SRA:
DESCRIPTION OF THE SOLOMON SYSTEM AS USED FOR.
MUC- 5
Chinatsu Aone, Sharon Flank, Doug McKee, Paul Kraus e
Systems Research and Applications (SRA )
2000 15th Street Nort h
Arlington, VA 2220 1
aonecQ sra.coin
BACKGROUN D
SRA used a language-independent, domain-independent, multipurpose text understanding system as the core
of the MUC-5 system for extraction from English and Japanese joint venture texts . SRA's NLP core system ,
SOLOMON, has been under development since 1986 . It has been used for a variety of domains, and wa s
aimed from the start to be language-independent, domain-independent, and application-independent . More
recently, SOLOMON has been extended to be multilingual, beginning with Spanish in 1990 and Japanese i n
1991 . The Spanish-Japanese text understanding system that uses SOLOMON was developed for a dornai n
very different from the MUC-5 joint venture domain (cf . Aone, et al . [2]).
SOLOMON's principal applications have been in data extraction, but it is also used in a prototyp e
machine translation system (cf. Aone and McKee [5]) . The domain areas in which SOLOMON application s
have been developed are : financial, terrorism, medical, and the MUC-5 joint-venture domain . SRA has
significantly enhanced its capability to add new domains and languages by developing new strategies fo r
data acquisition using both statistical techniques and a variety of user-friendly tools .
MUC-5 SYSTEM ARCHITECTUR E
SOLOMON employs a modular, data-driven architecture to achieve its language- and domain-independence .
The MUC-5 system, which uses SOLOMON as a core engine, consists of seven processing modules an d
corresponding data modules, as shown in Figure 1, which will be described in the following sections .
Message Zoner
The Message Zoner uploads the SGML-annotated text file into the data extraction system . Input files are
assumed to have been proprocessed so that they contain only "rigorous markup" (cf. Goldfarb [8]) SGM L
tags and text ; however, we do not require sentences or paragraphs to be tagged. Japanese text is assume d
to be encoded in EUC, but tags must be ASCII .
All input, including tags, is tokenized using a simple, language-independent, regular expression recognizer .
The (multi-word) tokens are parsed into sentences, paragraphs, headers and documents using a simpl e
operator-precendence grammar (cf. Aho, Sethi and Ullman [1]) operating on punctuation and tags . The
tokenizer and parser are written entirely in lex .
207
Figure 1 : MUC-5 System Architecture
Sentence and paragraph boundries are inferred using a conservative algorithm and marked as inferred .
Inference is not performed if sentences and paragraphs are rigorously marked . The output is piped to a
post-processor, which does a fast lookup of each word in a btree gazetteer, and includes entry information
in the tokens of place names .
Preprocessin g
Preprocessing consists of two processors, the morphological analyzer and the pattern matcher, and associate d
data in the form of morphological data, lexicons, and patterns for each language. Its input is a tokenized
message, and its output is a series of lexical entries with syntactic and semantic attributes .
Declarative morphological data for inflection-rich Japanese and Spanish is compiled into finite-stat e
machines . The English domain lexicon was derived from development texts automatically, using a statistica l
technique (cf. McKee and Maloney [10]) . This derived lexicon also contains automatically acquired domain -
specific subcategorization frames and predicate-argument mapping rules called situation types (cf. Aone an d
McKee [3]), as shown in Figure 2 .
Pattern recognition handles a wide range of phenomena, including multi-words, numbers, acronyms ,
money, date, person names, locations, and organizations . We extended the Pattern matcher to handle multi-
level pattern recognition . The pattern data are divided into ordered multiple groups called priority groups,
and the patterns in each group are fired sequentially, avoiding recursive applications as much as possible .
This extension speeded up the performance of Preprocessing significantly .
Syntactic Analysis
The processor for Syntactic Analysis is a parser based on Tomita 's algorithm (cf. Tomita [11]), with modifi-
cations for disambiguation during parsing . Syntactic Analysis data consist of X-bar based phrase structur e
grammars and preparse patterns for each of the three languages, English, Japanese, and Spanish . Syntactic
Analysis outputs F-structures (grammatical relations), along the lines of Lexical-Functional Grammar (cf .
Bresnan [7]), as shown in Figure 3 . The Semantic Interpretation module is interleaved for disambiguatio n
208
(SWIM
((CATEGORY . V)
(IDIOSYICRACIES (THEME (MAPPING (LITERAL WITH)))) ; "swim with the big fish"
(OCCS 11)
(PREDICATE ANIMATE-OBJECT-ACTIVITY)
(SITUATION-TYPE ACTIVITY)) )
(STEP
((CATEGORY . V)
(IDIOSYNCRACIES (SOURCE (MAPPING (LITERAL FROM)) )
(GOAL (MAPPING (LITERAL 01 INTO))) )
(OCCS 36)
(PREDICATE CHANGING-EVENT)
(PROB 8.1 . 1)
(SITUATION-TYPE ACTIVITY)) )
(TEAM
((CATEGORY . V)
(IDIOSYICRACIES (THEME (MAPPING (LITERAL WITH))) )
(OCCS 31)
(PREDICATE ANIMATE-OBJECT-ACTIVITY)
(SITUATION-TYPE PROCESS -CAUSED-PROCESS)) )
(SWITCH
((CATEGORY . V)
(IDIOSYICRACIES (SOURCE (MAPPING (LITERAL FROM))) )
(OCCS 161)
(PREDICATE TURNKEY-CHANGE)
(PROB 2.1 . 1)
(SITUATION-TYPE CAUSED-PROCESS)) )
Figure 2 : Statistically Acquired Lexical Entrie s
of prepositional phrase attachment, conjunctions, and so on, by calling semantic functions, which are share d
by all three languages, from inside the grammar .
Preparsing takes the burden off of main parsing and increases accuracy, by recognizing structures such a s
sentential complements, appositives, certain PP's, etc . by pattern matching, and sending these to the parse r
as chunks. These preparse chunks are parsed prior to main parsing using the same grammars, and their
output consists of F-structures as well .
• Appositives: Or i~ "industry's largest Tokyo Kaijou "
• Sentences with certain verb endings:
' 7 X . ]I ~ .
	
WE
	
. I
• PP's: start production [in january 1990] with production of 20,000 iro n
In order to test the progress of grammar development and pinpoint trouble spots, automatic evaluatio n
of grammars was used . SRA adapted the community-wide program Parseval (cf. Black, et al . [6]) for use
in Japanese in addition to English . Testing on Japanese was limited, since there are not many brackete d
Japanese texts to use as answer keys .
Semantic Interpretatio n
Semantic Interpretation uses a language-independent processing module, and its data are predicate-argumen t
mapping rules for each verb, plus both core and domain knowledge bases . Semantic Interpretation work s
209
BRIDGESTONE SPORTS CO . SAID FRIDAY IT HAS SET UP A JOINT VENTURE IN TAIWAN WITH A LOCAL CONCERN AND
A JAPANESE TRADING HOUSE . . .
[ST: <S>
SUBJECT: [ST: <NP>
HEAD: IT]
PREDICATE: [ST: <VP>
TENSE : PRESENT
ASPECT: PERFECT
PREDICATE: (CREATE)
ROOT: SET
VERB-PARTICLE : UP]
OBJECT: [ST: <HP>
HEAD: A-JOINT-VENTURE]
PREP-ARGS: ([ST: <PP>
MARKED: WITH
HEAD: A-LOCAL-CONCERN-AND-A-JAPANESE-TRADING-HOUSE] )
ADJUNCTS: ([ST: <PP>
MARKED: IH
HEAD: TAIWAN])]]]
Figure 3 : Simplified F-Structure Output by Syntactic Analysi s
off of language-neutral F-structures in order to handle all the languages. It outputs semantic structures, i .e.
predicate-argument and modification relations, as shown in Figure 4 . The predicate-argument mapping rule s
(i .e. rules which map F-structures to semantic structures) are acquired automatically (cf . Aone and McKee
[3]) . Domain knowledge bases, on the other hand, were acquired manually . However, a new rapid knowledg e
acquisition tool called KATooI was used to link a lexical entry to its corresponding semantic concept in th e
knowledge bases (cf. Figure 5) .
If a full parse cannot be created, SOLOMON uses a fragment combination strategy . Debris Parsing
and its subsequent process, Debris Semantics, work together to obtain the best interpretation from sentence
fragments. They use as data the grammars and knowledge bases, and they output semantic structures jus t
like when a full parse is created . Debris Parsing retrieves the largest and most preferred constituents from
the parse stack . It then reparses the rest of the input, and creates debris F-structures with the best fragmen t
constituents. Debris Semantics relies on the semantic interpreter to process each fragment, and then fit s
fragments together using semantic constraints on unfilled slots .
Discourse Analysis
Discourse Analysis, which was redesigned and implemented this year (cf . Aone and McKee [4]), performs
reference resolution . Discourse Analysis uses a data-driven architecture to achieve language-independence ,
domain-independence, and extensibility . It employs a single language-independent, domain-independen t
processor, and several discourse knowledge bases, some of which are shared among different languages . The
output, of Discourse Analysis is a set of semantic structures with coreference links added, i .e. File Cards
(cf. Heim [9]). Discourse phenomena handled for the joint venture domain include name anaphora (e .g.
[ST: <S>
SUBJECT: [ST: <HP>
HEAD: BRIDGESTONE-SPORTS-CO .]
ADJUNCTS: ([ST: <NP>
HEAD: FRIDAY] )
[ST: <VP>
TENSE : PAST
PREDICATE : (COMMUNICATE)
ROOT: SAY
SENT-COMP :
PREDICATE :
210
BRIDGESTONE SPORTS CO . SAID FRIDAY IT HAS SET UP A JOINT VENTURE I I
TAIWAN WITH A LOCAL CONCERN AND A JAPANESE TRADING HOUS E
(COMMUNICATE-1176 (ISA (VALUE (COMMUNICATE)) )
(TIME (VALUE (FRIDAY-1178)) )
(AGENT (VALUE (COMPANY-1146)))
(THEME (VALUE (CREATE-1163)) )
(TENSE (VALUE (PAST))) )
(COMPANY-1146 (ISA (VALUE (COMPANY)) )
(QUANTITY (VALUE ((EXACT 1))) )
(UNIT (VALUE (NATURAL-UNIT)) )
(JAMES (VALUE ((BRIDGESTONE SPORTS CO)))) )
(CREATE-1163 (ISA (VALUE (CREATE)) )
(LOCATION (VALUE (COUNTRY-1144)) )
(AGENT (VALUE (THING-1166)) )
(THEME (VALUE (TIE-UP-EVENT-1164)))
(CO-THEME (VALUE (CONJOINED-COLLECTIOI COMPAIY)-1172) )
(ASPECT (VALUE (PERFECT)))
(TENSE (VALUE (PRESENT))))
((CONJOINED-COLLECTION COMPANY)-1172
(ISA (VALUE ((AID CONJOINED-COLLECTION COMPANY))) )
(HAS-MEMBERS (VALUE (COMPANY-1170 COMPANY-1168))) )
(COMPANY-1168 (ISA (VALUE (COMPANY)) )
(QUANTITY (VALUE ((EXACT 1))) )
(UNIT (VALUE (NATURAL-UNIT)) )
(LOCATION (TYPE (AND T PHYSICAL-LOCATION)) (VALUE (LOCAL))) )
(COMPANY-1170 (ISA (VALUE (COMPANY)) )
(QUANTITY (VALUE ((EXACT 1))) )
(UNIT (VALUE (NATURAL-UNIT)) )
(NATIONALITY (VALUE (JAPAN))) )
(COUNTRY-1144 (ISA (VALUE (COUNTRY)) )
(ENGLISH-GAZ-STRING (VALUE (Taiwan (COUNTRY)))) )
Figure 4: Semantic (Predicate-Argument) Structure
3\ v\~ .\J\l :a~~~il:X25.
	
\>
	
; x ,3 :33\MAt'VY
Figure 5: Knowledge Acquisition Too l
211
DISCOURSE : Classified $<DISCOURSE-MARKER DISCOURSE-MARKER-181>("BRIDGESTONE SPORTS") as DP-NAM E
DISCOURSE : Found an exact match ,
ante: $(DISCOURSE-MARKER DISCOURSE-MARKER-83>("BRIDGESTONE SPORTS CO .")
ref: $<DISCOURSE-MARKER DISCOURSE-MARKER-181>("BRIDGESTONE SPORTS" )
DISCOURSE : Classified $<DISCOURSE-MARKER DISCOURSE-MARKER-206>("BRIDGESTONE SPORTS") as DP-NAM E
DISCOURSE : Found an exact match ,
ante: $<DISCOURSE-MARKER DISCOURSE-MARKER-181>("BRIDGESTONE SPORTS" )
ref: $(DISCOURSE-MARKER DISCOURSE-MARKER-206>("BRIDGESTONE SPORTS" )
Figure 6 : English Discourse Trace Exampl e
=> IMLEA:%glIg
I)ISCOURSE : Classified #<DISCOURSE-MARKER DISCOURSE-MARKER-511>( "
	
1 #
	
, .Z*k.") as DP-NAME
DISCOURSE: Found an exact match ,
ante: #<DISCOURSE-MARKER DISCOURSE-MARKER-248>("
	
1
	
`
	
")
ref: #<DISCOURSE-MARKER DISCOURSE-MARKER-511>("at :t .")
*A 14 => ni
	
F7 ~
l.)ISCOURSE : Classified #<DISCOURSE-MARKER DISCOURSE-MARKER-573>(" 1#AE") as DP-NAME
DISC(.)URSE: Found an exact match ,
ante : #<DISCOURSE-MARKER DISCOURSE-MARKER-511>("
	
" )
ref: #<DISCOURSE-MARKER DISCOURSE-MARKER-573>(' , E")
Figure 7 : Japanese Discourse Trace Exampl e
"BRJl)GESTONE SPORTS" for "BRIDGESTONE SPORTS CO .") and definite NP's such as "THE NE W
('OMPAN l
The system traces for English and Japanese walkthrough examples are shown in Figure 6 and Figure 7 .
In the English example, the two instances of name anaphora for "Bridgestone Sports Co." are recognized ,
while in the Japanese example, all the references to "Tokyo Kaijou Kasai Hoken, " including appositives, ar e
resolved .
Pragmatic Inferencin g
Pragmatic Inferencing performs reasoning in order to derive implicit information from the text, using a
forward chainer and inference rules . Pragmatic Inferencing outputs semantic structures, with inferred infor-
inat ion added . It infers additional information from "literal" meanings as required for application domains .
For instance, in the walkthrough example, in order to infer "THE TAIWAN UNIT " is a joint venture
company frorr, the phrase "THE ESTABLISHMENT OF THE TAIWAN UNIT" the following rule is used .
(defrule rule-0009 ((?event) (?event) )
:example ("PNI and SRA established a new company .")
:if (and (establish ?event )
(theme ?event ?x )
(company ?x) )
:then (and (tie-up-event ?event )
(joint-venture-company ?x)
(joint-venture-company ?event ?x)
(in-jv-event ?x ?event)))
212
It is easy for developers to add, change or remove inferred information due to the declarative nature o f
the inference rules . For instance, to get an additional tie-up from "Company A and Company B tied wit h
Company C " , in ,t,ty-000''2, we just, had to add another rule to infer that. when companies "tie," they form a
tie-up.
(defrule rule-0017b ((?event) (?event))
:example ("PNI tied with SRA")
:if (and (tie-event ?event )
(not (theme ?event ?z) )
(agent ?event ?x )
(company ?x)
(co-theme ?event ?y)
(company ?y) )
:then (tie-up-event ?event) )
Extract
The Extract module performs template generation, translating the domain-relevant portions of our language -
independent semantic structures into database records . We maintain a strong distinction between processin g
and data even in template generation . Thus, we use the same processing module to output in differen t
languages and to several database schemata, including to a flat template-style schema as in MUC-4 and t o
a more object-oriented schema as in MUC-5 .
To do the actual template filling, we rely on Extract data made up of kb-object/slot to db-table/fiel d
mapping rules and conversion functions for the individual values (e .g. set fills, string fills) . For example, th e
#nationality slot of an #ORGANIZATION object in our knowledge base corresponds to the Nationalit y
field of the Entity object in the MUC-5 template .
REUSABILITY OF THE SYSTE M
SOLOMON is designed for reusability . Each processing module is data-driven and reusable in other lan-
guages and other domains, as well as in applications other than data extraction (e .g. machine translation ,
abstracting, summarization) . A large portion of the data is also reusable in :
• Other languages and domains
- Core knowledge bases
• Other domains
- Morphological data
- General lexicons
General pattern data (e .g . date, location, personal name, organization name )
Grammars
Some of the discourse knowledge sources
• Other language s
- Domain knowledge bases
213
Figure 8 : Reusability of SRA 's MUC-5 System
– Some of the discourse knowledge sources
– Inference rules
– Extract (template generation) dat a
The data acquisition tools and techniques are also reusable in other languages and domains . The statis-
tical techniques used to derive lexical information can be reused for other domains . LEXTooI, the lexicon
acquisition tool, is multilingual and relies on system data files for category and morphological informa-
tion. KBTooI, the knowledge base acquisition tool, is language-independent just as the knowledge bases ar e
language-independent . KATool, the knowledge acquisition tool that links lexicon entries with the appropri-
ate knowledge base concepts, is entirely data-driven as well, and is therefore completely reusable . Figure 8
summarizes the reusability of SRA 's MUC-5 system.
TEST RESULTS AND ANALYSIS
Our MUC-5 results for the English and Japanese joint-venture domain task are shown in Table 1 . We spen t
10 .55 person-months for this task, most of which were devoted to data development for both languages (se e
Table 2) . The "other" category includes time spent on developing language-independent data such as a
joint-venture domain knowledge base, pragmatic inference rules, and Extract data for template generation .
We believe that the results do not indicate the potential of our system, since the system performance fo r
both languages was still improving after five months of development. Much of the work we did resulted in
long-term improvements to our overall text understanding capability, all of which will ensure a stronger base
system for future applications . This implies that although the development cycle for data extraction system
using a text understanding system may be slower in its current maturity stage, the potential for such a syste m
is still unknown and represents a most promising avenue for development . We are particularly pleased wit h
the success of our Japanese system : no other Japanese MUC-5 site is using the full understanding approach ,
but we did as well and our performance continues to improve )
Staff time was the major limiting factor . We needed more time to perform more testing and evaluation
l In the 18-month Tipster evaluation, the highest JJV F-measure was about 40 .
214
Englis h
ERR UND OVG SUB REC PR E
ALL OBJECTS
MATCHED ONLY
TEXT F ILTERIN(.
80
48
-
66
2 8
25
2 6
8
7
34
2 3
-
2 2
5 6
74
4 9
7 1
9 3
I'&R 2I'&R I'&2H.
F-MEASURE 30 .80 39 .56 25 .22
Japanese
ERR UND OVG SUB REC PR E
ALL OBJECT S
MATCHED ONLY
TEXT FILTERING
70
43
-
5 3
2 8
6
34
9
1
2 0
1 4
-
38
6 1
_
	
94
5 2
78
98
P&R 2P&R P&2R
F-MEASURE 43 .92 48 .74 39 .97
Table 1 : SRA 's Scores for the English and Japanese Joint Venture Domai n
task person-months
EJV 3. 2
JJV 2. 2
Testing 1 . 5
Documentation 0 .25
Other 3.4
Table 2: SRA 's Time Expenditure for MUC- 5
using the scoring program, and to finely tune Extract (template generation) mapping rules . We discovered
we were hampered by formatting errors, and in addition considerable information was "understood" by th e
system all the way through, but was not extracted by the template generator. Since the discourse modul e
was new, it would have been helpful to have additional time to test and expand it . In addition, we neede d
more time to fill the OWNERSHIP, REVENUE, and TIME objects, which we simply did not output .
CONCLUSION
Overall, the data-driven architecture in SOLOMON allowed for minimum work on processing modules whe n
working on different languages and domains. We ported the system to Spanish in a week for the demonstra-
tion given, at the MUC-5 conference .
Although we successfully acquired large amounts of domain data from domain texts in both languages ,
using both statistical methods and newly developed user-friendly knowledge acquisition tools, we recogniz e
the need to move even more quickly to new domains and languages . We plan to continue our work on
automatic acquisition of lexicons, knowledge bases, and links between them in multiple languages .
Tuning performance of each module (e.g. parsing, discourse analysis) as well as the' performance o f
the whole system to a particular task more rapidly is another research issue we identified . We believe that
developing automatic evaluation and training algorithms for such automated module/system tuning is crucial
to develop a data extraction system that produces optimal results .
21 5
ACKNOWLEDGEMENTS
We are indebted to Rajeev Agarwal, Debbie Sanders, and Vera Zlatarski for their hard work and dedicatio n
in data development, module testing, and more . We also gratefully acknowledge the contributions of Scot t
Bennett, David Garfield, and Hatte Blejer to the MUC-5 process .
References
[1] Alfred V . Aho, Revi Sethi, and Jeffry D . Ullman . Compilers : Principles, Techiniques and Tools.
Addison-Wesley, 1986 .
[2] Chinatsu Aone, Hatte Blejer, Sharon Flank, Douglas McKee, and Sandy Shinn . The Murasaki Project :
Multilingual Natural Language Understanding . In Proceedings of the ARPA Human Language Technol-
ogy Workshop, 1993 .
[3] Chinatsu Aone and Doug McKee . Acquiring Predicate-Argument Mapping Information from Multilin-
gual Texts. In Acquisition of Lexical Knowledge from Text : Proceedings of a Workshop Sponsored by
the Special Interest Group on the Lexicon of the Association for Computational Linguistics, 1993.
[4] Chinatsu Aone and Doug McKee . Language-Independent Anaphora Resolution System for Understand-
ing Multilingual Texts . In Proceedings of 31st Annual Meeting of the ACL, 1993.
[5] Chinatsu Aone and Doug McKee . Three-Level Knowledge Representation of Predicate-Argument Map -
ping for Multilingual Lexicons . In AAAI Spring Symposium Working Notes on Building Lexicons fo r
Machine Translation, 1993 .
[6] E. Black, S. Abney, D . Flickinger, C . Gdaniec, R . Grishman, P. Harrison, D . Hindle, R . Ingria, F . Jelinek ,
J . Klavans, M . Liberman, M . Marcus, S . Roukos, B . Santorini, and T. Strzalkowski . A Procedure fo r
Quantitatively Comparing the Syntactic Coverage of English Grammars . In Proceedings of the Fourt h
DARPA Speech and Natural Language Workshop, 1991 .
[7] Joan Bresnan, editor . The Mental Representation of Grammatical Relations. MIT Press, 1982 .
[8] Charles F . Goldfarb . The SGML Handbook . Oxford, 1990 .
[9] Irene Heim . The Semantics of Definite and Indefinite Noun Phrases . PhD thesis, University of Mas-
sachusetts, 1982 .
[10] Doug McKee and John Maloney . Using Statistics Gained from Corpora in a Knowledge-Based NL P
System . In Proceedings of The AAAI Workshop on Statistically-Based NLP Techniques, 1992.
[II] Masaru Toinita . Efficient Parsing for Natural Language . Kluwer, Boston, 1986.
APPENDIX
A ejv-0592 SRA's Original Respons e
<TEMPLATE-0592-1> :=
DOC NR: 0592
DOC DATE: 241189
DOCUMENT SOURCE : "Jiji Press Ltd. ;"
CONTENT: <TIE_UP_RELATIONSHIP-0592-3>
216
<TIE_UP_RELATIONSHIP-0592-2>
<TIE_UP_RELATIONSHIP-0592-2> :=
TIE-UP STATUS : EXISTIBG
ENTITY: <ENTITY-0592-6>
<ENTITY-0592-5>
JOINT VENTURE CO : <ENTITY-0592-7>
ACTIVITY: <ACTIVITY-0592-8>
<ACTIVITY-0592-8> :=
INDUSTRY: <INDUSTRY-0592-9>
ACTIVITY-SITE : (Taiwan (COUNTRY) <ENTITY-0592-10>)
<INDUSTRY-0592-9> :_
INDUSTRY-TYPE : PRODUCTION
PRODUCT/SERVICE: (67 "A JOINT VENTURE" )
<ENTITY-0592-5> :=
NAME: Taga CO
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-11>
<ENTITY_RELATIONSHIP-0592-11> :=
ENTITY1 : <ENTITY-0592-5>
<ENTITY-0592-6>
EHTITY2: <ENTITY-0592-7>
REL OF EHTITY2 TO ENTITYI : CHILD
STATUS: CURRENT
<ENTITY-0592-6> :_
NAME: Union Precision Casting CO
ALIASES: "Union Precision Casting"
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-11>
<ENTITY-0592-7> :=
NATIONALITY : Taiwan (COUNTRY)
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-11>
<TIfi UP_RELATIONSHIP-0592-3> :=
TIE-UP STATUS : EXISTING
ENTITY: <ENTITY-0592-14>
<ENTITY-0592-13>
ACTIVITY: <ACTIVITY-0592-8>
<ENTITY-0592-13> :=
NAME: Bridgestone Sports CO
ALIASES: "Bridgestone Sports"
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-15>
<ENTITYRELATIONSHIP-0592-15> :=
EBTITYI : <ENTITY-0592-13>
<ENTITY-0592-14>
REL OF EHTITY2 TO EBTITYI : PARTNER
STATUS: CURRENT
<ENTITY-0592-14> :_
TYPE: COMPANY
EHTITY RELATIONSHIP: <ENTITY_RELATIONSHIP-0592-15>
B ejv-0592 SRA's Corrected Response
<TEMPLATE-0592-1> :_
DOC NR: 0592
DOC DATE: 241189
DOCUMENT SOURCE : "Jiji Press Ltd. ;"
CONTEXT: <TIE_UP_RELATIONSHIP-0592-4>
<TIE UP_RELATIONSHIP-0592-3>
<TIE_UP_RELATIONSHIP-0592-2>
<TIE_UP_RELATIONSHIP-0592-2> :_
TIE-UP STATUS : EXISTING
ENTITY: <ENTITY-0592-7>
<ENTITY-0592-6>
217
JOINT VENTURE CO : <ENTITY-0592-8>
ACTIVITY: <ACTIVITY-0592-9>
<ACTIVITY-0592-9> :_
INDUSTRY: <INDUSTRY-0592-1O>
ACTIVITY-SITE : (- <ENTITY-0592-11>)
<INDUSTRY-0592-10> :_
INDUSTRY-TYPE : PRODUCTION
PRODUCT/SERVICE : (67 "A JOINT VENTURE")
<ENTITY-0592-6> :_
NAME: Taga CO
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-12>
<ENTITY_RELATIONSHIP-0592-12> :_
ENTITYI : <ENTITY-0592-6>
<ENTITY-0592-7>
ENTITY2 : <ENTITY-0592-8>
REL OF ENTITY2 TO ENTITY1 : CHILD
STATUS: CURRENT
<ENTITY-0592-7> :_
NAME: Bridgestone Sports CO
Bridgestone Sports
TYPE: COMPANY
ENTITY RELATIONSHIP : <EHTITY_RELATIONSHIP-0592-12 >
<ENTITY-0592-8> :_
NAME: Bridgestone Sports Taiwan CO
ALIASES: "Bridgestone Sports CO"
"Bridgestone Sports"
NATIONALITY : Taiwan (COUNTRY)
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-12>
<TIE UP_RELATIONSHIP-0592-3> :_
TIE-UP STATUS : EXISTING
ENTITY: <ENTITY-0592-16>
<ENTITY-0592-15>
JOINT VENTURE CO: <ENTITY-0592-17>
ACTIVITY: <ACTIVITY-0592-9>
<ENTITY-0592-15> :_
NATIONALITY : Taiwan (COUNTRY)
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-18>
<ENTITY_RELATIONSHIP-0592-18> :_
EHTITYI: <ENTITY-0592-15>
<ENTITY-0592-16>
ENTITY2: <ENTITY-0592-17>
REL OF ENTITY2 TO EHTITYI : CHILD
STATUS: CURRENT
<ENTITY-0592-16> :_
NAME: Union Precision Casting CO
ALIASES : "Union Precision Casting"
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-18>
<ENTITY-0592-17> :_
NATIONALITY : Taiwan (COUNTRY)
TYPE: COMPANY
ENTITY RELATIONSHIP : <ENTITY_RELATIONSHIP-0592-18>
<TIE_UP_RELATIONSHIP-0592-4> :_
TIE-UP STATUS : EXISTING
ENTITY: <ENTITY-0592-22>
<ENTITY-0592-21>
ACTIVITY: <ACTIVITY-0592-9>
<ENTITY-0592-21> :_
TYPE: COMPANY
ENTITY RELATIONSHIP: <EHTITY_RELATIONSHIP-0592-23>
<ENTITY_RELATIONSHIP-0592-23> :_
ENTITY' : <ENTITY-0592-21 >
<ENTITY-0592-22>
REL OF ENTITY2 TO ENTITY' : PARTNER
STATUS: CURRENT
218
<EITITY-0592-22> :_
NATIONALITY : Japan (COUNTRY)
TYPE: COMPANY
ENTITY RELATIONSHIP : <EITITY_RELATIOISHIP-0592-23>
C jjv-0002 SRA's Original Response
<7:/7IL– i--0002-1> :=
0002A'hrJ
El H: 850108
	
"VJf
	
IJ"{*J? : <M-0002-3>
<tM-0002-2>
<#j0002-2>WM: NH-
=%7474-:<=' 474--0002-4>{
jJ: <gM. fj-0002-5>
<W{j-0002-5> :=
: <-0002-6>
eel: (- <_ /7 4 7 4 --0002-7>)<M-0002-6> :=
<2 ' 4 ' 4 --0002-4> :=
Z ;/-7-4 .7- 4 –gg:
Z /7 4 -7- 4 –mg: <4 7 4- JfA-0002-8>
:/7 4 7 4 –f=-0002-8> :=
Z/744-z : <Z/7474--0002-4>
tZ f : i \– i` 1--
M; iM<#-0002-3> :=
#MR:'rr
Z:/7 4 4 -: <Z,/7 4 7 4 --0002-10>
<g-&M-0002-5>
<Z
	
4 -7- 4 --0002-10> :_
	
=;/7 4 7 4 – ,:
	
±
Z%7474- q:
Z~7 4 7 4 –Mfg: <X 4 7 4 - x.-0002-11 >
<2 i7 4 7 4 -Jf-0002-11> :_
Z/7474–Z:<,I>' 474--0002-10>
tJ` Z f : i\–1- t
219
D jjv-0002 SRA's Corrected Response
<7 i7°L— F -0002-I> :=
0002
H : 850108
	
" B> 1
	
f1J„
Js-4: <4 0002-4>
X0002-3>
<M-0002-2>
4J -o002-2> :=
MUM: RH
Z %T 4 7 4 -: < Z /7 4 7 4 --0002-6>
<._ %7 4 74 --0002-5>
* jJ: <-0002-7>
4+tJ-0002-7>14
: <14-0002-8>
ty'): (- <-T..'/'7' 4 7 4 --0002-9>)
<
	
i-0(X)2-8> :_
IJ: 'H--tom
<2.:./7- 4 7 4 --0002-5> :=
z~7-r74 — : tJJAIT
=%T 44its `
.r :/=r 4 7 4 –M <x:/=r 4 7 4– f,-0002-10>
<J
	
4 4 —Mt-0002- 0> :=
2 ~T 4 4 — Z .: <z %7 4 7(--0002-5>
<X / 4 7 4 --0002-6>
4
	
f : ) 1–~--
4-FA AM
<12/-7- 4 T --0002-6> :_
I%7474- K
=%7 4 T 4
	
<-T.:/7 T 4 -M-0002-10><Wt,--0002-3>
:_
&WM: 41i7
..%7474-:<=/T 44--0002-12>
tTh 1j: <ggf:M-0002-7>
<~ %7 4 4 --0002-12> :=
z/74 4
4 -7- 4 --n PEA
z%T 4 7 4
	
<I:/-7- 4 5 4 –M-0002-13 >
<z:/T 4 - 4 —f-0002-13> :_
1%7474 —Z: <Z:/74'7'4--0002-12>
EPftZ, f :
	
I-1--
M., :_
Mk%Tff
x/7474 -: <.=;/7474--0002-15>
<[iTgu1 -0002-7>
<I:/7 4 7 4 --0002-I5> :=
I/-r4 ---r47 , .t t
° 7L"
x:/-7- 4 4
	
4 4 –IM-0002-16>
<1:/-7- 4 T 4 —-0002-16> :_
1%7474-z.:<=/7474--0002-15 >I~ ( {
: R–1-- -1--
Wk. MA
220
