Exemplar-Based Sense Modulation 
Mohsen Rais-Ghasem (mohsen@scs.earleton.ea) 
Jean-Pierre Corriveau (jeanpier@ses.carleton.ca) 
School of Computer Science; Carleton University 
Ottawa, ON, K1S 5B6 Canada 
Abstract 
A great deal of psycholinguistic findings reveal that context highlights or obscures 
certain aspects in the meaning of a word (viz., word sense modulation). Computational 
models of lexicon, however, are mostly concerned with the ways context selects a 
meaning for a word (word sense selection). In this paper, we propose a model that 
combines sense selection with sense modulation. Word senses in this proposal consist of 
a sense-concept and a sense-view. Furthermore, we outline an exemplar-based approach 
in which se~se-views are developed gradually and incrementally. A prototype 
implementation of this model for sentential context is also briefly discussed. 
1. Introduction 
The main focus of this paper is the effects of 
context in modulating the meaning 
representation of an unambiguous noun as it 
appears in different contexts. The role of 
context in many existing computational lexicons 
ends once one reading, out of a number of 
contrasting readings, ~s selected. However, 
many psycholinguistic findings indicate that 
context affects the storage and retrieval of 
concepts. In this paper, we will argue that 
lexicon models must account for both sense 
selection and sense modulation (see Cruse, 
1986). 
We first motivate our work by emphasizing 
the inability of existing lexicon models to 
account for contextual effects at a level finer 
than the sense level. We then discuss our 
proposed model, which represents word senses 
as pairs of sense-concepts/sense-views. We will 
also describe how sense-views evolve from a 
number of exemplars. The paper concludes with 
a brief description of an implemented prototype 
and results of some experiments. 
2. Motivations 
Elsewhere (Rais-Ghasem, 1998), the first author 
has reviewed existing computational lexicon 
models and showed that, despite their 
differences, they all subscribe to the same 
meaning theory, namely sense enumeration 
(Seidenberg et al., 1982; Simpson and Burgess, 
1985). Fundamental to this theory are two 
assumptions: 1) the possibility of listing or 
enumerating all possible meanings for each 
word, and 2) the existence of a selection process 
in which one of these meanings is selected for a 
given word. 
One of the main disadvantages of such 
enumerative lexicons is their inability to account 
for a phenomenon generally known as semantic 
flexibility (see Barclay et al., 1974; Barsalou, 
1993; Greenspan, 1986). In short, semantic 
flexibility concerns changes that context causes 
in representation of concepts in memory. Many 
psycholinguistic findings have shown that 
context seems to highlight or obscure certain 
properties of a single concept as it appears in 
different contexts. For example, in an early 
experiment, Barclay et al. (1974) demonstrated 
how the interpretations of familiar, 
unambiguous words vary with context. For 
instance, they argued that the choice of 
attributes for piano is affected by the verb 
selection in The man (l~ed) (tuned) (smashed) 
(sat on) (photographed) the piano. They then 
provided evidence that the prior acquisition of a 
sentence like The man lifted the piano (vs. The 
man tuned the piano) influences the 
effectiveness of cues like "something heavy" 
(vs. "'something with a nice sound") in recall. 
They concluded that context can affect the 
encoding of concepts in memory. Similarly, 
Barsalou (1982) reported that subjects verify 
contextually relevant properties significantly 
faster than contextually irrelevant properties 
Witney et al. (1985) also report similar 
results. Their work is particularly interesting 
since they distinguished between two levels of 
property activation: 1) the functional level 
(useful in activities such as sentence 
comprehension and classification), and 2) the 
semantic access level (corresponding to the 
information that is actually accessed upon 
85 
seeing a word). They used intervals of 0, 300 
and 600 ms.-and found that all properties of a 
word were initially activated (accessed), 
regardless of the context. However, the 
contextually irrelevant ('low-dominant' in their 
terminology) properties would die off rapidly, 
and hence play a negligible role in the overall 
sentence comprehension. 
Greenspan (1986) studied the effect of 
sentential context on concrete nouns. He 
examined the presence of central and peripheral 
properties of a noun in different contexts. For a 
given a noun, Greenspan presented his subjects 
with a pair of sentences where in each sentence 
a different type of properties was emphasized. 
For example, consider the noun basket. Being a 
container is a central property for basket 
whereas being made of straw is a peripheral one. 
Each of the following sentences focuses on one 
of these properties. 
The butler placed the letter in the basket. 
(Container) 
Sally took several days to weave the basket. 
(Straw) 
Later he examined subjects' recall in various 
memory tasks and found that the central 
properties were activated regardless of the 
context, but peripheral properties were activated 
only if they were emphasized by the sentence. 
He further showed that the emphasized central 
properties were more activated than 
unemphasized central properties. He concluded 
that the interpretation of a concrete noun is a 
function of both the sentential context and the 
noun's central properties (Ibid.). 
Anderson et al. (1976) demonstrated that 
general terms are contextually instantiated to 
more specific terms. For example, they used 
fish in contexts like the following sentences and 
hypothesized that it was likely to be instantiated 
respectively to goldfish, trout and shark. 
The grocer stared at the fish in the bowl. 
The grocer stared at the fish in the stream. 
The fish attacked the swimmer. 
They examined their hypothesis in a number of 
experiments and found that an expected 
instantiated term was a better cue for the recall 
of a sentence then the general term itself, even 
though the general term had appeared in the 
sentence and the particular term had not. 
To the best of our knowledge, none of the 
existing computational lexicon models have 
adequately dealt with semantic flexibility. 
Having subscribed to the sense enumeration 
theory, the effects of context in these models are 
limited to selecting of one sense. Any changes 
in the characteristics of a selected sense either 
has to come in the form of a new sense or is 
ignored. 
This requires the ability to foresee any context 
that a word can appear in and define a separate 
sense for it. Obviously, this is impractical, if not 
impossible (see Corriveau, 1995, chapter 2). 
Thus, any lexicon model must support a 
combination of sense generation and sense 
selection. 
Of the various computational lexicon models, 
this issue is specially problematic for symbolic 
lexicons (e.g., Dyer, 1982; Wilensky, 1978) and 
lexicons based on semantic networks (Lange & 
Dyer, 1989; Waltz & Pollack, 1985). This is due 
to the fact that word senses in these models are 
coded as discrete entries. 
Owing to their adopted continuous 
representation, connectionist models, however, 
are potentially capable of dealing with 
contextual effects at a finer level. In fact 
McClelland and Kawamoto (1986) reported an 
unintended yet interesting result. They had 
presented their model with The ball broke the 
vase. Although throughout the training ball was 
always associated with the microfeature soft, in 
the output it was associated with the 
microfeature hard. They attributed this result to 
the fact that breakers in their experiment were 
all hard and the model had shaded the meaning 
of ball accordingly. 
However, the connectionist approach also has 
some disadvantages. First of all, connectionist 
lexicons presuppose a set of universal and fixed 
microfeatures and demand that every sense be 
characterized in terms of such microfeatures in 
advance. This constitutes a serious problem for 
any real world application. 
But what is even more important is the 
difficulty to separate patterns of contextual 
effects from the representation of a word sense. 
For instance, consider breakers in McClelland 
and Kawamoto (1986). It is impossible to 
examine this category of objects by itself. Thus 
we cannot determine 1) what qualifies an object 
to be a breaker or 2) what else can be a 
breaker? 
We believe separating patterns of contextual 
effects from internal representations of context 
is important. Firstly, such patterns can be 
thought of as ad hoc categories, categories built 
by people to achieve goals (Barsalou, 1983). For 
instance, the breakers category can be 
instrumental in achieving the goal of "breaking 
a window". Secondly, from a learning point of 
view, such patterns can be very useful. Rais- 
Ghasem (1998) has shown how a concept can 
evolve (i.e., acquires new properties) from such 
86 
patterns as it appears in various contexts. Also, 
Rais-Ghasem (Ibid.) has employed such patterns 
to implement a metaphor understanding system 
that interprets metaphors as class inclusion 
assertions (see Gluksberg & Keysar, 1990). 
3. A Lexicon for Sense Modulation 
In this section, we propose a lexical model that 
not only selects senses for input words, but also 
contextually modulates the selected senses. 
Examples used in this section are from a 
prototype implementation of this model for 
sentential contexts. 
3.1 Two-Tiered Word Senses 
Cruse (1986) specifies two ways in which 
context affects the semantic contribution of a 
word: sense selection and sense modulation. 
Sense selection happens in cases of lexical 
ambiguity where one sense is chosen among a 
number of contrastive senses (see Pustejovsky, 
1995) associated with a word. In sense 
modulation, the semantic characteristics of a 
selected sense are modulated according to its 
surrounding context. Cruse describes two types 
of sense modulation: a) highlighting or 
backgrounding, and b) promotion or demotion 
of features. In the former, context underlines 
certain aspects (i.e., properties) of the concept 
selected for a given word while dimming others, 
as it happens for piano in the following 
examples: 
The musician played the piano. 
The musician broke the piano. 
The musician lifted the piano. 
Context can also promote, or demote, certain 
aspects of a word's associated concept. For 
instance, teacher, by definition, is gender 
neutral. However, in a context like the 
following, the feature gender is promoted for 
teacher. 
The teacher stroked his beard. 
Similarly, certain aspects can be demoted, as is 
solidity, a property usually associated with 
butter, in this context: 
Mary poured the butter into the jar. 
To account for both sense selection and sense 
modulation, we propose to structure word 
senses into two tiers: sense-concept tier and 
sense-view tier (see Figure 1). While the sense- 
concept tier specifies the core meaning of a 
word in a given context, the sense-view tier 
specifies how this meaning is to be viewed in 
that context. It is our contention that this simple 
structure is capable of accounting for various 
types of sense modulation. 
Sense-Concept J 
Figure 1: Two-tiered word sense. 
Here is an example of the word sense generated 
by the implemented prototype (see section 4) for 
piano in The musician moved the piano. 
\[Sense-Concept(s)\] PIANO 
\[Sense View\] 
W EIGHT(0.666667)---->:H EAVY, 
STATE-OF-MATTER(0.756)---->:SOLID, 
IsKindOf-ARTIFACT(0.7047)---->: 
COLOR(0.7047)---->: 
AGE(0.7047)---->: 
OPERATED-BY(0.7047)o--->:HUMAN, 
As shown, this word sense consists of two parts: 
a sense-concept and a sense-view. Since 
concepts to a large extent are conventionalized, 
sense-concepts are specified only by means of 
the concepts they represent (concept PIANO in 
this case). Sense-views, however, are specified 
as a set of properties (e.g., WEIGHT) and their 
possible values (e.g., HEAVY). Each property is 
also accompanied with a number between 0 and 
1 indicating the property's weight (or centrality) 
in a given sense-view. The implemented system 
relies on the Mikrokosmos ontology (Mahesh 
and Nirenburg, 1995) to specify properties for 
sense-concepts and sense-views j. 
Now, compare the above word sense with the 
word sense generated for piano in a different 
context such as The musician played the piano. 
\[Sense-Concept(s)\] PIANO 
\[Sense View\] 
IsKindOf-MUSlCAL-INSTRUMENT(0.81 )°--->: 
WORK-EQUIPMENT-OF(0.81)---o>:MUSICIAN, 
IsKindOf-ARTIFACT(0.729)---->: 
STATE-OF-MATTER(0.729)---->:SOLID, 
COLOR(0.729)---->: 
AGE(0.729)---->: 
OPERATED-BY(0.729)---->:HUMAN, 
IsKindOf-INANIMATE(0.6561 )---->: 
I It must be noted that our model is not dependent on any 
particular ontology or set of properties. The choice of 
Mikrokosmos is primarily justified because of its 
availability and relative comprehensiveness. 
87 
Note that the two word senses share the same 
sense-concept. They differ, however, on their 
sense-views. Because of the sense-view, the first 
word sense portrays piano as an object of 
moving (by highlighting properties such as 
weight and artifact). Conversely, the second 
word sense underlines a different aspect of 
piano, namely the fact that it is a musical 
Instrument. 
As examples of property demotion/promotion, 
consider the word senses respectively generated 
for mttsician in the above two contexts. In the 
first context, we get: 
\[Sense-Concept(s)\] MUSICIAN 
\[Sense View\] 
GENDER(0.885367)---->:MALE, 
IsKindOf-H UMAN(0.8187)---->: 
IsKindOf-P RIMATE(0.54)---->: 
Notice how all properties specific to musician 
are demoted in this example, since they are 
irrelevant to the role played by musician in this 
context (i.e.. being a mover). This is not the case 
in The musician played the piano. 
\[Sense-Concept(s)\] MUSICIAN 
\[Sense View\] 
WORK-EQUIPMENT(1)---->:MUSICAI- 
INSTRUMENT, 
AREA-OF-INTEREST(1)---->:FIELD-OF-MUSIC, 
IsKindOf- ENTERTAINM ENT-ROLE(0.9)---->: 
IsKindOf-ARTISTIC-ROLE(0.81 )---->: 
IsKindOf-SOCIAL-ROLE(0.729)---->: 
IsKindOf-HUMAN(0.6561 )---->: 
GENDER(0.6561 )---->: 
IsKindOf-SOCIAL-OBJ ECT(0.6561 )---->: 
3.2 Exemplar-Based Sense-Views 
In this section, we outline a lexicon model in 
xvhich sense-views are developed gradually and 
incrementally. That is achieved by exploiting 
exemplars 2. Exemplars are previously processed 
contexts that exemplify the usage of word 
senses. In other words, every exemplar consists 
of a number of word senses, each of the latter 
formed by a pair of the form (sense- 
concept/sense-view). 
We first define the alike relationship: 
Definition 1: Sense-concepts SC, and SC 2 are 
alike if they appear in similar contexts, and they 
share the same thematic role. 
For example, according to this relationship, 
piano, fridge and computer are alike in the 
following inputs: 
They used a crane to move the stove. 
2 Similar to semantic relations in Cruse (1986). 
John carried the computer. 
Four men moved the piano. 
In other words, from the view point of "being 
moved", these representationally different 
sense-concepts can be classified together. 
This parallels the two different roles that 
Franks (1995) proposed for concepts. He 
distinguished between the representational and 
classificatory functions of concepts. While the 
former is used to discem instances of one 
concept from others, the latter specifies how an 
instance of a concept should be classified. 
Franks (Ibid.) argues that, depending on context, 
fake gun could be classified along with a gun, a 
toy, a replica, and a model. 
Given that all alike sense-concepts share the 
same sense-view, we can define a sense-view as 
the intersection of the properties of the sense- 
concepts that are connected to it. Figure 2 
illustrates this situation. 
Figure 2: A shared sense-view. 
A sense-view is initialized to the properties of 
the first sense-concept to join it. As more sense- 
concepts join, the sense-view evolves to better 
reflect what is common between all those sense- 
concepts. For example, the following is the 
content of the sense-view object of moving 
initiated with only one exemplar: 
The man moved the stove. 
\[Sense View\] 
Thematic Role: Object - No. of Exemplars: 1 
WEIGHT(1 )---->:heavy, 
IsKindOf-COMPLEX-MECHANI S M(0.9)---->: 
IsKindOf-MECHANISM(0.81 )---->: 
IsKindOf-DEVlCE(0.729)---->: 
STATE-OF-MATTER(0.729)--->:SOLID, 
MADE-OF(0.729)---->:PLASTIC,METAL, 
I sKindOf-ARTIFACT(0.6561 )--->: 
COLOR(0.6561 )---->: 
AGE(0.6561 )---->: 
OPERATED-BY(0.6561 )---->:HUMAN, 
IsKindOf-COOKING-AP PLIANCE(0.9)---->: 
IsKindOf-COOKING-EQUIPM ENT(0.81 )---->: 
IsKindOf-EVE RYDAY-ARTIFACT(0.729)---->: 
This definition gradually becomes more 
representative of this sense-view as new alike 
88 
sense-concepts are included. After adding 
computer, as it appears in: 
The student moved the computer. 
we get: 
\[Sense Viewl 
Thematic Role: Object - No. of Exemplars: 2 
IsKindOf-DEVICE (0.7695)--->: 
STATE-OF-MATFER(0.7695)---->:SOLID, 
MADE-OF(0.7695)---->:PLASTIC,METAL, 
IsKindOf-ARTIFACT(0.69255)---->: 
COLOR(0.69255)---->: 
AGE(0.69255)---->: 
OPERATE D-BY(0.69255)---->:H UMAN, 
This trend continues with piano in: 
The musician moved the piano. 
\[Sense View 1 
Thematic Role: Object - No. of Exemplars: 3 
WEIG HT(0.666667)---->:heavy, 
IsKindOf-DEVICE(0.513)---->: 
STATE-OF-MATTER(0.756)---->:SOLID, 
MAD E-OF(0.513)----> :MATERIAL, 
IsKindOf-ARTIFACT(0.7047)---->: 
COLOR(0,7047)---->: 
AGE(0.7047)---->: 
OPERATE D-BY(0.7047)---->:HUMAN, 
Notice how the picture of "a heavy artifact" is 
emerging 3. The modification of a sense-view 
continues until the sense-view reaches a certain 
level of stability (for more details, see Rais- 
Ghasem, 1998). 
3.3 From Words to Senses 
Having described the two-tiered word senses 
and our adopted exemplar-based approach to 
develop sense-views, we can now overview the 
models overall behavior. 
Input to the system is a context, presented as a 
number of input words and along with their 
syntactic categories and case markers (Delisle et 
al., 1993). For example, a sentence like The 
musician played the piano would be presented 
to the model as musician (noun, p-subj), play 
(verb, -), and piano (noun, p-obj). Case markers 
p-subj and p-obj respectively indicate 
"positional subject" and "positional object". 
The goal is first, to select proper senses for 
input words, and second, to contextually 
modulate those senses. The model attempts to 
accomplish both tasks by looking for an 
3 This process also meets, to some extent, another 
requirement, namely idiosyncrasy of word senses. For 
example, the property hem 3 ' is present in this sense-view 
only because of its strong presence in the sense-concepts 
added to this sense-view so far. That may not be the case 
for another reader and may not hold over time. 
89 
adaptable exemplar. Adaptability between a 
stored exemplar and an input context is defined 
as follows: 
Definition 2: Adaptability between an exemplar 
context C~ and an input context C, is a function 
(f) of the compatibility between the sense-views 
associated with C~ and their thematically 
corresponding concepts in C,. Compatibility 
between a sense-view and a concept is defined 
based on the degree of overlap in their 
properties. 
For instance, assume that the model already 
maintains the following exemplars: 
El. The musician played the guitar. 
E2. Mary played soccer. 
Also suppose the following input is presented to 
the model. 
John played the piano. 
To determine the adaptability of this input with 
E1 and E2, we should first find sense-views 
corresponding to the input words, and then 
measure their compatibility. Let us assume that 
we know that the input words John, play and 
piano respectively correspond to sense-views 
associated with musician, play and guitar in E1 
and sense-views associated with Mary, play and 
soccer in E2. 
Let's begin with E2. It is possible to select a 
sense for the input word John which is 
compatible with sport-player (the sense-view 
associated with Mary). The same is true for 
input word play and the sense-view playing- 
sport. However, finding a sense for piano which 
is compatible with the sense-view sport 
(soccer's associated sense-view) is not possible. 
Therefore, the input context and E2 are not 
adaptable. 
Conversely, E1 and the input context, are 
adaptable: John is compatible with music-player 
(musician's associated sense-view), play with 
playing-music, and piano with musical- 
Instrument. 
Adaptability, in fact, allows the model to 
decide if a set of word senses can be selected or 
generated for the input words. Put roughly, 
given some input words, the resulting word 
senses must be reusable together in an adaptable 
exemplar (see Rais-Ghasem, 1998, for more 
details). 
In Definition 2, correspondence between an 
input word and a sense-view was defined based 
on the thematic roles. Since thematic roles are 
unknown for the input words, the implemented 
system relies on case markers to find 
corresponding sense-views for an input word. 
Every sense-views maintains a list of case 
markers (which may evolve as new members 
join). Here is an example of the sense-view 
breakin,~-instrument. 
\[Sense View\] 
Thematic Role: Instrument - Marker(s): with-pp, p-subj, 
STATE-OF-MATTER(0.73305)--->:SOLID, 
MADE-OF(0.82805)---->: MATERIAL, 
The markers with-pp and p-subj indicate that 
breakers are usually marked either as subject or 
with-prepositional phrase in the input. Here are 
examples: 
The thief broke the windshield with a bat. 
The rack smashed the window. 
Once an adaptable exemplar is found, the model 
not only knows what senses must be selected for 
the input words, but also by associating these 
senses with their corresponding sense-views, the 
selected senses will be modulated. For example, 
John in the above example will be associated 
with the sense-view music-player (a case of 
property promotion). Similarly, the word play 
will be disambiguated _to playing-music (as 
opposed to plavin,~-sport in E2). 
4. Implemented Prototype 
The implemented system is structured in two 
marker passing networks. The bottom network, 
the ontology network, serves as the system 
knowledge base to define concepts in the second 
network. This definition includes concept 
properties and relationships between concepts. 
We used Mikrokosmos ontology (Mahesh & 
Nirenburg, 1995). 
The second network, the lexicon network, 
consists of four layers of nodes. Figure 3 
displays the schematic structure of the lexicon 
network. Lexemes, (displayed as squares) 
appear at the bottom. Connected to lexemes are 
concepts (rounded rectangles). Concepts are 
connected exemplars (double-lined rectangles). 
Exemplars constst of a number of sense- 
concepts (occurrences of concepts in 
exemplars). Associated with each sense-concept 
is a sense-view (displayed as banners). 
Figure 3: The structure of the lexicon network. 
90 
The process begins when input words are 
looked up the in ontology and their 
corresponding lexemes are found. Concepts 
connected to these lexemes are then activated 
which, in turn, leads to activation of all 
exemplars in which input words appear. 
Exemplars activated for a word, or more 
precisely, for the word's associated concepts, 
represent the model's knowledge, up to that 
point, of various ways that the input word can 
interact with other words. 
In the implemented system, determining 
adaptability is carried out simultaneously and 
concurrently by individual exemplars triggered 
by input words. Attached to each exemplars (in 
fact, attached to a .group of exemplars with 
sirnilar context) is an agent. Agents, 
(implemented as Java TM threads) receive 
activation and individually start measuring the 
adaptability of their exemplars with the input. 
More details can be found in Rais-Ghasem 
(1998). 
5. Experiments 
This section presents more examples of output 
generated by the implemented prototype. These 
examples intend to underline different aspects 
of the proposed model. 
5.1 Sense-View Development 
This experiment provides another example of 
sense-view development. The destination sense- 
view, initially exemplified by only one 
exemplar: 
Mary went to the office. 
This is how this sense-view looks like at this 
time: 
\[Sense View\] 
Thematic Role: Destination - Marker(s): to-pp, 
IsKindOf-BUILDING(0.9)---->: 
IsKindOf-PLACE(0.81 )---->: 
IsKindOf-PHYSICAL-OBJECT(0.729)---->: 
MADE-OF(0.729)---->:MATERIAL, 
WEIGHT(0.729)---->: 
$1ZE(0.729)---->: 
IsKindOf-OBJECT(0.6561 )---->: 
IsKindOf-BUILDING-ARTIFACT(0.81 )---->: 
IsKindOf-ARTIFACT(0.729)---->: 
STATE -OF-MATTE R(0.729)---->: SeLl D, 
COLOR(0.729)---->: 
AGE(0.729)---->: 
OPERATE D-BY(O.729)---->:HUMAN, 
IsKindOf-INANIMATE(O.6561 )---->: 
Notice both IsKindOf-Building and IsKindOf- 
Place are relatively central to office and 
therefore to this sense-view. The above set 
shrinks rapidly after processing the next input: 
The student went to the stadium. 
SENSE Generated for Input Word Stadium 
\[Sense-Corcept(s)\] STADIUM 
\[Sense Viev, i 
Thematic Role: Destination - Marker(s): to-pp, 
IsKindOf-BUILDING(0.8145)---->: 
IsKindOf-PLACE(0.73305)---->: 
IsKindOf-BUILDING-ARTIFACT(0.73305)--->: 
This trend continues with the following input, 
which leads to the following word sense: 
John went to the park. 
SENSE Generated for Input Word Park 
\[Sense-Concept(s)\] PARK 
\[Sense View\] 
Thematic Role: Destination - Marker(s): to-pp, 
IsKindOf-BUILDING(0.543)---->: 
IsKindOf-PLACE(0.7074)---->: 
Here, unlike previous case, IsKindOf-Place is 
more prominent than IsKindOf-Building. This is 
because park is not a building, but nonetheless, 
its effect is not enough to completely eliminate 
IsKindOf-Building from the sense-view. The 
next input, however, strengthens IsKindOf- 
Building and weakens IsKindOf-Place, mainly 
because this property is not immediately present 
for auditorium. 
The musician went to the auditorium. 
SENSE Generated for Input Word Auditodum 
\[Sense-Concept(s)\] AUDITORIUM 
\[Sense View\] 
Thematic Role: Destination - Marker(s): to-pp, 
IsKindOf-BUILDING(0.571275)---->: 
IsKindOf-PLACE(0.53055)---->: 
5.2 Property Highlighting/Backgrounding 
This experiment provides further evidence on 
how a single concept in this model can be 
viewed from different perspectives. Notice how 
the generated output for book changes in each of 
the following cases. 
The book broke the window. 
SENSE Generated for Input Word Book 
\[Sense-Concept(s)\] BOOK 
\[Sense View\] 
Thematic Role: Instrument - No. of Exemplars: 4 - 
Marker(s): with-pp,p-subj, 
STATE-OF-MATTER(0.73305)---->:SOLID, 
MADE-OF(0.82805)---->: MATERIAL, 
The student read the book. 
This is also a case of lexical disambiguation, or 
sense selection: read could mean announce or 
study for an academic degree. 
SENSE Generated for Input Word Read 
\[Sense-Concept(s)\] READ, IN-MIND 
\[Sense View\] 
Thematic Role: Action - No. of Exemplars: 3 
Marker(s): verb, 
MODE(1 )---->:IN-MIND, 
IsKindOf-ACTIVE-COGNITIVE-EVENT(0.9)---->: 
IsKindOf-COGNITIVE-EVENT(0.81 )---->: 
IsKindOf-M ENTAL-EVENT(0.729)---->: 
IsKindOf-EVENT(0.6561)--->: 
SENSE Generated for Input Word Book 
\[Sense-Concept(s)\] BOOK 
\[Sense View\] 
Thematic Role: Theme - No. of Exemplars: 3 ° 
Marker(s): p-obj, 
CONTAINS (0.666667)---->:INFORMATION, 
MADE-OF(0.666667)---->:PAPE R,INK, 
IsKindOf- P RINTED-MEDIA(0.6)---->: 
LOCATION(0.6)---->:ACEDEMIC-BUILDING, 
IsKindOf-VISUAL-MEDIA-ARTIFACT(0.54)---->: 
IsKindOf-DOCUMENT(0.57)---->: 
PRODUCE D-BY(0.57)---->:HUMAN, 
IsKindOf-LANGUAGE-RELATE D- 
OBJECT(0.513)---->: 
REPRESENTS(0.756)--- 
>:OBJECT, EVENT,LANGUAGE, 
And finally book as an object of moving. 
The musician moved the book. 
SENSE Generated for Input Word Book 
\[Sense-Concept(s)\] BOOK 
\[Sense View\] 
Thematic Role: Object - No. of Exemplars: 4 - 
Marker(s): p-obj, 
STATE-OF-MATTER(0.731025)---->:SOLID, 
MADE-OF(0.63475)---->:PLASTIC,METAL, 
IsKindOf-ARTIFACT(0.69255)---->: 
COLOR(0.69255)---->: 
AGE(0.69255)---->: 
OPERATED-BY(0.69255)---->:HUMAN, 
5.3 Property Promotion/Demotion 
This experiment provides an example of how 
one concept appearing in a context can be 
associated with properties not necessarily 
present in its original representation. Here is the 
input context: 
Mary reads physics. 
Because of its appearance in this context, MaD' 
(in fact, its corresponding concept, Female- 
Human) will be depicted as student. In other 
words, through the assigned sense-view, 
properties specific to student (e.g., being a 
social/academic role) will be associated with 
Mary in this context. The experiment also 
provides another example of the system's 
lexical disambiguation ability (read is 
ambiguous). Here is the output word sense for 
Mary. 
• 91 
SENSE Generated for Input Word Mary 
\[Sense-Concept(s)\] HUMAN-FEMALE 
\[Sense View\] 
Thematic Role: Agent - No. of Exemplars: 1 
Marker(s): p-subj, 
IsKindOf-ACE DEMIC-ROLE(0.9)---->: 
IsKindOf-SOCIAL-ROLE(0.81)---->: 
IsKindOf-HUMAN(0.729)---->: 
GENDER(0.729)---->: 
IsKindOf-PRIMATE(0.6561)---->: 
IsKindOf-SOCIAL-OBJ ECT(0.729)-->: 
IsKindOf-OBJ ECT(0.6561 )--->: 
5.4 Multiple Word Senses 
There are cases in which context does not favor 
any of the alternative readings of a word, and 
therefore the ambiguity must be maintained in 
the output. This experiment demonstrates the 
system's ability to handle such cases. In this 
example, both readings of bank are compatible, 
to some degree, with the destination sense-view. 
John went to the bank. 
Here is the output word sense for bank, with two 
sense-concepts, both linked to the same sense- 
view. 
SENSE Generated for Input Word Bank 
\[Sense-Concept(s)\] RIVER-BANK, BANK- 
BRANCH 
\[Sense View\] 
Thematic Role: Destination - Marker(s): to-pp, 
IsKindOf-BUILDING(0.51585)---->: 
IsKindOf-PLACE(0.5967)---->: 
5.5 Instantiation of General Terms 
This last experiment is inspired by the 
experiment conducted by Anderson et al. 
(1976). These researchers found that shark was 
a better cue than fish for subjects in 
remembering a sentence like the following: 
The fish attacked the man. 
They concluded thatfish was instantiated to, and 
encoded accordingly as, shark in the subjects' 
memory. 
Here is the word sense generated for fish in 
the above context. Notice how in the output, fish 
is associated with properties specific to shark 
(aggressiveness and black color). 
SENSE Generated for Input Word Fish 
\[Sense-Concept(s)\] FISH 
\[Sense View\] 
Thematic Role: Agent - Marker(s): p-subj, 
COLOR(1 )---->:BLACK, 
AGGRESSIVE(1 )---->: 
IsKindOf-FIS H(0.9)---->: 
IsKindOf-VERTEBRATE(0.81 )---->: 
IsKindOf-ANIMAL(0.729)---->: 
GENDER(0.729)---->: 
IsKindOf-ANIMATE(0.6561)---->: 
-5.6 Unknown Words 
Finally, here is an example of how sense-views 
can be used to establish some properties about 
unknown words. Here is the input: 
Mary went to the palladium. 
The word palladium is not defined in the 
lexicon. Nevertheless, the system associates it 
with the proper sense-view. Through this sense- 
view, some initial properties for palladium can 
be inferred. 
SENSE Generated for Input Word Palladium 
\[Sense-Concept(s)\] "" unknown °*" 
\[Sense View\] 
Thematic Role: Destination - Marker(s): to-pp, 
IsKindOf-BUILDING(0.51585)---->: 
IsKindOf-PLACE(0.5967)---->: 
6. Conclusion 
In this paper, we discussed a lexicon model in 
which the role of context is not limited to sense 
selection. Selected senses are further modulated 
according to their surrounding context. We also 
described the implementation of a prototype for 
sentential contexts. 
Acknowledgments 
Support from NSERC 
acknowledged. 
is gratefully 

References 
Anderson, R., Pichert, J., Goetz, E., Schallert, 
D., Stevens, K., & Trollip, S. (1976) 
Instantiation of general terms. Journal of 
Verbal Learning and Verbal Behavior 15:667- 
679. 
Barclay, J., Bransford, J., Franks, J., McCarrell, 
N. & Nitsch, K. (1974) Comprehension and 
semantic Flexibility. Journal of Verbal 
Learning and Verbal Behavior 13:471-481. 
Barsalou, L.W.(1993). Flexibility, structure, 
and linguistic vagary in concepts. In Collins, 
A., Theories of Memory (Ed.) Lawrence 
Erlbaum Associates. 
Barsalou L.W.(1983). Ad hoc categories. 
Memory and Cognition 11(3):211-227. 
Barsalou L.W.(1982). Context-independent and 
context-dependent information in concepts. 
Memory and Cognition 10(11):82-93 
Clark, H. & Gerrig, R. (1983). Understanding 
old words with new meanings. Journal of 
Verbal Learning and Verbal Behavior 22:591- 
608. 
Corriveau, J.-P. (1995). Time-constrained 
memory. Mahwah, NJ: Lawrence Erlbaum 
Associates. 
Cruse, D. (1995). Polysemy and related 
phenomena from a cognitive linguistic 
viewpoint. In Saint-Dizier, P. & Viegas, E. 
(Eds.) Computational Lexical Semantics. 
NY,Cambridge University Press. 
Cruse, D. (1986). Lexical Semantics, Cambridge 
University Press. 
Delisle S., Copeck, T., Szpakowicz, S. & 
Barker, K. (1993). Pattern matching for case 
analysis: A computational definition of 
closeness. ICCL, 310-315. 
Dyer, M. (1983). In-depth Understanding: A 
computer model of integrated processing for 
narrative comprehension. Cambridge, MA: 
MIT Press. 
Franks, B. (1995). Sense Generation: A 
"Quasi-Classical" Approach to Concepts and 
Concept Combination. Cognitive Science 
19:441-505. 
Gluksberg, S. & Keysar, B. (1990). 
Understanding metaphorical comparisons: 
Beyond literal similarity. Psychological 
Review 97(1): 3-18. 
Greenspan, S. (1986). Semantic flexibility and 
referential specificity of concrete nouns. 
Journal of Memory and Language 25:539- 
557. 
Lakoff, G. (1987). Women, Fire and 
Dangerous Things. The University of 
Chicago Press. 
Lange, T. & Dyer, M. (1989). Frame selection 
in a connectionist model of high-level 
inferencing. Proceedings of the 11 zh 
Conference of the Cognitive Science Society. 
Mahesh K, & Nirenburg, S.(1995). A situated 
ontology for practical NLP. Proceedings of 
the Workshop on Basic Ontological Issues in 
Knowledge Sharing. IJCAI 95. Montreal, 
Canada. 
McClelland, J. & Kawamoto, A. (1986). 
Mechanisms of sentence processing. In 
McClelland J. and Rumelhaurt, D. (Eds.) 
Parallel Distributed Processing: Explorations 
in the Microstructure of Cognition. Vol. 2. 
MrT press. 
Pustejovskky, J. (1995). The generative lexicon. 
MA, MIT Press. 
Rais-Ghasem,, M. (1998) An exemplar-based 
account of contextual effects (Ph.D. Thesis) 
Ottawa,ON: Carleton University, School of 
Computer Science. 
Seidenberg, M., Tanenhaus, M., Leiman, J, & 
Bienkowski, M. (1982). Automatic access of 
the meanings of ambiguous words in context. 
Cognitive Psychology 14:489-537 
Simpson, G. & Burgess, C. (1985). Activation 
and selection processes in the recognition of 
ambiguous words. Journal of Experimental 
Psychology: Human Perception and 
Performance 11(1):28-39. 
Waltz, D. & Pollack, J. (1985). Massively 
parallel parsing: A strongly interactive model 
of natural language interpretation. Cognitive 
Science 9:51-74. 
Wilensky, R. (1978). Understanding goal-based 
stories (Research Report), Dept. of Computer 
Science, New Haven, CT:Yale University. 
Witney, P., McKay, T., & Kellas, G. (1985). 
Semantic activation of noun concepts in 
context. Journal of Experimental Psychology: 
Learning, Memory, and Cognition 11 : 126- 
135. 
