A SYSTEM OF SEMANTIC PRIMITIVES 
Ray Jackendoff 
Department of English 
Brandeis University 
In a linguistic theory, the semantic 
representation of a sentence is meant to be 
a formal characterization of the information 
conveyed by the sentence. This information 
can be thought of as a set of claims abeut 
various individuals, properties, events, 
and/or states of affairs, and about the 
relationships among them. A satisfactory 
semantic theory must not only account for 
the information conveyed by a sentence; it 
must also account for how the sentence 
conveys that information, expressing both 
the syntactic and semantic generalizations 
of the language. A semantic theory is 
explanatory when linguistically significant 
generalizations are inherent in the choice 
of formalism, when the theory claims the 
language would be more complex any other 
way. Such a theory claims that the language 
learner does not have to learn these 
generalizations; rather they are determined 
by his innate capability to learn a human 
language and to conceptualize the world. 
This paper sketches a semantic theory 
for verbs, based on the analysis of Gruber 
(1965) and Jackendoff (1972), which 
expresses certain semantic generalizations 
in such a way that they correspond very 
closely to syntactic generalizations. The 
generality of the semantic analysis will be 
demonstrated by stating inference rules 
whose side applicability depends crucially 
on the form of the semantics. Thus it will 
be shown that the choice of semantic 
primitives made here is explanatory in the 
sense set out above. The theory is 
expounded in much greater detail in 
Jackendoff (1975), of which this is an 
extreme condensation. 
I. GO, BE, and STAY verbs. 
In sentences such as (I), 
(1)a. The train traveled from Detroit to 
Cincinnati. 
b. The hawk flew from its nest to the 
ground. 
c. An apple fell from the tree onto Ike's 
head. 
there is a common element of motion 
expressed. We will refer to the object in 
motion as the Theme of the sentence, to the 
Theme's initial position as the Source, and 
to its final position as the Goal. The 
semantic similarity between the sentences in 
(I) is expressed by assigning a common 
element in their semantic representation, a 
function GO(x,y,z). This function makes the 
claim that there has taken place an event 
consisting of the motion of x from X to ~. 
In other words, the first variable of GO 
corresponds to the Theme, the second to the 
Source, and the third to the Goal. 
24 
The semantic differences between the 
sentences are expressed in two distinct 
ways. First, the different Themes, Sources, 
and Goals are derived by inserting the 
interpretations of the various noun phrases 
in the sentences for the variables x, X, and 
z, according to the lexieal correlation of 
the verb's strict subcategorization feature 
with its semantic representation. The other 
semantic differences are described by adding 
a restrictive modifier expressing manner of 
motion, in these cases lexically determined, 
but in many cases due to an adverb in the 
verb phrase. So, for example, (Ib) is 
represented as 
GO(THE HAWK, ITS NEST, THE GROUND). 
Manner: THROUGH THE AIR 
(I use capitals to represent semantic 
markers for the corresponding English 
expression -- unsystematically, unless it is 
relevant to the present discussion.) Of 
course, a full explication of fly would 
involve further analysis of the manner 
marker; but what is relevant here is that 
the common element of meaning has been 
extracted from all verbs of physical motion. 
Next consider (2). 
(2)a. Max is in Africa. 
b. The cat lay on the couch. 
c. The statue stands on Cambridge Common. 
These do not describe a motion, but rather 
the location of an object relative to some 
object. The formal semantic representation 
of (2) will thus include a function BE(x,y), 
where X is the Theme (the object being 
located) and X its Location. As in (I), the 
differences of meaning among the sentences 
in (2) are expressed by substituting 
different markers for x and X and by 
attaching different manner markers as 
restrictive modifiers of the function BE. 
In addition to the verbs of location 
illustrated in (2), there is a second, 
smaller class of locational verbs with 
rather different semantic properties: 
(3)a. The bacteria stayed in his body. 
b. Stanley remained in Africa. 
c. Bill kept the book on the shelf. 
These differ from (2) in that (i) they 
cannot refer to a point in time, as can (2); 
(ii) they can serve as a complement to what 
happened was that, whereas (2) cannot. 
(4)a. The bacteria~were in his body at 6:00. 
r L'stayed 
b. The cat|lay on the couch at 6:00. 
\[*remained 
(5) What happened was that 
~tanley remained in Africa. 
~ill kept the book on the shelf. 
~Max was in Siberia. 
~The statue stood on Cambridge 
~ommon. 
The verbs in (3) will be represented in part 
as STAY(x,y), where x is the Theme and X its 
Location. 
I 
! 
I 
l 
i 
I 
I 
I 
The evidence from what happened was 
indicates that STAY verbs, like GO verbs, 
represent events, while BE verbs represent 
states of affairs. 
2. Position, Possession, and Identification 
The most important aspect of Gruber's 
analysis is his extension of the 
interpretations of GO, BE, and STAY to a 
wide variety of examples where the 
"position" (Source, Goal, or Location), of 
the Theme is not specified in physical 
terms, as it is in (I)-(3). Consider these 
examples: 
(6)a. Harry gave the book to the library. 
b, Charlie bought the lamp from Max. 
c. Will inherited a million dollars. 
(7)a. The book belonged to the library. 
b. Max owned an iguana. 
c. Bill had no money. 
(8)a. The library kept the book. 
b. The iguana stayed in Max's possession. 
c. The leopard retained its spots. 
In (6), the object described by the direct 
object of the sentence undergoes a change in 
whom it belongs to. By analogy with (I), we 
can call the object undergoing change the 
Theme, and the initial and final states 
Source and Goal respectively. (7) expresses 
states of possession; by parallel with (6) 
and (2), we will call the possessed object 
Theme and the possessor Location. (8) also 
expresses a single unchanging possessor; but 
at 6:00 may be added only to (7), not to 
(8), and what happened was may be prefixed 
only to (8), not to (7). 
Thus there is an important parallel 
between (6)-(8) and (I)-(3). Gruber chooses 
to represent this parallel by claiming that 
(6) are GO verbs, (7) are BE verbs, and (8) 
are STAY verbs. The difference between 
(6)-(8) and (I)-(3) is expressed by a 
modifier on the functions. For physical 
motion and location, the modifier is 
Positional; for possession, it is 
Possessional. (la), for example, is now 
represented as GO~oW,~ (THE TRAIN, DETROIT, 
CINCINATTI); (6a) is GO~os~ (THE BOOK, 
HARRY, THE LIBRARY). 
Another important "mode of location" 
besides Positional and Possessional is 
illustrated in the following examples. 
(9)a. The coach changed from 
young man into a pumpkin. 
b. The metal turned red. 
c. The ice became mushy. 
a handsome 
(10)a. The coach was a pumpkin. 
b. The metal was red. 
c. The pumpkin seemed tasty. 
(11)a. The poor coach stayed a pumpkin. 
b. The metal remained red. 
c. The redness persisted. 
The same three-way contrast obtains. (9) 
describes changes of state; (10) describes a 
state; (11) describes the persistence of a 
state. Of the two nonmotional cases, at 
6:00 may be added only to (10), and what 
happened was may be prefixed only to (11). 
Gruber proposes a modifier Identificational, 
which indicates that the Location or Source 
and Goal of the function to which it is 
affixed make claims about what the Theme is, 
rather than where or whose it is. Thus for 
example, (9a) is represented as GO~(THE 
COACH, A HANDSOME YOUNG MAN, A PUMPKIN). 
(For sentences such as (9a) in which one of 
the arguments is absent from the sentence, 
the semantic representation will contain a 
free variable.) 
By adopting the markers Positional, 
Possessional, and Identificational as 
restrictive modifiers on the functions GO, 
BE, and STAY, we capture important semantic 
distinctions and generalizations. The 
combination of the three markers with each 
of the three functions yields a particular 
class of verbs, accounting for the 
similarities and differences among the 
classes in a natural way. As evidence that 
this is the correct breakdown, we observe 
that many verbs occur in more than one 
locational mode, while preserving their 
classificaton as GO, BE, or STAY verbs: 
(12)a. The coach turned into a driveway. 
(Positional) 
The coach turned into a pumpkin. 
(Identificatonal) 
b. The train went to Texas. 
(Positional) 
The inheritance went to Philip. 
(Possessional) 
c. Max is in Africa. 
(Positional) 
Max is a doctor. 
(Identificational) 
d. Bill kept the book on the shelf. 
(Positional) 
Bill kept the book. 
(Possessional) 
e. The coach remained in the driveway. 
(Positional) 
The coach remained a pumpkin. 
(Identificational) 
In each pair, the same verb is used in two 
different locational modes. Since these 
uses are not a priori related, it is a 
significant generalization that a sizable 
number of verbs exhibit such behavior. In 
the present formalism the relationship 
between the uses is clear and nonaccidental: 
the verb stays fundamentally the same, 
changing only the value of the restrictive 
modifier denoting locational mode. 
The fundamental concept represented by 
sentences in the present theory, then, is 
giving the location(s) of an object at a 
particular time or during a particular 
interval; the richness of expression 
available to natural language comes in part 
from extending the concept of location to 
other than physical position. 
2B 
3. Causative and Permissive Agency 
In addition to the three functions 
discussed so far, there are two which 
describe different kinds of causation. 
Compare (13a,b,c). 
(13)a. The rock fell from the roof to the 
ground. 
Dick received the money. 
The bird went out of the cage. 
Noga stayed sick. 
b. Linda lowered the rock from the roof 
to the ground. 
Dick acquired the money. 
Laura took the bird from the cage. 
David kept Noga sick. 
c. Linda dropped the rock from the roof 
to the ground. 
Dick accepted the money. 
Laura released the bird from the 
cage. 
David left Noga sick. 
The events of the a. sentences are also 
described in the b. and c. sentences, but 
the latter two claim that the events are due 
to the agency of the subject, who is thus 
termed an Agent. In turn, (13b,c) differ in 
the kind of action performed by the Agent: 
(13b) is bringing the event about, or 
causing: (13c) is ceasing to prevent the 
event, or letting. We will symbolize these 
two kinds of agency as CAUSE (x,e) and LET 
(x,e) respectively. The first examples of 
each group in (13) receive the 
representations (14a,b,c) respectively. 
(14)a. GO@o~ (THE ROCK, THE ROOF, THE 
GROUND) 
b. CAUSE (LINDA, GO p~ (THE ROCK, THE 
ROOF, THE GROUND)) 
c. LET (LINDA, GOIBo~ (THE ROCK, THE 
ROOF, THE GROUND)) 
The last sentences in each group are 
(15a,b,c,). 
(15)a. STAY~ (NOGA, SICK) 
b. CAUSE (DAVID, STAY~ (NOGA, SICK)) 
c. LET (DAVID, STAY~ (NOGA, SICK)) 
CAUSE is quite familiar from the 
literature, LET less so. There are two 
interesting distinctions between them 
besides the inferences to be discussed 
below. First, CAUSE allows an expression of 
instrument, but LET appears not to: in (16) 
the with-phrase can be interpreted only as 
accompaniment. 
(16)a. Linda dropped the rock with a cable. 
b. Dick accepted the book with a $5 
bill. 
c. David left the bird in the cage with 
a lock on the door. 
Second, the final argument of LET may be 
either an event or a state of affairs, as 
seen from the following contrast. 
(17)a. David let Laura out of the room. 
b. David allowed Laura out of the room. 
26 
(17a) must be interpreted as David 
permitting Laura to gg out of the room; 
(17b) does not say anything about her 
movements, but only where she may be. On 
the other hand, CAUSE requires its final 
argument to be an event. All the causative 
locational verbs such as hold, keep and 
retain are of the form CAUSE (x, STAY(...)) 
rather than CAUSE (X, BE(...)). The verb 
cause, which does allow things like Dollie 
caused Martin to be happy, has a more 
complicated analysis than just CAUSE, as 
will be seen below in (36). (One should be 
immediately suspicious, since this example 
means "Dollie caused Martin to become 
happy" -- see (40b) below). 
4. Inference Rules 
We will now develop inference rules to 
derive logical entailments of sentences on 
the basis of their semantic representatins. 
An obvious rule of inference is that if an 
event is caused, it happens. This is 
formalized as (18). 
(18) CAUSE(X,E) -> E 
A typical inference derivable by this rule 
is (19). 
(19) Max gave Joe the money. 
CAUSE(MAX, 
GO~ (THE MONEY, MAX, JOE)) 
-> Joe received the money from Max. 
GO~s~ (THE MONEY, MAX, JOE) 
The related inference with LET is that if 
someone doesn't let an event happen, it 
doesn't happen: 
(20) NOT LET(X,E) -> NOT E 
A typical inference from this rule is (21). 
(21) AI didn't drop the pancake to the 
floor. 
NOT LET(AL, 
GO~o~ (THE PANCAKE,y, 
THE FLOOR)) 
-> The pancake didn't fall to the floor 
NOT GO?oW~ (THE PANCAKE,y,THE FLOOR) 
The inverses of these inference rules are 
not valid in general: something may happen 
even if no particular thing one can name is 
its cause; if something is permitted, that 
does not guarantee that it happens (although 
in certain situations such as drooping the 
inference does go through). 
The rest of the inference rules concern 
GO, STAY, and BE, and can be stated in terms 
of simple spatial intuitions. First, there 
is the principle that if someone stays 
someplace for a period of time, he is in 
that place at any instant during that time. 
(22) STAY (X,Y) FROM t~ TO t& 
-> BE (X,Y) AT t s 
Condition: ti~t~t ~ 
i 
I 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
! 
I 
I 
I 
I 
1 
I 
I 
I 
I 
i 
I 
i 
I 
I 
I 
22) permits inferences such as (23). 
23) Carl remained in the room from Tuesday 
to Friday. 
STAY~ (CARL, THE ROOM) 
FROM TUESDAY TO FRIDAY 
-> Carl was in the room on Wednesday. 
BEe~.~(CARL, THE ROOM) AT WEDNESDAY 
(18) and (22) work together in this 
inference (assuming proper times): 
(24) David kept Noga Sick 
-> Noga stayed sick 
-> Noga was sick. 
CAUSE(DAVID, STAY~ (NOGA,SICK)) 
STAY~L~% (NOGA,SICK) 
BE~k (NOGA,SICK) 
The inverse inference is that if 
someone doesn't stay someplace during an 
interval, there is a time during the 
interval when he isn't there. This can be 
formalized similarly. 
The parallel inference rule for GO is 
that if something goes from one place to 
another, it was at the first place first and 
the second place second: 
(25) GO(X,Y,Z) AT t~ -> 
for some times t~,~ such that 
t~<t~ <t~, 
BE (X,Y) AT t~ 
BE (X,Z) AT 
A typical entailment from this rule is (26). 
(26) Phil gave the bill to Lyn. 
CAUSE(PHIL,GO~.~(THE BILL,PHIL,LYN)) 
GO~$~(THE BILL,PHIL,LYN) 
~ Phil had the bill, and then Lyn had 
it. 
BE~(THE BILL,PHIL) AT t a AND 
BE~Q~(THE BILL,LYN) AT t~ 
for some t I , tm such that t~<t L 
There is no inference from NOT GO, since if 
someone didn't go from one place to another 
we can make no inferences about where he was 
at any time, without further knowledge. 
A very important inference rule is that 
if and only if someone is not someplace, he 
is somewhere else. We represent the sense 
"a place other than Z" as NOT Z, for reasons 
to become clear shortly. 
(27) NOT BE (X,Z)<->BE (X, NOT Z) 
Also, if and only if someone is someplace, 
he is not elsewhere. 
I (28) BE (X,Z)<->NOT BE (X, NOT Z) 
These rules play a role in inferences such 
as if John was not inside the house, he was 
I outside of it (Positional); if either John 
• or Bill had the book and John didn't, then 
Bill did (Possessional); and if Sue was 
sick, she wasn't healthy (Identificational). 
I The complete derivation of such inferences, 
however, involves other steps including 
I 27 
factual knowledge, and is too complex to 
include here. 
Similar to (27)-(28) are inference 
rules (which we won't formalize here) that 
if you're going someplace you're not staying 
anywhere, and if you re staying someplace 
you re not going anywhere. 
What makes the sort of inference rules 
proposed here of interest is the way they 
provide evidence for the explanatory power 
of the present theory of semantic 
description. For in this system, a rule of 
inference is simpler if it generalizes over 
all modes of location. The theory claims 
that it is no accident that the inference 
rules generalize in the way they 
do -- rather it is an essential part of the 
structure of the semantic description. I 
consider it a striking property of the 
present system that very simple principles 
about spatial understanding can be stated 
formally in such a way that they provide a 
rich variety of inferences in domains which 
bear no a priori relation to physical space. 
5. Circumstantial Location 
Consider the ~ interpretations of 
(29a,b). 
(29) a. Laura kept David in the room. 
b. Laura kept David working. (In 
deep structure, Laura kept David 
\[~David working\]) 
The interpretation of (29a) is 
CAUSE(LAURA,STAY~,~(DAVID, THE ROOM)). If 
the verb keep is to be essentially the same 
in (29b), we must provide (29b) with a 
similar interpretation, though clearly none 
of the modes of location discussed so far 
can provide one. We introduce a mode called 
Circumstantial: if an individual is in a 
Circumstantial Location, where the location 
is an event or state of affairs, this is 
taken to mean that the individual is 
involved as a participant in that event or 
state of affairs. Then we can assign (29b) 
the reading (30). 
(30) CAUSE(LAURA,STAY~,~(DAVID,DAVID WORK)) 
This claims that Laura caused David to 
continue to be involved in the situation of 
working, precisely the desired 
interpretation, and furthermore of precisely 
parallel form to its Positional analogue 
(29a). Other examples: 
(31) a. Linda kept Laura (away) from the 
cookie jar. 
CAUSE(LINDA,STAY?o~,~(LAURA, NOT THE 
COOKIE JAR)) 
~kept \] 
b. Linda|prevented~ Laura from 
screaming. 
CAUSE(LINDA,STAYaj~(LAURA, NOT(LAURA 
SCREAM))) 
kept \] 
(32) LauraLcontinued~screaming. 
STAY~,&~ (LAURA, LAURA SCREAM) 
(33) a. David avoided the beach. 
STAY~ (DAVID, NOT THE BEACH) 
b. David avoided playing checkers. 
STAY~,~ (DAVID,NOT(DAVID 
CHECKERS)) 
PLAY 
(34) a. Jim forced the ball into the hole. 
CAUSE(JIM,GO~(THE BALL,y,THE 
HOLE)) 
b. Jim forced Phil into confesing. 
to confess. 
CAUSE(JIM,GO¢,~(PHIL,y, 
PHIL CONFESS))' 
(35) Dick stopped the car from coughing. 
CAUSE(DICK,GO¢,$~. (THE CAR,y, 
NOT(THE CAR 
COUGH))) 
(36) Dollie caused Martin to be happy. 
CAUSE(DOLLIE,GO~(MARTIN,y, 
BE~(MARTIN,HAPPY))) 
(37) a. John allowed Fred in the room. 
LET(JOHN,BE~ (FRED, THE ROOM)) 
b. John allowed Fred to wash the 
dishes. 
LET(JOHN,BE~,~. (FRED, 
FRED WASH THE DISHES)) 
(38) Jack exempted Jim from fighting. 
LET(JACK BE~ (JIM,NOT(JIM FIGHT))) 
There are four points to observe about 
these representations. First, where a verb 
that takes a sentential complement has a 
Positional variant, the two versions of the 
verb have identical representatons but for 
the locational mode. Thus in the present 
system it is no accident that the verb 
occurs in two seeming disparate syntactic 
and semantic frames. 
Second, the NOT meaning "somewhere 
other than" in Positional contexts appears 
to generalize fully with the NOT of sentence 
negation in Circumstantial contexts. 
Third, complement type is related to 
semantic representation. Gerundives 
typically correspond to Locations or Goals; 
from-ing complements are negated Locations 
and negated Goals; to-infinitive complements 
are Goals of various sorts (including Goals 
of intentions, not discussed here); 
that-complements are typically Themes. The 
correspondence is hardly exact partly 
because there are far more semantic 
positions for clauses than there are 
complement types: but it is far from random 
either. 
Fourth, and most striking, the 
inference rules of the previous section can 
be applied to Circumstantial verbs quite 
freely. Given the special inference rule 
(39), which follows immediately from the 
definition of Circumstantial location, we 
can derive such inferences as (40) (omitting 
time dependencies). 
28 
(39) BEc,~ (X,E)->E 
(40) a. Laura kept David working. 
CAUSE(LAURA,STAY~,~.(DAVID, DAVID WORK)) 
(~ STAY~,i~ (DAVID,DAVID WORK) 
BE~,$~ (DAVID, DAVID WORK) 
~ David worked 
DAVID WORK 
b. Lyn caused Bob to be happy. 
CAUSE(LYN,GO~,&~(BOB,y,BEx&¢~ ~ 
(BOB,HAPPY))) 
~l~_~ GO~,~ (BOB,y,BEx~(BOB,HAPPY)) 
BE ~,.(BOB,y) and later 
BE~_(BOB,BE~mh~(BOB,HAPPY)) 
(~--~ y and later BE~(BOB,HAPPY) 
\[because Source and Goal must be distinct\] 
y / BE~(BOB,HAPPY) 
Bob was unhappy and later he was happy. 
BE~BOB, NOT HAPPY) and later 
BE~%(BOB, HAPPY) 
c. Jack didn't exempt Jim from 
fighting. 
NOT LET(JACK, BE~,~(JIM,NOT(JIM FIGHT))) 
NOT BEc~.~(JIM , NOT(JIM FIGHT)) 
(~ BE~ (JIM, JIM FIGHT) 
Jim fought. 
JIM FIGHT 
This kind of inference is characteristic of 
the "implicative" verbs described by 
Karttunen (1971). In the present system it 
follows immediately from the semantic 
analysis and the generalization of the 
inference rules for spatial location to the 
circumstantial mode. 
Lest the generalization from Positional 
to Circumstantial should still seem marginal 
and unmotivated, notice that it is quite 
pervasive in the language. For a few very 
clear random examples, consider t_go come t_go 
b__ge called Max, to lead someone to believe, 
t__oo drive someone t__oo confess, t__q bring 
oneself t__oo acknowledge something, and, among 
nominals, the way to find out. For an even 
more interesting example, consider the 
meaning of force in (34a). In addition to 
the semantic description given there, there 
is a marker of manner roughly paraphrasable 
as "by applying pressure against the ball's 
resistance." Here the concepts of pressure, 
applying pressure, and resistance are purely 
physical. But in fact the same manner 
marker "by applying pressure against Phil's 
resistance', is miraculously perfect for 
(34b), where the concepts of pressure, 
applying pressure, and resistance are much 
more abstract in nature. Surely this is no 
coincidence; it argues that the extension 
I 
I 
i 
I 
I 
I 
I 
i 
! 
i 
I 
Ii 
I 
I 
I 
I 
I 
I 
I 
from Positional to Circumstantial is highly 
determined in human cognition, and that the 
generalization of a verb's meaning follows 
innate lines of analogy. The present theory 
of semantic description is explanatory in 
that these generalizations fall out 
immediately from the choice of semantic 
primitives, which are in turn chosen on 
linguistic grounds. 
This result is consistent with Piaget's 
view (1947, 1970) that (nonverbal) knowledge 
of physical spatial relations is the .most 
basic sort of knowledge we have, and that 
all other kinds of knoweldge develop out of 
it. Within such a framework, the inference 
rules proposed here are not merely 
linguistic but conceptual, and correspond 
closely to Piaget's principles of 
conservation and identity. The growth in 
one's ability to handle abstraction then 
consists (in part) in understanding new 
modes of location and being able to 
generalize the rules of inference to a new 
system of relations. If this is so, the 
theory proposed here is a deep result not 
only for linguistic theory but for the study 
of human conceptualization. 
REFERENCES 
Chomsky, N. (1965), Aspects of the Theorz 
of Syntax. MIT Press. 
Gruber, Jeffrey (1965), Studies i__nn Lexical 
Relations, MIT dissertation, available 
from Indiana University Lingustics 
Club. 
Jackendoff, Ray (1972) Semantic 
Interpretation in Generative Grammar, 
MIT Press. 
(1976) "Toward an Explanatory Semantic 
Representaion," Linguistic Inquiry, 
7.1. 
Karttunen, Lauri (1971), "Implicative 
Verbs," Language 47, 340-358. 
Piaget, J. (1947) Psychology of 
Intelligence, Littlefield, Adams, & 
Co., 1972. 
(1970) Genetic Epistemology, Columbia 
University Press. 
29 
! 
