Presuppositions and Default Reasoning: 
A Study in Lexical Pragmatics 
Robert E. Mercer 
Department of Computer Science 
Middlesex College 
University of Western Ontario 
London, Ontario, CANADA 
N6A 5B7 
mercer@csd, uwo. ca 
Abstract 
Explaining how the meaning of words relate to the meaning of the utterance in 
which they are used is of utmost importance. The most common approaches view 
the meaning of an utterance as a composition of the meanings of it parts, which 
of course include the words used to construct the utterance. This approach is suc- 
cessful for entailments. However, similar approaches to explain the presuppositionai 
behaviour of utterances have for the most part failed. In this paper we describe the 
application of Default Logic to the representation and the generation of natural lan- 
guage presuppositions. The view is taken that the presuppositions of an utterance 
are conjectures made by the hearer based upon the assumption that the speaker 
is following Grice's maxims of cooperative conversation. These conjectures repre- 
sent information implicitly contained in the utterance which cannot be generated by 
classicM techniques. The compositional framework is maintained. The difference is 
that functional units rather than predetermined semantic units axe inherited by the 
meaning structure. The function's meaning changes depending on the contents of 
the meaning structure. Hence, we view the study of these functional units as iexicai pragmatics 
rather than lexical semantics. Default Logic is one formal method for 
performing default reasoning in the area of Artificial Intelligence called Knowledge 
Representation. Default reasoning attempts to fill with conjectures the gaps left 
by classical forms of reasoning. We suggest that the use of non-classical inferenc- 
ing techniques such as default reasoning will prove fruitful in the realm of lexical 
reasoning. 
1 Introduction 
Lexical semantics deals with various aspects of the meanings of words. One such aspect 
is entailment. The standard meaning of the logical notion of entailment can be naturally 
extended to natural language sentences (see \[Cruse, 1986\], for example). We can abstract 
away from the linguistic setting and formalize this linguistic entailment relation, by con- 
sidering meaning postulates and the logical entailment relation (usually as some form of 
deduction). 
Here, we are interested in the kinds of meanings that we can give to words in various 
contexts which puts the meaning postulate within the scope of a negation. We again wish 
to abstract away from the linguistic setting and formalize this process. We accomplish 
this using Default Logic. Exactly how this formal method transfers back to the practical 
linguistic setting is left for future investigation. 
We have coined the expression lezical pragmatics to emphasize the significant differ- 
ences between the semantic and pragmatic aspects of a lexical item's meaning. Whereas 
224 
the semantics of a lexical item remains constant in all contexts, and hence can be pre- 
computed, the pragmatics of a lexical item is determined by the context in which it is 
found. The pragmatics of a lexical item must then be represented as a function which is 
computed in context. The use of default rules shows precisely what the pragmatic notion 
needs to be. 
2 Background 
What is a natural language presupposition? Being implied by a simple natural language 
sentence and the natural (or preferred) interpretation of its simple negation is the primary 
quality that qualifies an inference as a presupposition. This evaluation of inferences is 
called the negation test. Presuppositions are generated from lexical and syntactic contexts. 
Those contexts which pass the negation test can be termed presuppositional triggers. The 
presuppositional trigger that is of concern here is the definition of words. In (1) the truth 
of the affirmative a-sentence always implies the truth of the c-sentence, and the truth of 
the negative b-sentence normally implies the truth of the c-sentence. 
(1) a. My cousin is a bachelor. 
b. My cousin is not a bachelor. 
c. My cousin is a male adult. 
The natural language presupposition has been a well-studied linguistic phenomenon. 
Because it has received so much attention, it has naturally been given a variety of defini- 
tions. So, before presenting the technical details of our work, it should be clarified that 
the point of view of Gasdar \[1979\] is assumed. In particular, presuppositions are consid- 
ered as part of the information conveyed by an utterance (not appropriacy conditions for 
making an utterance) given that the utterance was made cooperatively in the sense of 
Grice's maxims of cooperative conversation \[Grice, 1975\]. So, in the case of (lb) the view 
would be that what is being communicated is that 'the speaker knotvs 1 that my cousin is a 
married male adult.', not that it is inappropriate to utter the sentence in contexts which 
don't allow for the truth of 'My cousin is a male adult.'. Although the speaker-hearer 
view of communication is implicit in Gazdar's approach, we have made it explicit in our 
communication model. This has been done because we are interested in what information 
is needed and what forms of inference are required to enable a hearer to infer the correct 
presuppositions from an utterance. 
Presuppositions have the quality that there are linguistically natural means for indi- 
cating that a simple negation is not to be interpreted normally, that is, some form of 
"presupposition cancellation" is to be performed. (What is meant by the terminology 
in scare quotes also differs from method to method.) It is this latter problem, how to 
produce only the correct presuppositions of sentences, that has received most of the at- 
tention in the literature. It is called the projection problem. From a lexical point of view, 
the projection problem can be interpreted as how do the meanings of the lexical items 
contribute to the meaning of the utterance. 
Examples are given in (2). It is important to note that two methods are used to 
indicate that the negation is to be interpreted in an unnatural manner. (2a) illustrates 
1 See \[Gazdar, 1979\] for a discussion of the motivation for the epistemic qualification of the proposition. 
Gazdar uses the usual omniscient interpretation of the epistemic operator, Know. This causes some 
problems which are not important here. 
225 
that one method is to make the normal (presuppositional) inference inconsistent. (2b) 
indicates a second method. This sentence is actually a disguised form for 'My cousin is a 
bachelor and not a spinster (case 1) or my cousin is a spinster and not a bachelor (case 
2).'. The the speaker cannot commit himself to the natural interpretation of the negations 
because each of the cases commits the speaker to deny the presuppositions required by 
the natural interpretation. Details are given in the next section. 
(2) a. My cousin is not a bachelor. He is only three years old. 
b. My cousin is a bachelor or a spinster. 
Note that the sentences in (3) do interpret 'bachelor' in the normal way under normal 
circumstances. 
(3) a. My cousin is a bachelor or Mary will be unhappy. 
b. If your cousin is a bachelor then he will be able to attend the party. 
3 Presuppositions and Default Logic 
Throughout our work the projection problem has been approached from a knowledge 
representation point of view. A basic assumption has been that a hearer's interpretation 
of a natural language utterance should include the inferences that can be generated from 
three sources: the sentence uttered, knowledge about the world, and knowledge about 
language use. 
For the purposes of this paper the most important knowledge source is the lexicon. 
It is here that some of the knowledge about language use is stored (for our purposes the 
definitions of words). Other linguistic knowledge, such as Grice's maxims, has been placed 
in the general inference procedure. What information is to be stored in the lexicon and 
in what form has been the focus of this study into the phenomenon called presupposition. 
Previously, the standard "solution" to the projection problem has been one of three 
kinds: a set of rules that take the presuppositions of the clauses and remove the undesirable 
presuppositions as the sentence meaning is being composed \[Karttunen, 1973, Karttunen, 
1974, Karttunen and Peters, 1979, Soames, 1979\]; a set of rules, invoked after the sentence 
has been fully interpreted, that cancel the unwanted presuppositions from a complete set 
of potential presuppositions \[Gazdar, 1979\]; or a set of rules that embody both of these 
methods \[Soames, 1982\]. 
In the work discussed below, we have rejected these solutions and have replaced them 
with a method that views presupposing as a form of inference. Before giving details of 
the method, it would be appropriate to point out the philosophical differences between 
this new method and the previous ones, indicating both the need for default reasoning 
and the benefits that are obtained. 
Previous methods and the new one agree that the information concerning the presup- 
positions associated with words is located in the lexicon. Where they differ is what 
form this information takes. This representation has important ramifications on the 
kinds of knowledge needed outside the lexicon. More specifically, previous methods 
put extensional knowledge in the lexicon, that is, the knowledge of a particular lexi- 
cal item is designated as presuppositional and non-presuppositional (or what has been 
termed "what is said"). So in the case of 'bachelor', 'unmarried' would be designated 
as non-presuppositional; 'male' and 'adult', presuppositional. What this precomputing of 
226 
presuppositions requires is that knowledge be contained in a projection method which 
can eradicate the unwanted presuppositions in sentences like (2). \[Karttunen, 1973, 
Karttunen, 1974\], \[Karttunen and Peters, 1979\], and \[Soames, 1982\] exemplify this dis- 
tribution of knowledge. \[Gazdar, 1979\] did not have to resort to such a distribution 
of knowledge, however this method succumbs to other problems. The new method has 
evolved naturally from Gazdar's ideas but does not succumb to the same problems. 
The new method differs in the following important respect: the information about pre- 
suppositions is stored in the lexicon as intensional information, that is, in a pre-computed 
form. As an utterance is processed to derive its meaning, the intensional information for 
all of the presuppositional triggers is incorporated into the structure which represents the 
meaning of the utterance. Later, we will see that an appropriate structure is a default 
theory. Since the intensional information is then computed in the appropriate context 
(the structure representing the utterance's meaning), the presuppositional information 
is correctly generated. By formulating the problem of presupposition generation in this 
way, the importance of the complete context for the correct generation of an utterance's 
presuppositions dramatically exposes the pragmatic nature of presuppositions. Hence, 
the kind of information that is stored in the lexicon and the kind of processing used to 
connect the information to the structure representing the meaning of the utterance tends 
to suggest that it should be under the heading of lexical pragmatics rather than lexical 
semantics. 
This new method can be viewed as a reasonable extension to Carnap's meaning pos- 
tulates \[Carnap, 1956\]. The meaning postulate is extended to include the intensional 
information, which can be given a precise meaning. The extensional approach on the 
other hand cannot be an used to extend the idea of a meaning postulate since the notion 
of the universality of the meaning postulate is lost. 
Another important repercussion is that the previous methods are ad hoc, that is, 
they are not based on fundamental principles that have their foundations outside the 
particular method. On the other hand the default method can actually provide a better 
understanding of the phenomenon and leads to a definition based on foundations which are 
not intimately connected to the phenomenon and which are not complicated by operational 
features. 
Other benefits are obvious. We are dealing with a very large task when we confront 
the natural language understanding problem. We don't need to compound the difficulty of 
this task with purported solutions that lead to knowledge needlessly distributed amongst 
various parts of the solution. The default reasoning method has succeeded in making what 
seems to be a most natural break, that is, the knowledge associated with the lexical entries 
is located in the lexicon and the knowledge concerning the basic inference mechanism is 
located in the inference engine. 
3.1 Details of the Representation 
The work on presuppositions and default reasoning began with \[Mercer and Reiter, 1982\] 
reporting on its embryonic stages. Here we looked at representing presuppositions as 
default rules in Reiter's \[1980\] Default Logic. More recently, we have viewed this method 
as a natural way to extend Carnap's notion of a meaning postulate. This extension differs 
from the original in two significant ways. Firstly, we make the change from a one-way 
material implication in the meaning postulate to a logical equivalence. So, we are not 
only accepting that the right hand side of (4) are necessary elements of the definition 
227 
of 'bachelor', we are also accepting that these are sufficient, as well. Much debate has 
focussed on this issue. We agree that it is probably impossible to arrive at a necessary and 
sufficient set of conditions for the definition of a lexical item. However, this is not a theory 
about the correct definition of lexicai items. Rather, it is an agent-centred method about 
how an agent uses his knowledge to derive other facts. Hence, we do take as given that 
at any moment an agent does have necessary and sufficient conditions defining a lexical 
item. This definition may be subject to change. How an agent discovers that his definition 
needs modification, and how this is performed is not part of our immediate concern. If (4) 
were a one-way material implication, the contrapositive of (4) would say that the negation 
of any of the three right hand side predicates is sufficient to negate 'bachelor'. However, 
it does not say that the negation of 'bachelor' necessarily requires that the negation of 
one of the right hand side predicates is the cause. What having necessary and sufficient 
conditions allows an agent to do is determine that one of the right hand side items in (5) 
is causing the negation of 'not a bachelor' although it cannot be determined which one. 
(4) Vx.BACHELOR(z) ~ MALE(z) A ADULT(z) A-,MARRIED(z) 
(5) Vz.-,BACHELOR(x) ~_ ~MALE(x) V ~ADULT(z) V MARRIED(x) 
Secondly, we add more information about the lexicai item in question. This extra 
information concerns the presupposed ~ parts when the lexicai item occurs in the scope of 
a negation. This extra information takes the form of default rules and implicitly adds a 
new form of inference, as well. So for example, we add to the representation of 'bachelor' 
the two default rules (6) and (7). 
(6) -,BACHELOR(z) A LF(BACH ELOR, m): MALE(m) 
MALE(x) 
(7) ~BACHELOR(z) A LF(BACHELOR, z): ADULT(z) 
ADULT(x) 
A more detailed discussion of Default Logic is given in the next section. For the present, 
an interpretation of these two default rules is sufficient. (6) is to interpreted as stating that 
if the non-bachelorhood of a is provable from the default theory (x will be substituted 
appropriately throughout the rule with the appropriate constant) and if the linguistic 
form (that is, the sentence) predicates bachelorhood about a, then if it is consistent in the 
default theory to infer that x is male then infer that x is male. (7) says the same thing 
about a and the property of being an adult. Default rules will be used to determine which 
of the right hand side elements is causing the negation. So, under normal circumstances 
(6) and (7) will generate MALE(x) and ADULT(a) which together with (5) gives the 
result that MARRIED(x). 
2 Earlier work has called these the noncriterial parts. The one part negated by the preferred interpreta- 
tion was called the criterial part. This terminology was used because we were concerned with describing 
the negated and unnegated parts of the preferred interpretation without referring to presuppositions. Since 
this terminology differs from the use of criterial in \[Cruse, 1986\], say, (each of male, adult, and unmarried 
would be crlterial), and since it is not so critical here to be so pedantic about the term presupposition, 
we will use this latter terminology. 
228 
3.2 Details of the Generation Method 
The study of the method using Default Logic as a replacement for the projection meth- 
ods is the main theme of \[Mercer, 1987\]. A summary of this work appears in \[Mer- 
cer, 1988b\]. An important contribution is the connection made between presupposi- 
tions and the context, in particular Gazdar's formalization of Grice's maxims. More 
precisely, default theories corresponding to a well-defined set of cases which can be de- 
rived from the hearer's knowledge base (KBH) and the utterance (represented as G(u) 
to convey Gazdar's view of the Gricean maxims) generate the appropriate presupposi- 
tions for sentences such as those found in (2) and (3). The precise details for com- 
puting the set of cases (which can be found in \[Mercer, 1987, Mercer, 1988b\]) are not 
important for this paper. Instead, some examples will be provided in order that an 
understanding of how the Default Logic method generates the appropriate presupposi- 
tions for a variety of sentences. The default theories that correspond to the cases men- 
tioned above will be given without any explanation. Interested readers are encouraged 
to seek further explanation in the references given above. More recently, \[Mercer, 1988a, 
Mercer, 1990\] have solved problems left open in \[Mercer, 1987\], specifically, how negation 
interacts with sentential adverbs in the context of conditionals, and a generalization of 
Gazdar's view of clausal implicatures. The default method now captures all of the known 
problems for any of the previous methods plus the problem of disjunctive presuppositions 
which none of the previous methods could handle. 
Before embarking on a study of the application of Default Logic to the problem of 
natural language presuppositions, a brief introduction to Default Logic is in order. The 
inference rules of classical logic produce safe deductions. Deductions are safe if they are, 
in some sense, already known. Examples of this type of inference include: deducing that 
a particular individual has a property from the knowledge that all individuals have that 
property; deducing the truth of two conjuncts given the truth of the conjunction; and 
deducing the consequent of a material implication given the truth of the antecedent. 
Default Logic, on the other hand, allows (in a principled manner) the conjecturing 
of new information in the face of incomplete information. Remembering the discussion 
surrounding (5), the representation of a negated lexical item does not provide complete in- 
formation, and since a more complete interpretation of the negated lexical item is required, 
conjecturing of the extra information is necessary. 
A default rule is a rule of inference as shown in (8), 
(8) .° 
where are all first order formulae whose free variables are among those 
of ~ = ml,..., ram. Intuitively, a default rule can be interpreted as for all individuals 
zl,..., zm, if a(~) is believed and if #(~) is consistent with our beliefs, then w(g) may 
be believed. The a, #i's, and w are referred to as the prerequisite, justifications, and 
consequent of the default rule, respectively. 
Normal defaults are default rules with the properties n = 1 and/31 (~) = w(~:). Since 
all the default rules used throughout this paper are normal, discussion is restricted to 
this class. In addition to this structural property, normal defaults have other important 
properties which are discussed below. Closed defaults are default rules that contain no 
free variables in any of the first-order formulae. Open defaults are default rules that are 
not closed. An open default is meant to represent the set of closed defaults obtained 
229 
by replacing all its free variables by ground terms. Default rule schemata are meta- 
default rules that represent the set of open default rules obtained by replacing all the 
recta-variables with the appropriate substitution instance (for example, predicates, or 
formulae). A default theory, A, is composed of a set of first-order formulae W, and a set 
of default rules, D. The default rules can be viewed as extending the first-order formulae 
with the consequents of the default rules. An extension, E, of a closed default theory is a 
constructable fixed point having the properties listed below. 
1. WC_E 
2. if E ~- a, then a E E (E is logically closed, that is, Th(E) = E, where Th is a fixed 
point operator defined by 
3. for each default, a(g) : ~(~) E D, if a(g) E E, and -~(i) ~ E, then #(~) E E 
Normal default theories always have an extension. Multiple extensions are produced when 
two defaults conflict. Normal defaults conflict when their consequents cannot simultane- 
ously be in the same extension. These multiple extensions are orthogonal, that is, pairwise, 
they differ on the truth value of at least one formula. The third point in the definition 
of E is very important. Limiting the discussion to the effect that this definition has on 
the application discussed in the remainder of this paper, since the first-order part of the 
default theory and all its entailments are in E, if a conflict occurs between a default and 
anything in the logical closure of W the consequent of the default rule is in no extension 
of A. 
Although Default Logic has a proof theory, it will not be described here. Rather all of 
the discussion regarding presuppositions will use the definition of extensions just given. 
We are able to do this since the logic has a kind of soundness and completeness result. 
That is, the proof theory cannot generate any formulae that is not in some extension, and 
if a formulae is in an extension, then the proof theory can generate it. 
For purposes of all the following examples, the definition of 'bachelor' is represented 
by the first order sentence (4). Then the negation of 'bachelor' would be represented by 
(5). Similarly the definitions for 'spinster', and its negation, are given by (9) and (10), 
respectively. 
(9) Vx.SPINSTER(x) ~- FEMALE(m) A ADULT(m) A-,MARRIED(x) 
(10) Vx.~SPINSTER(x) _-- ~FEMALE(x) V -,ADULT(m) V MARRIED(m) 
The pragmatic rule for generating presuppositions from all lexical items can be captured as 
a default rule schema (11), where PRE represents information about the presuppositional 
features of the negated lexeme. The knowledge of presuppositional features could then be 
part of the lexicon and could be represented as (12). 
(11) -uP(x) A iF(P, x) : Px(x) Pl(x) , if PRE(P, P1) is in the lexicon 
(12) PRE(BACHELOR, MALE) 
PRE ( BA CHEL OR, A D UL T) 
PRE( SPINSTER, FEMALE) 
PRE( SPINSTER, AD UL T) 
230 
This method is just an economical form of representation. Since (11) is a schema it 
really represents the set of its instances. These instances can be obtained as needed 
by appropriately substituting for P and P1 throughout the schema. For the case of 
'bachelor' we have already seen the instances as (6) and (7). The instances for 'spinster', 
are similar. All of the examples below will use the already substituted instances rather 
than the schema. Before turning to the examples, we first need a more precise definition 
of presuppositions based on defaults. 
3.2.1 A Definition of Presuppositions 
In addition to the broad coverage, the default method provides a definition of presup- 
position not forthcoming from the previous methods. Before giving the definition of 
presuppositions, we require the prior definition of the consequents of the default theory. 
Definition 1 If D is any set of defaults then 
CONSEQUENTS(D) = {w(m) Ia(x) : #'(z)'''"/~n(x) E D 
For purposes of the next definition, the only defaults in KBH, are the presupposition- 
generating defaults found in the speaker's lexicon. 
Definition 2 Let u be a sentence uttered by a speaker, S, in accordance with Grice's 
Maxims of Cooperative Conversation. Let KBH be the bearer's knowledge base before the 
utterance, and let the default theories ~ucasel,... , AUcue n be the first order cases of 
the theory KBtt U {~(u)}. A sentence c~ is a presupposition of u with respect to KBH if 
and only if 
(i) a is in all extensions of AUca,ei, for i = 1 .... , n, 
(ii) a is in the deductive closure of CONSEQUENTS{D}) 
(iii) a is not an entailment of KBH W {G(u)} 
(iv) oe is not inferable from KBH 
3.2.2 Example 1 -- Simple Negation 
In the situation in which 'My cousin is not a bachelor.' has been uttered, and the logical 
form created maps 'my cousin' onto the constant cl, the representation of the sentence 
would be that given in (13). However, included in the hearer's lexicon is the information 
provided by the extended meaning postulate for 'bachelor' represented by (14)-(16). These 
four items taken together provide the default theory that represents the meaning of this 
utterance. 
(13) "-BA CHEL OR(c1) 
(14) -.BACHELOR(x) A iF(BACHELOR, x): MALE(x) 
MALE(x) 
(15) -,BACHELOR(z) A LF(BACttELOR, x) : ADULT(x) 
ADULT(x) 
231 
(16) Vx.~BACHELOR(x) ~ ..MALE(x) V ~ADULT(x) V MARRIED(x) 
It can easily be seen that all extensions of the default theory consisting of (13)-(16) (there 
is only one) contains MALE(c1) and ADULT(cl). By the definition given in the previous 
section, these are the lexical presuppositions of the utterance derived from 'bachelor'. 
3.2.3 Example 2 -- Negation with Qualification 
In the situation in which 'My cousin is not a bachelor, because my cousin is just three years 
old.' has been uttered, and the logical form created maps 'my cousin' onto the constant 
cl, the representation of the sentence would be that given in (17). Again, included in 
the heater's lexicon is the information provided by the extended meaning postulate for 
'bachelor' represented by (18)-(20). These four items taken together provide the default 
theory that represents the meaning of this utterance. 
(17) -BACHELOR(ct) A-~ADULT(cl) 
(18) -~BACHELOR(x) A LF(BACHELOR, x) : MALE(x) 
MALE(x) 
(19) ~BACHELOR(x) A LF(BACHELOR, x): ADULT(x ) 
ADULT(x) 
(20) Vx.~BACHELOR(x) ~ ~MALE(x) V ~ADULT(x) V MARRIED(x) 
It can be easily seen that all extensions of the default theory consisting of (13)-(16) (there 
is only one) contains MALE(c1) but does not contain ADULT(c1). The reason is that 
-~ADULT(cl) is an entailment of this default theory which defeats the justification of the 
default rule (19) thereby not allowing its consequent into the extension. Hence MALE(c1) 
is a lexical presupposition of the utterance, but ADULT(c1) is not. 
3.2.4 Example 3 -- Multiple Cases with No Cancellation 
In the situation in which 'My cousin is a bachelor or Mary will be unhappy.' has been 
uttered, and the logical form created maps 'my cousin' onto the constant cl, and 'Mary' 
onto the constant ml, the representation of the sentence would be that given by the two 
cases (21) and (22). As stated earlier, the computation of these cases is not given here. 
However, it should be obvious that these two cases are the two disjuncts of the exclusive 
or representation of the utterance. 
(21) 
(22) 
(23) 
(24) 
(25) 
BA CHEL OR(c l) A --1 UNHA PP Y (ml) 
--BACHELOR(el) A UNHAPPY(ml) 
~BA CHELOR(x) A LF( BA CHELOR, x): MALE(x) 
MALE(x) 
-.BACHELOR(x) A LF(BACHELOR, x) : ADULT(m) 
ADULT(x) 
Vx.BACttELOR(x) ~ MALE(x) A ADULT(x) A --MARRIED(x) 
232 
(26) Vx.~BACHELOR(z) ~ -,MALE(x) V -ADULT(z) V MARRIED(x) 
Together with the lexical information that is available for 'bachelor', which is given in 
(23)-(26) we obtain MALE(c1) and ADULT(cl) in all extensions of all cases. So, these 
are presuppositions of the utterance. 
3.2.5 Example 4 -- Multiple Cases with Cancellation 
In the situation in which 'My cousin is a bachelor or a spinster.' has been uttered, and 
the logical form created maps 'my cousin' onto the constant cl, the representation of the 
sentence would be that given by the two cases (27) and (28). 
(27) BA CHEL OR(c1) A -,SPINSTER(el) 
(28) -BA CHEL OR(cl) A SPINSTER(cl) 
-BACHELOR(z) A LF(BACHELOR, z): MALE(z) 
(29) MALE(x) 
(30) -,BACHELOR(x) ^ LF(BACHELOR, z) : ADULT(x) 
ADULT(x) 
(31) Vz.BACHELOR(z) _= MALE(z) A ADULT(z) A -~MARRIED(x) 
(32) Vz.-,BACHELOa(z) =_ -,MALE(x) v -~ADULT(z) v MARRIED(z) 
(33) -,SPINSTER(z) ^ LF(SPINSTER, z): FEMALE(x) 
FEMALE(z) 
(34) -SPINSTER(z) A LF(SPINSTER, z): A D ULT(x) 
ADULT(x) 
(35) Yz.SPINSSTER(z) ~ FEMALE(x) ^ ADULT(z) A -,MARRIED(x) 
(36) ¥z.-~SPINSTER(x) _= -~FEMALE(x) V -~ADULT(z) V MARRIED(x) 
(37) Vz.MALE(z) D ~FEMALE(x) 
The first case, represented by (27), together with the lexical information represented 
by (29)-(37) form a default theory which allows MALE(cl) and ADULT(cl) to be de- 
rived as entailments. Consequently, no extension of this theory contains FEMALE(cl), 
the consequent of the default rule (33), since the justification is defeated by MALE(el). 
ADULT(c1) can also be derived using the default rule (34). 
Similarly, in the second case, represented by (28), together with the lexical infor- 
mation represented by (29)-(37) form a default theory which allows FEMALE(cl) and 
ADULT(cl) to be derived as entailments. Since the justification of default rule (29) is 
defeated by FEMALE(cl), no extension of this theory contains MALE(cl). ADULT(c1) 
can also be derived using the default rule (30). 
Since not all extensions contain MALE(cl) and not all contain FEMALE(cl), neither 
of these formulae are presuppositions of the utterance. Although ADULT(c1) is in all 
extensions of all the cases it is not a presupposition of the utterance since it is entailed 
by it. 
233 
3.3 Lexical Pragmatics 
As mentioned earlier, we consider that this enterprise falls under the heading of lexical 
pragmatics. Now that a short description of this study of presuppositions has been given, 
it is appropriate to make more precise what we mean by the term, lexical pragmatics. 
The key difference between semantics and pragmatics has always been the role of con- 
text. It is evident from the previous discussion that for natural language presuppositions 
the role of context is paramount. There are explicit negations which produce different 
outcomes depending on the (non)existence of further qualifications. Sentences which have 
exactly the same syntactic structure, 'A or B.', have different presuppositional behaviours 
depending upon the content of 'A' and 'B'. Although the role of context is fundamental, 
here we would like to focus instead on the functional units, the default rules, in order to 
make more direct connections back to the source of these ideas: knowledge representation 
and default reasoning in Artificial Intelligence. 
Once the semantic meaning of 'bachelor' has been agreed upon, (38) say, this meaning 
remains unchanged no matter what context it is put in. This property is referred to 
as monotonicity. Monotonicity requires the behaviour of the inference mechanisms to be 
localized. All classical inference mechanisms are monotonic. So, for instance, the maleness 
of a person is always the case given that the person is a bachelor. Consequently, these 
attributes can be precomputed and associated with the lexical item 'bachelor'. 
(38) MALE(x) A ADULT(x) A--MARRIED(x) 
On the other hand the meaning of 'not a bachelor' has two components: the semantic 
component, (39), and the pragmatic component, which is most commonly defined in terms 
of what is normally the case. Again, the semantic component is unchanging. It states 
that at least one of the disjuncts is true whenever the person is 'not a bachelor'. On the 
pragmatic side, our approach is to reject the normal case only if something contradicts 
it. This non-monotonic form of reasoning is demonstrated by Default Logic. The default 
rules make explicit what it is that can defeat the generation of the consequence of that 
rule. These rules are intensional objects whose meanings are made explicit only in context. 
However, these rules can still be attached directly to the lexical item in exactly the same 
manner as the precomputed semantic components. 
(39) -~MALE(x) V -~ADULT(x) V MARRIED(x) 
The contextual nature of the presupposition inference as captured by default reason- 
ing indicates that one difference between semantics and pragmatics revolves around the 
monotonic/non-monotonic character of semantic and pragmatic inferences. This view 
adds a new dimension to pragmatics. We believe that it is this view that has the most to 
offer the lexicon. 
4 What Remains 
How to apply this method in a practical implementation still needs to be investigated. 
Some of the remaining problems are in the domain of research in Default Logic. Default 
Logic is not even semi-decidable in the first order case. In the propositional case it is 
intractable even for syntactically very restricted forms of theories. One issue which has 
recently become a focus of attention in the default reasoning area is the notion of scope, 
234 
that is, when deciding whether the default succeeds, what part of the theory should he 
investigated for the potentially defeating information. \[Etherington et al., to appear\] The 
interest in scope is twofold: to make default reasoning more tractable, and the commonly 
held belief that defaults in general are somewhat localized in their influence. For our 
purposes here, the sentence presents a very natural scope. It is probably not coinciden- 
tal that most of the examples given above deal with intrasentential competition among 
presuppositional triggers and entailments. 
Although this view of scoping can be used to overcome a variety of problems, some of 
the very (in)famous examples of presupposition deal with noun phrases. Here, we come 
in contact with referring phrases, which again force us into potentially considering every 
object in our domain as a possible argument to defeat the default assumption. However, 
we believe that a proper reading of Grice and an application of Gazdar's epistemic view 
of communication can provide the needed relief from this problem (although the details 
have yet to he worked out). In particular, if the speaker uses a particular referring 
expression in the utterance, then the speaker is suggesting that it is under this description 
that the information is stored. Thus the speaker is also communicating that he knows 
the information being communicated is related to this descriptor rather than some other 
descriptor. Our task of deciding what the speaker is communicating is therefore limited 
to this descriptor. (This of course requires an epistemic operator that is not omniscient.) 
This limitation assumes that there is no commonly known synonym for the referring 
expression (which the speaker would not need to include in the utterance). If common 
synonyms exist, then we regard this as a problem of knowledge representation, that is, 
how do we deal with these localized equalities without recourse to a promiscuous equality 
relation. 
Earlier we suggested that we have been interested in formalizing the notion of pre- 
supposition. Inherent in this task is the use of a formal apparatus, in this case, Default 
Logic. The application of these ideas to the practical situation of lexicon design and use 
will require some more effort. For example, the lexical presuppositions that have been 
discussed are attached to particular lexical items. So, for instance, the presuppositions 
of 'not a bachelor' do not hold for the logically equivalent expression 'not an unmarried 
adult male' nor does 'female' which logically entails 'not a bachelor' allow the default as- 
sumption of 'adult' to hold. In order to prevent these unwanted inferences, we have used 
guards on the default rules (the LF predicate) to limit them to appropriate contexts. 
Another method would be to allow defaults to enter the inferencing process for limited 
periods of time rather than being constantly available. How this method compares with 
the standard view of default rules in a Default Logic remains to be investigated. Once 
these questions are answered, the attachment of the defaults to lexical items in the lexicon 
and their proper use in computing the meaning of an utterance can be achieved. 
Lastly, it is of some importance to see what questions can be asked given this view 
of lexical features. Figure 1 represents a small portion of an ISA hierarchy of some 
hearer, H. The arcs have been labelled with features and the nodes have been labelled 
by the term which is associated with the set of features defined by the path from the 
root node. The deciding criterion for a negated lexeme's meaning is associated with the 
last feature which labels the last arcs in the paths from the root node to the negated and 
unnegated lexical nodes. In this instance the non-presuppositional feature is MARRIED 
-- "-,MARRIED. The emboldened path in Figure 1 represents the preferred interpretation 
of 'not a bachelor', that is, 'a married adult male'. Given this simplistic taxonomic 
structure two questions arise. Can the lexicon be organized in such a way that the 
235 
s ° 
HUMAN ~'" 
" A E HUMAN" 7~ "FEMALE HUMAN" M L • .m 
.. 
HUSBAND of \e BACHELOR 
Figure 1: Portion of H's ISA hierarchy. 
presuppositional and non-presuppositional features are given implicitly by the structure 
rather than having to be explicitly represented? Negation would normally modify only the 
last arc. Given that this type of representation can be achieved are there any interesting 
answers that arise from the structure, that is, are presuppositions given by the way we 
structure our linguistic (world) knowledge? 
5 Conclusion 
We have investigated a formal approach to the definition of natural language presupposi- 
tions. It has involved defining the presuppositions of lexical items as default rules together 
with the use of Default Logic proof theory in appropriate contexts to generate the pre- 
suppositions of utterances containing the lexical items that generate presuppositions. 
This is but one use of defaults in the lexicon. Other uses are fairly obvious -- and 
they are obvious in a sense that makes the boundary between the lexicon and classical 
knowledge representation very fuzzy. For example, the notion of canonical traits is the 
classical example given to support the use of default reasoning in knowledge represen- 
tation: Tweety is a bird, therefore it (probably) flies. As well, what happens when a 
standard concept is modified, for example, dog vs toy dog. The issue of what gets negated 
is very similar in nature to the problem of presuppositions: one item in the concept dog is 
negated everything else (by default) stays the same. So, in the toy dog situation the alive 
property is negated but toy dogs still have four legs, two ears, and a tail, it barks, and 
so on. These features may need to be appropriately modified (for example the barking is 
imaginary). However, the uses in the lexicon of techniques from knowledge representation 
seem open-ended. 
6 Acknowledgements 
This work was partially supported by the Natural Sciences and Engineering Research 
Council of Canada grant 0036853, and the Institute for Robotics and Intelligent Systems, 
a Canadian Network of Centres of Excellence. 
236 

References 
\[Carnap, 1956\] R. Carnap. Meaning and Necessity. Chicago University Press, 1956. 
\[Cruse, 1986\] D. A. Cruse. Lezicai Semantics. Cambridge University Press, 1986. 
\[Etherington et al., to appear\] D. W. Etherington, S. Kraus, and D. Perlis. Nonmono- 
tonicity and the scope of reasoning. Artificial Intelligence, to appear. 
\[Gazdar, 1979\] G. J. M. Gazdar. Pragmatics: Implicature, Presupposition, and Logical 
Form. Academic Press, 1979. 
\[Griee, 1975\] H. P. Grice. Logic and conversation. In P. Cole and J. L. Morgan, editors, 
Syntax and Semantics, v.3, Speech Acts, pages 41-58. Academic Press, 1975. 
\[Karttunen and Peters, 1979\] L. Karttunen and S. Peters. Conventional implicature. In 
C-K. Oh and D. A. Dineen, editors, Syntax and Semantics, v.11, Presuppositions, pages 
1-56. Academic Press, 1979. 
\[Karttunen, 1973\] L. Karttunen. Presuppositions of compound sentences. Linguistic In- 
quiry, 4:169-193, 1973. 
\[Karttunen, 1974\] L. Karttunen. Presupposition and linguistic context. Theoretical Lin- 
guistics, 1:181-194, 1974. 
\[Mercer and Reiter, 1982\] R. E. Mercer and R. Reiter. The representation of pre- 
suppositions using defaults. In Proceedings of the ~th Biennial Conference of the 
CSCSI/SCEIO, pages 103-107, 1982. 
\[Mercer, 1987\] R. E. Mercer. A Default Logic Approach to the Derivation of Natural 
Language Presuppositions. PhD thesis, Dept. of Computer Science, University of British 
Columbia, 1987. 
\[Mercer, 1988a\] It. E. Mercer. Solving some persistent presupposition problems. In Pro- 
ceedings of the 12th International Conference on Computational Linguistics, pages 420- 
425, 1988. 
\[Mercer, 1988b\] It. E. Mercer. Using default logic to derive natural language presupposi- 
tions. In Proceedings of the 7th Biennial Conference of the CSCSI/SCEIO, pages 14-21, 
1988. 
\[Mercer, 1990\] It. E. Mercer. Deriving natural language presuppositions from complex 
conditionals. In Proceedings of the 8th Biennial Conference of the CSCSI/SCEIO, 
pages 114-120, 1990. 
\[Iteiter, 1980\] R. Iteiter. A logic for default reasoning. Artificial Intelligence, 13:81-132, 
1980. 
\[Soames, 1979\] S. Soames. A projection problem for speaker presuppositions. Linguistic 
Inquiry, 10:623-666, 1979. 
\[Soames, 1982\] S. Soames. How presuppositions are inherited: A solution to the projection 
problem. Linguistic Inquiry, 13:483-545, 1982. 
