LE AND 8PACE COMPLEX (HM8C) 
AT 1808Z ON 20 8avur.MBER 1988. THE BR02 LAUNCH SYSTEM WAS U8ED TO 
PLACI THE ~TELJJTE INTO A 3 DEGREE ORBIT. WSJ OBJECT NUMBER 
8018 HA8 BEEN ASSIGNED TO THE PAYLOAD, 
However, when hi8 Informat~n Is Impremiao and hie 
Imowledge can add little to it, he must rmmrt to the 
qualified or mota-Gommented typos of messages 
dlm(~rJbed above. 
In order to accommodate such qualified end mete- 
~mmectary types of Informatkm, each event templets 
may have emaciated with It one or more mote templates 
contA~ing Interpretive or evaluative Information. Thus, 
88 represented In Figure 5, an Imitantiated launch tent- 
plate produced frmn an obaervaUonal event and a primi- 
tive event (a zeroth order reporting event, ea illustrated 
in Figure 4) may have several additional quallficetlorm 
(exemplified by, hut not Ilmlted to, the mete templates 
llh~streted In the figure). So, for example, a meta 
evaltmttve template 2 mmoclated with • laundl template 
exl=reales the Epalion reporterJs degree of belief or 
m=nfidenco In the launch parameters he rep(x~a: the 
object in the event template 18 belaeved by the Epalion 
reporter to be a CE (or Crop Enhancement) satellite from 
the Infometion presented In the observational relx~ by 
the sensor, end from hht owvl knowledge of pest 
occurrences of CE satellite launches, as wail as expea- 
teth:me of pG88iblc replacement ieunchee, etc., during 
p~rtlculer Use Intervals. All or none of the listed 
psi'easters for • launch event may be qualified In this 
way. Thus, in Figure 5, the Epalk=n reporter believes 
that, to the best of hhl knowledge, the space object 
Involved In the major launch event 18 a "probable" CE or 
crop enhancement satellite, and that the time of launch 
IS "approximate\[y" 113OZ. 
Each mete template ha8 field8 which Identify the source, 
am weft ms the Use and date of the Interpretive Informa- 
till. As opposed to the "infosouroe= parameter of these 
mote templates -- which shows the ulUmate source of 
the Information contained in the Instantlatad templets -- 
2. The template and mats template structures shown in 
this figure are Intended to be Illustrative only: for 
example, the object, date/time group, and deorbit 
information constitute embedded templates linked to 
the main "launch" event template by pointers. In 
addition, there are several alternatives for more 
economical Internal representation of mete template 
Ioformation, which are currently under review for the 
~'~uai design and Implementation o1' this information 
within MATRES and within the Actfvs/Irrm~pective 
InformaUon System (a knowledge-based Intelligent 
88alatsnt, as menUonod above), which MATRES 
feeds. 
• Termx 584") Is NY'r, the new8 agency of the Delta 
Corffederetiofl, Indicating that this InformaUon came from 
in NYT proml wmounoement quoted (and Interpreted) by 
an Epalion reporter. This distinguishes such Information 
frmu that relmremented by the "mmignn mete template, 
where the Epellon Republic reporting staff ~8algn an 
Identmcetlon number of their own to the satellite pay- 
load far futwe referMme. 
Another significant analytical ~ of the Epallon reporter 
In thio suemmlo 18 the ¢oem=eratlve evaluation, XUuatrstod 
by the w~'omperQm mete template. These comparisons 
Involve overate which have takes place before, in thhs 
¢:880, ImJnch evento, and/or obJecto Involved In such 
htun~ea. Ao In the example shme. In Figure 5, the com- 
perim~ nuty specify an ewmt Involving the continuation 
of • satellite In am active status, where other such 
israelite8 an now Inactive (Jew\[led compsr18on): e.g., 
~'oft~Jx 584 141 the on~ first generation crop enhance- 
met sutoUlta which 18 mm~mtiy active." 
An I~ fINtOUolt of mite templates Is to represent 
pimdkrtJve In/~N~tJons I.o., descriptk:m8 of events 
expe,~ed In the future, based on other events which 
have ocmHyed In the pelt, or are currently In process. 
The "expect e template in Figure 5 expresses the 
presumable or expected pMsJlrleters of mission duration, 
and conaoqueRtly, the deorbit event which 18 mltlclpated 
for (~tober 2~ 
To summarize, the function of the mete templates Is to 
Identify and delimit evaluative commentary, which iso- 
lates the factual InformaUon presented In most zeroth 
order mote event reports, and Identifies Information per- 
taining to credibility of the event occurrence, compere- 
bllll~/ with otilcr aim\[Jar entitles and events, prediction8 
of fUtlKe related ovento, etc. 
On the other hand, In addiUon to distinguishing the vsrl- 
o4J8 Iovehl of event occuffenGe, observation, and report- 
Ing, the fonotJon of the mete event structure Illustrated 
in Figures 2 and 4 Is to clearly demarcate the "Oetta 
versus Epallon" (in terms of the scenario described 
above) aspects of the messages. The reporters of the 
Epailoll Republic "assign" "Specold" and "WSJ" IdeITtlfl- 
cation numbers for space object Inventory purposes; the 
De\[iN "denlgnate" their own apace objects with par*,Jc- 
tdaJ, clanaea of object names, e.G., "Tarrex 5~59". They 
• launch", "put Into orbit", "deorbit", "recover", etc., 
while the Epailon reporters "assess", determine 
"active" vs. "Inactive" cactus, attribute satellite 
59 
m 
.. ¢) U 'I~ ~; .. 
~, = ~ :,.=..=. ~'~ .. \[i \[i-=l?, , 
iii el v ~ ll ,111 ,Ul *11 
14 
~=.~ 
--- '~,- u cl..l~ .. 
=I ~I"' 11414141 _~ ~ ~ • 
-~'a:o oli~.ii!i 
ii oEq, 
i 
~1.. ~..~ ~g | ~ ~ I~ ~:~. .~1~ ~ ~=~. = ~= I I: =~=. ;~ 
IJ Q.*. ~, Q. II III "11 I~ r "" .... a. - . II ". ~ 
t. m m Q U Q,~ .* - ,. , ~, "I~ .- 
~1~_~ ~ ~- -~ .', =----=-~-='~1-" =- -i" I m ~ - ~ '=- .'-" _~i 
r~ 
60 
=programs = and "medntenance" of such progreums, at(:., to 
the Delten. 
Alll~1ough the detailed ImplementaUon of some of theme 
noUon= remedna to be worked out -- In perttGular, the 
Interflclm with the Acthte/InU'oepectJve Informatlon 
System -- we expect that the essentleda of the analyl;I- 
r.4d ~oproech dls(:u~sed In this paper will be demonstr- 
able on the MATRES teethed toward the end of 188;). 

References

O~llik~ D. L, end G. Silva \[1981\] 
A/8 Data 8ass Generation. 
RADC-TR-81-4;), Rome Air Development Center, Grilffl=m 
Nr F~,-~e 84me, New York, April° 1981. 

Kuhn=, J. L~ end C. A. Montgomery \[1973\] 
EM Roco~ SpeciflcatJon System C~rlcep¢# 
Preliminary Notlon~ 
Operating Systema, Inc., N73-007, August, 1973. 

r~,,,. J. L \[1==;,=\] 
Syntaeele of Inference rectmique~ An InterprateO 
Symax for the Logic=# De.gcrl pUon of Events. 
Operating Systems, Inc., N74-003, May, 197& 

Kahn=, J. L, Montgomery, C. A., end O. K. Whelched \[1975\] 
ERGO -- A System for Evem Record Generalon w~ 
O¢gentzatlon. 
RAOC-TR-75-61, Rome Air Oewdopment Center, @riflqs= 
• Jr ~'~ce S~m, New York, March, 1975. 

Montgomery, C. A., and F. H. Ruspini 1981
The Active Information System A Data-Driven System for 
the Analysis of Imprecise Data 
Proceedings of the Seventh International Conference on 
Very Large Data Bases, pp. 376-385, IEEE Computer 
Society Prima, 1981. 

Ruspini, Endque H. \[1982\] 
Possibili W Theory Approe~he: for Advanced InformeR/on 
S'yatwns. 
IEEE Computer, Volume 15, Number 9: pp. 83-91, 
Seotembar, 1982. 

Silva, G., end C. A. Montgomery \[1977\] 
Autom~eO I&W File Generation. 
RAOC-TR-77-194, Rome Air Development Center, Grlfflss 
Air Force B~se, New York, June, 1977. 

Silva, G., Owlggins0 D. I.., end J. L Kuhns \[1979a\] 
A Knowledge-Based Automated Message Underst~lng 
Matltodology for an A@vanced /r, glcat/ons System. 
RAOC-TR-79-133, Rome Air Development Canter, Grlfflse 
Air Force Base, New York, June, 1979. 

Silva, ~, Owigglns, D. L, end C. A. Montgomery \[1979b\] 
Satellite etxY Missile Data Generation for AIS. 
fined Technical Report, Operating Systems, Inc., R79-037, 
