TIIE ROLE OF PERSPECTIVE 
IN RESPONDING TO PROPERTY 
MISCONCEPTIONS 
• MS-CIS-85-31 
May 1085 
Kathleen F. McCoy 
Department of Computer & Information Science 
University of Pennsylvania 
Philadelphia, PA 19104 
This work is partially supported by the ARO grant DAA20-84-K-0061 and by the NSF 
grant #MCS81-07290. 
This paper appears in The Proceedings of IJCAI-85, August 18-23, 1985, University of 
California, Los Angeles, Ca. 
190 
Abstract 
In order to adequately respond to misconceptions involving an object's properties, 
we must have a context-sensitive method for determining object similarity. Such a 
method is introduced here. Some of the necessary contextual information is captured by 
a new notion of object perspective. It is shown how object perspective can be used to 
account for different responses to a given misconception in different contexts. 
191 
1. Introduction 
As a user interacts with a database or an expert system, s/he may attribute a 
property or property value to an object that that object does not have. For instance, 
imagine the following query to a database. 
U. Give me the HULL-NO of all DESTROYERS whose MAST-HEIGHT is above 
190. 
If a system were to evaluate such a query, it might find that there are no such ships in 
the database. The reason for this is that the user has queried a value of the property 
MAST-HEIGHT that it cannot have for the object DESTROYER. I term this error a 
property misconception. Upon encountering such a query, even a very cooperative 
system could only respond: 
S. There are no DESTROYERS in the database with a MAST-HEIGHT above 
190. Would you like to try again? 
In most cases, however, this is not the way a human would respond. A study of 
human/human transcripts reveals that a human conversational partner often tries to get 
at the cause of the misconception and offer additional information to correct the wrong 
information. The additional information often takes the form of a correct query that is a 
possible alternative to the user's query. In this paper I describe some of the knowledge 
and reasoning that are necessary for a natural language interface to a database or expert 
system to mimic this human behavior. 
In the above query, since there is an object similar to a DESTROYER that has the 
value of HULL-NO given, the user's misconception may result from his/her confusing the 
two objects. Hence a reasonable response would be: 
S. All DESTROYERS in the database have a MAST-HEIGHT between 85 and 
90. Were you thinking of an AIRCRAFT-CARRIER? 
Notice the strategy used to correct the misconception is to (1) deny (implicitly) the 
property/value given, (2} give the corresponding correct information, (3) suggest an 
alternative query containing the object the user may have confused with the 
misconception object. 
In other situations, a reasonable alternative query might involve the same object 
the user asked about, with a different property/value pair. This is the case in the 
following query. 
U. Give me the HULL-NO of all DESTROYERS whose MAST-HEIGHT is above 
3500. 
192 
S. All DESTROYERS in the database have a MAST-HEIGHT between 85 and 
90. Were you thinking of the DISPLACEMENT? 
This response is similar to the one given above except that the alternative query suggests 
an attribute rather than an object which may have been confused. 
In general, there can be two major reasons why a wrong attribution may occur. 
Either (1) the user has the wrong object - that is, s/he has confused the object being 
discussed with a similar object or has reasoned (falsely) by analogy from a similar object; 
or (2) the user has the wrong attribute - that is, s/he has confused the attribute being 
discussed with a similar attribute. If one of these two can be seen as likely in a given 
situation, then a revised query can be suggested which mentions the similar object or the 
similar attribute. 
To propose alternative queries, a system must have a method for determining 
similarity of objects and attributes. In this paper I will focus on responses involving 
object confusion; thus I will examine a similarity metric for objects. In the next section 
such a similarity metric is introduced. The following section introduces a new notion of 
object perspective which is needed to provide the similarity metric with some necessary 
contextual information, in particular, attribute salience ratings. Finally, an example of 
how perspective information and the similarity metric can be used to give reasonable 
responses to misconceptions involving object properties is given. 
2. Object Similarity 
As was shown above, in order to respond effectively to property misconceptions, we 
must have a method for determining object similarity. Object similarity has previously 
been shown to be important in tasks such as organizing explanations \[6\], offering 
cooperative responses to pragmatically ill-formed queries \[2\], and identifying metaphors 
\[9\]. In the above systems the similarity of two objects is based on the distance between 
the objects in the generalization hierarchy. One problem with this approach is that it is 
context invariant.* That is, there is no way for contextual information to affect 
similarity judgments. 
However, Tversky \[8\] proposes a measure of object similarity based on common 
and disjoint features/properties of the objects involved, which enables contextual 
*See \[5\] for additional problems and discussion of this point. 
193 
information to be taken into account. Tversky's similarity rating for two objects a and 
b, where A is the set of properties associated with object a and B is the set of properties 
associated with object b, can be expressed as: 
B(a,b) = Of CA CI B) - af(A - B) - ~(B - A) 
for some 0, ~, ~ )~ 0. This equation actually defines a family of similarity scales where 
0, a, and ~ are parameters which alter the importance of each piece of the equation, and 
f maps over the features and yields a salience rating for each. The equation states that 
the similarity of two objects is some function of their common features minus some 
function of their disjoint features. The importance of each feature involved (determined 
by the function f) and the importance of each piece of the equation (determined by 0, a, 
and ~) may change with context. 
Previous work \[4, 7\] has discussed the effect of °focus" on the prominence of 
objects. Focusing algorithms can be adapted to set the values of 0, a, and ~. For 
instance, if object a is "in focus" and object b is not, then the features of a should be 
weighted more heavily than the features of b. Thus we should choose a ~ ~ so that the 
similarity is reduced more by features of a that are not shared by b than vice versa. 
The problem then is to determine f. Other work \[3, 9\] has hand encoded salience 
values for the attributes of individual objects in the knowledge base, effectively setting 
the f function once and for all. This approach, however, is not sufficient since salience 
values must change with context. The following examples in which two objects 
(Treasury Bills and Money Market Certificates) are compared in two different 
circumstances, illustrate the importance of context on the similarity rating. 
Consider someone calling an expert financial advisor to see if she can better invest 
her money. She begins by telling the expert where her money is: 
U. We have $40,000 in money market certificates. One is coming due next week 
for $10,000... I was wondering if you think this is a good savings... 
E. 
Vo 
E. 
Well, I'd like to see you hold that $10,000 coming due in a money market 
fund and then get into a longer term money market certificate. 
l-lm.., well I was just wondering, what about a treasury bill instead? 
That's not a bad idea but it doesn't replace your money market certificate in 
any way - it's an exact duplicate. They're almost identical types of 
instruments - so one, as far as _l'm concerned, is about the same as another. 
Now consider how the same two objects can be seen quite differently when viewed 
194 
in a different way. Imagine the following conversation: 
U. I am interested in buying some US Government Securities. Now I was 
thinking of Money Market Certificates since they are the same as Treasure 
Bills. 
E. But they're not - they are two very different things. A Treasury Bill is 
backed by the U.S. Government: you have to get it from the federal reserve. 
A Money Market Certificate, on the other hand, is backed by the individual 
bank that issues it. So, one is a Government Security while the other is not. 
In the first example the objects are viewed as savings instruments. This view 
highlights attributes such as interest-rates and maturity-dates that are common to 
Treasury Bills and Money Market Certificates. This highlighting causes the two 
instruments to be seen as "identical'. In contrast, the second example views the objects 
as instruments issued by a particular company or organization. In this case attributes 
such as issuing-company and purchase-place are highlighted. Since these highlighted 
attributes are different for the two objects, the objects are seen as being quite different. 
As the examples illustrate, a context-free metric of similarity is not sufficient; 
contextual information is needed. A notion of object perspective, introduced below, can 
capture the needed contextual information. In particular, perspective accounts for how 
the f function (the assignment of salience values to various attributes) changes with 
context. 
3. Perspective 
\[4, 1\] note that the same object may be viewed from different perspectives. For 
instance a particular building may be viewed as an architectural work, a home, a thing 
made with bricks, etc. According to this work, an object viewed from a particular 
perspective is seen as having one particular superordinate, although in fact it may have 
many superordinates. The object inherits properties only from the superordinate in 
perspective. Therefore different perspectives on the same object cause different 
properties to be highlighted. 
Although this notion of perspective is intuitively appealing, in practice its use is 
rather difficult since it hinges on the use of a limited inheritance mechanism. The 
problem is that attributes may be inherited from the top of the generalization hierarchy, 
not just from immediate superordinates. So, an object's perspective involves not just one 
superordinate but a chain of superordinates. Therefore one must not only determine 
what perspective a particular object is being viewed from, but also what perspective its 
195 
superordinate is viewed from, and so on. As one continues up the hierarchy in this 
fashion, the definition of perspective as viewing an object as a member of a particular 
superordinate becomes less and less appealing. 
In addition, this notion of object perspective says nothing about the density of the 
generalization hierarchy. That is, in some situations the immediate superordinate of an 
object (and the properties it contributes} may be ignored. For example, even though a 
whale is a cetacean (a class of aquatic mammals including whales and porpoises}, this 
classification (and all attributes contributed by the classification} may be ignored in some 
situations in which the important attributes instead are inherited from a superordinate of 
cetacean, say, mammal. In other situations, the class °cetacean ° may be central. The 
notion of object perspective outlined above has no way of determining whether or not 
certain superordinates should be ignored or included. 
Here I introduce a new notion of perspective which is able to handle both the 
assignment of differing salience values and the density problem. In this notion, 
perspectives sit orthogonal to the generalization hierarchy. Each comprises a set of 
properties and their salience values. A number of perspectives must be defined a priori 
for the objects in a particular domain. The specification of perspectives, just like the 
specification of an object taxonomy, must be done by a domain expert. Knowledge of 
useful perspectives in a domain then, is part of the domain expertise. 
With this new notion of perspective, when an object is viewed through a particular 
perspective, the perspective essentially acts as a filter on the properties .which that object 
inherits from its superordinates. That is, properties are inherited with the salience 
values given by the perspective. Thus properties of the object which are given a high 
salience rating by the perspective will be highlighted, while those which are given a low 
salience value or do not appear in the perspective will be suppressed. The density 
rroblem is handled by ignoring those superordinate concepts which contribute only 
~,.ibutes suppressed by the current perspective. 
4. Using Perspective to Determine Responses 
Perspective information can be used with Tversky's similarity metric to help 
determine alternative queries to a query containing a misconception. To see how this 
works, consider a domain containing the following three objects with the attributes 
shown: 
196 
Money Market Certificates 
Maturity: 3 months 
Denominations: $1,000 
Issuer: Commercial Bank 
Penalty for Early Withdrawal: 10% 
Purchase Place: Commercial Bank 
Safety: Medium 
Treasury Bills 
Maturity: 3 months 
Denominations: 81,000 
Issuer: US Government 
Purchase Place: Federal Reserve 
Safety: High 
TreasuryBond 
Maturity: 7 years 
Denominations: $500 
Issuer: US Government 
Penalty for Early Withdrawal: 20% 
Purchase Place: Federal Reserve 
Safety: High 
and the following perspectives: 
Savings Instruments 
Maturity- high 
Denominations- high 
Safety- medium 
Issuing Company 
Issuer- high 
Safety - high 
Purchase Place - medium 
Notice that the perspective of Savings Instruments highlights Maturity and 
Denominations, and somewhat highlights Safety. This indicates that when people are 
discussing securities as savings instruments, they are most interested in how long their 
money will be tied up and in what denominations they can save their money. The 
perspective of Issuing Company, on the other hand, highlights different attributes. 
When securities are discussed from this perspective, things like who the issuer of the 
security is and how safe a security issued from that company is, become important. 
Suppose the perspective is Savings Instruments and the user says: 
197 
U. What is the penalty for early withdrawal on a Treasury Bill? 
This query indicates that the user has a misconception since s/he has attributed a 
property to Treasury Bills that they do not have. One reasonable correction to the 
query would contain an alternative query which to replaces Treasury Bills with another 
object that has the property specified and is similar to Treasury Bills. The system may 
reason that both Money Market Certificates and Treasury Bonds have the penalty 
specified, and so check to see if either of these objects is similar to Treasury Bills. 
Notice that the Savings Instruments perspective highlights attributes common to 
Treasury Bills and Money Market Certificates (they have the same Maturity and 
Denominations), as well as attributes disjoint to Treasury Bills and Treasury Bonds (they 
have different Maturity and Denominations). Using these salience values, the similarity 
metric will find that Money Market Certificates are very similar to Treasury Bills while 
Treasury Bonds are very different. Thus Money Market Certificates will be deemed a 
probable object of confusion and the following correction may be offered: 
S. Treasury Bills do not have a penalty for early withdrawal. Were you thinking 
of a Money Market Certificate? 
Notice that if the perspective had instead been Issuing Company, which highlights 
attributes common to Treasury Bills and Treasury Bonds and disjoint to Treasury Bills 
and Money Market Certificates, the most reasonable response would be: 
S. Treasury Bills do not have a penalty for early withdrawal. Were you thinking 
of a Treasury Bond? 
Selecting the appropriate perspective is in itself a difficult question which is 
currently under investigation and will be reported in \[5\]. Certainly important in the 
selection procedure will be the attributes that have entered into the conversation so far: 
these attributes should be of fairly high salience in the selected perspective. Other clues 
to the selection process include the objects under discussion, the superordinates which 
contribute the attributes under discussion to these objects, and the current goals of the 
USer, 
5. Conclusion 
In this paper we have seen that a context-dependent similarity metric is needed in 
order to respond adequately to misconceptions involving the properties of an object. 
Such a metric has been suggested and a notion of perspective has been introduced to 
account for some of the contextual information required by the metric. These notions 
have been shown to account for differences in the way a particular misconception is best 
198 
corrected in two different circumstances. 
6. Acknowledgements 
I would like to thank Julia Hirschberg, Aravind Joshi, Martha Pollack, Ethel 
Schuster, and Bonnie Webber for their many comments and discussions concerning the 
direction of this research and the content and style of this paper. 
7. References 
\[1\] Bobrow, D. G. and Winograd, T. SAn Overview of KRL, a Knowledge 
Representation Language." Cognitive Science 1, 1 (January 1977), 3-46. 
\[2\] Carberry, Sandra M. Understanding Pragmatically Ill-Formed Input. 10th 
International Conference on Computational Linguistics & 22nd Annual Meeting of the 
Association of Computational Linguistics, Coling84, Stanford University, Ca., July, 1984, 
pp. 200-206. 
\[3\] Carbonnell, Jaime R. & Collins, Allan M. Mixed-Initiative Systems For Training and 
Decision-Aid Applications. Tech. Rept. ESD-TR-70-373, Electronics Systems Division, 
Laurence G. Hanscom Field, US Air Force, Bedford, Ma., November, 1970. 
\[4\] Grosz, B. Focusing and Description in Natural Language Dialogues. In Elements of 
Discourse Understanding, A. Joshi, B. Webber & I. Sag, Ed.,Cambridge University 
Press, Cambridge, England, 1981, pp. 85-105. 
\[5\] McCoy, K.F. Correcting Object-Related Misconceptions. 1985. Forthcoming 
University of Pennsylvania doctoral thesis 
\[6\] McKeown, K. . Generating Natural Language Text in Response to Questions 
About Database Structure. Ph.D. Th., University of Pennsylvania, May 1982. 
\[7\] Sidner, C. L. Focusing in the Comprehension of Definite Anaphora. In 
Computational Models of Discourse, Michael Brady and Robert Berwick, Ed.,MIT 
Press, Cambridge, Ma, 1983, pp. 267-330. 
\[8\] Tversky, A. "Features of Similarity. = Psychological Review 84 (1977), 327-352. 
\[9\] Weiner, E. Judith. "A Knowledge Representation Approach to Understanding 
Metaphors." Computational Linguistics 19, 1 (January- March 1984), 1-14. 
199 
