DISCOURSE REFERENTS ~ 
Lauri Karttunen 
University of Texas at Austin 
0. Introduction. 
Consider a device designed to read a text in some r~tural 
language~ interpret it, and store the content in some manner~ say, 
fop the purpose of being able to answer questions about it. To 
accomplish this task~ the machine will have to fulfil\[ at \[east the 
following baste requirement. It has to be ab|e to build a file that 
consists of records of-all the individuals~ that is~ events~ objects~ 
etc. ~ mentioned in the text~ and~ for e~eh individual~ record whatever 
is said about it. OF couPse~ fop the time being at \[east~ it seems that 
such a text interpreter is not a praetica\[ idea~ but this should not 
discourage us from studying.in abstPaet what kind of capabilities the 
machine would have to possess~ provided that our study provides us 
with some insight into natural langL~ge in general. 
In this paper I intend to discuss one particular feature a text 
interpreter must have: that it must be able to recognize when a novel 
individual is mentioned in the input text and to store it atong with its 
characterization fop future reference. Of course~ in some P_~ses the 
problem is trivial. Suppose there appears in some sentence a proper 
name that has not been mentioned previously. This means that a new 
-2- 
pePson is being intPoduced in the text and appPopPiate action must be 
taken to PecoPd the name of the pePson and what is said about him. 
OthePwise, the pPopeP name is used to PefeP to an individual already 
mentioned and the machine has to locate his file in the memory with 
the help of the name. This pPoblem of identification wilt be mope 
difficult where a definite descPiption--a definite noun phPase such as 
the man Bill saw 7estePda~,--is used, since thePe will, in genePal, 
not be any simple look-up procedure fop associating the descPiption 
with the Pight individual. With definite noun phrases there is also the 
pPoblem that it is not possible to tell just from the noun phPase itself 
whetheP oP not it is supposed to refer" to an individual at all. Fop 
example, it is cleaP that the phPase the best student is not used 
PefePentiaUy in a sentence such as Bill is the best student. -\[-here ape 
thus two problems with ordinary definite noun phrases: (i) Is it a 
definite descPiption at all? and (ii) How to match a definite descPiption 
with an individual already mentioned in the text? The fiPst question is 
clearly of the kind linguists can be expected to solve, but it will not be 
discussed here. The only aspect of definite descPiptions that interests 
us hepe is the fact that they caPPy an existential presupposition: to 
call something "the ... " presupposes that there be some such thing. 
While it is in genePal a stPaight-for~vaPd matter- to decide 
-3- 
whether or not a.proper name in a text introduces a new individual, 
indefinite noun phrases pose a more difficult probtem. To put the 
question in a general way: Given an indefinite noun phrase, under 
what circumstances is there supposed to be an individual described 
by this noun phrase? This need not be understood as some sort off 
ontological question subject to philosophical speculation, in this 
paper" \[ intend to approach it from a purety linguistic point of view. 
It is in just those cases where the appearance of an indefinite NP 
implies the existence of some specific entity that our hypothetical 
text interpreter shoutd record the appearance of a new individual. 
What \[ have in mind can perhaps be made clear w ~:h the hetp 
of the following examples, it is a watt-known fact about language that 
indefinite noun phrases cannot be interpreted as refer'ring to expres- 
sions when they appear, in the predicate nominal position. 
(l) Bitl is not a linguist. 
(l) is obviousty a statement about one individual. It is not a statement 
about some linguist and \[Bill. It is also weir-known that in generic 
sentences singutar indefinite noun phrases play a peculiar rote. 
(2) A tion is a mighty hunter`. 
In its gener'ic sense, (2) is a statement about tions in general, 
not about any tion in particular,, unless we want to postulate a 
-4- 
hypothetical entity 'the typical lion' of whom all generic statements 
about lions ape predicated. It is clear that indefinite noun phrases 
have a very special Pole in (l) and (2) and it is not difficult to decide 
that they could not introduce any new individuals into a discourse. It 
is out of question that a text in which (1) appears would contain a 
later" reference to 'the linguist which Bitt is not' or that (2), in its 
generic sense, would justify a later reference to 'the lion who is a 
mighty hunter'. 
But consider the following example. (3a) may be followed by 
any of the sentences (3b-d) that give us more information about a 
specific cap first mentioned in (3a). 
(b) It is black. 
(3) (a) Bitt has a car'. (c) The car is black. 
(d) Bill's cap is black. 
On the other hand, (4a) cannot be followed by any of the alternatives 
(4b-d). 
(b) *It is black. 
(4) (a) Bill doesn't have a car. (c) *The cap is black. 
(d) *Bill's cap is black. 
The above examples show that just in case of (3a), the text 
interpreter has to recognize that the appearance of the indefinite NP 
a cap irnpties the existence of a specific car- that can be talked about 
again by referring to it with a pronoun or" a definite noun phrase. But 
no cap is introduced by (4a). The alternative continuations (4b-d) are 
inappropriate, since they presuppose the existence of something that 
is not there. To show that this is a linguistic and not an ontological 
-5- 
fact one only has to point out that examples (5) and (6) behave just 
like (3) and (4). 
(5) Bill saw a unicorn. The unicorn had a gold mane. 
(6) Bill didn't see a unicorn. *The unicorn had a gold mane. 
Let us say that the appearance of an indefinite noun phrase establishes 
a discourse referent just in case it justifies the occurrence of: a 
corererential pronoun or a definite noun phrase hater in the text. In 
this paper we will try to find out under what circumstances discourse 
referents are established. We maintain that the problem of corefer-- 
ence within a discourse is a linguistic problem and can be studied 
independently of any general theory of extra-linguistic reference. 
The present study was inspired by the notion of 'referential 
indices' in transformational grammar. Following a suggestion by 
Noam Chomsky (1965), it has generally been assumed that the base 
component of a transformational grammar associates with each noun 
phrase a referential index, say, some integer. The purpose of 
Chomsky's proposal was not so much to account for the meaning of 
sentences, but to augment the notion of noun phrase identity. It 
seemed that the notion of 'referential identity' was needed in addition 
to the two other types of identity, 'structural identity' and 'morphemic 
identity', for the structural descriptions of certain transformations. 
According to the standard theory, referential indices are merely 
-6- 
formal indicators of coreference with no further semantic significance. 
They amP not meant to imply the existence of discourse referents in our 
sense. This notion of cor~fempntiality has played an important role in 
recent syntactic arguments. It led to the study of pronoun-antecedent 
relations, largely ignored by traditional grammarians, which has 
revealed intricate constraints that have great theoretical impor~tance. 
What we are studying in this paper can be looked at as further con- 
straints on compFerentiality that extend beyond the sentence level, 
\]. Case studies 
l. 1 A note on specificity 
\[n the following we amP going to examine case by case certain 
aspects of sentence structure that play a role in cletermining whether 
an indefinite NP establishes a discourse referent. In the examples 
that ape discussed~ there is a possible ambiguity that has to be men- 
tioned in advance, although it will not be discussed until later. In 
general, indefinite noun phrases have both a specific and norr-specific 
interpretation. Example (7) can be interpreted to mean either (8a) or 
(Bb). 
(7) 
(8) 
Bill didn't see a misprint. 
(a) 'There is a misprint which Bill didn't see' 
(b) 'Bill saw no misprints' 
-7- 
\[1 = (7) iS Understood in the sense of (8a), we say that the 
indefinite NP a misprint is interpreted specifically. (Sb) represents 
the non-specific interpretation. OF course, not all indefinite noun 
phrases are ambiguous in this way. We could disambiguate (7) by 
adding the word certain ("a certain misprint") or an appositive 
relative clause ("a misprint, which I had made on purpose"). These 
changes would allow only the specific interpretation (8a). The addi- 
tion of the word ~ ("a single misprint") would allow only the 
sense (Sb). There are also cases where the verbs involved partially 
disambiguate the sentence by making one interpretation 
Far more plausible to the reader than the other. For example, the 
NP a piano in (9a) is naturally understood non-specifically, that is, as 
meaning 'any piano', white the same noun phrase in (9b) suggests the 
l interpretation 'a certain pianO'. 
(9) (a) John tried to find a piano. 
(b) John tried to lift a piano. 
(but he didn't succeed in 
finding one\] 
\[but he didn't succeed in 
lifting it\] 
It is something about the verb lift that suggests that a piano describes 
some specific object. On the other hand, (9a) is easily understood to 
inform us only about the kind of object John was trying to find. We note 
in passing that, if interpreted in the above manner, (9b) establishes a 
-8- 
discourse referent, i.e., 'the piano John tried to lift', but (9a) 
certainly does not justify a later reference to 'the piano John tried 
to Find'. Example (7) establishes a referent in its specific sense 
'the misprint which Bitt didn't see', but fails to do so in the sense, 
of (Sb). 
I_et us forget, for the time being, that indefinite noun phrases 
can also be understood specifically and consider first only non-specif- 
ic interpretations. 
1.2 Complement clauses 
As pointed out above, an indefinite noun phrase does generally 
establish a discourse referent when it appears in a simple affirmative 
sentence. But if the sentence is negated, a non-specific NP felts to 
establish a referent. Let us, tentatively, accept this finding for 
simple sentences and took at cases where an indefinite NP belongs to 
a complement clause. There are many other factors that play a rote 
here besides negation. 
\]. 21 Modal verbs 
The following examples are anomalous in the intended sen&e, 
although there is no negation involved. 
(10) (a) You must write a letter to your parents. *They are 
expecting the letter. 
(b) Bill can make a kite. *The kite has a tong string. 
-9- 
Tr`aditionalty~ sentences with a modal auxiliary have been considered 
as simple sentences. However~ it has been argued convincingly by 
Ross (1967a) and others that modals should be analyzed as main ver"bs 
o£ higher` sentences. Therefore~ let us assume that~ even in the above 
examples~ the indefinite NPs originate in a complement clause~ just 
as they do in (l l). 
(l\]) (a) John wants to c~toh a fish. *Do you see the fish from 
here? 
(b) Mary expects to have a baby. *The baby's name is 
Sue. 
There is a great number` of verbs that behave like want and e_.~pect in 
this respect~ e.g. ~ try~ ptan~ ~ hope~ etc. What is common to 
all of them is that the complement sentence by itself is understood to 
represent a yet untrue proposition at the time specified by the tense 
and time adverbials in the main clause. The present pr`oblem~ is in 
fact~ another point in favor` of the view that modals originate in a 
higher" sentence~ because it enables us to acknowledge the similarity 
of the anomaly in (10) and (\]1). The conct'usion is that non-specific 
indefinites do not establish discourse referents when they appear` in a 
complement of a modal verb. 
l. 22 I mplicatives 
There is a class of verbs that~ if they are not negated~ imply 
the truth of the proposition represented by their" complement sentence. 
Let us call them implicative verbs. 2 In English~ this group includes 
-lO- 
vePbs such as manage, r'emembeP, venture, see fit, etc. An indefi- 
nite NP in the complement of an implicative vePb establishes a refer- 
ent, as shown by the following examples. 
(12) (a) John managed to find an apartment. The apartment 
has a balcony. 
(b) Bill ventuPed to ask a question. The lectuPer- 
answePed it. 
But if the implicative vePb in the main sentence is negated, a non- 
specific .indefinite fails to establish a PefePent. 
(13) (a) John didn't manage to find an apaPtment. *The apar't- 
rr.~nt has a balcony. 
(b) Bill didn'tdaPe to ask a ,question. *The lecturer" 
answe Ped it. 
There ape also verbs that inhePently have a negative implication. In 
English, this type includes ver'bs such as foP~t, fail, and neglect. 
Consider" the following anomalous discour"ses. 
(14) (a) John for-got to wr"ite atePm paper". *He cannot show 
it to the teacher". 
(b) John failed to find an answer". *It was wr"ong. 
These implicative ver"bs have the ver"y interesting pr"oper"ty that, if 
i 
there is double negation, the implication is positive, and an indefinite 
NP does, after" all~ establish a referent. 
-11- 
(14) (a) John didn't fail to find an answer. The answer was 
even right. 
(b) John didn't remember" not to bring an umbrella, 
although we had no room for" it. 
This pr`operty distinguishes clearly verbs with negative implication, 
such as for`ge.___, tt, fr`orn modal verbs discussed above, although both 
types deny the tr`uth of the proposition represented by the complement 
sentence. 
1.23 Factive verbs 
There is a group o£ verbs, called facttve verbs (Kiparsky 1968), 
that presuppose the truth of the proposition represented by the corn- 
plement. For" example, know, realty_e, and regret are factive. It is 
not surpr-istng to find out that an indefinite NP does establish a refer'ent 
in a complement of a factive verb, of course, provided that the com- 
plement itself is affirmative. 
(15) John knew that aar`y had a car', but he had never" seen it. 
In contr"ast to the implicative ver"bs discussed above, negation in the 
main sentence has no effect at art. 
(\]6) Bitl didn't realize that he had a dime. It was in his pocket 
The truth of the embedded preposition is presupposed even if the fac- 
rive ver"b itself is negated. Oonsequently, (16) is quite acceptable as a 
continuing discourse. 
-12- 
l. 24 Non-factive verbs 
The class of verbs commonly called non-factive (Kiparsky 
\]968) includes such verbs as believe, thin_~k~ say~ claim~ doubt. In 
general ~ nothing is presupposed about the truth of the embedded prop- 
osition. Notice~ however~ ~¢hatl~efollowing discourse would be 
contradictory. 
(17) I doubt that Mary has a car. *Bill has seen it, 
On the other hand, ther~ ism~thing v~ong with the following example. 
(18) Bill dou~ts-that Mary has a car. I have seen it. 
What makes these verbs difficult ±o handle is that there are two per- 
sons involved: the ~l~r and the subject of the non-factive verJo -- 
these roles may~ of couPse~ coincide. The speaker is not committed 
to any view whatsoeverabout the truth of the L~rnbedded proposition, 
although he may imply what his beliefs are as the discourse continues. 
For exampie~ in (18), the speaker--unlike Bill---must hold that the 
complement is true. The non-f-active verb is binding for the speaker 
only in case he is talking in the first person as in (\]7). BUt even in 
case that the speaker withholds judgment or disagrees altogether~ an 
indefinite NP in the complement of a non-factive verb that implies 
positive belief does establish a referent of a peculiar sort. It can be 
I 
referred to again in a complement of a similar non-factive verb that 
has the same subject. 
-18- 
(19) Bill says he saw a lion on the street. He claims the lion 
had escaped from the Zoo. 
What this amounts to is that a text interpreting device will have to sort 
out what belongs to 'the world as seen by the speaker' and" 'the world 
as seen by X'. The same referents need not exist in all of these 
worlds. 
A nor~factive vePb that implies positive belief (clairr~, think, 
believe, say, etc.) allows an indefinite NP in the complement to 
establish a referent as far as the world of the subject person is corr- 
cer'ned but need not have the same effect in the speaker's world. A 
non-factive verb with negative implication (doLLb._~t) may still allow that 
a referent is added to the speaker's world, albeit not to the world of 
the subject person. There is a spirited study of 'other worlds' by 
Lakof~ (1968b). 
l. 25 Genera\[ remar'ks 
We can now genePalize the previous' observation about single 
sentences to cover also complement clauses. A non-specific in- 
definite NP in an affirmative sentence (single sentence or a comple- 
ment) establishes a discourse referent just in case the proposition 
represented by the Sentence is assePted~ implied or" presupposed 
by the speaker to be true. A non-specific indefinite in a negative 
-14- 
sentence establishes a referent only if the proposition is implied to 
be False. This latter stipulation is needed because of negative 
implicatives discussed in ~ 1.22. In general, discourse referents 
exist in the realm 'world as seen by the speaker'. However, 
the non-factive verbs discussed in § I. 24 establish referents 
in other realms and are ambiguous as far as the speaker is 
concerned. 
In order to decide whether or not a non-specifEc in- 
definite NP is to be associated with a referent, a text interpreting 
device must be able to assign a truth value to the proposition 
represented by the sentence in which the NP appears. It must 
be sensitive to the semantic properVcies of verbs that take 
sentential complements, distin~ish between assertion, 
implication, and presupposition, and f\[nally, it must distinguish 
what exists for the speaker from what exists only for somebody 
else. 
l. 26 An apparent counterexample 
There is an interesting group of verbs that seem to 
provide a counter'example to the general rule. Cons{der the follow- 
ing discourses. 
(20) (a) l need a caP_. *It was a Mustang. 
(b) Seymour wants a knife. *It is sharp. 
-15- 
(c) John pr'omised Mary a bracelet. *The bracelet was 
very expensive. 
(d) The casting dtmector- was rooking for- an innocent 
btonde. *She was fr-om Bean Btossom, Indiana 
Provided that the indefinite NPs ape tnterppeted non-specificatty, 
art examples in (20) are anomalous, although they took super- 
ficiatty identical to those in (2t), which behave as expected. 
(2l) (a) ! owned a car'. It was a Mustang. 
(b) Seymour- imagines a knife. It is shar-p. 
(c) John bought Mar-y a br-acetet. The bracelet 
was very expensive. 
(d) The casting director- was looking at an 
innocent blonde. She was from Bean 
E3tossom, Indiana. 
in (20), what appear-s to be an omdtnar-y non-specific object NP 
fails to estabtish a ~,'efer-ent, atthough the ~ntence is affirmative 
asser-tion. There are many other- verbs in addition to those tn 
(20) that have this peculiar" consequence, for` exampte: ask for`, 
destr'e~ expect, hope for`, propose, ~quest, sugge~, watt for`, 
yearn for-. It seems significant that most if not all of these 
ver-bs, in addition to or`dinar`y noun phrase objects, atso take 
sententiaI complements, as the fottowing examples show. 
i6- 
(22) (a) Seymour wanted to have a Imife. 
(b) I propose that you eat a bagel. 
(c) John promised to give Mary a bracelet. 
(d) Mary expects John to b W her a braoelet, 
In fact, these are the same modal v~s discussed above (§1.21) 
that imply that the proposition represented by the complement is 
not yet true. We can thus account for tile l~culiarity of (20a-d) 
by assuming that, in spite of the simplicity-of the su~ce 
structure, the ordinary noun phrase objects of these verbs 
are derived from underlying r~~ons .which contain 
sentential objects. This is clearly one of those cases where 
semantic problems can be simplified by assuming a more abstract 
deep structure. But it is not entirely clear what kind of embedded 
sentence should underly the surface object. There seems to be little 
evidence for deciding this question beyond the observation that 
it cer~cainly .~uld be some type of existential or possessive con- 
struction. This is because of many noar paraphrases of the 
following type. 
(23) (a) John wants a car. -- John warts to heve a car. 
(b) I suggest an immediate halt ir~ the bombing - 
I suggest that there be an itrwr~diate halt in the 
bombing. 
- t7- 
(c) I expect no change in the situation. - I expect 
there to be no change in the situation. 
In some cases an existential paraphrase seems more natural, in 
• other" cases one prefers a possessive interpretation. Qbserve the 
difference between (23a) and (24a) and the two kinds ol = promising 
in (24b) and (24c). 
(24) (a) John wants a -revolution. - John wants there 
to be a ~evotution. 
(b) John promised Mary a bracelet. - John 
promised Mary that there will be a mir'acte. 
(c) John \]gromised Mary a miracle. - John 
promised Mary that there will be a miracle. 
Whatever- the correct solution is with regard to the exact nature o1= 
embedded sentence, there is no ~eason to consider the exceptional 
nature of verbs such as want, need, etc., as a serious counter ~- 
example to the general theory o1= discourse referents. 
l. 3 Short term Peferents 
In the preceding sections, it was tacitly assumed that 
discourse referents are ~_able entities that are established once 
and for all. But we have to recognize that an indefinite N P that 
fails to establish a permanent referent may never{heless permit 
-18- 
the appearance of coPefePential noun phrases within a limited 
domain. ConsideP the following examples. 
(25) (a) You must write a letter to your parents and mail 
the letter right away. *They ape expecting th__e_e 
\[etter. 
(b) John wants to catch a fish and eat \[.t fop supper. 
*Do you see the fish over there? 
(c) \] don't believe that MaPy had ~ and named her" 
Sue. *The baby has mumps. 
In (25a)~ it seems that the indefinite NP a letter may serve as 
antecedent fop a coPefepential definite NP the letter provided th~f 
the latter is contained in a conjoined complement sentence~ but 
not otherwise. Outside the scope of the modal must~ 'the letter t 
ceases to exist. Similarly, in (25b)~ thePe {s no fish that could 
be talked about outside the scope of want. Within the paiP of con- 
joined sentences in the complement it is a different matter. 
In oPdeP to take cape of these phenomena~ a text inter- 
pPeting device apparently has to process complex sentences 
i 
starting from the inside. Fop example~ in case of (25c)~ it fiPst 
has to consider the parec "tvlaPy had ~ and named her Sue ~', 
On the basis of ths first member of the conjunct~ it can s tenta- 
lively, set up a referent corresponding to the NP a bab~, and 
accept her" in the second sentenC~ as coPefePent{al. After 
lb 
u~ 
- t9- 
considering the whole sentence beginning with "\[ don't believe 
that... ", it then may decide that there is no such baby, 
after all. In short, a text interpreter must keep track of the 
status of referents it has established and delete them when 
necessary. 
Notice also that the life-span of a short term referent is 
not always so neatly bound as the above examples suggest. 
Sequences 0f the following type are quite common. 
(26) You must write a letter to your" parents. It has to be 
sent by air'matt. The letter must get there by 
tomorrow. 
At least incase of modals (and the future will), it is possible to " 
continue discussing a thing that actually does not yet exist, 
provided that the discourse continues in the same mode. In this 
case, every successive sentence is prefixed by the same type of 
modal. Even the following example is possible. 
(27) Mary wants to marry a rich man. He must be a 
banker. 
Under the non-specific interpretation of a rich man~ there is no 
specific individual yet that Mary wants to marry--and there may 
:never be one. By continuing with another modal, however, it is 
possible to elaborate on the attributes of this yet non-existing 
- 20- 
individual. 3 In the following sections we will present other oases 
where the tile-span of a short term referent may be extended. 
I. 4 Suppositions 
Another way to talk about what is not is to suppose that it 
is. Consider the following discourses. 
(28) (a) Suppose Mary had a car. She takes me to work 
in it. " I drive the car too. 
(b) If tv~ary has a oar~ she will take me to work in it. 
I can drive the cam too. 
(c) If Mary had a cam~ she would take me to work in it. 
I could drive the car too. 
(d) I wish Mary had a car. She would take me to work 
in it. I could drive the car too. 
(e) When Mary has a car~ she can take me to work in 
in it. I can drive the car too. 
All of the above examples elaborate a hypothetical situation that 
is based on the counter1:actual or dubious premise that Mary has 
a car. The difference between the first and the second pair is 
that in (28c--d) the condition is implied to be unrealizable or hard 
to realize. There are clearly several ways in which a supposition 
may be introduced in a discourse. Essentially~ however~ all of the 
- Z{ - 
above examples reduce to the form IF S O THEN S l . $2 • • • Sn" 
Whatever referent is introduced by S O exists fop the sequence 
S\[-Sn, which apparently has no fixed length, although there 
obviously ape certain conditions that all sentences belonging 
to it have to fullf/l\]. The following discourse would be 
anomalous. 
(29) I wish Mary had a car. *I will drive it. 
That is~ f/ctitious individuals may be referred to ~naphoPically 
only as long as the proper fictitious mode is sustained, but when 
the illusion is broken, they cease to exist. 
As the above examples show~ a text interpreter must also 
be able to cope with short tePr~ referents that owe their existence 
to some condition that in reality is not fulfilled. It must catch 
a supposition in whcx~tever foPn3 it comes and recognize where 
the supposition ceases to be in force. Neither of the two tasks 
is likely to be easy. FoP example, what looks like oomman d may~ 
nevertheless, be a supposition. 
(30) Lend him a book and heql never return it. 
1.5 Commands and Yes-No Questions 
It is to be expected that indefinite noun phrases in commands 
and Yes-No questions fail to introduce referents. The proposition 
- 22. 
corresponding to an interrogative or imperative sentence ordinarily 
is not assumed to be true. Thus there is something missing in 
the following examples. 
(3l) (a) Does John have a car? *It is a Mustang. 
(b) Give me a hotdog, please. *It looks delicious. 
But it is again possible to have coPeference within the imperative 
or interrogative sequence itself. 
(32) (a) Does John have a car and is it a Mustang? 
(b) Give me a hotdog~ please, but don't put 
any mustard on it. 
There are, however, ways to interpret the following examples as 
acceptable. 
(33) (a) Did you write a letter? Let me see it. 
(b) Give me a hotdog, please. I will eat it. 
For example, the interrogative sentence in (88a) need not be 
taken as a true question at all, but as an expression ol = 
surprise prompted by a preceding assertion. (33b) could 
be understood as elliptic. What is implicit is "You will 
give me a hotdog. " Discourses such as in (33) clearly ape 
not counterexamples, since their acceptability is not due 
to the appearance of an indefinite NP in a command or 
Yes-No question, but to other considerations. 
- 23 - 
l. 6 Quantil=iers 
IndeFinite noun phrases are generally ambiguous in 
sentences that contain quantifier-like expressions. The 
Following examples can be understood at least in two ways. 
(34) (a) Ha hvey courts a girl at every convention. 
(b) Most boys in this town are in love with a 
go-go dancer. 
(84a) can mean that, at every convention, there is some 
girl that Harvey courts, or that there is some girl that 
Harvey courts at every convention. Let us call the above 
paraphrases the non-specific and the specific interpretation 
of the N P a~rl, respectively. (See the note on specificity 
in 6l.t.) in the spectCtc sense, Harvey always courts the 
same girl, in the non-specific sense, it may be a 
di~erent girl each time. Similarly, a go-go dancer in (3413) 
also has two interpretations. However, the Following discourses 
leave no room for such ambiguity. 
(35) (a) Harvey courts agirl at every convention. 
She is very pretty. 
(b) Most boys in this town are in love with ago-go 
dancer'. Mary doesn't like her at all. 
- Z4- 
in (35), only the specific interpretation is possible. There must 
be a unique girl and a unique go-go dancer. This fact indicates 
that a non-specific indefinite fails to establish a discourse referent 
in case there is a quantifier'-like term in the sentence, in spite of 
the fact that the sentence is an affirmative assertion. 
But notice that the following example is ambiguous again. 
(36) Harvey cOurts a girl at every convention. She always 
comes to the banquet with him. The girl is usually also 
very pretty. 
(36) admits both the specific and non-specific interpretation of a girl. 
The reason for" the anomaly of the non-specific interpretation in (35) 
and its acceptability here is apparently that, in (36), every successive 
sentence continues to have a similar quantifier--like term: "at every 
convention", "always", "usually". There is also nothing wrOng with 
~he non-specific interpretation of the NP a book in (37). 
(37) Every time Bill comes here, he picks up a book and 
wants to borrow it. I never let him take the book. 
We have to say that, although a non-specific indefinite that falls into 
the scope of a quantifier fails to establish a permanent discourse 
referent, theme may be a short term referent ,within the scope of the 
quantifier and its life-span may be extended by flagging every succes- 
sive sentence with a quantifier" of the same type. 4 
- Z5 - 
2. Specificity 
Let us now return to the problem of specificity that was First 
introduced in § 1.1. As we already pointed out, many of the examples 
above that were judged anomalous in the intended sense can also be 
given another interpretation that makes them perfectly acceptable. 
0 
Although 'non-specific' indeFinites do not permit eoreFerence in (88), 
there is nothing wrong with these examples provided that the indefi- 
nite NP is understood 'specifically'. 
(88) (a) Bill didn't find a misprint. Can you Find it? 
(b) John wants to catch a fish. You can see the fish 
from here. 
How should we represent this distinction? As the terms 'specific' and 
'non-specific' imply, transformational grammarians have traditionally 
assumed that there is a feature \[+specific\], just as there is a feature 
F+deFinite\], and that indefinite NPs are to be marked with respect to 
specificity. Let us call this view, that goes together with Chomsky's 
original proposal that Core£erence be marked with integer--type indices, 
the classical theory. There is also another approach to these, prob- 
lems suggested by Emmon Bach (1968), James O. McOawley (1967), 
George l_ako£-£, and others. The essential feature ol = their proposals 
is that referential indices are variables, bound by quantifiers that act 
like quantifiers in symbolic logic. What corresponds to the indefinite 
- Z6 - 
article is, of course, something very similar to the existential 
quantifier in predicate calculus. (Bach calls it 'the some operator'.) 
Base structures resemble formulas in symbolic logic. This approach 
to syntax has now become known as 'generative semantics'. 
It is easy to see that in the framework of generative semantics 
e there is no justification nor need for a feature such as \[+specific\]. 
The ambiguities in question ape naturally accounted fop by the fact 
that the quantifier binding variable that underlies some indefinite 
noun phrase may be placed in different positions in the base 
structure. Specificity thus becomes a ITiEtter of the scope of 
quantifie rs. 
As far as the problems discussed in this paper ape relevant 
to choosing a theoPetical framework~ they seem to argue in favor of 
adopting the Bach-McOawle~ proposals, It is rather difficult to see, 
how one could achieve an adequate description of the facts in the 
classical theory, FoP example~ consider the following case. Both 
(39a) and (39b) are ambiguous with respect to specificity. 
(39) (a) Bill intends to visit a museum. 
(b) Bill visits a museum every day. 
In the tspecifict sense~ both examples establish a discouPse referent. 
q 
It would make perfect sense to continue with a descP~:~tion of lthe 
museum Bill intends to visit t or fthe museum Bill visits every day'. 
- 27- 
In the 'non-specific' sense, there is no such museum at all. So far 
so good, we can say that the NP a museum can be r+specific\]. But 
what about example (40)? 
(40) Bill intends to visit a museum every day. 
It is clear that (40) is ambiguous in many ways. For example, 
the quantified time adverb every day could be assigned either to the 
complement or to the main clause, let us now consider only the For- 
mer case. The remaining ambiguities should be attributable to the 
indefinite NP a museum, in Fact, we should have a two--way ambiguity 
between the specific and nort-specific interpretation. But example 
(40) is stilt ambiguous in more than two ways. It could be inter- 
preted to mean (41a)~ (41b), or (41c). 
(41) (a) 'There is a certain museum that Bill intends to 
visit every day. ' 
(b) 'Bill intends that there be some museum that he 
visits every day. ' 
(c) 'Bill intends to do a museum visit every day. ' 
It is easy to see why this happens. What the feature F±specific\] 
accomplishes in case of (39a) is that it clarifies the relation between 
the indefinite NP a museum and the verb intend in the main sentence: 
Is Bill's intention about some particular museum or not? In (39b), 
we employed the same device to characterize the relation between 
- Z8. 
the quantified time adverb~ and the indefinite noun phrase: 
Is it the same museum every day or not? To do the work in (40) we 
would need two features~ one to characterize the relation between 
intend anda museum, another for the relation between a museum 
and every day. Under the interpretation (4\]b), for exampte, _a 
museum would be non-specific with respect to the verb intend but 
specific with respect to the quantified time expr-ession. But to say 
that there are several varieties of specificity is a way of saying that 
there is no feature r+specific\] at alt. The ctassicat theory clearly 
is not sufficient to account for the multiple meanings of (40).5 
On the other hand, in the Bach-McOawtey framework we are 
able to account for the ambiguities in a straight-for~Nard way. The 
three senses of (40) discussed above might be represented roughly as 
in (42). 6 
(42) (a) (~ x)r museum(x), intend(Bitt, (every day)visit(Bitt, x))\] 
(b) intend(Bitt, (~x)r museum(x). (every day)vtsit(Bilt, x)\]) 
(c) intend(Bill, (every dayX~x)\[ museum(x), visit(Bill, x)\]) 
Another advantage of generative semantics is that there is an explana- 
tion ready for the fact that (40) establishes a discourse referent under 
only one of the three interpretations we have considered, narnely (42a). 
I 
The rule is that an indefinite NP establishes a permanent referent just 
in case the proposition to which the binding quantifier is attached is 
- 29- 
assumed (asserted, implied, or presupposed) to be true, provided 
that the quantifier is not itself in the scope of some higher quantifier. 7 
The First part of the rule accounts For the difference between (42a) 
and (42b-c), the second part is needed to explain why (39b) estab- 
lishes a permanent referent only under one of the two possible inter- 
pretations. Notice that, in (42a), the quantifier underlying the NP 
a museum is attached to the main proposition. Since the main 
proposition is asserted to be true and there are no higher quantifiers 
involved, (2~2a) establishes a referent corresponding to the NP a 
museum. Now, consider the other two interpretations of (40). The 
verb intend is one of the modal verbs discussed in (1.21). We know 
that the complement of a modal verb taken by itself is not implied or 
presupposed to be true. In (42b) and (42c), the quantifier underlying 
the NP a museum is attached to the complement. Therefore~ the 
above rule correctly predicts that no referent corresponding to a 
m 
museum is established under these two interpretations. 
From the point of view of a text interpreting device, the classi- 
cal theory has tittle to recommend itself. The problems studied above 
clearly argue in Favor of the Bach-McOawtey framework, in proces- 
sing a sentence, a text interpreter apparently has to associate an 
indefinite NP with a variable and attach the binding quantifier to some 
sentence above the NP using whatever clues there are present to 
- 30- 
assign the scope with as little ambiguity as possible. Clues that 
reduce scope ambiguity include the presence of an appositive relative 
clause or of special words such as "certain"~ "single"~ and "some" 
in the noun phrase itself and the surface order of quantifiers~ nega- 
tion, and articles in the rest of the sentence. Secondly~ the inter- 
preter has to keep track of the truth value of the proposition repre- 
sented by the sentence to which the quantifier is attached. The 
following example demonstrates some of the diFFiculties that are 
involved. Let us start a discourse with (43). 
(43) Mary may want to mammy a Swede. 
Highly schematicalty~ the underlying structure of (48) is something 
like (44). 
(44) 
NP VP 
I I ~ may 
Mary want N P I 
Mary marry x 
J The quantifier that binds the variable underlying the NP a Swede 
may belong to any of the three sentences, Sl, S 2, and $3, which 
causes (43) to be ambiguous at least in the following three ways. 
/ 
- 31 - 
(45) (a) 'There is some Swede whom Mary may want to 
marry. ' 
(b) 'It may be the case that there is some Swede whom 
Mary wants to marry, ' 
(c) 'It may be the case that Mary wants her future 
husband to be a Swede. ' 
Of the three sentences involved, only S l is asserted by the speaker 
to be a true proposition. The two other sentences, S 2 and S8, are 
both commanded by a modal verb (may and want), therefore, their" 
truth is not implied or presupposed. The indefinite NP a Swede 
establishes a discourse referent just in case its binding quantifier is 
attached to S 1" This can be demonstrated easily by pointing out that, 
if the speaker continues the discourse with (46), the preceding sentence 
(48) can only be understood in the sense of (4.5a). 
(46) She introduced him to her mother yesterday. 
However, the following continuation, .where the pronoun its' stands for" 
$2, permits both (4,5a) and (45b). 
(47) Suppose that it is tr`ue, then she will certainly introduce 
him to her mother. 
As a final example, after some thought it should be obvious that a 
disoourse consisting of (48) and (48)~ where the first \[t in (48) stands 
for S 2 and the second {.t fop S 8 is three ways ambiguous just as (48) 
- 3Z- 
by itself. Since all three component propositions of (44) are now 
either asserted or supposed to be true, there is no way of resolving 
the inherent scope ambiguity by looking at the coPefePentiality of a 
Swede and him. 
(48) Suppose that it is true and that she does it, then she 
will certainly introduce him to her mother. 
Although the argument against the traditional fea±ure \[-+specific\] 
should leave no doubt about its uselessness in discussing anything but 
the simplest kind of scope ambiguity, it does not necessarily mean 
that the familiar terms 'specific' and 'non-specific' should be 
rejected. They have proved quite useful and no harm is done, pro- 
vided that they are understood in a relative sense and not as denoting 
some absolute property inherent in indefinite noun phrases. For 
example, consider interpretation (45b) of (4.3), which assigns the 
quantifier to S 2 • One might ,want to say that, with respect to the 
verb want the indefinite NP a Swede is specific. On the other hand, 
if the quantifier is attached to $3, as in (45c), a Swede could be called 
non-specific with respect to want. In general', let us call an indefinite 
NP specific with respect to a given verb (or quantifier, or negation) if 
the latter is in the scope of the quantifier associated with the NP. It 
is non-specific in case the verb commands the quantifier. This kind 
of definition seems consistent with the way these terms have been used 
- 33 - 
in recent literature, and there is no reason to stop using them as 
tong as the relative nature of specificity is understood. 
3. Summary 
It is time to review the situation. We started by asking the 
seemingly na|be question: "When is there supposed to be an individual 
associated with an indefinite noun phrase?" Na|k/e as it may be, it 
must be answered in case there is ever going to be a device for inter-- 
preting written texts or everyday conversation with anything approach- 
ing human sophistication. There is also another reason to be 
interested in the subject. From a linguistic point of view, it is a 
problem of coreference constraints of a somewhat different kind than 
those studied under the label 'Pronominalization'. The present type of 
constraints are even more basic. It would seem that the question 
whether two noun phrases can be coreferential at all must precede the 
question whether a pronoun-antecedent relation may hold between them. 
Secondly, if relative clauses are derived transformationally from con- 
joined sentences by 'Retativization', as many linguists believe, the 
constraints discussed here are also a prerequisite For that transforma- 
tion. For these reasons~ the problems studied in this paper are of 
some theoretical interest quite independently from whether the results 
lead to any practical applications. 
- 34- 
We found that, in simple sentences that do not contain certain 
quantifier-like expressions, an indefinite NP establishes a discourse 
referent just in case the sentence is an affirmative assertion. By 
'establishes a discourse referent' we meant that there may be a 
coreferentiat pronoun or definite noun phrase tater in the discourse, 
indefinite NPs in Yes-No questions and commands do not establish 
refe rents. 
In studying more complicated examples, it was found neces- 
saP)/to replace Chomsky's integer-type referential indices by bound 
variables. In this frarnework~ the traditional problem of specificity 
is treated as scope ambiguity. We studied several types of verbs 
that take complements and their semantic properties. We concluded 
that~ in general~ an indefinite NP establishes a permanent discourse 
referent just in case the quant\[fier associated with it is attached to a 
sentence that is asserted, implied~ or presupposed to be true and 
there ape no higher quantifiers involved. 
There ape a couple of special problems: 'other worlds' and 
short term referents. Although discourse referents ordinarily exist 
for the speaker~ there is a class of fworid--cr@ating t verbs~ such as 
believe r that also establish PefePents of another kind. These exist 
fop somebody else~ not necessarily for the speaker. ThePefoPe~ we 
need to distinguish between the speakerts world and other realms 
- 35- 
and allow for the possibility that they are not populated by the same 
individuals. Secondly, there are short term referents, whose life- 
span may be extended by continuing the discourse in the proper mode. 
What this proper mode is depends on the circumstances. FoP 
example, every successive sentence may have to contain (i) a modal 
as the main verb, (ii) a quantifier of a certain type, or" (iii) be in the 
counterfactual mood. That is, it is possible to elaborate for" a~vhite 
on situations that are known not to obtain or that may or should 
obtain and discuss what sometimes or always is the case. 
FOOTNOTES 
* This work was supported by the National Science Foundation 
Grant GU-1598 to the University of Texas at Austin. 
1 These examples are due to C. LeRoy Baker 1966. 
2 I am indebted to Robert E. Wall for suggesting the term 'implicative' 
to me. 
3 What remains unexplained here is the fact (pointed out to me by 
John Olney) that must in (27) has two meanings depending on the 
specificity of the NP a rich man in the preceding sentence. If the 
first sentence is about a specific man, then must in the second 
sentence is interpreted in a rather weak sense: 'It is likely that he is 
a banker'. But tf the NP a rich man is non-specific, the second 
sentence means: 'It is necessary that he be a banker'. 
4 George Lakoff (forthcoming) has suggested that quantifiePs and 
negation be analyzed as verbs (predicates) instead of giving them a 
special status, as is usually done in symbolic logic. It is yet unclear 
to me whether there is any st,,bsta,.-tive issue involved or whether he 
is only proposing another no~-'_ .-,. 
5 There are other good aPgurr~nics against the feature \[ *_ specific\] in 
Janet Dean 1968. Unfortunately, they did not persuade the author 
herself. 
6 The complement of intend is what W. V. O. Quine calls 'opaque 
context'. \] ignore here his view that one should not be permitted to 
quantify into such a context. It seems to me that the objections he 
raises have to do with the double Pole names play in such contexts 
and only call fop moPe sophisticated linguistic analysis. Notice that 
Qu~ne approves of (i) while rejecting (ii) as meaningless '(Quine 1960, • 
p. 166): 
(i) (~x) (Tom believes x to have denounced Catiline) 
(it) (~x) (Tom believes that x denounced Oatiline) 
From a linguistic point of view, however, there is nothing but a 
superficial differenoe between (i) and (ii) due to 'Subject raising' that 
l~s applied in (i) but not in (ii). 
7 By 'higher" quantifier" I mean quantifter's such as al.._.t, each, many, 
and few, in fact, everything except the quantifier" associated Mth the 
singular" some and the indefinite article. The reason for" making this 
distinction is the fact that, if there are two indefinite singular NPs in 
the same sentence, both establish a referent no matter what their" 
order` is. 
(i) A dog was kitted by a car`. 
The above example, of course, justifies a later reference 
both to the dog and the car'. 

REFERENCES 

Bach, Emmon (1968) "Nouns and Noun Phrases. " in Bach and Harms 
(eds.) .Universals in Linguistic Theory. New York: Holt, 
Rinehart and Winston. 

Baker, C. LeRoy (1966) Definiteness and Indefiniteness in Engttsh. 
Unpublished Master's thesis. University of Ittinois. 

Chomsky, Noam (1965) Aspects of the Theory of Syntax. Cambridge, 
Mass. : MIT Press. 

Dean~ Janet (1968) "Nonspecific Noun Phrases in English" in Kuno, 
Susumu (ed.) Report No. NSF-20. Harvard University. 

Kiparsky, Paul & KiparCsky, Carol (\]968) "Fact. ' ' I in Biervvisch and 
Heidolph (eds.) Recent Advances in Linguistics. The Hague: 
Mouton and Co. 

Lakoff~ George (1968) "Counterparts~ or the Problem of Reference 
in Transformational Grammar. " Paper presented at the 1968 
Summer meeting of the LSA. 
(forthcoming) "Generative Semantics. " 

McCawley, James D. (1967) "Where Do Noun Phrases Come From?" 
Unpublished paper, to appear in Jacobs and Rosenbaum (eds.) 
Readings in Transformational Grammar. 

Quine> Willard Van Orman (1960) Word and Ob)ect. Cambridge, 
Mass. : MIT Press. 

Ross, John Robert (1967) "Auxiliaries as Main Verbs. " Unpublished 
paper. MIT. 
