American Journal of Computational Linguistics 
Microfiche 44 
COMPUTER UNDERSIANDING OF 
METAPHORICALLY USED VERBS 
SYLVIA WEBER RUSSELL 
Forest Park 9-E 
Durham, New Hampshire 03824 
Copyright o 1976 
Association for Computational LinguistiCs 
A major problem conrronting computer pro:lram:. driven 
by nnturol-1 ngu ge input consists of thc interprctntion of 
1inguh.l ic cxprossiobs for which the int endcd literal 
meaning is not e:q~licitly given by the lc~ioi~l coml>onents 
of the expresL;ion. ~'tn ccxi~mple is thc "extended useu of the 
verb 'leapt in 'the country leapt to prosperity'. buch 
extended usages--whether cunsi-as inilntcd or original-- 
czn be considered metaphorical to the cktcnt th-,t they are 
based on analogies. This paper establishes a framework for 
interpretin? netaphoric,ll expres ionk by analysis of under- 
lying cbstrgct cor:~ponents --such 9s "trnns  tio on" and "inten- 
'i'h~s is in contrcst to pre- sity" for the nbovc example. 
viouc: epproaches which rely on 3 a~mbcr ~f word senses 
intended to represent metnp oricL21 usi ges directly. 
An eexperimentcl rro )ram find, literal inter7rctations 
for inyut rcprc~snting a simyle scnt~l ce m \.hit!. the 
"verb21 concepttf (action, state or cttributc) is used meta- 
phorically. This in ut has thc gcner 1 conAguration 
1 ~uB JCfl Vc,& OBJECT ;~UXCE/GO.IL 
or 2 UBJLCT PiLDICA TE- 
ADJECTIVEf. The interpretation are ~ive? in the f om of 
primitive ;;nglisFI pcraphrases . 
The-e parsphrases , which 
are intended merely to illustrate the informcrtisn which 
czn be extrrcted from metr.?:~orical inmt , dre bL sed on 
scmr.ntic representations which are convertible to structures 
specified by SchL.nkt s conceptual deaendency theory. The 
interpretation of mctaphori~nlly used verbs thus represents 
a particul..r case of the general tusks of disambiguation 
and interpretation encountered by the conce-tual dependency 
parser, 
The hpproxim~tion to the literal mc, ning of n metaphor- 
ical verb is achieved through reference to semantic descrip- 
tions based primarily on n srn.111 number of. conceptual features 
and absrrrct structures. THese descriptors are specif icd Lor 
classes of those concepts which ore c:qrcosed in dnglish by 
nouns, verbs, adjectives .:nd prepositional phrases. The 
complete set of values L-or the clescriptors or verbal concepts 
is represented as a multi-dimensional matrix containing the 
defined conceptd. This matrix, which is only p~rtially 
described in this paper, exhibits relatlonshipb and analogies 
wldch underlie metaphorically used verbs. 
The relative independence and cbutrcct chnracter of the 
basic sernuntic descriptors render rhe system easily extensible 
to f urthcr capabilities, such as more conclusive interpreta- 
tions or the tre tment or' more chnllenging expressions. The 
emphasis on systematic uescriptions ,nd primitive concepts to 
produce sim~le p raphrases is viewed as reilecting hwn 
understanding or novel linguistic expressions an& providing 
a model to explore questions related to such under-tanding. 
Contents 
1. Approach 
1.1 halogies 
1.2 Conceptual dependency interpretations 
2. Characterization of Verbal Concepts 
2.1 Levels 
2.2 States 
2.3 Structures 
2,4 Features 
3, Characterization of NOMLNALs 
3.1 Features 
3.2 Function descriptors 
4. Nethod of Interpretation 
4.1 Conditions on metaphorical extension 
4.2 Operational context 
4.3 General procedure 
4.4 Operation of routine 
4.5 Tests and criteria 
5. Examples 
5.1 Level shift 
5.2 Category shift 
5.3 R-O switch 
5.4 Intra-level feature shift 
5.4,L Actor-feature shift 
5.4.2 Object-feature shift 
5.5 ~oun compounds 
6. Conclusion 
Metaphorical usages have often been regarded ns "special 
ccscou to vhich the particular language analysis mcthod under 
discussitm did not apply. This pnpcr prcsdnts n mcthod for 
comsutez* undcistnnding of a class of phrases in which the 
verb is used 9netnphoricollytt, but which ignorer: the dis- 
tinction between rrextcndedlv and "assi~nil tedtr usages. This 
approach provides flexibility in handling previously unseen 
usages, The assumjtion underlying this approach is that 
analogies ore involved in language understanding to a greater 
extent than speakers consciously realize. 
1.1. - Analogies 
Analogies arc the means by which we substitute, extend 
or borrow concepts. In the use of an analogy, a word is 
borro?+erl Jrom its usual context to exprcs3 some component of 
meaning sh'rrcd by the concept underlying the borrowed word 
in its literal sense and the concept which the borrowed word 
is to reprclsent. This results in an extendeu or netaphorical 
use or the word. The system to be described is intended to 
show the analogy comprehension necessary for the interpreta- 
tion of metaphorical usages of verbs. 
The problem of determining the meanins of a metaphorical 
expression is one of knowing the critical similarities and 
differences which a borrowed sense of a word has with respect 
to the orilginol sensc. In somc case.; an es::cntic,lly meto- 
?horic,lL usage ccoses to be tl~ought of as borror~cd, nnd 
acquires on idiomatic scnsc of its own. r!owcvcr, ir thc 
sirr,il.;:rit.ies and dir'r'erencu which ontcr into :rtnphoric.;l 
usnses c.In be idcntificd, vr? c:?n still ht ndlc ~uch nn ex- 
prcssian as wc uo those exnressions which :.rc p:encrally 
viewed as iact n~~horicnl . Concid~r t hc L xk3l.~ples 
L) L'he ilousc 1;illeck the bill 
2) 
I scc wh~t you mean 
Here the first example oprlears to be ncraphorical, the second 
not. A langu,~ge nnaLy..er prepared to handle only non-meta- 
pl10ri.c~ 1 input might achieve the correct interpret2 t ion of 
'1 see1 in the sensc of '1 un~crst'nd I. :io,:evcr, it would 
succcecl only if lsectswere listed' in the dictionall) as enuiv- 
alent to tunaerut..i~d' in one sensc. Such n solution ignores 
the cc:~pahilities which humrns hhve for correctly interpreting 
such sentences without h ving lecrned this s:.nonymity. A 
parser which lacks this ability, i.e. to interpret without 
relying on ad hoc aids, will n..t have the flexibil'ity required 
to a,-proach similrr problems in r~hich such nids are missing 
due to the prejudices of the aerson who deir'ines verbs for 
the lesicon. 
In this sense, we should be able ro unucrst-nd mcta- 
phorical sentences on the basis of an analogy to the ordinary 
or literal sensc of thc words involved. The excl~-nlea 
1) 
The idea of growing their own radishes was born , 
2) He hid his embarrassment about the honey pot 
3) He relinquished his hopes 
4) 
Her painting said something to me 
are all metaphorical in different ways with respect to the 
ordinary sense of the verb: 
the literal effect of 'hidn' is 
visual; thnt of 'relinquisht has to do with control of a 
phyeical concept; thnt of say1 has to do with Linguistically 
expressed information. 
But in each case there is an analogy 
between the ordinary .mnti the metaphorical usage of the verb. 
The analop*.-d~onsists of the similarity of the which 
occur in the mn-metaphorical and metaphorical usages: 
I.),, 
The idea (= to grow their own radishes) was born 
The baby (= Percy) was born 
Effect = new idea (baby) can be related to 
2) He hid his embarrassi~~ent about the honey pot 
Xe hid the honey pot 
Effect = Others are not visually aw-ire of his emborrass- 
ment (honey pot) 
3) He relinquished his hopes 
He relinquished the presidency 
Effect = IIe no longer has a certain ottkibute 
4) Her painting mid something to me 
Her book araid somet1 ng to me 
She said something to me 
Effect = I have a~new mental concept to consider 
Information derived from such metaphorical expressions 
should at least include analogous effects of this kind, 
dhich represlmt the "rcsultlt component of: tho meaning of the 
expression. (A related problem of extracting conceptual 
inferences is d&scussed by Schank and Xieger (8 ). ) 
This task requires a verb description system which 
cittcgorines verbs* by two criteria : 
1) the identitic.ition of lm. underlying dtructural 
component which is similar for verbs which arc used 
analogously in linguistic expressi. ns , and 
2) the identification of a certain level at which the 
verb applies, such as "physicallt . 
Each verb will thus be classified, not in terms of a single 
category, but in terms of two type9 of variables having 
values according to these two criteria. Thus levels and 
structure-concepts must be determined which can be used as 
a basic form of description of verbs in the dictionary. 
1.2. Conceptual dependency interpretations 
In addition to such verb descriptions, which serve the 
analysis task, the form of "targetu representations, i.e. of 
the literal interpretations must be considered. 
The basic 
asstqtion underlying a choice of representation is that a 
"translation" from a metaphorical to a corresponding literal 
ex:~ression cannot be achieved by manipulaf ion of components 
at any syntactic level. 
What is needed is an tvinterlinguatt, 
which deals with relationships between concepts at the 
cognitive level. 
The conceptual represent at ions which apply 
to thie interlingua are not dependent on the original lexical 
form (or language) of the input, and can be used to generate 
paraphrases of the input into the same or other languages, 
given the concept-to-syntax mapping rules for that ~ongu$ge. 
The chdea of a form of conceptual representation must be 
guided by the extent to ~hich it shows rel,ltionships brtween 
concepts at the cognitive level. 
The conceptual dependency 
theory of Schank (6-9) provides such a- representation in 
terms of predicntive and qualifying dependencies between 
conceptual categories and is assumed as the context of the 
method presented here. 
In a dependency, according to this theory, a concept 
of one conceptual category is dependent on, qualifies directly 
or serves to describe a concept of another conceptual category 
according to rules of conceivability. These unambiguous, 
language-free dependencies are word-independent, although the 
concept symbols on occasion map directly into some lexical 
term expressing these concepts. The nature of conceptual 
dependency representations, as well as their suitability for 
metaphor analyses, can be conveyed by a simple example. 'The 
ink stained the floor' can be represented conceptually as 
. . ink 
color: Xj . The significance of thls 
colors xi 
representation lies not in the particular hotation adopted, 
but in the components of meaning which it rcve~ls. The 
dots (... ...) indicate tht1t the ink is not necessnrily the 
agent, but is merely somehow involved in thc action. 
The 
~~dk@tbn arrow!' ( m) indicates n causal reLn tionship as 
opposed to thc eesclmple 'the ink hit the floorv. 'The 
notation indicntcs n change of state of tfloorl, ar mare 
specificnlly, an Inalienable PUT of the floor. The 'NEJ~ative' 
notation is a iconnotationtt (5) wl~ich is secondary to the 
purely objective representdtion of 'stainT. 
If it is assumed that the use of metaphor relies on some 
similarity of semantic components between an ordinary and an 
extended sense, it c.ln be seen that a representation of this 
type, ref Lecting a conceptually-oriented sen-ntic theory, is 
ndequcte to the task at hand. By reference to the abstract 
components of causation, change of state, pert vs. %,hole and 
negativeness reve~led by the iibove cunceptu~l structure for 
*staint, a paraphrase for the metaphorical 'his business 
activities stained his reputationT is essily npproximated; 
'his activities csused a neg~.tive change in (part of) his 
reputation1. There is no dependence on complex trznsforrna- 
tions or muktiple word senses, which might in fact fail in 
the: case of novel f oms of expression, such as more rtcreative" 
metaphor. 
The components of conceptual dependency representation 
can be briefly described as follows. 
The conceptual categories 
between which the various conceptdal dependencies exist are 
ACT, PP (lfpicture produceru) and PA ("picture assistcr~), 
At tho eyntaoti~ level, these categories are sometimes em 
premed in the English language by verbs, nouns and adjec- 
tives respectively. However, such correspondence does not 
always occur. 
For example, many nouns can be expressed 
directly in terms of verbal or attributive concepts ('the 
'that which,..'). Such nouns would not be statc of... , 
mapped directly into PPs. 
The dependencies which hold between the specified 
categories ot the cognit ive level must ult imittely be given* 
by a uconceptual grammarM which reflects their conceivability 
and theref ore their comprehensibility. Such n grammar, 
independent 6f actual word-construct usage, would include 
information such as what kind of concepts can be related by 
a specification of position in time. Our concern here, 
however, will be mainly with the lower-level and more detailed 
information contained in a bfconceptuolizationl', or simple 
conoeptual s tructure. 
The general conceptual dependency format which has been 
established for the conceptualizations which will be referred 
to takes one of the following forms (semantic terms which 
are irrelevsnt to the metaphor problem, such as tense, 
will be ignored): 
<relation> 
PP (object) PP (object) 
ON 
e,g. ink floor 
The ink is on the floor 
The ink is in contact with the floor 
PP (object) < PA (attribute) 
e-g. ink <+> COWR: black 
The ink is black 
The ink has a black color 
(attribute-value (new)> 
(change of state) 
(attribute-value (old)) 
CCLOR: black 
e.g. ink 
COLOR: ? 
The ink turned black 
The ink changed to a black color 
(goal> 
PP (actor) <$ r-lm (action) +PP (object +- 
C <source, 
I 
(continued) +<instrumental conceptualization> 
Eiary 
e.g. John<=> ATRANS +- C0NTROL:ink 
I 
'32 John 
"C 
Nary ( IPART :hand) 
(cont.) t John PTW +ink + 
John( IPART: hand) 
John gave Siary the ink by handing it to her 
John handed Mary the ink 
CP(mrg 1 
or JohneEPPRwS +$ 
? "-C CP (John) 
John communicated the ink story to Mary 
Mary heard about the ink from John 
The actual relevancc and character of some of the 
components of the latter type depends on which ACT is present. 
The list of i&Ts is: MOVE GRASP PTR hNS 
FROPEL SPEAK bXflCRiNS 
INGEST Am'ENL ATEtUS 
EXPEL El5 UILD 
'Ilhe source-goal component is irrelev nt to tile ACT GkLlSP, 
for exclmple. For FTkiNS (physic.11 tmnsition), the object , 
source and goal must be spec if ied and are phys iccl . 
For 
WCY\NS fmental tmnsition), the object is itself n concep- 
tualization and the sourue and goal arc the rnentl-1 processors 
of human or at least anirn:lte beings: - Conscious - Processor, 
Long Term Memory and Immediate iv!emory. 
For MTRANS (abstract 
-.I - - - - 
transition) the object is a form of control and the source 
and goal re animate beings. Each of these three foms of 
transition involves a type of t'conceptual casetf: 
PTR.,NS 
takes the Directive or Locative case ( + 
"C 
, and l+ll'&U?S 
and ATWS the hecipient or possessive case ( +- RE d. 
The object which is dependent on an AL'~ in that it is 
0 
"acted uponf1 is in the Objective case ( t PP (object)). 
There are a number of other conceptual connectives 
and modifiers which apply to such conceptualizati ns. 
These 
cdn be referred to in (6). The r.ost important of these as 
concerns the representat ion of the concepts considered in 
this paper is the element of causation: 4 causing ccnceptualization> 
m 
orle e=> DO 
This component underlies verbs such as 'make ' ( 
one (=} 1)0 
m 1 
<EP><=+ be 
and color ( <PP>(&c CDI,UK: <new vnluc) 1. 
COLOR: <bid value> 
Llifierentiated types of cautintion and the conditions for their 
applicability ore given in (6). Intcndcd causation or purpose 
I? 
will be designated in the present rvork as ,,, 
It is proposed that verbs be-represented us entries in 
a multi-diti~ensional m::trix which shows the dn\ilC.rities and 
difzerenccs mentioned. .is a char~ctcrisazion of 
verb.1 concepts is desired, re~ardless 02 whether these are 
realided 1c::ically as verbs, odjcctivcs or prepositional 
relations, such concepts will be rc~orred to as s , as 
opposed to the lexical tvcrbsl. 171e column headings of this 
matrix ~ive the char.~ctcristic t~structurea" of the Vi.;Bs, 
eithrr explicitly or as conii ur.ltions or' feature.;, ilnu thc 
row henclings ,Ire "lev~la~~, "pl. nestl or 'Irr. ~~eworks gf the 
VG3s. Each entry then represents a l~cstc~,ory'' of m?rbs 
which sxtisfy column- and row- (and further di:xens.+imn- ) 
values. Conceptual :+GI's as introduced in the prevhns 
section are also subject to uetinition in terms of his 
matrix. ~di's re considered ta be sufficient 2s e basis 
for describing all actions underlying langucgc, sraless 
of hoo this action is expressed in a p rticular t#npu~l:;e. 
Wr inst:ince, the ACT 114TLUbt underlies the verbs 'tell', 
'forget ' and other 'verbs of mental transition. 
It is these 
primitive concept8 rather than :my snecif ic lexical verb 
which will be retrieved from the matrix as output of an 
operational metaphor routine. 
2.1, Levels 
The four levels postulated for vcrbnl concepts are: 
PIIYSJCAL (e.g. 'touch1) 
mNT& (em& 'think' ) 
SENSOR' (e.g. 'seet) 
CONTROL (e.g. 'donate') 
The PHYSICAL Level includes verbs which predicate the 
existence, attributes or associations of objects with 
spatial (material) aspects, 
The PENTAL - level is distincf in that 
on this 
level ere representations of objects, or of other represen- 
tations in a recursive manner. It is thus the level through 
whlh thought and communication take place. 
Verbs of thought 
have been analyzed in (8). 
Since l;Ch'T& objects are not 
real-world objects or situations, but rather pointers to 
such objects, they cannot be concqtually dependent on 
non-3ENTAL concepts other than (usually human) mental 
processors. 
The SENSORY level includes VWs of perception, or the 
reception of %utgesl1. Concepts on this level prnvi.de the 
link fr~m the physical world to the consciousness of a 
language user as well as to other animate beings. 
SEXMRY 
concepts could be analyzed in PHYSICAL (spatial and temporal) 
terms. However, this kind of dctail seems to have little 
relevance to the linguistic groblbms under considerhtion. 
The CCNTRLL level refers to relationships which express 
possession or corltrol by an animate being. An object on 
this level is a form of control, or a "potential for actionfr. 
CONiiiOL mRBs basically consist of conditions attached to the 
actions of an animate being and are sometimes ewressed 
lexically through modal auxiliaries, for example as 'can', 
'may * or 'must *. Possession defined as a CONTlWL relation- 
ship is thus distinguished from purely PIIYSILiL or locative 
ralntionships. The verb 'have (a physical object)', for 
example, is defined in terms of the CON?R\L Level rather 
tbn the PHYsICAL level. 
Each of these levels has a few sublevels (e.g. SENSORY: 
eye, ear) which are sometimes specifically referenced in 
metaphorical extensions. These are described in (5). 
2.2. States 
Given the matrix format of the verb descriptors, the 
specified levels (row components ) can best be clsrified 
by conoiderntion of the Its implestw structure (column component) 
as it applies on each level. 
This structure is referred to 
as a STATE or as a ST.iTIC structure, and represents in general 
terms "existence, with or without an nttribute, and with or 
without association with another objectft. ST.iTBs are pre- 
sented here in two forms which represent the '+I and *-* 
values of one of the "featurestr (Section 2.4) which further 
dif fercntiate VERBS, These two feL~-!. urc v lues are termed 
uactual?t and "p~tential", according to whether the given 
STATE has the feature value '-HY~othet ical or '+Hypothetical '. 
STkbEs in these two forms represent primitive concepts to 
which further feL!tures can be applied to obtnin more complex 
VERBS. &n explanation of verb entries which are ex~mples 
for the two forms at various levels should give some idea 
of the scope and basis of the verb on.4ysi.s. Space consid- 
eratiuns limit the discussion to two levels. 
If other feature values arc ignored for the moment, 
-/+~~~othetical STATES can be thought of as the first two 
colums of the matrix. At the EENTAL level we have: 
ACTW POTENTIAL 
Percept ion 
0-Preuiczt ion: 
R think SUN: believe 
OBJ: know 
VAL TV=+: be true 
cont, ACTUAL POTENTIAL 
R en joy SUBJ: like 
OBJ: appreciate 
0 be in CP please 
VAL AV= t : AV=+ : bc nice 
"be funtr 
Volition 
R 
0 
VAL DV=+: be good (to do) 
The BLRTM, level is divided into two sublevels to correspond 
with the faculties of perception and vulitioc. Perceptian 
in turn has two forms--the predicction -_ _ _ __- of the existence of 
the - Object, and the valued perception of a presu~osed Object. 
Only the former type is examined here. 
It is first noted that this and each sublevcl allows 
for an R-, 0- and VU-form or' a VLB. 0 = Object, R = 
Recipient or experiencer ( "loc~tion" , "sollrcelt or "goal tt 
of O), dnd VAL = VALW of 0. The labels R, 0 and VXL indicate 
whether the lexical entry which n~ps into the slcts headed 
by these lzbels expresses a verbal concept from the pcint of 
view of R or 0, or expresses a value 02 0. Syntactically, 
the "point of view" of R or 0 is reflected by a verb having 
a noun ~ith'~'role" R or 0 respectively as syntactic subject. 
For cxLlrnt?le, the R-role 'see corresponds tb the 0-role 
'appear1. 
If no verb for o given slot comes to mind, a 
phrase is given which s imply ref lscts the conceptUcl represen- 
tation of this STrdE. Thus the O-role entry corrcspondlng 
to tbelievcl is tO<=propbsition) be in - Long I Term - f.kmoryl. 
The vnlue iml-,osed by 'believet on the Object, which for this 
sublevcl is a Truth Vnlue (TV), is positive ( *+ ). 
Verbs or predicate adjectives ip the 'VdW row express 
a possible value of the Object as onposed to thc relrtionship 
of the Object to on R which is givcn by 0-role vcrbs. 
Uthough the verbs and adjectives given as examples ~11 presume 
a positive vnlue of 0, other degrees of positiveness on the 
vaLue scale could underlie othrr vcrbs or ad jcct ives. For 
instance, 'be indifferent to' means that the bject lies midway 
between the '+' arlu - .I) Attitude or - Aesthetic vdue for the 
expcriencer. 
aince a IENTrlL aT.bE expresses an attitude townrds a 
IvENTAL object which may or may not correspond with that of 
'loutsiue observersf1, verbs may express cither a SJBJect ive 
or an 'OBJective* MENTAL STATE, as shown. Th~t is, a speaker 
says 'he knows that.. . ' to mean 'he believes that.. . , and it 
is true '. 
The difference between the ACTUAL and POTLNTIAL columns 
can best be explained in terms of the present examnle. 
Generally, *actual1 refers to the fact thct the relationship 
underlying the verb is presently "in operationM, lfreali~edv 
or wexpressedtt. 'Potentialt denotes that tllc object can be 
retrieved in order to create [In 'actual' relati L nship. Thus 
'thinkt in the sense of mentd activity ('think about') 
involves an l1activeW abject :md says <ornething about the 
present state of the thinking person, but 'believet or 'know1 
represents a llstorcdu rather th'tn an active object. This 
diLierence is expressed through representation of 1; as CP 
(Conscious 1hcessor) and UIEi for ' think an3 'believe 
respectivcly. There is no value ssigned to cl for this sense 
of 'thinkv, slnce o truth value is not assigned to a I.LE;T.~L 
object e:;cept in the process of rorning n bclief or making 
an assumption. 
The SL~?-~ORY level can be illustrated bficfly by reference 
to the rel,rcscntation of conccptual attributes in terr~.s of 
this level. 'Be jut if ul ' is dm'ined by prir-itivc components 
on the 'visualt sublevel: 'oE:'hOK\; (cye) VAL: AV=+' or, 
'a visually perceived, ac.;thetically positive attribute of 
an objectt. 
All three non-?:iYSIGJ, levels involve objects G which 
are non-mterial, i.e. not PPa. Rather the object is a form 
of inf ormction, inage or control for the IIZNTAL, SLYSOX and 
CONTR)L levels rcspectivcly. .JL of thebc objects, which 
night bt thought of conceptually as verb: 1 or attrizutive 
concepts, hove a l'relationshipU only to a true exeriencer , 
i.e. an animate B. .it the r'hY&IC& level, on the other hand, 
R need not be animate. The PHYbXCAL level reflects only the 
physical nspcct of the relationship expressed by a VE1.B; R 
may h.,ppen to be animate, but the animate aspcct is irrel- 
evant to khis lcvel. This rnezins thLtt 'have in the genr;e 
John has the ncwspapert is nssicned to the CON'IXOL lcvel 
rather thi~n to the PIT~~uAL. However, a IJIIYsItiAL relation- 
 hip, as exoressed by ' John has the newspaper on his head1 
or '...in front of himt, could be derived 3s an inference 
of the CONTJiO L- level * have *. 
Representative verb forxs for LtlA.:"Gs at the PHY,IL~AL 
level follow: 
ACTUAI, POTENT I& 
R have as part 
contain 
have on 
0 be connected to 
be in 
be on 
be at be ne.tr 
VAL be almost be 
be -value> 
e,g. be red 
The R- and 0-VERBS correspond to relctions between PPs 
identified in (5) (IN, ON, AT, PRDX), whereas the VAL VERBS 
are conceptual attributes--PA dependencies on IPS, 
The ACTUAL/ liQTWIAL distinction as described above 
does not strictly apply to the IJIIYbIC~& level, for in one 
sense all P:IYSIC;LL relationships are nactuellf. Fowever, an 
analogy suggested by this analysis in comparison with the 
other ltvels is discussed in (5). 
2.3, Structures 
Structures reflect abstract ions of verbal concepts, i,~ 
clerncntn of conceptual states or actions which humans 
recognire independently of whether any matter or object 
involved is visible, A VEd structure consists of an 
"ef~cct corn)onentl1 :md, if the concept of chnn~<e is inplied 
in tho Vd<B, a "ccluse componenttt. If both 3 cause and, im 
efEect con~ponent are present, they are connected by a causal 
If: --as for lpiactice (for) '. EEiect structures causal link-- ,,, 
toke one of the following forms: 
0 VAL --attributive bTLITE 
1 
+c ST~iTlil( (new) 
--transition of loc:tiw? STATE 
STATER (0 ld ) 
STAl'EVAL (new) 
--tr,~nsition oi' ~ttributive STAiTE 
ST:LT EVAL (0 ld ) 
STATES, which underlie verbs sbch as 'think1, 'watcht, tc~ntrolt, 
'havet and 'be', have been introduced above. In the verb 
definitiuns to follow, STATES are represented mnemonically 
as '(0 AT R)' and '(0 BE)? or '(0 BE (VAL): )I. 
The transition-arrow should reflect the assumption 
that language-users usdlly focus on a certain aspect of the 
chdnge uuality, e.g. 'start sthtet or 'stop state1, even 
though s toppi~g one state alw.iys me-ins starting another. 
Thus there are three types of change-of-state effects: 
a) TUNs.tTION (complete) mu 1: 
I--* 
b) start TRiNSITION (leave STA'IE) (TR-L) ' 
I 
I-- 
C) finish TILINSITIUN (enter STATE) (TR-E) , 
I-, .$ 
These structures underlie verbs such as a) 'give1, 'pass 
to'; 
Features might be thought of either 1) as - each 
providing an additional dimension of thc matrix in terms of 
its set of vnlues or 2) as applying to structures in various 
combinations of vnlues to f om configurations of feature 
values. 
In the latter case, the ~~nfigurntions provide the 
values (columns ) of one (horizontal ) dimension. In either 
case, the vnlues of the following binary features indicate 
whether a certain conceptual element is present in n VERB. 
The above structures implicitly presume n negntive value for 
all Eeatures except CONTINUOUS, which is positive (c.f. 
Fillmoret s l-bbmentary (1)). Explicitly stat& feature 
values expand the idormdtion given by these structures in 
a WkB descriprion. The fentures,witk '+' and -* examples, arc: 
+/- AGENTive e.g. break (vases)/(vases) break 
Hypothetical believe/contemplate 
SHARED agree/believe 
CONTINUOUS iivddie 
REP eated beat/nit 
VOLitional (voluntary) look/see 
TRY (tried without off er/give 
implied success) 
it is relatcd to the role specifiqntion of tlw vcrb a; dc- 
scribed in section 2.2. The syntactic subject of +.IcYLNI' 
rcrb hns role L\C;:;NT ,~nd thc ob jcut role ;i or O. The feature 
itself is clciincu in a rcbtrictive sense; +/&G ,i':ll refers to 
,d~ethcr ':n qpnt which is external, i.e. other th..n L: or 'ir, 
I-Yl y\ 
i.4 involvhkb. 'thus ' tell ,mu 'give nrc + h (one who 
tolls = .\GAZPT; recipient of inEonl<lti;n = h; infomat on 
received = O), but 'recall1 (in 1: e ucnsc ol 'refi~cnber ) 
2nd ttalcet e -A (one w'lo rec~lls = one ~ho Mreceives!f 
= K; inf cirnl: tion lvreceivedll = 0). 
In order to know when vcrb sub~titutions along o vertical 
dimension 02 thc n trix. c.m be mCde 'Fn~e.~ningfullytt, wc need a 
descripaion sys tat:\ for flNO?~IS.d.,s tt, i. c. concepts thnt serve 
as uobjcctslr at one of the pbavc levels, which governs 
possibilities of dcpcndtncies of these N~I%.IGLLS on the VaBs. 
Furthermore, in ordcr to allow more Elexibility in h~ndling 
the inherently vo;uc problem of. what is neaningf ul, it is 
useful to refer to a tvo-level hierarchy oi Irdegree of 
rcs trictiveness in judging whether such depcndcncies re->re- 
sent nct~phorical phr, ses. In the verb dei'initions, this 
infomation is given in term of s ~ecificotions on the 
NOMIN4Ls which a!)penr in thc dictiont:ry definition or the 
verb. The two dcgrres of retjtricti~n cre rnitrkcd 'B (Broad)' 
and 'N (Narrow) ; the specifications the~nselvcs con,:ist of 
either sgeciiic Ncjla.IN.Ls or fe.ltures of NOhLNALs. Thc~e 
descriptors arc illustrated in ~ectivns 4 ~nd 5. 
3.1. kieatures 
f e* turc-oriented system of clcscrl ption for NOl!lKi~s 
is described in (5). Hcre the defining elements LIL' NOIIINALs 
are presented without olaboraiion, merely to show the tcni~s 
in whicl~ NOT41K'iL de?endencicct on verb 1 eoncc~~ts nre- 
specified. ii configur:.tion of levels f cr SLIP IEJILLs htfs been 
 devise^ ~(r:nch is not i.:cntical to but is rcllitcd to that 
b,tfllY~IXL, TIM,, . Ilowevcr , for chis Limitea dis- 
cussion, VLU3 l~vels rrill be ns.;umrd for NQbIIN,Ls, with 
The feiitwes arc presented in three groqx, altholrgh 
this clividion is not significant to the iimlementntion of the 
theory. The first groun expresses t opo1o:~icnl or b .sic 
physicd properties : 
+/- PART roof, step / house, proof 
bWPE rainbow, idea / log, geograql~y 
CONTAIN shoe / pencil 
FIXED field, tree / bira, ball 
1-UII~WbiONAL fence, streak / ball, flash 
2-UIIZNSIUN.~, ocean, tL:ble / pole, st tue 
FI;UID trpLurr:llr concc~t, river , (some) time 
/ ice, moment- 
It might be seen by the examples given that these fet:turcs 
arc considered :I, rbbstrnct !~rcopcrtics which i?nb c:;ten.~e\. 
to levels other Lhl1n tl~c 1'IIYdIbJiLr 
Thc u~contl group c~nsi~t:; of: 
+~~mtiN will bc c'lns idcrod t o itnpl:; +.'iXliAi11'2. 
1'h~ third group iocu,,cs on thc "n~~::.ninC;" of a conce;-t 
rsther than on an, objective properties: 
+/- LIE k.y , motor, story / boy, stone 
U~TNAI~IC boy, motor, story / key, stom 
The uYNLUAI(: featurc refe-*; not to the rcsencc or ~bsence 
ol;: rn~ ving part;, but rather to whether the concc~t has some 
kind of l~continuous existence" by ~taelf, sther than mere 
spatial presence. The dif fcrence between 'story ' and 'motort 
t 
on one hdnd and key on the ot'lcr is th~t a key is an inert 
object nr71ich is used ~-,~ssivcly for D single o .erntion, .fter 
which it .iguin becviacs mcrel a *~iece of ncta2. A motor 
(like an cnimcte being), once stnrtcc;, r3pe:rs to function 
by itself. Likewise, o story cnu in fact m-st mcntnl concepts 
can be thought Q as having an efeect or ltcontinuous functionn 
for those peo;,le who come in contact with these conce-ts. 
ThL significance of this f eciture is suggcstca by the m~ny 
cases in which peoale s ~eak of +UYN.tUiIC concepts as being 
"al;Lvetl or effective in t hemselvea, 
These features are all essentially binary (1+1, t-t) 
with a possible variable vnlue ('?' ) for sum features. 
The 'FUELt feature, for example, is ltv~riab.bleu. A flower 
is +FlXEU in its natural state, b ut -FIXEU when in a vase or 
in many other circumstnnccs. 
3.2, Function descriptor8 
In addition to c~nccptual features which aetern~ine the 
thcre are specific non-conceptual function associ;\tions which 
apply to may NObiI&sIs, especia~ly +bLii~bE ones, which serve 
as dcrining elements. Also ''sizeM criteria for dependencies 
arc recognided in the form of a 0-5 scale vL.lue for physical 
objects. Although these descriptors ore more ilnnortont for 
problems not dealt with here (see (5) ) , they also enter the 
question of metcphorical interpretations. iior example, the 
knowledge thL:t the functions of both a ship o tractor in- 
cluue the notion of 'goingt or 'movingt is of use in recog- 
nizing the substitution of ip' ior 'tractort in 'the ship 
plowed the seat vs. the Liternl 'tile tractor plowed the field'. 
The function can then be incorporated into 2n approximation of 
the mekning of the former exarn~le. 
aeveral types of function have been identified, according 
to the conceptual roles which the object plays in the action 
which represents the realization of this functi on. The 
type which is probably referred to most extensively in meta- 
phorical intetpretations is l-GXTMN L1, meaning that the 
functional object appears as an qxternal (to the actor) 
object in the conceptual representation of an action which 
serves as an instrument to some rasult. In this experimental 
inq~lcmcntntion an clbbrcviated functidn rc~rascntntfon is 
used: 'kni~e (EN: T (cut))'. 
Method of Interpretation 
r\s imintcd out nt the bc::innin=, of this paper, if n 
dofinition of rnetb~p!lor it: restricted to include only thse 
usages which strike the speaker of the ,ivcn 1nngun;:e as 
p,>etic or colorful, that ucfinition will be uous , 
for 
lansuage is constantly changing with rlapect xo what is- 
considered "originr.1" vs . \ghat is an established word 
sense or idiom. This phen.menon c~uld prove to be a 
quandary for anyone defining verbs or other lexiccl items 
for entry into the dictionary. 'rhc L-uastion af 16a: s~nse 
of il verb is literal md wt1.d is metaph~ric~~l can be cmectec 
to vary not only from one individual to another, but also 
over time, 
In order to alleviate this problem, it is suggested 
that a definition 02 a metaphorical us~:ge include any verb 
which is "borrowed" from arrot\er level, whether or not 
speakers are still conscious of this borrowing. For 
instance, the word 'destroy' is easily conceived of as 
applying tc all levels ('destroy house, image, idea, 
privilege ) . IIoweVer, t Flis system ass ips it to the 
PHYSIWL level, from which it can be borrowed by extension 
to other levels. A verb is simply always deLined as applying 
ltnoru&llylt only to a certain base Level (which in case of 
dbubt con be considered to be the PHYSICAL level, if that 
level is one of the alternatives 1. A human editor wed not 
worry about whether usages of the verb at other levels Bra 
metaphorical. 
Thus the proposed procedures rest on the. essmption 
that the ltmetaphorical senseM of a verb is not in the 
lexicon as such1 the semantic component should exhibit the 
analogy comprehension of humans, who do not need to hove 
such senses explained to them. If we accept that analogies 
refhr to the sharing of a conceptual component, and are 
therefore reflected in our tllevelsv, which share one or more 
columns of our matrix, then the most significant way in 
which the verb description system can be applied is evident: 
glven e verb which is defined in the rn'itris by ~?n entry ig a 
given column (structure) and row [level), a metaphorical 
sense of this verb is represented by a Vm with the same 
strwture but a different level. This type of extension 
can be referred to as "level shiftu. A second type of 
extensioqwhich abstracts the effects of animate acti~ns 
and applies them to inanimate objects is aescribed in 
Section 5.4, 
4.1. Cond-itions on metaphorical extension 
Identification of a meta;>horical usage rec uires the 
knowledge that semantic restrictions od che dependency 
context of' the verb as used on the base level re being 
violnteiL, but that this violation re!>rcc;cnts o. conlprchcnsible 
metaphoricLll substitution rot!ler than &In vanomalnusM case 
which must bc pasb cd to some subsequent routine for inter- 
pretation. In other words, interpretations ror mrtnphorical 
ex!~reslsions must satisfy ccrtclin notions of conceivL*bility, 
just as conccpts underlying liter.*l ushgea do. In terms of 
depqndcncies hctween an object , its location and/or its 
attributes as described above, the most im.\ort<-tnt c~ndition 
for 'tconceivability"l is that the IiENTAL and P!Xs IG-IL levels 
can never tPmixtl across the uepenuency links relrting an 0 
and an R, though both types of levels rn,-y coexist in a P; which 
maps into one of these role-concepts (e.g. 'book1 ). In other 
words, in my metaphorical usage, as in a literal one, some 
correspondence between the typea of comi~oncnts within a con- 
ceptnalizetion nust exist. in terms ofdsyntax, if a direct 
object is conceptually a 3I;NTAL object, thcn the verb must 
be either liZ3TAGlevel or used metaphorically on the iGWAZ, 
level. Thus the dependencies on r verb. 1 concept in a 
rnetaphoricnl use do not conform to the level o'f the verb in 
its literal sense. 
The noun which mat.$ into the object 02 a conceptuali- 
zation determines the level to which other elements of the 
conceptualization conform. In genercl , non- FIiYSIC,>L objects 

FIWD, on the other hand, is less im:~ortant than CoNTAlN 
in the dete mination of a met:r:)hcrical expression. 
In juclg ~ng the consec,uences of the unccrtc~inty ~hich 
may arise in the definition of these criteria, tlnc should 
keep in mind that the aistlnction between c rnetaph~ric~ 1 and 
an incornprehcnsible ex lression is also vnguc i:nd in Iborder- 
line1' cn.:cs mcy vary from onc inuividuol to another. The 
problcn in language understanding is more oftcn to l'ind an 
interpretation rather than to esclude tlstmngew constructs. 
h lexicon editor, theref ore, may In cese cf doubt rezm nably 
adopt a policy cf minimizing the 'Broadt restrictions on 
the NOhINtiLs potevt ially dc .endent on r he verb which is 
being ue~ined. 
A related probl-ern of definitmn is t'7.e inker-3rctation 
of the feeturos in terms of which rhe forcgoin~ restrictions 
=ire ciefined. 'l'he meaning 01 , .e prcscntcd features has been 
brierly describecl for the PtIYalCj\L lcvcl; a m~re comdcte 
interpretation of at ese fe, tures for other lev 1s should 
eventually be concisely ddscribed. For exam le , at t -e 
PHYSICAL level ttlere is a Listinctionbetween 'contain' 
in the sense of 'surround' and 'containt in the sense of 
'consist of'. At the IEhPTAL level these senses merge, or 
ratiyer the former sense seems to lose is relevnncc. 
In addition to the abovc criteria, there are sen~antlc 
criterL1 governing t!~e "tar~cir re->rescntztionw ~?b.lch ensure 
that the interpretation given as output satisfies the general 
requirements of conccp tual dctjendcncy for any conceptual 
structure. Sincc these conditiuns are not peculiar to the 
problem of metaphor itszlf, it is noted here only thnt two 
labels exist w111ch ini~icate Ithow seriously such criteria 
must be takentt. The so-tisf nctiun of tunc~nditi~nal criteria 
ind icntec thnt the resulting interprc t ntion should bc 
accepted in any cas& 'Condition.11 raieru to criteria which 
support a "last resort" interl)rctotion--an interpretation 
to be considered if no better alternativ: s ,Ire available to 
the parser. The in lemented ~~rocedurcs~ do n,?t yet exhibit 
this discriminntioh in their output. 
4.2. Operotiunal context 
%he parser with vrhlch the n~ctnphor interpretation 
procedure is intendcd to f unction (Riesbcck (4) ) operates 
on tl-e basis of scmantic expectations. To a large extent, 
the.;c expectations are concerncu with finding in the sentence 
being pdrsed an object which conf~ms to basic sem: nric 
re uirements ~ovcrning t'-e depenuenc of that object on a 
verb which has appeared in the sentence. If there is more 
than one possible scnse of t .e verb ~~hich has b een found, 
the choice of sense depends on what kind of cn object is 
founc. 
This object is described by a few fer-tur:.s such as 
PHTISZCAL and ANILTE. As the perser presently is uphysically 
orientedM, expecting physical objects for verbj which 
ordinarily are interpreted in a physical sense, it is not 
able to find nn interpretidtion for e::tondcd us.il;cs in which 
the only candidate for on object is non-PHYaICAL. 
l~ore smcif icnlly, appose that the parser finds the 
verb 'dropt i.2 the course of a sentence analysis; that only 
one sensc QL the verb is given in the diction: rynpnrt from 
idiomatic usngcs such as 'drop someone o line' ; and that the 
rniniwl requirements for its object incluLLc thc. specification 
'I?HYSIG.LL~. If this restriction is not satisfied, the porser 
must turn to the metaphor routine for an interpretation. 
Thus if 'ideai vere the only candidate for an object of 
'drop', the parser w ould note thc t c PHYSICAL-level specif i- 
cation (which could be represented as a +I)IIYhICAL feature- 
value) is missing from the definition of *idcat. It would 
then chcckwith the metaphor routine, passing as information 
the c'indidnte for an object ('idcat), the verb sense of 
'dropt which would have been selected, had the object pos- 
sessed o +PIIYSICJ, feature, and any potential dro~per, source 
and/or goal. 
As output the metaphor routine returns a representation 
for each level at which the verb can be interpreted. 13is 
representation, w'iich is based on the serncntic components 
o= 
introduced in Section 2 ('TR-L (0 AT R) (VOL +) ... , 
'idea', R is +ANITWTE, e.g. 'het), provides the information 
to build the corrcc t conceptual structure (or to form im 
o?pro::im te paraphrase according to t *o crop m ucscribcd 
bolo;:J. ih:lt i, this informxtion cont:ims mc trix ilimcndion 
l,ointer; \: icli lcitd Lo he catc' ory oi the inv lvcd oci:ion 
or bTAit~'t: ;.nd to thc .rJT or conccptucil notiition vhich under- 
lies thi:; action or S'l.rTS rr:;pccLivcly. 
For < ur ex: mnplc, 
ij\c und, ?lying conceptu;ll information nssociatcd wit11 the 
nbovc scm ntic coml3oncnt s at the IblA:li'i~L lcvcl 
u detcrmined 
he (-9 ~~~'ILLNS idea t Or, 
the climinc3t ion of a cotnpuncnt of the $1 ;WJ'AL STLiL' 
of :in 
inuivi~ual. 
(Additional notation reprc.;entinz she concc7t 
underlying '+VOL1 is discussed in (7).) 
4.3. -- p~neral procedure 
The g enernl. nctl~od \.E thc metaphcr routine for under- 
stondin:, 1.1et~lph~~rical expres ions c n be spccificd as 
f ol~ows. Thc routine cxamlnes th~ .; emL~nt-ic descril~tor:; of the 
illis scrnantic inf omction c-:nb e obtained directly from the 
diction .ry entry for that verb, or indirecrly in case the 
entry is re!~resented in term of another verb and certain 
f cature va1uc.s. It notes the s;~ecif ied NOBIINAL dependencies, 
including the 'Narrowt s~ecifications on these NOIIIR .Ls, 
if any. The sat isf action-of t'lcso specif icatisns by the 
NuPI~NALs which actually occur in Lhe ini~ut would indicde 
th,-t C-I bcl;e interpretation is available. ;'he routinc thus 
contains the capability of determining such intcrprctations; 
howcvor, in actual 017cri1tion it will be nlisurWd that the 
parser 1) has unsucccssiully checked for t hc :?ossibility 
of base intcrprctaticmz bcf ore turnins to the mt.,phor 
routine, or 2) has found n b.lsc rctprc.;cntntion, but is 
intcres ted in ,mss ible t~letnphoric~ll interprct:rtions. 
Case (2) reflect$ the fdct rh,lt the idcntificxion of a 
base intc;.pretction ~rccludes cnonaly buc not the possibility 
that a m~tnphoric~ll interpretation was ;c.ually-intended. 
This is pL~rticulc:rly likely in the case th,::: the NZllIN Ls 
involved have features which ?lace than on nor2 than one 
level, with the metaphoricnl levdl being more lrusunlll than 
the base level. rm exilmple of 1 type to bc considered 
is tGuropC and hcrico ore drifting apart (. 
In either CDSC, the t,:sk of the routine 1s to determine, 
on the bnsls 02 the zuidelincs of hecCion 4.1, whether there 
are rL~etap'i.~riczl internretat Ions Lor the g iv :n inl>ulr, and, 
if so, to return r e7rcs~-ntations i t  en. Intenretations 
for all possible levels should ultirxtely be ziv-n 73riorities. 
No del'initc met'lod has been stabllshe~~ tor detcrrnining 
priorities in i~olztion from the context of discourse. 
Presumably such context vould bc thL dommating fcctor m 
establishing the lev21 of the expression. T -us if the 
ncti ns of hmns are bein discussed, t+.uropet would be 
interpreted in its institutional or AN1M;X sense rathex 
than its geographical PHYsICAL-~~V~~ sense. 
If the expression is accepted as metap'1oric.11, its 
meaning remains to be repre.mntcd. 
In order to a rrivc at 
the verbal c onccp t which expresses i he tteefect 'I underlying 
the analogy amploycd , the ramgram uses t l~c s tructural o loments 
underlying the input verb as a ltroadmnpi' through the tnlltrix 
to obtain the corresponding target verbal concept at the 
t~esired levcl. That is, the structural elements or feature 
values cim b~ thought of as vclluc; of dimcnuions of the 
matrix which specify an entry. This entry, which may consist 
of a 7rimitive AC1i', for exzmple, con then be inserted into 
the representation which gives an zpproxirnti~ion of he 
meaning of the phrase. 
dong with structurcll elements, any magnitude descriptors 
present, i.e. rULUNT or INA ..UaI1lY: > , < ore carricd l~ver to 
the targelrepresentation, since it is ire uently these com- 
panents which ;re Socused on in a mt:taphoric.J expression 
( 'he jumped (IidLIZjITY: > ) to conclusionst ). However, the 
program reterred to here doe; not yet include this mec!~anism. 
4.4, Operation of routine 
The procedure to be described has been irwlemented in 
an extended version oi F&TRHN IV, which wal the only 
language conveniently accessible at the time. 
me outline 
given here represents the 
rocedure nc~ually followed in 
the implemcntntion, which was dcsi,gned only Lur test 
a) 
Input: Yhc in,>ut: con~ists ol: two or thmc lexical items 
in Lheir "roottt forms in the order 'noun verb (noun)'. This 
group regrescnts a synt nc Li~ conr'igurat ion d~tcrmincd 
tcntotivcly by the pnrsor as 'subject verb1 pr 'subject 
verb object '. In terms of roles, the first case may 
re2resent 'AGENT VV.LGt or '0 (OBJECT) VLABt; the second 
VB 0 or '0 VUG R (S~WC or GOAL)*. Theoret- 
ically, then, the entire role conf $swation 'AGENT WRB 
OD JCCP &3;11R08 and/or GO&' need not explicitly b e provided 
for in the fnput, since t'lis configuration is covered by the 
two component configurations just given. 
b) Dic~ionary definitions : First, the ucmnnric definitiom 
of all items are retrieved from the uictionary. Examples : 
(noun) ship ((PIES) T -1 (GONT +) (1 -1 (XU +) (2D -1 
(S&iPL +) (SILZ 3) (FLUIU -) (AN111 -) 
(>a1 +) (FN: EXT (sail) 1)) 
(verb) plow ((PIIYS) TR-E (bTATZ (0 BE bTJ 9E: )) (.G.,NT +) 
(WLE 0) (I?'STR: TLIXST.&sTE (0 AT K))) 
(0 (hBW land) (BRU (2D +) (FIXED +)I) )) 
Control is then gassed to the 'subject verb1 (SV) or 
'subject verb objectt(SV0) routine for determinction of roles. 
C) 
Roles : At this point of the procedure, roles to b e 
assigned are only temporary; a test for the lfR-O switchTr 
type of metaphor (Section 5.3), for mnstance, may determine 
thct the role configuration ex,)ected on the basis of syntactic 
informi~dion has been altered in the extended use. 
The tentative ro1.e~ are assigned nkcording to role 
informat &on !;hen in the definition OX the verb: 
FO~ SV: Role oi verb (F; or 0) is assigned to subject. 
- .L 
I) 
For SVO: If verb is +AGXNZ': 
- 
AGdNI' is assigned to subject and role of 
verb is assigned to oFjoct. 
- - 
If verb is -AGENlt: 
Role of verb (K or 0) is assigned to subject 
and the 'tither role (O or B respectively) 
is assigned to - object. 
d) ~nterpretations: Control is then passed to other 
routines, depending on which role configuration is present: 
RV 
ov 
RVO 
OVR 
Am 
AVO 
These routines return any interpretations found, a ccodin;, 
to the criteria to follow. In this version the interpretations 
are expressed as pseudo-paraphrases, i.e. p.lraphrases which 
ignore certzin syntactic details such as word suffixes and 
tenses, in order to allow for some measure d judgment as to 
;he extent to rhich the meaning of the metaphcricel phrase 
is captured. However, in actuzl operation, the target 
representation will be a c-0ncc~:tual one, which co~ld be 
operated on by a dialogue program or by a paraphrase program 
4.5. Tests and criteria 
The following  test:^ with corrosnoncung criteria for appli- 
c,ition ern{ for success represent procedures which hme been in- 
plernented. Each test (b through d) refers to a certuin type of 
metaphor 03 shown. Tho discussion of relev nt exk mnles in the 
next section complements these speci~ications by bdicating the 
rationt:lc used in the oprjroech to finding mctnphoric 1 inter- 
prctations. .\ teat for n b~se-lcvcl intcrpr~t~ction (a) has been 
inclurred for ilurposes of cort~parison with examples seen as either 
metaphorical or (with respect to the given toot ) anornolous. 
a) 
- Base Level (alrunys tried) 
1) 
All EU'CHIEAU are consistent with bnse level of verb, ie. : 
level of 0 is base level of the verb; 
R for any -PHYSIC.iL verb is + iBIi -.Ti3 or 
has an .WIE,L~TC function (e.g. co~~uter') ; 
R for any +pW6IC.L verb is +,FYoIC.rL; 
2 ) A11 NOELINHLS fulfill Narrow specifications found 
in the dcf inition OF' the verb. 
Interpreted: IIe drank the ink 
The ship Lisintegrntcu 
Not interpreted: The chzir arsnk the ink 
He closcd his mind 
b) Intra-"level (PWalCAL) Feature Shift (tric~ lf cll items 
have PZIY,ICld, level, but base interpret tion fL:ils): 
Actor-fecture shift : 
1) Verb specifies +ANIU. TL fe~ture for R; 
2) R is +PHYaICsL but not +ANIMATE; 
3) 0 Eulr'ills *Narrowt specif ioations found in verb 
definition, or 0 is absent. 
Interpreted : 
The ch~:ir drank the ink 
Not inter~~reted: l'he ship plowed the sea 
~b jeer- f eaeure shift : 
~ubject and object fulfill the 'Broad' but not 
necessarily %he 'Narrowt spoc ificr tions by 
verb definition, 
Interpreted : 
The ship plowed t hc sea 
The skier plowed the sea 
(Thc 'Broad1 specification for the subject here 
is ltsometkiw which goes", i. e . "something 
which changes location: 'TR (0 AT R) ' . ) 
interpreted: The ch~.ir p',owed the sea 
C) Levcl Shift (from MIY~ICAL level only, at present) 
(tried for each possible level of the object when no 
base interpretation is found or icr all levels when 
input f ormht is (subject (+ANIMdL) verb)) : 
1) R is either absent, +ANIB;'iTE or 
an ANIbL.ZdE (Xnalieneble ) PART ,' i. a. 
4 PP 7ANIM TE~ (IilART: <R +'IENTAL >) or 
<PP +ANIMATE> (IPART: (R +~;-.N~u.S> (eye, tta\t, etc.) 
for ~LNTAL and SXNSOKY leva s respectively; 
2) R and 0 fulfill 'Broad1 specifications by 
verb definition. 
Interpreted: He closed I~is mind 
Not interpreted: He closed his ~rosperity 
c') Category Shift (tried ::hen no base inter~retation 
is found): 
1) O is some attribute of R or oL a lexically absent 
conceptu a1 NOMINAL; 
2) R fulfills condition tc-l*; 
Interpreted: ~osperity disintegrated 
His indif ferenco collapsed 
Pro.qm5.ty came to the country 
Not interpreted: rrosperity was occupi~d 
Prosperity cimo to :-he choir 
d) 
Level - Shift with R-0 Switch 
(tried when nc> base 
-- 
interpretation is found ur when implied source (goal) 
is not explicitly present): 
1) 
Source or goo1 (temporarily assigned rolc kt) 
has level bLNT"U,, sF:Kx?&Y or CQKti'RGL; 
2) Temporary 0 is +~ULSE; 
3) Source or goal iulf ills ' BroL.dv specifications 
for 0 given in verb definition. 
1nterpretp.d : The country leapt t~ ?ros;\erity 
Not interpreted: Thc chair leapt to ;>rosperii.y 
bme SC~IF~~S of interpretations are $hen in Figure 1, 
which rei)resents actual output. In ut datci is ~ivcn in 
Figure 2. Particular details or' the procedures used ire given 
along with ciiscussion of these cnd other ex, mples . s they 
occur in the following expositian cf the various types of 
metaphor. 
5.1. Level shift 
Not 211 extensions are made from the PHYSIC& to the 
non-PHYSIC;& levels. The KNTAL, s ,NSOl.Y and CQXrRCIL levels 
sometimes serve nj a base irom which metaphorical extensions 
can be made. The examples which rollow indicnte certain 
extensions (eotnc of which have evolved into idioms) which 
can be made between levels. Some types of extension are 
obviously more frequent or interesting than others. 
Examples for specified extensions are: 
PHYSICAL - $EN?!&: He closed hls mind, 
Protests rained upon the govemmcnt, 
Euro!~o and America are dri~ting apart. 
Kohoutek's tail points to its origin. 
(.lmbiguous between PIIYSICAL and 
MENTAL levels. On M:NTALI level, 
a refers to 'in£orrnc9tion 2. bout 
Kohoutekls tail1 and 'origin* to 
tinformtion about origint.) 
PHYSICAL - SENSORY: Music floodcd the room. 
PHYdfCA - CONTROL: 
The privilege of cleaning the erdsers 
landed in his lap. 
Control of the situation slipped away, 
MLFlT& - PHYSIW: That chocolate didnt t agrce with me. 
SENSORY - MEW&: I searched for an cnswer. 
Let us x-ray this political party. 
SEIQSORY - CONTROL: Their rights disappeared one by one. 
CONl'ROL - PENT&: She offered him an idea. 
CONTROL - SENSORY: Her hat usurped his view. 
'He closed his mind! appears 
in the output of Figure 1. 
The base--i.e. PEIYsICAL--definition of fcloset is one of the 
more complicated verb definitions, s ince the syntactic object 
is either a space - e an object containing the space (which is 
filled or eliminated), and its complete representation will 
not be discussed here. However, the "effectw portion of 
the semantic representation ior tim scnsc in which the ~b ject 
is a space is th-t nothing can pass into or cut of the object 
containing the space. The rclev.~nt: portion oz the "pass intoH 
internretiition is renresented by the nested ueLinitLn: 
(close ( (PHYs) (TR-L) (STdL; ((iIYP +) Tk-E (S'I'JL'TE (0 IN R)))) 
{ROLE R) (riGdNT %) . 
(R (NR\T (0 +) (BKD C (CONY +I)) )). 
It is noted that the *HY2 Value refers to the i~otentiol charnc- 
ter of the outemnost ST.iTE; the HYP vl\lue fen. the innermost 
STATE is negative, honsistont with the obscrv~ition in Section 2.2 
th'. t all pt~ysicijl relationships (excluding separdtion) can be 
considered ltactuallr. 
The role routine determines th\ t, since *close1 is R-role, 
the uirect object of 'close , i.e. 'his mind l , maps into R, But 
the base-interpret<!tion routine then discovers that ii is +MENTAL 
and not +WIalC .L as requireu by 'close'. The vroSr;im therefore 
attempts an interpretation ct the lAdX~' L lcvcl, she level of 
'mindt. R = 'mindt is an ANIId1'L P (*he CI:.d'I': nind = 
+) in fulfillment oE criterion -1 It can therefore 
serve as a tllocation'' at the 2, :NTIIL level. 0 is unspecified 
in the sentence and thus does not impose any level- or other 
restrictions. In checking to see til-t the 'Eroau' sl,ecifica- 
tions by *c~ose' are satisfied by *mindt, the progrm finds 
that 'mind does h;rve the +CGSI'-;IN fedture 2s require~l. 
'fie verb can therefore be interxeted st the XENTAL level. 
The innermost structure--TR-E (STA." (0 .Z R))--is extracted 
and the ~.:ZNTIIL level is substituted for the Y"HY,>IC&* The 
absence of 8 VE~~UC indication is interprerea as + The 
4s 
ttroodmapw through the matrix  ort ti on given in ~~iigure 2 then 
consists of the dimensions r 'EUNTiiL (lcvcl) P, (sublevel) 
R (kol*) SGZI'E (structure) -1HP (feotura] + (vLLlue) '. For pur- 
poses ~f pi~r.~nhmsing directly out 
of the matrix, the entry re- 
sultins from this s o,.rch is the hnglish verb v think1, which 
x~ould corres!~ond to a concq)tuc~l structure 0 (=> ~,L~LOC: (Cp(K) ), 
i.e. be (mentally) located in the conscious processor of Kt. 
kith closer attention to the sublevel(s) of vmind', a more spe- 
cif ic e%yression could be deternitled. For instance, an asso- 
ciation of both the P and V sublevrls with !tnin(J yiclds 
'think about- the truth of.. . * and ' thinlc about doing.. . '. 
In order to complete the parilphrase, the progrcm assumes 
thiit ther hds the silme referent as and notes from the 
clef inition of 'mind1 that 'mind1 is an IPAR'J! of this referent. 
It then picks up those conceptual elements uf tcloset other than 
those describing the innermost underlying aT4%TE--(TH-L aT.bE 
(IHP +) 'SR-E)--and submits the entire 
list of elements (he 
(IPART: mind) Tit-L STATE (HYP +) TLE think) to a simple mapout 
routine. This gives the rlparaphroselr 'he (Ii'AIRT: mind) STOP 
WSSIBUXTY-OF START thinkt. An actual generator could arrive 
at paraphrases such as 'he stopped thinking '. 
21 correspondtng 
0-role interpretation would be 'nothing ccn stcrt to b e in 
(i.e. enter) his mindt, which is also a reasonable approximation. 
The aspect of tclosel whieh implies th~t ncthing can leavc 
rather than enter would yield 'he started to keep everything in 
his mindT. This might be uwerstood- in terms of not forgetting 
or not expressing oneself. 
The inability of the described 
- 
method to identify exactly which meaning is intended is the 
price of its objectivity and flexibility in being able to 
arrive at an interpretgtioi~ with no previous knowledge of 
what such phrases refer to in a givcn culture. 
In this 
respect the model represents the abilities of o Language-user 
who is unfamilinr with the idioms of the s?eskers of his 
environment. 
5.. 2. Category silift 
A "category shiftu refers to the feet that instead of a 
PP or a conceptual NOMIN L, an attributive or verbal concept 
in the form of a noun appears as the concept which has been 
assigned role 0. liore specifically, these concc,~ts are either 
l'conceptual attributesu (of objects) as repredented by 
'color1, 'truthi, 'bec.utyt, 1 , etc., or attributes of 
aninxtte 11s \;hich r.~ight be described as iLhTAL-, bENSQ1-Y- or 
CONTROL level VEKBs in which the focus is on R rather than 
on the relctionship between R and some 0, as represented by 
tconfitiencet, !perception ', 'pos;ession', etc. t;etnpl~orical 
uses involving either type of concept often involve level 
shift, as in 'its vaLue der'latedt or 'she built up his 
confidencet. In this sense they resemble that class of 
expressions designated simply as "Level shift". Also, the 
procedures for interpreting the (level- and) category-shift 
'they decimated his joyt and the level-shift 'they decimated 
his version of the accidentt are similar. 
However, the designation of a cata2,ory shift allows 
for the interpretation of the metaphorical ?hLs smile 
disintegrated*, which docs not involve n level shift in the 
sense described above. In 8 ddition, this d~~sign tion 
preserves the theoretical distinct lon bctwcen ' joy1 as an 
(anirnmtb) attribute and the NCMINtiL '~crsion~ ('story', otc.) 
as a concept \:hich is isolated f Porn its anint te source, a 
distjnction which is reali~ed in the definition of 0 in each 
cade. Rather than being defined as a Nu~~~NAL, 'joy' or the 
noun *smilet is dcf ined as a noun with a bilsic VAB structure 
and level. 
NOMINAL fec~tures, which are not as criticL.l for non- 
PIIYdIW as for PHYSIC ,L NOMINALS, ore even loss distinguish- 
abLe for attributes. The .&NIh TE feature divldes the class 
of attributcs as described abvvc; in general, however, 
feature v.ilues nrc presc~~tly ignored US possible rcstrictions 
on metaphorical uses of attributes. 
A few examples of clitegory shift in which the toi.ical 
focus is on an qttribute rather than on a human experiencer 
K, illustrate auditional points concerning metap'lorical 
inter~retations. One of the general problems of meta- 
phorical interpretation is to show in the re wesentation of 
a phrase the analogy to a conceptual object , as well as 
'%hat is re3119 happeningg". 
Theb in7ut ex.irn,>le 'his sb..ile disintegratedt should 
produce the same representation as that of 'he stopped 
smilingt. Yet the :~bilit-y to thus relate these expressions 
must be b.~se~ on some underlying similarity with intuitive 
appeal. 
This tnslc rcrluires a verb def initi.cn procedure such 
as the one presented here, r:hich rcsts on n sm 11 number of 
conceptual elements. The primitive element of TRiNS ITIuN 
underlies 'he sto~ped smiling , 'his smile disintcgm~ edt 
and l his smile left himt, even though on tho surf ncc it 
appears only to underlie the latter (third) form of ex- 
pren\>ion. Phis element is expressed in our semnntic repre- 
sentation as TR-L, From there the TR-L structure could be 
incorporated into a conceptual diagram in a number of ways, 
e 
e.g. by n transition arrow I _I , by a ttcause-to-notn 
structure 9 or by a "f inish-ACTtt notation 
L, 
Lf 
, Thc latter notation is the one actually used in 
conceptual dependency for exampleaof the type @en. This 
notation does not express any relationship between the three 
forms of the exxn7le given above. Howevef, it is rnnpyed out 
oL TR-L, which does show this relatianship and is referred 
to in the following analysis of 'his indifference disintegratedv 
Briefly, 'disintegrate ' ~s defined as changing from 
existence to non-existence of an object, on the PHYSICAL 
level: ((PH) TTt-L (STATE 0 BE)...). since 'indifferencev 
is not consxstent wit11 the PHYSICAL level, a base-level 
interpretation fails. Since 'he ' is +ANIIL?LTE, the IEmAL- 
level noun 'indifferencet can serve as an attribute of 'het 
as ex,eriencer A. 
There are no feature specifications which 
mu;t be fulfilled by the attribute * indiffercncc ' ; theref ore 
there crc no obstacles to an intc?rprat,ition on the PiJY'l'AL 
level. Thc delfinition of 'indiffcrcncet yields the descriptors 
'((1.C A) (sT.:C& (0 rrf R) (VAL +-)))', where 'A' represents 
iVAL, i.e. the LGNTd, sublevel 'Attituder, and +- is the 
value ..or 'neither positive nor negnt ive'. 
'mdifferencel can be defined in terms OL either the 
LTM or thc dP, i.e. either as +HYP or -1IYP; +IiYP is arbitrarily 
assumed ftir non-PlNSIQi'L concepts. 
The program thus enters 
the matrix with dimension informarion (b& A K bTL!1'E +H +- ) to 
obtain a corres1,onding K-role VEI\D. It finds 'be-indif f erent- 
to1, uses the R nlrenriy determined as subject and adds struc- 
ture element TR-L given by the ver,b to obtain 'he STOP be- 
indifferent-to...', leaving n slot for the object of the 
indif iercnce. 
The procedure is similnr for 'his smile disintegrated '. 
The mogrcm determines o cdtegory ~hift and nccet*s the TR-L 
structure for 'disintegratef with the Itloss of existcncell of 
0 (0 BE) interpreted as che ttloss of state1' of ii (0 AT R). 
Thus the resulting R-role representation is the sane as thdt 
for 'he stopped smilingt. (It might be noted thst the sub- 
stitution of a concept sucha s 'smilet for n physical object 
could be represented as a PHYSIU-to-ACTIVC,sbift, if an 
ACTIVE level ispostulated for HO3lIN~iLs snd VERBS (5). 
pursuit of this approach would designate this example, like 
the other exmples of this section, as n case of levL 1 s hift. ) 
T~Q excin~plc 'truth burned up ' is dismissed by Katz (3) 
as som.,ntic:llly lous. However, if hwns ctm unders tend 
sentences involving verbs which e:-lpnrently v iolnte selectional 
restrictions, tlxen such expresstons are i:lw zubject to 
computer understanding. In terms oC the matrix, 'truth ', 
a conceptual attribute, represents a positive v.:lue of on 
attribute of n (lexically absent ) EC4N'i'tU ub joct, which is 
in turn dominated by a (lexically absent) R. kince 'burn upt 
differs from Tdisintegratet only in the me.nz or manner of 
thc aation, ehe endysis of this ex~+mple is similcjr to that 
of 'his indifference disintegrated*, with a shift to the 
TVAL- rather than to the sublevel. By Lllo~ing for the 
assumed R and 0, the output routine con obtain the opproxi- 
motion 'one STO: know.. . ( tpeople storlped knowing ) or 
one STOP ialriTE-TO true inf orrna tion ( 'people s topped 
h'iving or telling the truth ). Thus lilthough contextual 
restrictions on 'burn upt would inuicnte a +PiTYSlC=rL 
NOMINAL as actor, the program st ill lluricierst andstr the usage 
while recognizing that it is not i' base or ltnormalv usage. 
Thls is possible because the system is01 tes rhe ~ri~itive 
structure of a verb from its ordir.sry select ional restrictions. 
5.3. R-0 switch 
B-0 switch isexemplif ied by 't le country leapt to 
prosperity' in that 'prosperity ' rather than 'countryt 
sppeors to be the goal :>nd is thus initially oadigned role R 
r. ther th~n 0. This kind or metaphor may actually includc 
B category shift (which itself may include n levcl shift), 
and is unad to express a change of state (of 'countryt) ns 
a transitioh (of lcountry' ). 
Looking 1,113 prosperity in the dict iotlLbry, tho program 
find 1 (prosperity ((GO I3 W) (hOLC R) (ST.;I'E (t) (&*IT 3) ) 
(R (Hum +I) (BRL (?IUCLN +)1) 1). 
That is, 'prosperity1 maps into on attribute on the CL.NTH)L 
levcl (Extrinsic 1L control of - PHysicol concepts), is positively 
valued, of o great MiOUNT, and dependent on any 4HULW concept. 
'Country' has tho f rliture ref uired for an R on the CUNTIUL 
level (+ANIF!), and further, it satisfies the +HWJ;N 
specification demimded by 'prosperity ' ; 'leap to specifies 
no particular restriction &or li other than +PHY,ICAL. 
\.e thereiore wish to take over the structure for (leap tot, 
but to indicate the CONTRPL rather than the FHY: ICAL level. 
The structurc essentially is TR-C as found in the definition 
of the 0-role varb 'leap-to '. The conce:~t of 'start to b eq 
or 'become' which underlies TR-E at any level is transformed 
to 'start to have' in m R-role ep ression. The object of 
the control involved, in 'proq erity , which is given as 
'PHYJICAL', can b e mapped into the word 'rnaterizlt for 
purposes of generation. Thus our represent at ion yields 
the R-role Tcountry SPART have-materialT in the implementation, 
and could be the basis for other non-mctnphorical paraphrnses 
such a; tl~c b-role 'the country 13: s bcsorning prosncrous I. 
.;nothcr esrmplc, intcrestin:, bccLruc;c potcntinlly all 
lcvcl; ~lrc involvcd in its mctaphoricol internict:~t ion n, is 
liurope ancl Ancrica arc drifting apart1. Subject to the 
contest of the ~iiscour;c, thc ~nct,t,)horical intcrnrr tit tion 
in this cnsc 111.y turn out to bc n more likely interprctation 
than thc bt se (:IIYoI~.'~L) one. 'drift apart1 is defined in 
the ~ictionnry as a symmetric, i.e. +&WU verb on the 
I? YbI&iL level: 
((PIES) T&L (bT TE (0 AT R)) (NIL 0) (::Gtlrn' -)...(AKSZIJ +I). 
Si-.ce the syntactic joint actors, tl;uro!>e anSl '2merica1, are 
both deiined ~13 having a +Pi;YdIGiiL comnonent, i.e. their 
geographical areas, we have the PINaI(;& interprctction thot 
the continents of Lurope and mcric': .are in the process of 
going away iron one another. That is, ,uro?e or America or 
both ore loalng the loCation they once ;ih:)red. 
Since 'driftt notentially takes a source or @a1 as 
indicatca by 'r-iT Rt, the excmple satisfies thc condition for 
testing for an R-0 switch. The mis~ing NOSiINAL or attribute 
(which rrould correspond to 'prosperityt in the previous 
exzmple) imnlieri in the sentcnce czn hzve any level, since 
it is not emlicitly given. '2uropet and ' mcricaf as 
institutions fulfill the +ANINATE condition for R: 
(~uro;~e/lbrerica ( ,.. (ANIII +) .. . ) ) The level of the 
nissin~ NOhlINAT, or attribute from wh-lch they are drifting 
is mknom. 
Thus the. program determines that interpret at ions 
on the ILNTAL, LGN;N~CJRY and CONi'ROL levels arc also possible. 
On the PdNTA level, the abovc structurc for 'drift apart' 
is the structure which underlies a possible poratjhrase gen- 
crntion of 'Europe and iuncrica no Longer zqgrcet; on the 
SLNSORY love1 it is the structurc for 'Lurope clnrl America 
no lon~er perceive the same things ' ; onti on the LONTIiOL level 
it is the structurc lor 'Curope and .dnerico no longer have 
the same rights, responsibilities or types of control!. 
5.4.. Intra-level feature shift 
In the level shifts descrised above, a verb is usually 
borrowed frbm bne level znd aaplied at the level of the object 
with which it will be used. In intra-lcvei ieature shifts, 
all components conform to the same level, usually the PINSIC&, 
but o apccification(s) or t'eature(s ) of the ob jept is viol~ted. 
h%en the +HNLIvGiTE feature of an actor is violated, n kind of 
personification or anthropomorphic behavior results, as in 
'the chair drank the ink1. This ccm be referred to as an 
Itactor-£ eature shiftu. 
(1% corresponding ex-le on the IENTkU 
level might be 'that painting ssys something to me1, where 
the psinting does not literally say anything, but the result of 
looking at the painting is the same as if something had been 
said. ) If, however, it is the object which docs not meet 
the snecif iczt ions of the verb definition md yet the 13hrnse 
is ~c~rn~~~ehcnsible~, thcrc is on 'tab joct-ieature +if t". 
An cxcmple is the ship plowed t he e eo r. 
5.4,l. kictor-Eenturc shift 
As stl~tcd above, there is no change in lcvcl this 
typc of mctt>p3~or, but t he +.it!.:IPI .T,: rcstr.icti\m on tho actor 
is violated. Thus 'tl~c chair drcnk.tb:c inkT i. .n example 
of intra-lcvcl hilt, b ut ' ' t!~e boy uranul: in the ->oetry ' is 
not, as it involves an estension to a d1r"ierent lcvcl. In 
general, the semantic ret uirements on the object of such an 
eq~rcssion nrc the scme as tho.;c in r n~n-mctap:~orical usage. 
m 'tibe chair dz.nk the inkt, both the 'c:nirt 2nd the *inkt 
arc ordins ry phy- ical concepts , althoush the use of 'drink 
is not quitc the ordire ry one. 
m e::;.~.~~nation of this 
exa~xple by Lhc jcta nt ic corn-oncnt rcvcnls nothin unusual 
.~bou; 'drank the inkt ; inkt is +P!;oibd~L and +FLUId as 
re< uireu by thc NarTowt speciiications of 'drinkt. t clIc * t 9 
though ++A~Y~~CLIL, is noted to lack the +LiNIiiiTZ feature value 
speciiied by 'drinkt, so the ordinary sense is rejected, 
while the conditions ior an actor-feature shiC t cre satisfied. 
The 
determination of a metaphorical interpretadon 
iup1i.e~ that the effects or Linguistic inr'hrencc; deriv~ble 
from the underlying conceptualization are sirnil& r to those 
deriv~ble iron n conccp t ualizat ion containin; the litaral 
4 
sense oi tarink', which is 'to 1KGG;SST a +PHYSIC.Ly d?LUII) 
substancet, Since the in13ut: example is already in R-role 
form, i.e. with tho Recipient as subiect, an 0-role form 
is given as pmaphrase in the output. Becausc thc structure 
is a TR3 one, with 0 = 'ink', it is known that the ink was 
removed Errlrn somewhere and is now in the chair. The inf or- 
motion given as o result is 'ink STAiXT BR in chairt. Con- 
sidering other vari:~tions on this input, we note th.lt we could 
not readily interpret 'the blotter drank the chrir', since 
'chair is -r'ZUIU, 
5.4.2. Object-feature shift 
The example 'the ship plowed the sea' fails n literal 
interpretation on the basis crf the definition of tpl~wt : 
(PLOW ((1~3X3) TR-I: (bTlbE (0 BE SHAPE: )) (ROLE 0) (AGENT +) 
.., (IasTRt TR (STATE (O L~T R))) 
(O (NR~ land) (B!U (212 +) (~1ml.1 +)I) I), 
since 'scat is not a synonym for and However, 'seat 
fulrills the '8rond1 specifications of 'plow' for 0. 'Shipt, 
the syntactic subject, is assigned the role c.f AGXNT. In 
(S), it is explained that an AGENT, i.e. n NCEIINAX which has 
some role in a causative action, either 1) is +;th'lbL.TE, 
2) itself represents an action and therefore has the ACTIVE 
level, or 3) hac a specific func'iisn whicil entor:. into the 
causation. Since 'shipf isneither +.rNIB2~'l'i: nor ACTIVE-level, 
it is assumed to have a functional role in the causing 
conceptualization. The program checks to see that certain 
ranuirements oi an instrumental involvement of 'ship1 in 
lplowing' nro fulfilled. The function Qr' vshipl is given 
as 'sail. '. The strucf ure ufidcrlying Js.?il is TK (:iT.;TE 
(0 AT R)). Although the noun vnlo~vt might be giwn as the 
cx )licit instrument of tho verb .'olow , the program ignores 
the failure to cgree with such specific infom:a~tion, just ;AS 
it ignores Narrow restrict ions on' ob jocts when considering 
metnphoricnl cx 3ressions. On examinlbg the gcncral structure 
given for the instrumental concoptunlizotian of '?low', the 
4 
program finds TR (bT.SE: (0 AT R) ) , which aI;rees vith the 
verb structure of the Punc tion of s . In othsr words, 
although only o 'plow' can truly '?low1, in a rneta~horical 
interoretation anything' xhich tt~hysically goesu can con- 
ceivably h..ve a llplow-like" ef fcct. i'hc ?rosrarn the=£ ore 
arrives at the roq:h intcrprctotion 'ship DO (sen 3 dT 
(BE SIIAPE: ) . Conceptual dcpen..ency rules would then 
transform 'ship DO' into a structure corresnonding to 
'one o7ernte ship'. 
~~nsider now a vzrb- - 'kill --w?~ich is subject: to 
metaphorical use. but in a non- straight£ orword way, since 
a level shift and/or-an - object-feature sh -ft may k involved. 
Iq the exam2Les 'John Bilged the cat' and 'the House killed 
the bill ', the ordinary object of 1 ( cct ) is no more 
a mere ;d~ysical ob j&ct than the metcpY.oricc1 object (tbillt ) 
is a mere mental object. The P~?YSI~-to-i~iiW.~L extension 
in the aecond example is obscured by the simultaneous 
presence of an object-feature shift. 
To clarify the role which each type of shift plays, a 
similar example is first presented which involves only 
ob ject-feat- s hift: 'he killed thcmotor I. This example 
could be more explicitly paraphrased as 'he did something 
which caused the motar to die*, The interpretation depends 
on what it mewrr-fot the cbject. *motdrt, to 'die1 or 'be 
dead'. It would be deo%rablc 'to obtain the interpretation 
'he stopped the opekathn or rning of the motorv, while 
rejecting a sb-s intcrplrtation for 'he killea the stone1. 
*Motor* and 'stone * are,both RIYaICAL NObIINC\Ls; no level 
shift is involved. 
Rather, the +ANII~IYZ specification on 
the object is violated, yielding a mctaphoriccl, interpreta- 
tion in the first case and no interpretation in the second. 
?Motort is a medningful object of 'kill1 because it is a 
~~IC NOMIN L-, its function king to run. ( It is recalled 
from Section 3.1 that the +DYNAMIC feature value specifies a 
function which can be identiEied with the particular meaning 
of a NOMINAL.) When a motor is ?killeda, this function 
attribute is eliminated-- a consequence which. differs from 
e.g. the disintegration of the motor which might represent 
its being 'destroyed'. 
The proceaure of the program operating on the first two 
examples canbe outlined as follows. The semantic repre- 
sentation for 'kill is : 
(kill ((PHYS) TR-L (STATE (0 <FN(O)>)) (ROLL 0) (IGDIT +) 
(TRY-1 .a. 
CO (N1W (AN'IM t))) (BRD (UYN +)) )). 
For the exmple 'John killed the cat I, the progrim will find 
that all specifications on the NOMIN &s by the definition of 
the verb ere met by the wrds of the in3ut. 
In p.lrticular, 
'u 'Pttrd'n'tt is + '.~TIBk'~TE; thdt is, in terms of DYNAMIC FUNCfIOK, 
it 'livest in a literal sense, Thus the Liter,til sense of 
'kill' 'is accepted. If the inbuttJohn killed the stonet is 
encountered, the program notes that ' stone has no r.'~NIEtiTE 
feature vCllue and therefore fails a base interpretation. 
Furthermore, 'st one isno t +DYNAMIC and theref ore does not 
satisfy the 'Broadt specifications necessdry for a mcta- 
phorical interpretation. 
The input 'the House killed the bill presents a more 
interesting case, '~ouse' in the sense of '~ouse of Repre- 
sentative~~ or 'Lower HouseT h:*s the +ANIPLiTX feature 
preferred by the target representation conditions on an 
AGE;NT as specified above- with respect to ' shipt; however, 
(bill1 does not hove the +ANINATE feature value as required 
by 'killt. Thus a base-level interpretation is rejected. 
IIowever, tbill does have the +DYNAPlIC v~lue, corresponding 
to the observation that it has a ucontinuous effectly on people. 
Thus the basic components are satisfied for an interpretation. 
since the dispensable 'Narrow +ANI~$~TE descriptor, i. e. the 
literal live' function, is viol~ted, b ut the minimal, i. e. 
'Brood1 requirements are fulfilled, the employment of 'killt 
is considered a metaphoricill extension from the l+IIYSfCAL to 
the EZ1:NTAL level, The structure TR-L (CL" 0 NO ) ) 
(AGENT +) then yields the paraphrase lHouse &TOP bill become 
law', where 'lawt is n COXI'WL concept re~~reacntcd in terms 
of lone mustt and lone nmyt. 
Gy noti= that which il; co\~-unon to both the base sense 
and metnphorical senses of 'kill', we c.ln cotl-tp rc the m&nn- 
ings of thcoc srnscu. 
The underlying structurc oE the verb 
itself spe~izies in $11 cases that on 2ction was successfully 
tL4ken to elbin~te the 1)YNUIIC: function or effectivaness of 
the object. 'fie cffect component oi this structurc says 
that the Object no longer ~xists in its previous State, for 
this is t71e interuretntion assigned to the TR-L structure. 
Thus the ccit no longer lives; the otutor no longer runs; 
consideration of the bill stops, and the - intended result, 
A. 
r; I 
defined conceptually as is prevented--thus the prohibi- 
tion, order or permission contained in the bill is never 
realized, 
5.5, Koun compounds 
This type of metaphor andysis can also be applied to 
noun compounds in which the nouns are defined in terms of 
verb~l concepts. The further ~.evelopment of the above rnech- 
animsms must precede an implementation of the more coclplex 
now-compound metaphor analysis ; however, the =.p,n-oach to 
interpretation of such constructs can be indicated. An 
example is given by the noun compound idea factoryt, \.hich 
is close in nleoning to the verb-noun cdmnound 'think tankt. 
If the +I'llfolU~L objects or mc:tter usuL.lly asvociatcd with 
f :;ctory or 'tank1 arc ignored, noun-coml~ound intcr~xxt,~tion 
pr0cedurc.s (5) can be used to arrive at 'institution which 
m~kes $deasl or lcnvironment in which one thinksq respectively. 
Ncre tllc verb 'think and the noun 'idea ' , which is an object 
ot thought, retain their litcrcll srnsa, 'c~!~ereus the functions 
underlying 'factory ( tmalcc ) and ' t nnk ( be in ' ) undergo an 
nb~;tr,~ction prqcess similar to that involved in level shift. 
Consider also the esarnple 'the foreign-born may hold the 
\nzite IIousc key socnt. It: is possible to uncierstnnd *I..%ite 
House key1 in it8 metaphorical sense because: 'keyt 5s a 
NOBIINAL described ,is having the iunction 'own ' ; ' opening 
t 
implies the possibility of enteringv (cT. 'close ', section 
5.1 and ' bite Hou;c * is not only a AIT~L&& building, but 
is also defined with the fectures of $n institution, which 
includes ANIbdfE beings. Thus the f rclmework exists ior 
handling some metapi~orichlly used noun constructs with 
underlying verbal and/or ettribvtive concepts. 
The examples of Sections 5.1-5.4 are representative of 
the v~rious metaphor mech:tnisms which have been identified. 
The (;uestion arises as to the extent to which such mechanisms 
hold for any metaphorical use of a verbal or attributive 
concept. An assessment of the validity of the nnClysis 
method for such metaphorical uses depends upon 1) the com- 
pleteness of tho identified categories, i.e. whether such 
categories cover all types of verbs in the class under 
consideration; 2) whether ~uch categories cre based on the 
most "importantH component which enters into metaphorical 
extensions; and 3) the extent to ~hich v:.riotions within a 
category af~ect the plausibility of an interpretcticn. The 
first two conditions are concerned.r:itk thc quootion of 2 
"rninirn4lM interpret'.tion, i. e. thc exclusion of n :!lscfl 
interpretation, the third with an "adequate" interpret ation. 
With tespect to the first point, the Verb description 
system presented has intenticnnlly focused on the breauth 
or scope of the categorization rdher thm on 2. more detailed 
il~ustr~tion of any one category. Such an ovcrview must have 
prior consideration bccciuse the trensl; tion of a mctaphoriccll 
verb requires compzris~n with other vrrbs, which themselves 
must be assigned a ~tlocction'l within thc system bclr'orc ~ny 
refinement of inter~retat ions can b enin. The ~ivcn system 
outlines t is cc7tegorization in terms of three rel~ted prim- 
itive structures-= &TATE, ENTER-LT-bE and L&iVL-STATE, which 
are subject tc. embeduing, as in thc cuse o 'close' (Section 
5.1). Roles define the ap lication of rhcL e structures to an 
"objectw, a "locationtt of an object and an '?a&entft of any 
change, the result of 1:hich is represented by such a structure. 
These roles, which ore few in nut.iber and reliltively simple to 
identi~y  or ilny given verb, are adec ucte to rclcte ilny lex- 
ical verb rorm to i:n undcrlyin structwc. This structure- 
rolc description divides the class or prcdicc: ivc concepts 
b 
with the csccntion of "logiccltt terms such as 'ilnplyT or 
'equate '. Thus the field of verb(11 and attributive concepts 
is covered by this minimill classification b<lsed on intuitive 
abstract concepts. These abstract. structures ciln be primi- 
tively realized in the literal output paraphrases as 'be 
(or not be) in a certain stcteT, 'sti;rt to be in a ccrtnin 
stzte' and 'stop being in n certain st.:ter. 'Ae nature of 
the 'stk\tet can then be described to tke extent allowed by 
the levcl/sublevel definition of the Object of the -41rase. 
The i~cntificd lzvels which define the ficld of mctar~horical 
extension c:.n olweys be exp.,ncled or refined to ~i.ve more 
inform, tion, since they LO not af iect those comnonents-- 
structures and ltstructurnltl fcntures exclu~ling HYP and VOL-- 
which remain cons tc:nt in an extens ion. 
It is clLairneci t:-~~t these structures arc thc mast basic 
characteridin& eler~~ents of a verb in the sense th. t the 
identified primitives ;nd rnech'nisrns are those which can 
also be recognized as ~n~~erlying conceptunlly simpler ling- 
uistic constructs not usually thought of as metc~horical. 
~lthough a phrase such as 'he reached >rosperityt is not 
obviously metaphorical, there is c "tr~nsl~ ti:ntt between it 
and the phrasc 'he became prosperous which is similar to 
the translation between the more coLoriul 'he leapt to 
prosperityt and 'he became (suddenly) prosperoust. In 
each cnsc, the primitive TK-E representing 'start to1 
relhtes the two forms of the expression. 
'l'his simi1:lrity 
rests on the fact that - all linguistic expressions which 
treat abstractions (lprosperous') as objects ('prosperity') 
might in a sense be considered metaphorical. 
It seems 
reasonable to ap roach the problem of metaphor with an 
anelysis valid for the simplest form of such expressions. 
The analysis represented by the structural descriptors is 
trivial but basic in th'lt it is a prerequisite to any more 
complete interpretation, and in that it relates expressions 
exhibiting varying degrees of metaphor without resorting to 
ad hoc definitions or rules. 
Although they provide a basic interpret~t-iirn, the 
structures and features whic7? render an extension meaningful 
are not necessarily, the focus of a metaphoric21 expression. 
The focus moy be an attribute which, while provided for and 
broauly classified by the structure-lcvcl dciinition of the 
verb, ittielf rul~nins to bc dof ined. To tr:kc n r. ther difficult 
exan~,?le, the verb tbleachl can h . ciincu :I!; '+AGJNT ll'&,L 
aLdi"I'L; (0 2 1: ) o)', or, 'to Ci7udc ,m object to 
Vg $1 dE of Qkc~i:'. If 'blench l i., u;ea ~n~tuphorici~lly, 
as in :,he bleached the storyt., this dciinition gives the 
mini!~mL informiltion that solnc L:ttribute of the tory dis- 
appears. This is the nod btli;ic ,or nccc.-.,:*ry ;I. rt of the 
interpretation, but is not'very interestins. It would also 
be desircble to know how the attribure itself enters into 
the metaphor, i.e. what the color or loss or c~lor signifies. 
For quantit, tive attributes, i.e. those ~cith magnitudes 
as values, the prii~itives . . and I+ arc in- 
cluded in the deiinition at the PI':YAILAL levcl rnd &:= 
easily ostcndcc to other lcvcls. The ch~racterizotion of 
qun1it:tivc attributes, such CIS 'with or without (a certnin) 
colort, is more diPiicult. A 3ug;cstcd dspro~ch (5) assigns 
PC~I"~~~/N,~G :.'I'I~ value lVconnot tionsgt :o 
ttributes wI-.ere 
they suggest themselves ; for exarqle .'bri;ht : 202, 
tflt7t: NSG~ (but 'even: POS' j. ~hcsc assignncnts can be 
expected to rcelcct culturrl diff~??cnccd in uncresstanding 
metaphor. 
 or the   resent essmL*le, even tb; Q .ninFi.mL def ini- 
tion is diflicult, because literal bleaching can bc uone ror 
different purposes : 
bleaching ni,ht bc qerccived as rdG.\TIVC 
in the sense of lrepuvin~ colort, but ,lOaI'.iIVE in the aense 
of 'launuering or Irer.~oving stains' (1- ' I3 (0 ( IP AT: ) 
i3Z COLuR: G IIowcvcr, thc rcsulfin~ ~~mbiguity in any 
mcta;~horicnl intcrnretntion lmgcly rcflcxtk the ambiguity of 
any literal use of the verb. In both cases n lcnowlcdge of 
- - 
the linguistic or situational context is required 2or e 
correct unc~erstanding of the use of 'blench ( thc bleached 
tho report of thc war cnsualticci, blcnchcd the anecclotes'). 
Thi.: example points out the acco~aplishments and limits 
of the ny;tcm in defining components :signif icnnt to metaphor. 
Vhut it uoes is to s--ecify S: structural Erlls~cworlc in terms of 
which those propcrtieu of predic.tive concepts relev..nt to 
inetophoric~l usages ct.n be n\cthodiccllly dcr'incd. In other 
words, the sys tern riistin.guishes the conceptual structure 
cornponcnt underlying a vcrb form i'rom thdse s ernantic attri- 
butes vhich are ;lnon- s tructureu". Tl~us 'the s structure unucr- 
lying 'bleach' is autmtticnlly specified, as obovc, and 
3roviJes llslo:sla such as PU~/KGG for the attribute repre- 
sented by iwldtel or 'without color ' . The s-,eciEications 
Lor quclitative gttributes nust reraain f le:tible, subject to 
the associations ~rhich a culturc or subculture tlssitns - to 
s'uch attrcvutes; the symbolic value of dn ab: traction such 
as 'black * i; dif Eicult t~ .define in i3 gcnerbll n nncr, apart 
from cny c5ntext. In thLs sense the rcuucri~n OF a verb use 
to quantit~tive primitives IU clualitative attribute-values 
may.re?re;ent the limit to which metaphorical imelyses can 
be generalized, 
Given thClt this system of reproscntntion pro\.uces minimal 
interpretations wl~crevcr possible, ;is opnosed t o the nltcrna- 
tive oL disminsin:: an cxprcssion ns ..nom.~lou::, it rrl.\ni~~s to 
consiuck thc third condition listc~l nbnvc, n mcly tile ade- 
quocy of the mctl~od as anpl:.:d t~ v\rbtll cancc-3ts which .ire 
-4 
clni:.~cd ti- fnll within -- n category, i.c. which .~c Li:;~i ncd 
thc sal,\e conLi$ur.~tiurt of ~lcscril~toro. ~t i~ ,!ssu\\\ed that 
the primitive structure unuorlying o verb is ;lw,lys carried 
ovdr in LI m~tnphoric.il usage, ,:lthou h it nI:y bc magnitude 
(which is llo~~cd for in tern~s of AWQUKT and IKJENalTY) or 
some other as~ect of ~st$lett (which is not- woviucd for) which 
I 
is cmphcisized. T'*us 'leap to', 'drift mtot. 'land on1, 'hit' 
cnd 'plow into or 'plow through to ' (which share the s~.me 
structure '2nd fet~turc vulues except EoP VOL) all lend to 
similar interpret tions, given a common go@, c.p. prozperity '. 
All yield 'the beginning ol: a prospcrdus state1; with incor- 
porati.cn or' the INTEKITY descriptor, 'leap to1 and 'hitt 
yield 'sudden beginningt. Treating (one sense of) 'hit' and 
'leap tot as ne; rly synonymous (TR-E &TAX (0 ON R for 'hit ' , 
0 AT R for 'leap to ) (INTENaITY: > ) (&OLE 0) ), which they 
are not, entdils some loss of information, of course, but the 
resulting approximation is useLul. In the case of r~Low 
through to l, on the other hand, the lack of the in£ ormation 
that a wlaboriows ef f art" is involvcd weakens the interpre- 
tation to L~ greater degree; this kind oi style, which depends 
on the specific mechanics of the action or the attitude of 
the actor, is difticult to incorporate into o systematic 
chornctcricntion. IIowc\ler, sucfi intopma t ion, independently 
determined, could be added to the verb dhscrit~tion. For 
example, the descriptor tINTIIMbITY:>t appcnclcd to the 
feature vtLlue t+VOLt could be assigned to thc verb tplow'. 
These dcocriptors would be cnrriod over to the incom>lete 
but more iMormnti~e tnetaphorical 'lintcrpretationtt: 'he 
very consciously did something to become prosperoust. 
Thus it can be concluded that tlte method prcsentcd 
covers a major class of predicativc concepts, 2nd that the 
resulting opproximi.tion to the meaning of on expression is 
reason..ble but varies in the amount of information conveyed. 
It is oc' sibmificance that the emphasis on inclusive classes 
together with t specific suggested format allows for extension 
of thc system. Inter~rctationsproduced on the basis of rela- 
tively rninimd information will not ~lways be completely 
satisfactory, nor will they 7rovide all the nuances of ling- 
uistic expression. However, the possibility that interpre- 
tations of L lerge class of metap~oricol exjressions can be 
approximated by a systematic analysis of tlze concepts involved 
ensures further opportunities to aevelop computer understanciing 
of novel expressions. 
F1GU"iXE 1 
OUTPUT INTERPRETATXOXS 
Fa rmat : < IN1)UT 131111ASE~ 
<TYPE OF hIErrAP1.IOR> ~INI'SSIZL ROLE COSFI(7URATL'OH > 
r OUTPUT PlSRAl~l~~lA SE 3 
( 1IF DRINK INK ) 
13A SE RVO 
( HE DRINK INK ) 
( EIE CI-OSE INK ) 
( NO INTGRPREJ'ATION ) 
( HE CLOSB 311ND ) 
I.EVEL-SIIIFI' R VR 
( IIE ( IPART: YIXD ) STOP POSSIl3ILII'Y-OF STAHI' I'HINIC ) 
( slrIP PLOW SE-4 ) 
FEATURE-SIIIFT AVO 
( SHIP DO SEA START ( BE SlLZPE: ) ) 
( SIIIP PLOW CHAIR ) 
( NO INTERPRETATION ) 
( CIt4IR PLO\q SEA ) 
( NO INTERPRETATION ) 
FIGURE 1- -Continued 
( saIr 1nsI~rsam'rE ) 
RASE OV 
( SHIP DISINTBGlWTE ) 
( COUNrRY LEAP-TO PROSPERITY ) 
110-SIIXFI' 0 V11 
( COUNTRY START 1IAVE-?LATERIAL ) 
( PROSPERI ry DTSINTEGRATE ) 
CATEGORY-SITIFI' OV 
{ HE STOP IfAVIS->!ATERIAL ) 
FIGu2F: 2 
SANI3Ll3 INPUT DATA 
D+ctionnry Entri~s: 
(CHAIR ((~11) (L~ART -) (CORT -) (FTXBD -) (ID - (~11~1'B r) (>1%6 2) 
(FIUID -) (CX -) (,I$LH -) (>I>! +) (FS: 1.0~ (ox)) 1) 
(rs~,((rl~) (11nnr -) (cov+r -1 (FIXED +) (111 1) I ) (PI~~ID t) 
(CX -1 A -1 +) (FS: ISST (\i~nrl:)) 1) 
Z 
(IIB ((1111) I - (I - (PIYKD -) (ID -) (s~-i,u'h' +) (ST%).: 2) 
(EI,U~D -) (CX -) (ANIM +) ()IZI -) (DYN - - +) (FX: AC~IPR (LIYS)) 1) 
(NIND ((zI~S P) (PAIIT (ANIN +j) (COST +) (FIYSD +) (11) +) (SII=IPE +)' 
(FI,UID -) (CX +) (,<XIM t) (1 -) (DYX +) (FN. (T~IIUK)))) 
[SHIP ((PII) (~~111~ -) (CONT +) (FISI:~~ -) (ID i) 5(2~ -) (s!I,'IF& +) 
(sIzS 3) (FL-UID -) (CY i) (ASI?I -) (>I?! t) (FN: ESr (SAT].)))) 
(PLOW ((mi) (A - (COST -) (FIXED -) (ID -) s 1 (51tB 2) 
(FLUID -) (CX -) A -) (\\?I +) (FN IiYT (1~)) ( -) )) 
(I.AND ( (PIT] 1 -) (CONT +) (FlX13D +) [ 1 4) (2D +) (~11.11'~ -) 
(SIZE ) -3) (F~,UID -) (CY -) IS - I - (UYN t) 
(F: ACTIVE (I)RODUCS) ) ) ) 
(SEA ( (~11) (P-~RT -) (CONT +) (FIS~<II +) (2~ i-) (2~ t) (SITAPT +) 
(SIZE 4) (FLUID +) (CX -) (.~IJI -) (3~s -) 
(DW t) (FN: LOC (IE)) )) 
[COUNTRY ( (PII) (PART -)   CON^ +) (FIXED +) (ID t-) (2~ +) (S!L!PB +) 
(sIzS 4) (FLUID -) (CX +) (ITUW.IN i) (I t) 
(>iX +) (FN: LOC (IN AT)) (DYX i) ( ) )) 
(IKDIFFERENcE E A) (STATE (0 AT R) (v.4~ +-) ) ) 
(~'ROSPERITY ( (CO B PlI ) (STA~E (0 AT R) (VAL +) (XYT >) ) ) 
FIGURE 2- - Cont inucd 
(DISINTEGRATE ((PI{) TR-L (STATE (0 BE)) (ROLE 0) (AGBNT -) 
(0 (NRW (BRD 1) 1) 
(LEAP-TO ( (PII) TR-E (STATE '0 AT R) ) (INTNS ) (~01.~ 0) (AGENT -) 
(INSTR FN: INT (LEGS ) ) ) ) 
(DISAPPEAR ((SE EYE) TR-L (STATE (0 SENSED-UY R)) (ROLE 0) 
(AGENT -) (0 (NRW (BRD 1 1) 1 
(CLOS$ ( (PII) (TR-L) (STATE ( (INP +) TR-E (STATE (0 IN R) ) ) ) .I 
(ROLE R) (AGENT t ) (R (NHW (CONT +) ) (~RD (CONT +)) ) ) ) 
('DRINK ((pH) TR-E (STATE (0 IN R)) (ROLE R) (R (NRI~ (AN ) NIL) 
(O (NRW (FLUID +)) (BRD (FLUID +))) (AGENT -) )) 
(SAIL ((PH) TR (STATE (0 AT R)) (ROLE 0) ) ) 
(PLOW ((PI]) TR-E (STATE (0 BE SHAPE: ) ) (AGENT +) (~01.13 0) 
(INSTR: TR (STATE (0 AT R))) (0 (NR!~ LAND) (RRD (2~ +) 
(FIXED +))I -1) 
MaC'rix Secrmcnb : 
2- 
NB (P (R (STATB (-H (t THINK - ()) 
+H (+ BELIEVE - DISBELIEVE)) TR-L () ) 
o (STATE (-N (+ (IN CP) - 0) 
+H (+ (IN ~~31~1 - 0)) TR-L () )) 
(R (STATE (-N (+ ENJOY - XOT-ENJOY t- RE-UNAFFBCTBD-BY) 
tH (t LIKE*- DISLIKE +- BE-INDIFFERENT-TO)) 
TR-L (% 1) 
0 (STATE (-A (+-(IN CP) - ( ) +- {) ) 
+H (+ (IN LT~IP) - () t- () )) 
TR-L ( m, 1) 
FIGURE 2--Continued 
+Ir (+ luve-wralrrhr. - ( ) +- ( ) ) ) 
TR-L (-11 ( ) 
tH (+ 1.OSE - GAIN +- ) )) ) 
1 i nt ~r Pc\r'npllrn so \fords 
--- - -LL -- - .. - --- -- L. - 
(& 
(STATE TR-E TR-I. (RYP +) ) (RE STAliT STOP ~'oSSII~TLITY-OF) : 
Input .. Phrases; 
- 
CHAIR DRINK INK; 
HE CLOSE MIXD: 
SHIP PLOW SEA; 
SHIP PI,OW CHAIR; 
CHAIR PI,O\J SEA; 
SI-IIP TIIS IhTEG RATE ; 
ISDIFFERENCE DISINTEGllATE ; 
COUNTRY LEAF-TO PROSPERITY; 
PROSPERITY DISINTEGRATE: 

References 
(1) Fillmore, C., "Types of Lexical Innf~rmation~~, 
Studiesin Syntax and Semanttcs , ed. F, Kiefer, 
D. Reidel, Porarecht, The Netherlands. 
(2) Goldman, N., Riesbeck , C., "A Conceptually Based 
Sentence Paraphraser", AIM- 196, Qomputer Science 
Department, Stanford University, Stanford, California 
(May 1973). 
(3) mtz, J., "Semantic Theory and the Meaning of t~oodgfl, 
The Jourmtll of Philosophy, Vol. XI, No. 23' (1964). 
(4) Riesbeck, C., "Computer Analysis of Natural Language 
in Contexttt, Ph. D. thesis, Computer Science Department, 
Stanford University, Stanford, California (1974). 
(5) Russell, S. . llComputer Understanding of ConceptuaLly 
Complex Phrasesu, ph. D. thesis, Stanford University , 
Stanford, California (March 1975). 
(6) Schank, R. , tlCausality and Reasoningf1, Technical Report 
No. 1, Istituto per gli Studi Semantici ecognotivi, 
CastagnoLa, Switzerland (1974). 
(7) Schank , R. , 'The Fourteen Primitive ~ctions and their 
Inferencesu, AIM -183, Computer Science Department, 
Stanford university, Stanford, California (1973). 
(8) Schank, R., Goldman, N., Rieger, C., Riesbeclc, C., 
Vrirnitive Concepts Underlying Verbs of Thought 
A1M162, %omputer Science Deportment, Stanford 
University, Stanford, California (1972). 
(9) Schtlnk, R. ! Meger, C., "Inference and the Computer 
Understandxng of Natural LanguageH, AIM-197, 
Computer Science Liepartment , Stenf ord University, 
Stanford, Californk (1973). 
