Concept and Structure of Semantic Markers for Machine Translation 
in Hu- Project 
Yoshiyuki Sakamoto 
E\] ectroteehni cal 
Imhoratory 
~:kur a Iilura, 
Niihari {9.m, 
Ibaraki , Japan 
Tetsuya :\[shikawa 
Univ. of Iibrary & 
\]:nformati on Science 
Yatabe maehJ, 
Tsuknba gun 
Ibarak5 , Japan 
Masayuki £~t.oh 
Japan \]:nformation 
Center of Seience and 
Technology 
Nagata eho, Chiyoda ku 
'fokyo, Japan 
O~ Ahsttact 
This paper discusses the semantic features of nouns classified into categories Jn 
Japanese-to-English translation, and proposes a system for" semantic markers. In our 
system syntactic ana\]ysis Js carried out by eheckJnc; the semantic compatibility between 
verbs arid nouns, The semantic structure of a sentence can be extracted at the same time 
as its syntactic analysis. 
We also nse semantic markers to select word. °, Jn the transfer phase for translation 
into I!:n~\] J sh. 
The system ef the Semantic Markcr.~; for Nouns consisi.~ of 13 eouceptiona\] Facet..'.-; 
including one facet for "Others" (discussed \].ater ), and is made tip of 4.9 filial s\]ots 
(semantic markers) as terminal;;. We haw'. tested about 3,000 sample abstracts in science 
and techne\]ogJ.cal fJ.{d.(Is, Onr research has revealed that our' niethod is e×trenicly 
effective in determinin{,; the iileanings of \]¢o(JO verbs (baste Japanese verbs) which have 
broader concepts \] ike En{,j ish verbs, %ilake" , "get" , "take" , "put" , etc. 
LIntroduc!.i.ott 
~manl::ie features are introduced to ensure the 
IilaXillRllll possible accuracy of syntactic ana\]ys.is, 
transfer and goneration. We a.im at a well balanced 
usage cf syntax and semantics throughout the whole 
process of l\]laehJ.ne\] trans\].ation. 
The presc, nt paper introduccs semantic cot\]copes 
for nouns classified aeeor(ling to facets and stets 
~rhJch we cal\]ed semantic markers. Then we show how 
these semantle inarkers are written in the., respective 
lexicons for ana\].ys\]s, transfer and ~;eneratJon, and 
how effective they are in improving the qua\] try and 
accuracy of the machine translation system in each 
phase of analysis and transfer. Therefore, semanti.c. 
features are analyzed by the structure embedded into 
the case frame in Japanese syntactic analysis: these 
features play an important role when selecting words 
in the transfer phase from Japanese into Pkqglish. 
Semantic features are more word specific. Pairs 
of deep cases and nouns should be written in the 
lexicon. }k)wever, due to the huge number of nouns, it 
is more effective to include pairs of the deep cases 
and semantic markers in the lexicon instead of nouns. 
The Ms-project. is a Japanese national project 
supported by the S'l'A(Scienee and Technology Agency) 
"Research on a Machine Translation System(Japanese 
Eng\] ish ) for Scientific and Technologica\]. 
Documents. "* 
~Transfer Apprpach to Machine Translation 
We are currently restricting the domain of 
translation to abstract papers in scientific and 
technological, fields. The system is based on a 
transfer approach and consists of three phases; 
analysis, transfer and generation. 
In the first phase, morphological analysis 
divides sentence into lexical items, then syntactic 
analysis is carried out by syntactic and semantic 
analyses of' Case Grammar in Japanese. In the second 
transfer phase, lexical features are transferred and 
at, the seine time, the syntactic structures are 
transferred by maldn{,; them match tree patterns 
between Japanese and English, Here, we use semantic 
features Lo select words for translation into 
English. In the final generation phase, syntactic 
stru¢%urcs are generated by the Phrase Structurc 
Grammar and the morphological features of ErR;lish. 
The fo\] \]owJne; describes the processing fnnct.Jons 
employed in our system. 
Morpholoo:Jeal analysis and generation program 
are described in L\]:SP, which is adequate for 
morphological proeess in Japanese and English, whi\]e 
syntactic analysis, transfer and generation programs 
are written in GRADE (Grammar DEscriber). Such 
process written Jn GRAI)E are independent of natura\]. 
languages in machine translation. GRADE allows a 
grammar writer to write grammars using the same 
expression in al\] three phases. 
Grammati ca\] rules written in GRN\]E (Gl~mmar 
DEscriber) are trans\[.ated into internal forllis, which 
are expressed by S-expression in LISP. This 
trans\].atJon is performed by GRADE trans\].ator. 
3. Concept of a. Dependency Structure baaed on 
Case Grammar in Japanese; 
In Japan, we \[lave come to the cone\] usion that 
case grammar is the most effective one for" Japanese 
syntactic analysis in machine translation systems. 
This type of grammar has been proposed and studied by 
Japanese linguists before Fillmore's presentation. 
As the word order" is heavily restricted in 
F~iglish syntax, ATNG(Au~nented Transition Network 
Grammar) based on CFG(Context Free Grammar) is 
adequate for syntactic; analysis. However, Japanese 
word order is almost unrestricted and Kok~j_q ,~hi 
(postpositonal case particle) play an important role 
as deep eases in Japanese sentences. Therefore, case 
grammar is the most effective method for Japanese 
* This project was carried out with the aid of a special grant for the pronlotion of science and technology 
from the ~-ience and Technology Agency of the Japanese Government. 
13 
syntactic and semantic analyses. 
In Japanese syntactic structure, the word order 
is unrestricted except for predicates(verbs or verb 
phrases) which will be located at the end of 
sentences. In Case Ora,mar, verbs play a very 
important role in syntactic analysis, and the other 
parts of speech act only in partnership with or 
subordinate to verbs. 
That is, syntactic analysis is made by ehecking 
the semantic compatibility between verbs and nouns. 
Consequently, the semantic structure of a sentence 
can be extracted at the same time as syntactic 
analysis. 
1) Morphological Analysis:Segmentation of a Japanese sentence by Lexicon Database 
Ex. Input sentence'~l~(~-~P~. 'is segmented as follows 
2) Syntactic Analysis: The item-to-item relationship of the sentence is analyzed 
to give syntactic features for the respective items. 
~x. 
(v) 
(~) (pp) 
3) Lexical features are transferred and the syntactic structure are transferred 
by matching patterns Between .Japanese and English. 
i 
4) Syntactic generati0n:The ~ord order in English i8 c0tivertcd a~c01cli;:i{ k;: 
Phrase Structure grammar. 
~x. ~e translate text My computer 
5) Morphological generation: Inflectional features such as tesse, aspect 
are attached. 
~x ~e translated texts by computer. 
e\[c 
Eigure~,L__Process of_MaQhin~Traslaki~n.__.in 
Mu-Projeet. 
4.__C~. se Fra~erned by_.~9.pcjep_ 
The case frame governed by Yogi using 
t~okt{j_~.~hi, Case Labels(deep cases) and semantic 
markers for nouns are analyzed to illustrate how we 
apply Case Grammar to Japanese syntactic analysis in 
our system. 
.~)~e~t consists of verb, K~i~N shi(adjective) 
and \](ei~ouc~QR-Sl~_(adjectival noun). !fak~o:~ 
includes obligatory case and optional case markers in 
Japanese syntax. But a single l(c~k~jo.~slt~ corresponds 
to several deep eases: for instance, <Ic> 'hi" 
corresponds to more than ten deep cases including 
SPAce, Space-TO, TIMe, ROLe, MANner, GOAl, PARtner, 
14 
CONponent, CONdition, RANge, etc. We have ana±s~. 
relationships betweenK~j~ ~L!_ and case labels, and 
written them out manualiy according to the sample 
texts of 3,000 abstraets. 
As a result of categorizing deep eases, 84 
Japanese case labels have been determined as shown in 
Table 3. I. 
Table 3. I Case Labels for Case Frames 
Japanese Label English Examples 
(I) -2~.~ SUBject ~ 
(2) ~. OBJect -~ 
(3) ~@~ RECipient ~1~-~-,4,. 70 
(4) ~:~- ORigin ~'5~, ~-) 
(5) ,~-~- 1 PARtner ~-~, Nt2~ 
(6) ~195~-2 OPPonent ~f)~)~'d~-~-~6, ~f/l~-~6 
(7) B# TIMe 1980 igl~ 
(8). ~ • ~..~., Time-FRom 5 fi\] fJ~ 
(9) ~ • ~,~, Time-TO Jl~l~ "~ 
(I0) ~P~ DURation 5 ~r)~J\[l..f~.,~-~6 
(12) L~N " /~,.~,, Space-FRom -;~d9'7~26 
(13) t~-J~ • $~,~ Space-TO ~"x:~7o, ~ I~:~lJ~-7$ 
(14) NN" }'I~1 Space-THrough --~70, J~?i~'~NA~ 
(15) ~d~\]~ SOUrce \] 5.5 ,~\]~ ~ 6 % "~-~I ~ Jl?f'~ 
(17) N~ ATTribute Ng~I~-N~2. ~O75, LL~" 
(19) ~-~ • il~\[,~ TOOl 4 ,'P 7~'~, V ~\] ;t,'e 
(20) }~'~'\]" MATerial -'¢.-7, b "~t"gTo 
(21) ~ COMponent ~,~. ~:~fij~-~ 
(22) J~'£~ IvlANner i~l\]l~, 10m/sec"~ 
(23) :.~{@: CONdition ..~." .~..~f~.~ 76 
(24) ~¢\] PURpose ~IC~'~c'Y~, ~.i.g, ,~',~\]~¢ 
(26) ~'~fl~,~ COnTent "~\]~A¢..~-"<Ta. ~2-~- 
(27) ~,-~.,~2 RANge -;~:'~"-C.-I~fk.qL'C 
(28) ~..,~r TOPic - l~, - ~ I~ 
(29) ~.,~,i VIEwpoint _~_~ ~. - a),,~,-~ 
(30) ~2.~" COmpaRison ~ d:: ~ ;,~ ~ (, '. - IC *~ .~ 
(31) \[~\[~ ACOmpaniment -&&61~, ~c~-~-C 
(32) /!~ DEGree 5%~'~\]I\[l'zj-7~. 3.4-~'~'~ 
(33) J,~.\[~ PREdicative ~'~ 7~ 
(34) ~ o) {tl~ ETC 
Note: The capitalized letters are used as 
abbreviations 
To write the semantic markers for nouns in the 
case frmne of the verb lexicon, reference is made to 
the noun lexicon for these nouns. 
Note that we write only the semantic markers for 
these nouns appearing in the context of our samples. 
Ko$=uj~hi as surface cases and case labels as 
deep cases are described for YxzuN_en. Then semantic 
markers for nouns preceeding to ~Ho-shi are 
described. 
~SemanticMarkersfoK N_ou~na 
This section describes what the system of 
semantic information for nouns is and what the 
concept of semantic markers is and how semantic 
markers are attached to nouns. 
5.1 System of semantical information for nouns 
1 ) Study 
in the primary stage of our study, we thought 
that all nouns were symbols to display the following 
concepts recognized by humans. We set up four 
concepts :in heighest level ; "Concrete objects", 
"Abstract concepts", "Phenomena", and "lluman 
actions". Concrete objects are the selfsame objects 
in the world. Abstract concepts are the standards 
which fix intellectual activities of hunlankind. 
Phenomena include both social, phenomena and natural 
phenonlena. \[\[unlan actions are the selfsame acted by 
humans. Wc assigned facets to these four concepts. 
Then we further extracted the feature of a park from 
these facets and assigned a new facet "Parts ° . 
Similarly, another coneept of "Attribute" was 
extracted from °Phenomena ° and "lluman actions". This 
feature is crucial especially for action nous. Thus 
we added two faeeLs; "Parts" and °Attribute". Nouns 
also include concepts of measurement, space & 
topography and time. So we added three facets; 
"Measurements" , "Spaces & Topography", and "Time". 
We classified into lnore concepts as follows. 
The concept of concrete objects arc- classified 
into "Nations & Organizations', °Animate objects' arid 
"inanimate objects" which cons t\] Lute three 
independent concepts. The concept of Ihmlan actions 
was elassifJ ed t.wo facets, "Sen:~e & Feel ing" and 
"Actions °. 
We rallied the scope formed by the concept 
"conceptual category". It is difficu\]k to define the 
conceptual scope e.xplicitly. The concept which can be 
defined explicitly in the conceptual category is 
called a facet. The facet i.s subc\]assifJed into a 
number of semantic slots. This relation is 
illustrated i.n fir:ure 5.1. 
/. "'-~ ..... "" "-. /a semantic slot 
,' I t_-----,--~--~ sei~lantic marker 
, ~ ........... " cateqory 
a facet \[ 
facet name 
Figure 5, I Relationship between Facet and 
~em_ant ie Mm:kerN 
2) SubclassJfication of Facets 
Facets, for example, were subclassified into 
slots as follows, the facet of Animate objects was 
subclassified as semantic slots °humans", "animals ~ , 
and "plants ° The facet of Phenomena was 
subelassified as slots "natural phenomena", "physical 
phenomena ° , "power and energies", "physiological 
phenomena", °social phenomena ° and "social systems 
and customs'. We then set up an "others" slot in each 
facet, for these words which cannot he assigned to 
any slot. The use of these slots is explained in 
section 5.3. We will study "others" slots through 
semantic analysis for nouns; new slots or facets may 
have to be assigned. 
These semantic slots and facets are named 
semantic markers. The System of Semantic Markers for 
Nouns is shown in figure 5.2. The system of semantic 
markers for nouns is made up of 13 conceptual facets 
including ~Others ° markers, and 49 fiiial slots as 
term i nal s. 
We also use Special. Semantic Markers for 
"functional words" which represent some patterns, 
syntactic or semantic information. For exmnple, the 
word "comparison" presupposes more than two 
nouns(arguments); comparison between "A" and "B ° . 
Then, "WK(Relation)" as a special semantic marker is 
attached to the word °comparison". The word "time ° 
assumes time case. These features suggest an 
effective device for semantic analysis. 
5.2 Concept ef ~mant.ic Harkers 
The following descri.bes concepts of 12 facets in 
the System of Semantic Markers for Nouns ('Other" (ZZ) 
not. included ). 
l ) Natiens and Organizations (OF) 
This concept.us\] facet includes words related to 
such functional human groups as nations, parties, 
corporations and organizations. Words in this facet 
can occur with volitional verbs, when used as 
subjects. 
2) Ani mate Objects (OV) 
This conceptual facet includes such names as 
that of man, animal and plant, llowever, names of 
organs of the animate objects are included in the 
slot of "Organs or Components" (EL) under the facet of 
"Part.". Names of diseases are included in the slot of 
"Physiological phenomena" (PB) under" the facet of 
"Phenolllerla" . 
3) Inanimate Objects (CkS) 
This eoncepLual facet only includes words 
related to concrete objects in the inanimate objects, 
such as natural substances, parts and materials of 
products, artificial substances and J.nstitutions. The 
objects which do not exist as concrete objects are 
included in the facet of °Intell.eetua\] Objects". 
4) Intellectual Objects(IO) 
"IO" includes words related to theories, 
abstract tools and materials, intellectual products 
that are created by hunlan intellectual activities. 
5) Phenomena (t~) 
"I~" includes words related to natural 
phenomena, physical phenomena, power & energies, 
physiological phenomena, social phenomena and 
systems/customs. Words having causal properties are 
attached to words under this facet using plus minus 
signs (° i-" and .... ). Sign "4" denotes desirable 
conditions(e.g, success), while sign ° " indicates 
undesirable conditions (e. g. suicide ) 
6) Sense and Feeling(SO) 
°(IS ° includes words related to human mental 
phenomena such as feeling, reaction, recognition and 
thinking. 
7) Actions(H)) 
"IX)" includes words related to human 
such as human actions and movements. 
activities 
8) Parts (EO) 
"tO" includes such words related to parts and 
e~nponents of concrete objects as parts, components 
and organs. 
15 
\[\] • i~M • gi~il 
(NATIONS & ORGANIZATIONS) 
~o~ 
(ANIHATE OBJECTS) 
ft.(HUMANS) 
~'XANIHALS) 
~.~(PLAHTS) 
{e)~(OTHERS) 
(INRNIHATE OBJECTS) 
Q~'~(NATBRAL SUBSTANCES) 
I~6~\]~(PARTS & HATERIALS) 
A~(ARTIFICIAL SUBSTANCES) i~ 
I~t(INSTITUTIONS) 
~)(~(OTOERS) 
(THEORIES, RULES) 
I~0~ ~ ~l~.~ • P~:,(ABSTRACT TOOLS) 
(INTELLECTUAL OBJECTS) ~ I S~ ~II~I~JTv~4CABSTRACT MATERIALS) 
I Gj ~OI~I~5~I~Z~(INTELLECTUAL PRODUCTS) 
{ #){L~ (OTHERS) 
~-~ ~ • ~(PART & BLF31ENTS) 
(PARTS) (ORGANS, COMPONENTS) 
L-~X! -~ e)i~,,(OTHERS) 
(ATTRIBUTES) 
~A_~ ~I~E-~(ATTRIBUTE NAMES) 
~R~ B~I)~(RELATIONS) 
~: ~#/\[\](SHAPES) 
,~(CONDITIONS) 
A T~ ~5~(STRUCTURES & CONSTRUCTIONS) 
~{~(NATUBES & PROPERTIES) 
~¢)~(OTHERS) 
(PHENDHERA) 
--~ ~5~(NnTURAL PHENOMENA) 
-~ ~(PHYSICAL PHENOMENA) 
~-P~ ~b • -~;~--(POHER & ENERGIES) 
~9~(POYSIOLOGICAL PHENOMENA) 
--~P~S\] ~gP.~(SOCIAL PHENOMENA) 
~P~ ,~J~ .~(SYSTEHS & CUSTOMS) 
~\] ~ ~){~\[OTHERS) 
(SENSE & FEELING) 
~5~.~B~3,(FEELING & REACTIONS) 
~gBI-~(RECOGNITION & THINKING) 
~(OTHERS) 
~ ~(ACTS) 
(ACTIONS) ~ ~ (MOTIONS) 
~M~ ~/(NUMERAI_S) \[ 
F~ ~2~(Nt~IES OF NUMERALS) 
i M_oj 
|~ ~ *~(STANDARDS) 
(HEASURMENTS) ~ 
II~(UNITS) 
~, ~ ~)/~(OTHERS) 
(SPACES ~ TOPOGRAPHY) 
B~ ,~L(TIHE POINTS) F 
I~:~T~. 4 ~ ~PJIP-'J~(TIME DURATION) 
(TIME) - T~ ~NI(TIME ATTRIBUTES) 
~__~ -~ O)(B(OTHERS) 
Special semantic markers for "functional words" 
y ~W~ T I M E 
--\[W~ DURAT I ON 
~W~ CAUSE 
--~-W~ RESULT 
I LW~ MANNER 
- W~W~ PURPOSE 
~WA~ APPOSITION 
---!WC~ COND I T I ON 
~IWL~ PARALLEI, 
RELATION 
~'W~ DERIVATION 
-~ EXPRESSION 
~ ~IL~ (OTHERS) 
Eigure~.2.__S:c~te~of S~mantic Horkers for No_uas 
16 
9) Attributes (AO) 
"AO" includes words related to attributes of 
concrete objects and abstract concepts. Their slots 
consist of attribute's names, and attribute's values 
with causal relations, shar~s, structures, 
constructions and nature. 
For exampl.e, the word "color ° i.s the attribute's 
name(then, marker in AO), words such as "red" and 
"white" are attribute's valne.(AC). 
10 ) Measurements (MO) 
"MO" includes words related to numerals, name 
for numerals, standards, and units for measurenlent. 
Examples are "argument", "fee °, "standards", and 
"ki 1 crueler". 
11 ) Space and 1'opography(SA) 
"SA" includes words referri.ng to spatial 
extentien of concrete objects and abstract concepts. 
Examples are direction, area, orbit and Brazil 
1 2 ) Time (TF) 
"Tf" includes such words related to time points, 
time duratJ.on and tinle attributes, as "autumn ° , "for 
a week", "every day° and "life time'. 
5.3 flow te attach semantic markers to words 
\]'he semantic markers for' nouns are deternlined in 
the following steps. 
I) Attach semantic markers to the following 
nouns. 
Propcr houri 
Connnon noun 
Action noun l (Sahen mei shi ) 
Action noun P,(exeept action noun 1) 
Adverbial noun(only when the words include the', 
concept of "Time" or "l.ocation" ) 
Interrogative pronoun 
Persona\] pronoun 
Demonstrative pronoun (only when the words 
include tile c.oneept of "\[,oeation") 
2) Attach semantic markers to the: words 
according to the definition, semantic scope and 
examples given in the "definition table" of semantic 
markers. 
3) Do not attach semantic markers to the 
following words : 
Molecular formulas 
Arithmetic expressions 
Names of product models 
4) \]if a word belongs to multiple slots in the 
same facet, attach all relevant markers. 
5) if a word belongs to a facet but this word 
not belongs to any appropriate slot in the facet, 
attach "others" marker in this facet to that word. 
6) if a word is equal to a facet name itself, 
attach the semantic marker of that facet name to the 
word. 
7) If the concept of a word is not included in 
any facet, attach "Others ° faeet(ZZ) to that word. 
8) For compound words consisting of more than 
one word, attach the markers putting into 
consideration semantic information of the compound 
words themselves; do not always attach the marker 
only to the last element of the compound. 
6. Sen!anti.c Informationfor Adverbs 
In our system, adverbs are subelassified 
fol lows. 
I ) Adverb of condition(Joukyou fuku shi) 
2) Adverb of degree(Teido fuku shi ) 
as 
3) Adverb of stateJnent(Chinjutsu fuku-shi) 
4) Adverb of quantity(Suuryou fuku shi) 
Besides, the aspects of verbs are c\]assified', 
"Mood", "Aspect ° , "Tense ° , and "Degree". They 
contrast specific adverbs. Then semantic information 
for adverbs is used to ensure more accurate 
translations. Semantic information for adverbs are 
defined according to the concept aspects as follows. 
I ) Semantic information on mood determined by 
the adverb of statenlcnt 
Subjunctive(e.g. if), Interrogation(e.g. when), 
Negation (e. g. not always), Desirability (e.g. 
possibly), Entireness (e. g. entirely), Concession (e.g. 
kindly ) 
2,) Semantic information on aspects of verbs 
determined by the adverb of condition 
Complet ~on (e. g. finally ), Progression (e. g. 
rapidly ), Repe ti tion (e. g. repeatably ), 
ConventJ on (e. g. accordingly) 
3) Seinantie information on tense determined by 
adverb of condition and statement 
Past (e. g. yes terday ), Present (e. g, now ), 
Future(e.g. tomorrow) 
4) ~mantic infornlation on degree when the 
adverb or adjective can be modified by the adverb of 
degree and quantity 
Scale(e.g, serJous\].y), Degree(e.g. fairly) 
7, Examples of Semantic Marker~ \[Jscd i~AnQlysi, s 
7.1 DeDernfination of tile Usage of Verbs by Case 
Patterns 
Ca~:e patterns are used to deternfine the usage of 
verbs having broader concept. This is especially an 
effective nlethod in determining the meanings of !~9o 
verbs(basic Japanese verbs) having broader concepts 
like Ene~\]ish verbs, "make", "do", "take', °put', etc. 
We take I{o9o verb " ~hTzZ, "as an example and show 
the difference of the meanings of verb " ?'1kS " by 
mean of' case paDtern (a), (b), (c). Furthermore we 
show the semantic markers which co- occur to each 
casc. 
e g ~g0 verb ~57~7~ (hit, strike, nmlerstand, treat, he, engaged in, 
he equal, correspond, be appropriate, etc ) 
(a) ~'i~ 70 in the concept of ~¢oDxZ~, ~/~!ll-~) ~& (hit, strike, reflex, 
collide, ctc. ) 
{',ase pattern(a): A<ohject, physical phenomenon>~ll<obj0ct, place> t~ 
(verb) 
I!x. 1 A<$:\[(stonn)> ~_ B<~\]~j f.(glass)> ~ ~'/1,~@~ (A stone hits glass) 
\[!× ~, A<)I(\](I ight)> ~ \[\]<~q.ifl~ (s lope) > JZ_ ~'/?c~ (I,ight hi ts the slope) 
\[!x. :{ A<'d_~J~;~'~-(electr ic wave signal)>j~ IKII\](mnuntah/)>~. ?h/2~'E 
J~,\[J~; (An electric wave signal hits the mountain and reflexes) 
( )"17~.£a in the cuncept 0l ~(:J:j",/a ( n ertake, h engaged in, 
deal with, el.c.) 
Case pattern(b): A<ohjcct with "witl">_aiB<action>~.; (verb) 
I!x. 1 A<~(patrol boat)>_/#~ II<;I~IDj (I saving)> J_<_ ~-tgTz & (A 
patrol boat is engaged in life saving) 
\[!x. 2 A<AL£Oar. A)>~ I}<~l!(command)> S_ >'q~c ~ gtr. A is in charge of 
commaiuti ng) 
I!x. 3 A<A&L (Company A) > D! B<','J3,~,{~.!f{ (inspect i on and re, pai rs) > I~ ?h # 
&(C0mpany A deals ~ith inspnction and repairs) 
(c)?"1f<75 in the concept 0f a~"lJ~(be nqual, c0rresp0nd, 
he appropriate, etc.) 
gas{: pal, teiu (c) : A<Mman> 2~ B<human> J~_ (verb) 
A<placn> /O~i II<place> l__~_~ (verb) 
A<t ime> D 4_ B<t ime> i~- -(vet h) 
A<measuremen tun i t>~)i II<measurement uni I,> j~__ -. (v0rb) 
A and I\] can take variable values, Iml. shouh\] not take different values 
in the same sentence. 
17 
',<. ::,P <d ~fc,- 
gx. 1 9-4 J:/(Saig0n)O I/<~iNO (Iinza Ramiki Avenue)> ~ ~aG 
A<~_-,. V-~O (CMdo Avenue)> (thud0 Avenue in Saig0n corresponds to 
Ginza Nar, iki Avenue) 
Ex. 2 A<-~'v H ((today)>~!~a 5 &"(just) 1<--~1 (first year)> I~ N~z,5 
(Today just corresponds to tile first year) 
I!x. 3 A<14 Y4-(1 inc0>~j_B<2 54cm> 1£N~(1 inch is equal to 2.54 
cm) 
fix. d l</~O 7" 5 e\] x (w0men' s b 10uses) > ~ NCz,5 A<~, "2 ~, 'y (open 
necked shirts)> (0pen-necked shirts corresponding to women's blouses 
as office ~ear) 
In this way, we can determine the usage of verbs 
by means of ease patterns and the semantic markers. 
7.2 Interpretation of Optional Cases 
One tCqkt!jo:shi (surface case) often plays the 
role of different deep eases in Japanese. Often, 
various optional cases are included in this deep 
ease. Fact optional, case is determined by the 
combination of K(gcujo:sltt(surface case) and the 
semantic marker of the noun which co occurs with it 
in the dictionary. In the process of transfer, 
appropriate English prepositions must be specified 
according to each determined optional deep case. 
For example, take kuk~jo-shi"~ ", The optional 
case is deternlined from the semantic marker of noun 
and the semantic marker which co occurs with the 
surface case of the verb. 1'hen the case frame in 
English is selected arid the preposition "in" is 
determined. This process is shown as follows. 
Ex. (1) Source sentence: J-& L-<J\]t~Eft~D~\]tLz~'6 
~2'l~fl/il~lik"\] 7,%~ 7< & O-$;~Ic.->t, ,-<a~lIJI<, 
Translated sentence: The numerically controlled 
superhigh speed drilling machines are explained which 
are active mainly now in markets. 
Explanation of Example (1): Let us observe the 
Japanese Analysis Dictionaries for Noun and Verb 
shown in Figure 7.1. The noun < fl~N (market)> has 
semantic markers (SA and PS) according to Noun 
Dictionary, while verb < ?~\[~8 (be active)> has (SA 
or OF) for the noun according to case slot2 of case 
pattern in the verb dictionary. °Market" and "be 
active" match with each other with respect to 
smnantic marker "SA". Thus , the surface case < ~ > 
in the Japanese input sentence is determined to have 
the surface case "SA", which corresponds to case 
label < ~> SPAce+ 
Ex. (2) Source sentence : ~flillIgT~l,IIz 2. & t---~N~IIZ%¢ilc-) 
Translated sentence: Problems are solved bg two 
handling methods about laser absorption terms by the 
inverse damping radiation, and numerical solutions 
are obtained. 
Explanation of Example (2) : Using these 
Dictionaries, the noun < IN O ~I,'P~ (handling 
methods)> has semantic markers(!C and AN) while verb 
< ~g< (solve)> has (DA, IT, J_C or TS) for the noun 
preceding the Kaknjo-shi. "handling methods" and 
"solve" match with each other with respect to "IC ° . 
(($.ZIJ, UII:~- "0 O O 0 0 O,OI,) 5 6 0 5") <$II II 
¢ .l~.,~g4~.-iffN ~,~...~,I -'g) ($iill'~II*ll 1)b¢$~iff(4I I)) 
¢ 13< - ,g ll.,k lI N 
(, ,I, iel ~II 3,} lip I i~_ G iI ) 
( galln)/-e-\].s SA P S))) 
{ rt of speech ~-- ~elmntic marker 
Lexicul uni f 
_(~) Contents of the _No.un.. Dietiona rg _for 
< ~t~ (market~ 
18 
(<$~tlb~ -V 0 0 0 7 6 0 O-O i") 
($m-~ e) ~ Lexical 
( $ "~g 1 2) ! Ob%r~mtion 
( $ ~*~5. "D,-~ © < -t5")> j 
( $ ~JJ illO\] flI £ ~)" t) Morphological 
( $ ilt~lffN a; informatioa 
V 1 ~attern i ($-rx-<~/ b ,},l~'e) 
$ A.£ tIl~ ) t ($*~:~EfffN Case s-l©,,/\]) , 
Surface case -2 (( $ ~N$8 to<) ~ ~1' i 
'Dee ccl,'e --~) ( $ ig N ~45 .~{$) ) ii .... i $~,,1 ..... := - OV) 
~e,i~,n,.~c ,mr,cer < $ g,~l,Z. I~,. v, d 
i ($£,N'1¢ 
o)))))) ' ! 
Case slot2 j' 
6b) Contents _.of the Verb Dictionary for 
Figure Z~I Contents of t.he ,!apane6e/Lna!ysi.s 
Dietionaeg 
Thus, the surface case <<'> i.n the Japanese 
sentence is determined to have the surface case "IC", 
which corresponds to case label < ~-~xFR"'Td 4s,~,~ > TOO\].. 
8, Exanp\]es of Selantic Markers Use(l in Transfer 
Process 
in the transfer' phase the English verb is chosen 
by examining the semantic markers for nouns filling 
the deep case slot of" the condition part of the verb 
transfer dictionary. 
Two examples of selecting the traslation for the 
Japanese verb," ~'" are as foil©us: 
Ex. (1) Source sentence:*l-I ~,~J:U~6£x2 yg 
Translated sentence: The radiation and 
convection models of the vertical mercury arcs which 
contain the sodium and the scandium iodide. 
gx. (2) Source sentence:~6~iE~f~K 
~J- a £ 14 ~ b N~ea~No 
Translated sentence: Lemmas are verified by 
showing specific constitution methods about 
constitution methods of the normal double orthogonal 
bases which i~clud~ given normal double orthogonal 
systems. 
Explanation of Example (1): 
Based on the Japanese to English Transfer 
Dictionary, Figure 8.1, both l J-b I O~ (sodium)> 
and < ~5£x~ ~99~ (scandium iodine)> in Example (1) 
have OM. According to the conditions of the 
dictionary, ~ matches one of the semantic markers 
(OSO~jPN PB PP PE) in correspond to the appropriate 
case slot (in this case <. ~ (object case)> ) of 
Japanese case frame in the dictionary for the verb 
<~O> (contain, include). So verb "contain" is 
selected. 
3aPonese Case .fr*dme -- 
UmMition_ of selectiag 
verb translations ~- 
(SEQ 360) 
(J LEX ~')¢ ,lat~nese lexical mtil Ja~mese surface 
(J CAT ~ ~) ~ Japalmse cateooru 
(USAGE 
( (J SURFACE CASE (J SURFACE CASEt ~) (J SURFACE CASE2 
(CASE-PATTERN VI) - - - 
(J DEEP CASE (J DEEP CASEI 3~) (J DEEP CASE2 ~)) 
(CONOITI-ON ...... 
((JSEM2 0 S O M P N P B P P 
((ELEX i n c I u d e | ))) 
(A 
~se 
yra.~ of "col~taia" 
Corre lai ioa o.f 
Japanese m~l \[ 
E~tglish case frames ~ 
i \[ 
(B 
cas \] 
P E) (E LEX c o n t a i n 1)) 
(E LEX C 0 n t a i n) 
(E CAT V) 
(E SURFACE CASE (E SURFACE CASEI S U ~ J ) (E SURFACE CASE2 0 B J i )) 
(CASE-PATTERN V1) .... 
(E DEEP CASE (E DEEP CASEi C P O) (E DEEP CASE2 0 B J )) 
(CORRESPONDENCE-(CORRESPONDENCEI -ji~3 (CORRESPONDENCE2 ~$j~))) 
(E LEX i n c 1 u d e ) 
(E CAT V) 
(E SURFACE CASE (E SURFACE CASEi S U B J ) (E SURFACE CASE2 © B J l )) 
(CASE-PATTERN VI) - -- 
(E DEEP CASE (E DEEP CASEI C P O) (E DEEP CASE2 0 ~ ,I)) 
(CORRESPONDENCE (CORRESPONDENCEI J~'I~) (CORRESPONDENCE2 ~}~)))))) 
Figure 8,1 C ont,~nta~f_ th~Japaneae _to Eng!ish .Transfer Dic_tio_nar~ 
As for Example(g), the semantic marker for 
< ~g~Y~:.~ ( normal double orthogonal system)> is 
IC, which does not match any of the semantic markers 
in the appropriate ease slot (in this case 
< ~'t~ (object case)> ) in the dictionary for the 
verb <.,~t2.>. Thus, verb "include", which is the 
default wdue of the English word, is selected. 
9~._Concl u..;ion 
Acknow \] edgmen 
We would like to thank Prof. Makoto Nagao, Prof. 
Junichi Tsujii, Mr. Junichi Nakamura(F~oto 
University), Mrs. Mutsuko Kimura (Institute for 
Behavior and Science), and Miss Masako Kume(Japan 
Convention Services, Inc.) and the other members of 
the Mu project working group for the useful 
discussions which led to many of the ideas presented 
in this paper. 
I) When semantic markers are attached by human 
operation, several problems arise. The first problem 
is simple mistakes made by humans. The second problem 
is a fluctuation of semantic analysis due to a large 
amount of data. So it is necessary to develop an 
automatic marking system to save time and to improve 
efficiency. 
P) When assigning semantic markers to nouns, we 
attached them without considering the relationship 
between nouns and verbs. That is, we attached 
semantic markers simply based on noun concepts. This 
is not adequate to handle nouns which are 
intrinsically related to verbs. One of the solutions 
to this problem will be to study the correlation 
matrix of the semantic markers for nouns in relation 
to the case frame of verbs. 
3) Our system of semantic markers for nouns has 
been designed for Japanese nouns. We have to design a 
syst~, of semantic mgrkers for English nouns. Since 
recognition for its concept in an Enlish word is very 
difficult for the Japanese, we are also studying a 
method of evaluation test to handle these data. 
4) Our system of semantic markers for nouns 
simply consists of the tree structure of facets and 
slots. Subclassification for these structures with 
deep tree structure is a significant problem in order 
to analyze the concept of nouns more in detail, but 
such a semantic marking operation will bec~ne more 
c~nplex and difficult for I), 2) problems. 
5) We suppose that a concept system for words is 
not a static structure, but various semantic networks 
constructed dynamically according to a given sUory. 
We must give thorough consideration to this 
prospective problems. 
~efereBces 

References

(1) Sakamoto, Y., SaLoh, M. and Ishikawa, T.: 
Lexicon Features for Japanese Syntactic Analysis in 
Mu .Project JE, COTING84, Stanford, 1984. 

(2) Nagao, Makoto:Struetural transformation in 
the generation stage of Mu Japanese to English 
machine translation system, Theoretical and 
methodological Issues in machine translation of 
nakural language, Hamilton, New York, 1985. 

(3) TsujJi, Jun ichi:Science and Technology 
Agency's Machine Translation Project, Proceeding of 
the International Symposium on Machine Translation, 
JIPDEC, Tokyo, 1985. 

(4) Nakai, H. and Satoh, M. ; A Dictionary with 
Taigen as its Core, Working Group Report of Natural 
Language Processing in Information Processing Society 
of Japan, WCNL 88 7, July, 19~3. 

(5) Nagao, M. ; Introduction to Mu Project, WGNL 
382, 196~3. 

(6) Sakamoto, Y. ; Yougen and Fuzoku-go Lexicon 
in Verbal Case Frame, WGNL 38 8, 1983. 

(7) Ishikawa, T., Satoh, M. and Takai, S.; 
Semantical Function on Natural Language Processing, 
Proc. of 28th CIPSJ, 1984. 

(8) Nagao, M., Tsujii, J. , Naka~lura, J. , 
Sakamoto, Y., Torinmi, T. and Satoh, M: Outline of 
machine traslation project of the Science and 
Technology Agency, Journal of Information Processing, 
Vol .28, No. 10, 1,985. 
