met*: A Method for Discriminating 
Metonymy and Metaphor by Computer 
Dan Fass* 
Simon Fraser University 
The met* method distinguishes selected examples of metonymy from metaphor and from liter- 
alness and anomaly in short English sentences. In the met* method, literalness is distinguished 
because it satisfies contextual constraints that the nonliteral others all violate. Metonymy is 
discriminated from metaphor and anomaly in a way that \[1\] supports Lakoff and Johnson's 
(1980) view that in metonymy one entity stands for another whereas in metaphor one en- 
tity is viewed as another, \[2\] permits chains of metonymies (Reddy 1979), and \[3\] allows 
metonymies to co-occur with instances of either literalness, metaphor, or anomaly. Metaphor is 
distinguished from anomaly because the former contains a relevant analogy, unlike the latter. 
The met* method is part of Collative Semantics, a semantics for natural language processing, 
and has been implemented in a computer program called meta5. Some examples of meta5's 
analysis of metaphor and metonymy are given. The met* method is compared with approaches 
from artificial intelligence, linguistics, philosophy, and psychology. 
1. Introduction 
Metaphor and metonymy are kinds of figurative language or tropes. Other tropes 
include simile, irony, understatement (litotes), and overstatement (hyperbole). 
Example 1 
"My car drinks gasoline" (Wilks 1978, p. 199). 
Example 2 
"The ham sandwich is waiting for his check" (Lakoff and Johnson 1980, p. 35). 
Sentences (1) and (2) contain examples of metaphor and metonymy respectively. Nei- 
ther sentence is literally true: cars do not literally drink nor do ham sandwiches literally 
wait. Notice, though, that the two sentences are interpreted differently. "My car" in 
(1) is commonly understood as resembling an animate drinker while in (2) "the ham 
sandwich" is generally interpreted as referring to the person who ordered the ham 
sandwich. 
Most of the considerable literature on metaphor and the smaller one on metonymy 
(see Van Noppen, De Knop and Jongen 1985; Shibles 1971) is from philosophy, lin- 
guistics, and psychology. On the whole, the two phenomena remain vague, poorly 
defined notions in that literature. In artificial intelligence (AI), detailed treatments of 
either metaphor or metonymy are relatively scarce. Moreover, most of those treatments 
are paper implementations that have not been coded up and run on a computer. 
* Centre for Systems Science, Simon Fraser University, Burnaby, British Columbia, Canada V5A 1S6 
(~) 1991 Association for Computational Linguistics 
Computational Linguistics Volume 17, Number 1 
The met* (pronounced "met star'0 method provides a means for recognizing se- 
lected examples of metonymy and metaphor, and also anomaly and literalness, in 
short English sentences. 1 The method is part of Collative Semantics (hereafter CS), 
which is a semantics for natural language processing. CS, and hence the met* method, 
has been implemented in a program called meta5 (so called because it does more than 
metaphor). The meta5 program is, as far as I know, the first system to recognize exam- 
ples of metaphor and metonymy. TO my knowledge, there is only one other working 
program that might be said to recognize instances of metaphor (Martin 1988; 1990) 
and two systems that appear to recognize cases of metonymy, TEAM (Grosz et al. 
1987) and TACITUS (Hobbs and Martin 1987). 
The rest of the paper is organized as follows. Section 2 surveys general issues 
and approaches in metaphor and metonymy, notably the distinctive characteristics 
of metaphor and metonymy, the relationship between metaphor and metonymy, and 
the relationship between literalness and nonliteralness. Section 3 presents the met* 
method, concentrating on the basic topology of the met* method algorithm. Section 4 
shows details of representations and processes used in CS. Section 5 gives examples 
of the meta5 program analyzing simple metaphors and metonymies. Descriptions get 
progressively more detailed from Section 2 through to Section 5. Sections 6 and 7 de- 
scribe some extensions to metaphor interpretation in CS and compare the met* method 
against other approaches to metaphor and metonymy, especially computational ones. 
A glossary of key terms is provided at the very end of the paper. 
2. Survey of Metonymy and Metaphor Research 
Metonymy and metaphor are so poorly understood that widely divergent views exist 
about them and their relationship to each other. This section reviews research on 
metaphor (2.1), metonymy (2.2), the relationship between them (2.3), and the more 
general relationship between literalness and nonliteralness (2.4). 
2.1 Metaphor 
Four views of metaphor are critically discussed: the comparison view, the interactive 
view, the selection restriction violation view, and the conventional metaphor view. 
Computational examples of each kind are included by Gentner, Indurkhya, Hobbs, 
Wilks, and Martin. Space does not permit discussion of other AI work on metaphor 
by, e.g., Russell (1976) and Weiner (1984; 1985). 
2.1.1 The Comparison View. According to the comparison view 
a metaphor is a comparison in which one term (the tenor or subject of the 
comparison) is asserted to bear a partial resemblance (the ground of the 
comparison) to something else (the vehicle), the resemblance being insufficient to 
sustain a literal comparison. As with any comparison, there is always some 
residual dissimilarity (the tension) between the terms involved in the comparison, 
but comparison theorists tend not to emphasize this dissimilarity (Tourangeau 
and Sternberg 1982, p. 205, their italics). 
What is crucial in the comparison approach, then, is finding the correct ground 
in a metaphor. According to Tourangeau and Sternberg, Aristotle proposed the first 
1 The met* method takes its name from a remark made by Yorick Wilks. He used met* to refer 
collectively to metonymy and metaphor: "*" is a match-anything symbol in the Unix operating system; 
hence, the token "met*" matches the two tokens "metonymy" and "metaphor." 
50 
Fass Discriminating Metonymy 
comparison theory and suggested several principles for finding the ground of a metaphor. 
Tourangeau and Sternberg reduce these principles to two basic ones: finding a 
category to which the tenor and vehicle belong and constructing an analogy involving 
them. 
Gentner's (1983) Structure-Mapping Theory, which has been implemented in the 
Structure-Mapping Engine (Falkenhainer, Forbus and Gentner 1989), closely resembles 
a comparison view of metaphor. The theory addresses literal similarity, analogy, ab- 
straction, and anomaly, which Gentner refers to as four "kinds of comparison." An 
algorithm compares the semantic information from two concepts represented as sets 
of properties. Properties are either "attributes," one-place predicates like LARGE(x), 
or "relations," two-place predicates such as COLLIDE(x,y). The four kinds of compar- 
ison are distinguished by the relative proportions of attributes and relations that are 
matched, and the forms of mappings established between them. Mappings between 
relations are sought before those between attributes. Pairs of relations are compared us- 
ing the "systematicity principle" that regular structural correspondences should exist 
between terms occupying the same positions in those relations. Mappings are purely 
structural and independent of the content of the relations (i.e., the predicates). 
Tourangeau and Sternberg (1982) list some problems with the comparison view, 
including the following: 
(a) that everything has some feature or category that it shares with everything 
else, but we cannot combine just any two things in metaphor; (b) that the most 
obvious shared features are often irrelevant to a reading of the metaphor; (c) that 
even when the feature is relevant, it is often shared only metaphorically; ... and 
(e) that metaphors are novel and surprising is hard to reconcile with the idea 
that they rely completely on extant similarities (ibid., pp. 226-227). 
Johnson (1980) also notes problem (a) with comparison theories, pointing out that 
as a result they cannot account for the semantic tension between the two terms of a 
metaphor: 
the comparison theory ... tries to circumvent the experienced semantic strain by 
interpreting metaphor as nothing but a way of comparing two things to see in 
what respects they are alike. And since any two things are similar in some 
respects, this kind of theory can never explain what is interesting and important 
about metaphor (ibid., p. 52). 
2.1.2 The Interaction View. The interaction view focuses more upon the surprise and 
novelty that metaphors create. According to Tourangeau and Sternberg (1982, p. 212), 
proponents of the interaction view include Black (1962), Hesse (1966), Miles (1967), 
Richards (1936), and Wheelwright (1962). 
Interaction theorists argue that the vehicle of a metaphor is a template for seeing 
the tenor in a new way. This reorganization of the tenor is necessary, because the 
characteristics or features of the vehicle cannot be applied directly to the tenor; 
the features they 'share' are often only shared metaphorically. As Black (1962) 
observes, the ground of a metaphor may itself be nonliteral. 'Men are wolves,' in 
Black's example, in part because both are predators; but they are predators in 
sharply different senses that may only strike us as similar when we interpret the 
metaphor. In Black's reading of this metaphor, we see competition in social 
relations as corresponding to predacity in beasts (Tourangeau and Sternberg 
1982, pp. 212-213). 
51 
Computational Linguistics Volume 17, Number 1 
A problem with the interaction view is that theorists have not provided much 
detail about the processes involved, though Black (1962) does make some suggestions. 
According to Black, tenor and vehicle.., each have a 'system of commonplaces' 
associated with them. These commonplaces are stereotypes, not necessarily 
definitional, not even necessarily true, just widely agreed upon. In interpreting 
'man is a wolf,' we 'evoke the wolf-system of related commonplaces' and are led 
by them 'to construct a corresponding system of implications about the principal 
subject (Man)' (Black, 1962, p. 41). In Black's view, then, interpretation involves 
not so much comparing tenor and vehicle for existing similarities, as construing 
them in a new way so as to create similarity between them (Tourangeau and 
Sternberg 1982, p. 213). 
One might distinguish, then, two main differences between the interaction and 
comparison views. First, similarities are "created" in the interaction view (accounting 
for the novelty and surprise in a metaphor) whereas only pre-existing similarities 
are found in the comparison view. Second, a whole system of similarities are evoked 
between tenor and vehicle in the interactions view, whereas the comparisons view is 
based upon finding a single similarity. 
One version of the interaction view is the domains-interaction view, set forth by 
Tourangeau and Sternberg (1982), who take the view that 
features 'shared' by tenor and vehicle are often at best only analogous features, 
each limited in its application to one domain or another. Of course, some 
features or dimensions are quite general, applying across the board to a number 
of domains (p. 218). 
Among comparison and interaction theorists, much attention had been paid to 
selecting the comparisons or interactions in a metaphor. The importance of analogy 
or correspondence in metaphor has been stressed by Gentner (1983), Ortony (1979), 
Tourangeau and Sternberg (1982), and Wilks (1978), among others. Various mecha- 
nisms have been advanced for highlighting certain comparisons or interactions, in- 
cluding relevance (e.g., Hobbs 1983b; Tversky 1977) and salience (Ortony et al. 1985). 
Among computational approaches, Indurkhya's (1988) Constrained Semantic Trans- 
ference theory of metaphor can be viewed as a formalization of Black's interaction 
theory (ibid., p. 129). Source and target domains are viewed as "systems of relation- 
ships." In metaphorical interpretation, an "implicative complex" of the source domain 
is imposed on the target domain, thereby shaping the features of the target domain, 
which in turn produces changes in the features of the source domain, hence the "in- 
teraction." It is assumed that a structural analogy underlies every metaphor (ibid., 
p. 129). 
A metaphor is identified with the formal notion of a T-MAP which is a pair / F,S / 
where F is a function that maps vocabulary of the source domain onto vocabulary 
of the target domain and S is a set of sentences from the source domain which are 
expected to transfer to the target domain. A metaphor is "coherent" if the transferred 
sentences S are logically consistent with the axioms of the target domain, and "strongly 
coherent" if they already lie in the deductive closure of those axioms (cf. Stallard 1987, 
p. 181). S is thus the "implicative complex" of the source domain imposed on the target 
domain. Every metaphorical interpretation of a given set of sentences is associated with 
a T-MAP. There may be several possible T-MAPs for a set of sentences. 
I would argue that Hobbs (1983a; 1983b) has also taken an interaction view of 
metaphor. Hobbs' goal has been to develop a unified process of discourse interpre- 
tation based on the drawing of appropriate inferences from a large knowledge base, 
52 
Fass Discriminating Metonymy 
which Hobbs sometimes calls "selective inferencing" (e.g., Hobbs 1980). Selective in- 
ferencing is concerned with drawing or refraining from drawing certain inferences in 
a controlled fashion (cf. Hobbs 1983a). He argues that many problems have the same 
or almost the same inferencing solutions. These solutions are found via four separate 
semantic operations that all draw inferences from text (e.g., Hobbs 1977). 
2.1.3 The Selection Restrictions Violations View. The selection restriction violation 
view has also been called "the semantic deviance view" (Johnson 1980, p. 50) and "the 
anomaly view" (Tourangeau and Sternberg 1982, p. 211). Johnson (1980) describes this 
view as a common one among linguists; Tourangeau and Sternberg (1982) list the 
following people as holders of this view: Beardsley (1962), Bickerton (1969), Campbell 
(1975), Guenther (1975), Percy (1954), Van Dijk (1975), and Wheelwright (1962). To this 
list one might add Levin (1977). Johnson (1980, p. 50) describes this view as where: 
metaphor constitutes a violation of selection restriction rules within a given 
context, where the fact of this violation is supposed to explain the semantic 
tension one experiences in comprehending any live metaphor. 
The theory of metaphor in Preference Semantics (Wilks 1975; 1978) consists of a 
selection restrictions view and a comparison view. In the theory, information about 
word senses is contained in knowledge structures called "semantic formulas." An 
algorithm matches pairs of semantic formulas, seeking satisfied or violated preferences 
between them. A satisfied preference indicates a literal semantic relation; a violated 
preference indicates either a metaphorical or anomalous one. This part of the theory 
is implemented in a machine translation system (Wilks 1973). 
To distinguish metaphor from anomaly, a different knowledge structure and a 
second algorithm are used. The algorithm, called projection, operates on a knowl- 
edge structure, called a pseudo-text, that contains lists of templates (a further kind of 
knowledge structure) linked by case ties. A brief example of projection is given for (1). 
Example 3 
"My car drinks gasoline." 
Projection operates only on preference violations. The best representation of (1) con- 
tains a preference violation, so projection is used. The algorithm compares the template 
representation for the sentence 
\[my+car drink gasoline\] 
against templates from the pseudo-text of 'car' seeking "the closest match," and selects 
\[ICengine (USE)#1iquid\]. (USE) is projected onto drink in the sentence representation 
which becomes 
\[my+car use gasoline\] 
Example 3 
"The rock is becoming brittle with age" (Reddy 1969, p. 242). 
53 
Computational Linguistics Volume 17, Number 1 
Example 4 
"Idi Amin is an animal" (Johnson 1980, p. 51). 
Example 5 
"People are not cattle" (Hobbs 1983b, p. 134). 
Example 6 
"No man is an Island" (John Donne, Meditations XVII). 
The main problem with the selection restrictions view is that perfectly well-formed 
sentences exist that have a metaphorical interpretation and yet contain no selection 
restriction violations (Johnson 1980; Ortony 1980; Reddy 1969); for example, in (3), 
there is a literal interpretation when uttered about a stone and a metaphorical one 
when said about a decrepit professor emeritus. Sentences (4), (5) and (6) also have 
twin interpretations. 
The existence of such sentences suggests that a condition that occasionally holds 
(i.e., a selection restriction violation) has been elevated into a necessary condition 
of metaphor (Johnson 1980). Moreover, viewing metaphor only in terms of selection 
restriction violations ignores the influence of context: 
We seem to interpret an utterance metaphorically when to do so makes sense of 
more aspects of the total context than if the sentence is read literally. Consider 
the simple case of the sentence All men are animals as uttered by Professor X to an 
introductory biology class and as uttered later by one of his female students to 
her roommate upon returning from a date. In the latter instance the roommate 
understands the utterance as metaphorical (ibid., p. 51). 
In a similar way, Ortony (1980) suggests that metaphor should be thought of as 
contextually anomalous. This means that a literal interpretation of the expression, 
be it a word, phrase, sentence, or an even larger unit of text, fails to fit the 
context (p. 73, his italics), 
so whether or not a sentence is a metaphor depends upon the context in which it 
is used: 
if something is a metaphor then it will be contextually anomalous if interpreted 
literally .... Insofar as the violation of selection restrictions can be interpreted in 
terms of semantic incompatibilities at the lexical level, such violations may 
sometimes be the basis of the contextual anomaly (ibid., p. 74). 
2.1.4 The Conventional Metaphor View. Lakoff and Johnson (1980) have popular- 
ized the idea of conventional metaphors, also known as conceptual metaphors. They 
distinguish three main kinds: orientational, ontological, and structural. Orientational 
metaphors are mainly to do with kinds of spatial orientation like up-down, in-out, 
and deep-shallow. Example metaphors include MORE IS UP and HAPPY IS UP. They 
arise from human experience of spatial orientation and thus develop from the sort of 
bodies we have and the way they function in our physical environment. 
Ontological metaphors arise from our basic human experiences with substances 
and physical objects (especially our own bodies). Some examples are TIME IS A SUB- 
STANCE, THE MIND IS AN ENTITY, and THE VISUAL FIELD IS A CONTAINER. 
54 
Fass Discriminating Metonymy 
Structural metaphors are elaborated orientational and ontological metaphors (cf. 
Lakoff and Johnson 1980) in which concepts that correspond to natural kinds of ex- 
perience, e.g., PHYSICAL ORIENTATIONS, SUBSTANCES, WAR, JOURNEYS, and 
BUILDINGS, are used to define other concepts, also natural kinds of experience, 
e.g., LOVE, TIME, IDEAS, UNDERSTANDING, and ARGUMENTS. Some examples 
of structural metaphors are ARGUMENT IS WAR and TIME IS MONEY. 
The ARGUMENT IS WAR metaphor forms a systematic way of talking about the 
battling aspects of arguing .... Because the metaphorical concept is systematic, 
the language we use to talk about the concept is systematic (ibid., p. 5). 
What Lakoff and Johnson fail to discuss is how metaphors in general, let alone 
individual metaphorical concepts, are recognized. Martin's (1988; 1990) work has ad- 
dressed this issue. He has pursued a conventional metaphor view using KODIAK 
(Wilensky 1984), a variant of Brachman's KLONE knowledge representation language. 
Within KODIAK, metaphorical relationships are represented using a primitive link 
type called a "VIEW." A VIEW "is used to assert that.., one concept may in cer- 
tain circumstances be considered as another " (Martin 1990, p. 59). In Martin's work, 
"metaphor-maps," a kind of VIEW (ibid., p. 64), are used to represent conventional 
metaphors and the conceptual information they contain. 
2.2 Metonymy 
Metonymy involves "using one entity to refer to another that is related to it" (Lakoff 
and Johnson 1980, p. 35). 
Example 2 
"The ham sandwich is waiting for his check." 
For example, in (2) the metonymy is that the concept for ham sandwich is related 
to an aspect of another concept, for "the person who ordered the ham sandwich." 
Several attempts have been made to organize instances of metonymy into cat- 
egories (e.g., Lakoff and Johnson 1980; Stern 1931; Yamanashi 1987) or "metonymic 
concepts," as Lakoff and Johnson call them. A common metonymic concept is PART 
FOR WHOLE, otherwise known as synechdoche. 
Example 7 
"Dave drank the glasses" (= the liquid in the glasses). 
Example 8 
"The kettle is boiling" (= the liquid in the kettle) (Waldron 1967, p. 186; Yamanashi 
1987, p. 78). 
CONTAINER FOR CONTENTS, another metonymic concept, occurs in (7) between 
'drink' and the sense of 'glasses' meaning "containers," and also in (8). In (7), 'drink' 
has an object preference for a potable liquid, but there is a preference violation because 
glasses are not potable liquids. It is not glasses that are drunk, but the potable liquids 
in them. There is a relationship here between a CONTAINER (a glass) and its typical 
CONTENTS (a liquid): this relationship is the metonymic concept CONTAINER FOR 
55 
Computational Linguistics Volume 17, Number 1 
CONTENTS. Below are examples of two further metonymic concepts (from Lakoff 
and Johnson 1980, p. 38, italics in original). 
PRODUCER FOR PRODUCT 
"I'll have a L6wenbrau. " 
"He bought a Ford." 
"He's got a Picasso in his den." 
"I hate to read Heidegger." 
OBJECT USED FOR USER 
"The sax has the flu today." 
"The BLT is a lousy tipper. "*.2 
"The buses are on strike." 
Example 9 
"You'll find better ideas than that in the library" (Reddy 1979, p. 309). 
Reddy (1979) has observed that metonymies can occur in chains. He suggests that 
(9) contains a chain of PART FOR WHOLE metonymies between 'ideas' and 'library': 
the ideas are expressed in words, words are printed on pages, pages are in books, and 
books are found in a library. 
Example 10 
"I found an old car on the road. The steering wheel was broken" (Yamanashi 1987, p. 79). 
Example 11 
"We had a party in a mysterious room. The walls were painted in psychedelic color" 
(ibid.). 
Example 12 
A: "I bought an interesting book." B: "Who is the author?" (ibid.). 
Example 13 
"He happened to die of some disease, though I don't know what the cause was" (ibid.). 
Yamanashi (1987) points out that basic metonymic relationships like part-whole 
and cause-result often also link sentences. According to him, the links in (10) and (11) 
are PART-WHOLE relations, the one in (12) is PRODUCT-PRODUCER, and the one 
in (13) is a CAUSE-RESULT relation. 
There has been some computational work on metonymy (Weischedel and Sond- 
heimer 1983; Grosz et al. 1987; Hobbs and Martin 1987; Stallard 1987; Wilensky 1987). 
The TEAM project (Grosz et al. 1987) handles metonymy, though metonymy is not 
mentioned by name but referred to instead as "coercion," which "occurs whenever 
some property of an object is used to refer indirectly to the object" (ibid., p. 213). 
Coercion is handled by "coercion-relations;" for example, a coercion relation could be 
used to understand that 'Fords' means "cars whose CAR-MANUFACTURER is Ford" 
(in Lakoff and Johnson's terms, this is an example of a PRODUCER FOR PRODUCT 
metonymic concept). 
2 A BLT is a bacon, lettuce, and tomato sandwich. 
56 
Fass Discriminating Metonymy 
Grosz et al. (1987) note a similarity between coercion (i.e., metonymy) and modifi- 
cation in noun-noun compounds, and use "modification relations" to decide whether, 
e.g., "U.S. ships" means "ships of U.S. registry" or "ships whose destination is the U.S." 
Hobbs and Martin (1987) and Stallard (1987) also discuss the relationship between 
metonymy and nominal compounds. Hobbs and Martin treat the two phenomena 
as twin problems of reference resolution in their TACITUS system. They argue that 
resolving reference requires finding a knowledge base entity for an entity mentioned in 
discourse (i.e., what that entity refers to), and suggest that the resolution of metonymy 
and nominal compounds both require discovering an implicit relation between two 
entities referred to in discourse. The example of metonymy they show is "after the 
alarm," which really means after the sounding of the alarm. 
Hobbs and Martin seem to assume a selection restrictions approach to metonymy 
because metonymy is sought after a selection restrictions violation (ibid., p. 521). In 
their approach, solving metonymy involves finding: \[1\] the referents for 'after' and 
'alarm' in the domain model, which are after(eo, a) and alarm(a); \[2\] an implicit entity 
z to which 'after' really refers, which is afier(eo, z); and \[3\] the implicit relation between 
the implicit entity z and the referent of 'alarm,' q(z, a). 
Like Hobbs and Martin (1987), Stallard (1987) translates language into logical form. 
Stallard argues that with nominal compounds and metonymies "the problem is deter- 
mining the binary relation which has been 'elided' from the utterance" (ibid., p. 180) 
and suggests shifting the argument place of a predicate "by interposing an arbitrary, 
sortally compatible relation between an argument place of the predicate and the ac- 
tual argument" (ibid., p. 182). Stallard notes that "in any usage of the metonomy (sic) 
operation there is a choice about which of two clashing elements to extend" (ibid.). 
Stallard's work has not yet been implemented (ibid., p. 184). 
Stallard (1987) also briefly discusses anaphora resolution. Brown (1990) is be- 
ginning research on metonymy and reference resolution, particularly pronouns. This 
should prove a promising line of investigation because metonymy and anaphora share 
the function of allowing one entity to refer to another entity. 
Example 2 
"The ham sandwich is waiting for his check" (= the male person who ordered the 
ham sandwich). 
Example 14 
"He is waiting for his check" (= the male person). 
This similarity of function can be seen in comparing (2), which is metonymic, with 
(14), which is anaphoric. 
2.3 Relationship between Metonymy and Metaphor 
Both metonymy and metaphor have been identified as central to the development 
of new word senses, and hence to language change (see, e.g., Stern 1931; Waldron 
1967). Some of the best examples of the differences between the two phenomena come 
from data used in studies of metonymic and metaphorical effects on language change. 
Nevertheless, there are widely differing views on which phenomenon is the more 
important. Some argue that metaphor is a kind of metonymy, and others propose that 
metonymy is a kind of metaphor, while still others suggest that they are quite different 
(see Fass 1988c). 
57 
Computational Linguistics Volume 17, Number 1 
Among the third group, two differences between metonymy and metaphor are 
commonly mentioned. One difference is that metonymy is founded on contiguity 
whereas metaphor is based on similarity (cf. Jakobsen and Halle 1956; Ullmann 1962). 
Contiguity and similarity are two kinds of association. Contiguity refers to a state 
of being connected or touching whereas similarity refers to a state of being alike in 
essentials or having characteristics in common (Mish 1986). 
A second difference, advanced by Lakoff and Johnson (1980) for example, is that 
metaphor is "principally a way of conceiving of one thing in terms of another, and 
its primary function is understanding" (ibid., pp. 36-37) whereas metonymy "has 
primarily a referential function, that is, it allows us to use one entity to stand for 
another" (ibid., their italics), though it has a role in understanding because it focuses 
on certain aspects of what is being referred to. 
There is little computational work about the relationship between metonymy and 
metaphor. Stallard (1987) distinguishes separate roles for metonymy and metaphor in 
word sense extension. According to him, metonymy shifts the argument place of a 
predicate, whereas metaphor shifts the whole predicate. Hobbs (1983a; 1983b) writes 
about metaphor, and he and Martin (1987) develop a theory of "local pragmatics" that 
includes metonymy, but Hobbs does not seem to have written about the relationship 
between metaphor and metonymy. 
In knowledge representation, metonymic and metaphorical relations are both rep- 
resented in the knowledge representation language CycL (Lenat and Guha 1990). 
2.4 Literalness and Nonliteralness 
Much of the preceding material assumes what Gibbs (1984) calls the "literal meanings 
hypothesis," which is that 
sentences have well defined literal meanings and that computation of the literal 
meaning is a necessary step on the path to understanding speakers' utterances 
(ibid., p. 275). 
There are a number of points here, which Gibbs expands upon in his paper. One 
point concerns the traditional notion of literal meaning, that all sentences have literal 
meanings that are entirely determined by the meanings of their component words, 
and that the literal meaning of a sentence is its meaning independent of context. 
A second point concerns the traditional view of metaphor interpretation, though 
Gibbs' criticism applies to metonymy interpretation also. Using Searle's (1979) views 
on metaphor as an example, he characterizes the typical model for detecting nonliteral 
meaning as a three-stage process: \[1\] compute the literal meaning of a sentence, \[2\] 
decide if the literal meaning is defective, and if so, \[3\] seek an alternative meaning, 
i.e., a metaphorical one (though, presumably, a metonymic interpretation might also 
be sought at this stage). Gibbs (1984, p. 275) concludes that the distinction between 
literal and metaphoric meanings has "little psychological validity." 
Among AI researchers, Martin (1990) shares many of Gibbs's views in criticizing 
the "literal meaning first approach" (ibid., p. 24). Martin suggests a two-stage process 
for interpreting sentences containing metaphors: \[1\] parse the sentence to produce 
a syntactic parse tree plus primal (semantic) representation, and \[2\] apply inference 
processes of "concretion" and "metaphoric viewing" to produce the most detailed 
semantic representation possible. 
The primal representation represents a level of semantic interpretation that is 
explicitly in need of further processing. Although it is obviously related to what 
58 
Fass Discriminating Metonymy 
has traditionally been called a literal meaning, it should not be thought of as a 
meaning at all. The primal representation should be simply considered as an 
intermediate stage in the interpretation process where only syntactic and lexical 
information has been utilized (ibid., p. 90, his italics). 
However, Martin believes that at least some sentence meaning is independent of 
context because the primal representation contains part of the primal content of an 
utterance and 
\[t\]he Primal Content represents the meaning of an utterance that is derivable 
from knowledge of the conventions of a language, independent of context (ibid.). 
2.5 Review Summary 
The metaphor literature contains many differing views, including the comparison, 
interaction, selection restrictions, and conventional metaphors views. AI research on 
metaphor includes all of these views. Of the AI research, only Martin's work has 
been implemented to my knowledge. Among the points raised are that metaphorical 
sentences exist that do not contain selection restriction violations and that metaphor 
requires interpretation in context. The much smaller metonymy literature stresses the 
selection restrictions view too. The TEAM and TACITUS systems both seem to process 
metonymics. 
The two main differences commonly noted between metonymy and metaphor 
are in their function (referential for metonymy and understanding with metaphor) 
and the kind of relationship established (contiguity in metonymy versus similarity 
in metaphor). No one to my knowledge has a working system that discriminates 
examples of metaphor and metonymy. 
3. met* Method 
In this section, the basic met* algorithm is outlined. The met* method is based on the 
selection restriction, also known as the preference. Metonymy, metaphor, literalness, 
and anomaly are recognized by evaluating preferences, which produces four kinds of 
basic "preference-based" relationship or semantic relation: literal, metonymic, metaphor- 
ical, and anomalous. Within the method, the main difference between metonymy and 
metaphor is that a metonymy is viewed as consisting of one or more semantic re- 
lationships like CONTAINER FOR CONTENTS and PART FOR WHOLE, whereas a 
metaphor is viewed as containing a relevant analogy. 
I agree with Ortony's remark that metaphor be viewed as contextual anomaly, but 
would suggest two modifications. First, not just metaphor but all of the preference- 
based relations should be understood in terms of the presence or absence of contextual 
constraint violation. Second, I prefer the term contextual constraint violation because 
\[1\] one of the phenomena detected by contextual violation is anomaly and \[2\] the 
selection restriction/preference (on which the met* method is based) is a kind of 
lexical contextual constraint. The section starts with an explanation of some of the 
linguistic background behind the met* method. 
3.1 Linguistic Background 
I have argued elsewhere (Fass 1989a) that understanding natural language (or seman- 
tic interpretation) be viewed as the integration of constraints from language and from 
context. Some language constraints are syntactic, while others are semantic. Some 
59 
Computational Linguistics Volume 17, Number 1 
language constraints are lexical constraints; that is, constraints possessed by lexical 
items (words and fixed phrases). Lexical syntactic constraints include those on word 
order, number, and tense. This section describes three lexical semantic constraints: pref- 
erences, assertions, and a lexical notion of relevance. 
Preferences (Wilks 1973), selection restrictions (Katz 1964), and expectations 
(Schank 1975) are the same (see Fass 1989c; Fass and Wilks 1983; Wilks and Fass in 
press): all are restrictions possessed by senses of lexical items of certain parts of speech 
about the semantic classes of lexical items with which they co-occur. Thus an adjective 
sense has a preference for the semantic class of nouns with which it co-occurs and a 
verb sense has preferences for the semantic classes of nouns that fill its case roles. For 
example, the main sense of the verb 'drink' prefers an animal to fill its agent case role, 
i.e., it is animals that drink. 
The assertion of semantic information was noted by Lees (1960) in the formation 
of noun phrases and later developed by Katz (1964) as the process of "attribution." 
Assertions contain information that is possessed by senses of lexical items of certain 
parts of speech and that is imposed onto senses of lexical items of other parts of speech, 
e.g., the adjective 'female' contains information that any noun to which it applies is 
of the female sex. 
Lexical syntactic and semantic constraints are enforced at certain places in sen- 
tences which ! call dependencies. Within a dependency, the lexical item whose con- 
straints are enforced is called the source and the other lexical item is called the target 
(after Martin 1985). Syntactic dependencies consist of pairs of lexical items of certain 
parts of speech in which the source, an item from one part of speech, applies one or 
more syntactic constraints to the target, another lexical item. Examples of source-target 
pairs include a determiner and a noun, an adjective and a noun, a noun and a verb, 
and an adverb and a verb. 
Example 15 
"The ship ploughed the waves." 
Semantic dependencies occur in the same places as syntactic dependencies. The 
(metaphorical) sentence (15) contains four semantic dependencies: between the deter- 
miner 'the' and the noun 'ship,' between 'ship' and the verb stern 'plough,' between 
'the' and the noun 'waves,' and between 'waves' and 'plough.' In each semantic de- 
pendency, one lexical item acts as the source and applies constraints upon the other 
lexical item, which acts as the target. In (15), 'the" and 'plough' both apply constraints 
upon 'ship,' and 'the' and 'plough' apply constraints on 'waves.' Semantic dependen- 
cies exist between not just pairs of lexical items but also pairs of senses of lexical items. 
For example, the metaphorical reading of (15) is because 'waves' is understood as 
being the sense meaning "movement of water," not for example the sense meaning 
"movement of the hand." 
Semantic relations result from evaluating lexical semantic constraints in sentences. 
Every semantic relation has a source (a lexical item whose semantic constraints are 
applied) and a target (a lexical item which receives those constraints). Other terms 
used to refer to the source and target in a semantic relation include: vehicle and tenor 
(Richards 1936), subsidiary subject and principal subject (Black 1962), figurative term 
and literal term (Perrine 1971), referent and subject (Tversky 1977), secondary subject 
and primary subject (Black 1979), source and destination (Winston 1980), old domain 
and new domain (Hobbs 1983a), and base and target (Gentner 1983). 
In CS, seven kinds of semantic relation are distinguished: literal, metonymic, 
metaphorical, anomalous, redundant, inconsistent, and novel relations (this list may 
60 
Fass Discriminating Metonymy 
not be exhaustive -- there could be others). Combinations of these seven semantic 
relations are the basis of (at minimum) literalness, metonymy, metaphor, anomal)~ re- 
dundancy, contradiction, contrariness, and novelty. Semantic relations belong to two 
classes, the preference-based and assertion-based classes of relations, depending on the 
kind of lexical semantic constraint enforced. The preference-based class of semantic 
relations, which are the focus of this paper, contains literal, metonymic, metaphorical, 
and anomalous semantic relations. The assertion-based class of relations are described 
in greater length in Fass (1989a). 
Figure 1 shows the met* method laid out as a flow chart and illustrates how the 
preference-based class of semantic relations is discriminated. A satisfied preference 
(diamond 1) distinguishes literal relations from the remaining three relations, which 
are all nonliteral. 
Example 16 
"The man drank beer." 
There is a literal relation between 'man' and 'drink' in (16) because 'drink' prefers an 
animal as its agent and a man is a type of animal so the preference is satisfied. 
Example 7 
"Dave drank the glasses" (= potable liquid in the glasses --* CONTAINER FOR CON- 
TENTS). 
Example 17 
"Denise drank the bottle" (= potable liquid from the bottle -~ CONTAINER FOR CON- 
TENTS). 
\ 
Figure 1 
The met* method 
61 
Computational Linguistics Volume 17, Number 1 
Example 18 
"Anne reads Steinbeck" (= writings of Steinbeck --, ARTIST FOR ART FORM). 
Example 19 
"Ted played Bach" (= music of Bach ~ ARTIST FOR ART FORM). 
Metonymy is viewed as a kind of domain-dependent inference. The process of 
finding metonymies is called metonymic inferencing. The metonymic concepts presently 
used are adapted from the metonymic concepts of Lakoff and Johnson (1980). Two of 
the metonymic concepts used are CONTAINER FOR CONTENTS and ARTIST FOR 
ART FORM. In (19), for example, Ted does not literally play the composer Bach -- he 
plays music composed by him. 
As Figure 1 shows, a metonymy is recognized in the met* method if a metonymic 
inference (diamond 2) is found. Conversely, if no successful inference is found then no 
metonymy is discovered and a metaphorical or anomalous semantic relation is then 
sought. A successful inference establishes a relationship between the original source 
or the target ("one entity'9 and a term ("another that is related to it'3 that refers to 
one of them. 
Like Stallard (1987), who noted that "in any usage of the metonomy (sic) operation 
there is a choice about which of two clashing elements to extend" (ibid., p. 182), the 
met* method allows for metonymies that develop in different "directions." A successful 
inference is sometimes directed "forward" from the preference or "backward" from 
the target, depending on the metonyrnic concept (more on this shortly). It is this 
direction of inferencing that determines whether the source or target is substituted 
in a successful metonymy. The substitute source or target is used to discover another 
semantic relation that can be literal, metonymic again, metaphorical, or anomalous. 
In Figure 1, the presence of a relevant analogy (diamond 3) discriminates metaphor- 
ical relations from anomalous ones. No one else (to my knowledge) has emphasized 
the role of relevance in the discovery of an analogy central to a metaphor though, as 
noted in Section 2.2, the importance of relevance in recognizing metaphors and the 
centrality of some analogy have both been discussed. 
Example 20 
"The car drank gasoline" (adapted from Wilks 1978). 
The form of relevance used is a lexical notion -- i.e., the third kind of lexical semantic 
constraint -- that what is relevant in a sentence is given by the sense of the main 
sentence verb being currently analyzed. Thus, it is claimed that the semantic relation 
between 'car' and 'drink' in (20) is metaphorical because there is a preference violation 
and an underlying relevant analogy between 'car' and 'animal,' the preferred agent of 
'drink.' A car is not a type of animal, hence the preference violation. However, what is 
relevant in (20) is drinking, and there is a relevant analogy that animals and cars both 
use up a liquid of some kind: animals drink potable liquids while cars use gasoline. 
Hence the metaphorical relation between 'car' and 'drink.' 
Metaphor recognition in the met* method is related to all four views of metaphor 
described in Section 2. Recognition is viewed as a two-part process consisting of \[1\] 
a contextual constraint violation and \[2\] a set of "correspondences" including a key 
correspondence, a relevant analogy. The contextual constraint violation may be a pref- 
erence violation, as in the selection restrictions view of metaphor. The set of "corre- 
spondences" is rather like the system of commonplaces between tenor and vehicle in 
the interaction view. The relevant analogy is related to the comparison and interaction 
62 
Fass Discriminating Metonymy 
views, which emphasize a special comparison or an analogy as central to metaphor. 
Moreover, the relevant analogies seem to form groupings not unlike the conceptual 
metaphors found in the conventional view. 
Example 21 
"The idea drank the heart." 
Anomalous relations have neither the semantic relationships of a metonymic rela- 
tion nor the relevant analogy of a metaphorical relation. Hence the semantic relation 
between 'idea' and 'drink' is anomalous in (21) because 'idea' is not a preferred agent 
of 'drink' and no metonymic link or relevant analogy can be found between animals 
(the preferred agent) and ideas; that is, 'idea' in (21) does not use up a liquid like 'car' 
does in (20). This is not to say that an anomalous relation is uninterpretable or that 
no analogy can possibly be found in one. In special circumstances (for example, in a 
poem), search for analogies might be expanded to permit weaker analogies, thereby 
allowing "ideas drinking" to be interpreted metaphorically. 
The topology of the flow chart in Figure 1 results from needing to satisfy a number 
of observations about the preference-based phenomena, particularly metonymy: 
1. literalness is distinct from the others, which are all nonliteral; 
2. metonymies can occur in chains (Reddy 1979); 
3. metonymy always seems to occur with one of the other three; and 
4. metaphor and anomaly are the hardest to tell apart (and thus require the 
most extended processing to distinguish). 
Hence a preference-based semantic relation can be either a single relation or a 
multi-relation. A single relation consists of one literal, metaphorical, or anomalous re- 
lation. A multi-relation contains one literal, metaphorical, or anomalous relation plus 
either a single metonymy or a chain of metonymies. All these combinations, but only 
these, are derivable from Figure 1. 
Note that in the met* method as presented in Figure 1, semantic relations are 
tried in a certain order: literal, metonymic, metaphorical, and finally anomalous. This 
ordering implies that a literal interpretation is sought before a nonliteral one (cf. Harris 
1976). The ordering results from thinking about discriminating the semantic relations 
in serial processing terms rather than parallel processing terms, particularly the serial 
order in which selection restrictions are evaluated and metonymic inference rules are 
tried: satisfied selection restrictions (indicating literalness) then metonymic inference 
(metonymy) then violated selection restrictions (metaphor or anomaly). 
Gibbs (1984) criticizes the idea that literal and nonliteral meaning can be discrimi- 
nated in ordered processing stages. My response is that if the met* method is viewed in 
parallel processing terms then literal, metonymic, metaphorical, and anomalous inter- 
pretations are all sought at the same time and there is no ordering such that the literal 
meaning of a sentence is computed first and then an alternative meaning sought if 
the literal meaning is defective. Gibbs' other main criticism, concerning the traditional 
analysis of sentence meaning as composed from word meanings and independent of 
context, will be discussed in Section 7. 
63 
Computational Linguistics Volume 17, Number 1 
4. CoUative Semantics 
CS is a semantics for natural language processing that extends many of the main 
ideas behind Preference Semantics (Wilks 1973; 1975a; 1975b; 1978; see also Wilks and 
Fass in press). CS has four components: sense-frames, collation, semantic vectors, and 
screening. The met* method is part of the process of collation. Fuller and more general 
descriptions of the four components appear in Fass (1988a; 1989b). 
Sense-frames are dictionary entries for individual word senses. Sense-frames are 
composed of other word senses that have their own sense-frames, much like Quillian's 
(1967) planes. Each sense-frame consists of two parts, an arcs section and a node section, 
that correspond to the genus and differentia commonly found in dictionary definitions 
(Amsler 1980). 
The arcs part of a sense-frame contains a labeled arc to its genus term (a word 
sense with its own sense-flame). Together, the arcs of all the sense-frames comprise a 
densely structured semantic network of word senses called the sense-network. The node 
part of a sense-frame contains the differentia of the word sense defined by that sense- 
frame, i.e., information distinguishing that word sense from other word senses sharing 
the same genus. The two lexical semantic constraints mentioned earlier, preferences 
and assertions, play a prominent part in sense-frame nodes. 
Sense-frame nodes for nouns (node-type 0) resemble Wilks' (1978) pseudo-texts. 
The nodes contain lists of two-element and three-element lists called cells. Cells contain 
word senses and have a syntax modeled on English. Each cell expresses a piece of 
functional or structural information and can be thought of as a complex semantic 
feature or property of a noun. Figure 2 shows sense-frames for two senses of the noun 
'crook.' Crook1 is the sense meaning "thief" and crook2 is the shepherd's tool. 
All the terms in sense-frames are word senses with their own sense-frames or 
words used in a particular sense that could be replaced by word senses. It1 refers 
to the word sense being defined by the sense-frame so, for example, crook1 can be 
substituted for it1 in \[it1, steal1, valuables1\]. Common dictionary practice is followed 
in that word senses are listed separately for each part of speech and numbered by 
frequency of occurrence. Hence in crook2, the cell \[shepherd1, use1, it1\] contains the 
noun sense shepherd1 while the cell \[it1, shepherd1, sheep1\] contains the verb sense 
shepherd1 (in a three-element cell, the second position is always a verb, and the first 
and third positions are always nouns). 
Sense-frame nodes for adjectives, adverbs and other modifiers (node-type 1) con- 
tain preferences and assertions but space does not permit a description of them here. 
Sense-frame nodes for verbs and prepositions (node-type 2) are case frames con- 
taining case subparts filled by case roles such as 'agent,' 'object,' and 'instrument.' 
Case subparts contain preferences, and assertions if the verb describes a state change. 
sf(crookl, sf(crook2, 
\[\[arcs, \[\[arcs, 
\[\[supertype. criminal1\]\]\], \[\[supertype, stick1\]\]\], 
\[nodeO, \[nodaO, 
\[\[it1, steal1, valuables1\]\]\]\]). \[\[shepherd1, use1, it1\], 
\[it1, shepherd1, sheep1\]\]\]\]). 
Figure 2 
Sense-frames for crook1 and crook2 (noun senses) 
64 
Fass Discriminating Metonymy 
sf(eatl, sf(clrinkl, 
\[(arcs, \[\[arcs, 
\[\[supertype, (ingest1, expenclt\]\]\]\], \[\[supertype, (ingest1, expendt\]\]\]\], 
(node2, \[node;t, 
\[\[agent, \[\[agent, 
\[preference, animalt\]\], \[preference, animal1\]\], 
\[object, \[object, 
\[preference, foocll\]\]\]} D. \[preference, drink1\]\]\]\]\]). 
Figure 3 
Sense-frames for eat1 and drink1 (verb senses) 
Figure 4 
The met* method (CS version) 
Figure 3 shows the sense-frames for the verb senses eat1 and drink1. In both, the agent 
preference is for an animal but the object preferences differ: the preference of eat1 is 
for food1, i.e., an edible solid, while the preference of drink1 is for drink1 (the noun 
sense), i.e., a potable liquid. 
The second component of CS is the process of collation. It is collation that con- 
tains the met* method in CS. Collation matches the sense-frames of two word senses 
and finds a system of multiple mappings between those sense-frames, thereby dis- 
criminating the semantic relations between the word senses. Figure 4 shows the use 
of the met* method in CS. Figure 4 is similar to the one in Figure 1 except that the 
diamonds contain the processes used in CS to check for satisfied preferences (diamond 
1), metonymic inferences (diamond 2), and relevant analogies (diamond 3). 
The basic mappings in collation are paths found by a graph search algorithm 
that operates over the sense-network. Five types of network path are distinguished. 
Two types of path, called ancestor and same, denote kinds of "inclusion," e.g., that the 
class of vehicles includes the class of cars (this is an ancestor relationship). Satisfied 
65 
Computational Linguistics Volume 17, Number 1 
preferences are indicated by network paths denoting inclusion, also known as "inclu- 
sive" paths (see diamond 1 in Figure 4). The other three types of network path, called 
sister, descendant, and estranged, denote "exclusion," e.g., that the class of cars does not 
include the class of vehicles (this is a descendant relationship). Violated preferences 
are network paths denoting exclusion, also known as "exclusive" paths. 
These paths are used to build more complex mappings found by a frame-matching 
algorithm. The frame-matching algorithm matches the sets of cells from two sense- 
frames. The sets of cells, which need not be ordered, are inherited down the sense- 
network. A series of structural constraints isolate pairs of cells that are matched using 
the graph search algorithm. Network paths are then sought between terms occupying 
identical positions in those cells. Seven kinds of cell match are distinguished, based 
on the structural constraints and types of network path found. Ancestor and same are 
"inclusive" cell matches, e.g. \[composition1, metal1\] includes \[composition1, steel1\] 
because the class of metals includes the class of steels (another ancestor relationship). 
Sister, descendant, and estranged are types of "exclusive" cell matches, e.g. \[composi- 
tion1, steel1\] and \[composition1, aluminiuml\] are exclusive because the class of steels 
does not include the class of aluminiums since both belong to the class of metals (this 
is a sister relationship). The remaining cell matches, distinctive source and distinctive 
target, account for cells that fail the previous five kinds of cell match. For more detail 
on cell matches, see Fass (1988a). 
A kind of lexical relevance is found dynamically from the sentence context. This 
notion of relevance is used in finding the relevant analogies that distinguish metaphori- 
cal from anomalous relations; it is also used when finding CO-AGENT FOR ACTIVITY 
metonymies. Relevance divides the set of cells from the source sense-frame into two 
subsets. One cell is selected as relevant given the context; the remaining cells are termed 
nonrelevant. Collation matches both the source's relevant and nonrelevant cells against 
the cells from the target sense-frame. A relevant analogy is indicated by a sister match 
of the source's relevant cell (see diamond 3 in Figure 4). 
Five types of metonymic concepts are currently distinguished. Examples of two 
of the metonymic concepts, CONTAINER FOR CONTENTS and ARTIST FOR ART 
FORM, have already been given. The remaining three are PART FOR WHOLE, PROP- 
ERTY FOR WHOLE, and CO-AGENT FOR ACTIVITY. 
Example 22 
"Arthur Ashe is black" (= skin colored black ~ PART FOR WHOLE). 
Example 23 
"John McEnroe is white" (= skin colored white --+ PART FOR WHOLE). 
In (22) and (23), the skins of Arthur Ashe and John McEnroe, parts of their bodies, 
are colored black (white). 
Example 24 
"John McEnroe is yellow" (= limited in bravery --* PROPERTY FOR WHOLE). 
Example 25 
"Natalia Zvereva is green" (= limited in experience ~ PROPERTY FOR WHOLE). 
In (24), for example, John McEnroe is limited with respect to his bravery, a property 
possessed by humans and other animals. 
66 
Fass Discriminating Metonymy 
Example 26 
"Ashe played McEnroe" (= tennis with McEnroe --~ CO-AGENT FOR ACTIVITY). 
These concepts are encoded in metonymic inference rules in CS (see diamond 2 in Fig- 
ure 4). The rules are ordered from most common (synecdoche) to least. The order used 
is PART FOR WHOLE, PROPERTY FOR WHOLE, CONTAINER FOR CONTENTS, 
CO-AGENT FOR ACTIVITY, and ARTIST FOR ART FORM. 
The first two concepts, PART FOR WHOLE and PROPERTY FOR WHOLE, are 
source-driven; the others are target-driven. The difference in direction seems to be 
dependent on the epistemological structure of the knowledge being related by the 
different inferences. PART FOR WHOLE metonymies are source-driven, perhaps be- 
cause the epistemological nature of parts and wholes is that a part generally belongs 
to fewer wholes than wholes have parts, hence it makes sense to drive inferencing 
from a part (source) toward the whole (target) than vice versa. 
In CONTAINER FOR CONTENTS (target-driven), on the other hand, the episte- 
mological nature of containers and contents is that the containers generally mentioned 
in CONTAINER FOR CONTENTS metonymies are artifacts designed for the function 
of containing -- hence one can usually find quite specific information about the typ- 
ical contents of a certain container, for example, some glasses as in (7) -- whereas 
the contents do not generally have the function of being the contents of something. 
Hence it makes sense to drive inferencing from the container, and the function it per- 
forms, toward the contents than vice versa. The same reasoning applies to ARTIST 
FOR ART FORM (target-driven). An artist has the vocation of creating art: that is 
his/her purpose. 
A further step in collation distinguishes metaphorical from anomalous semantic 
relations. Recall that a metaphorical relation contains a relevant analogy, as in (15) 
and (20), while an anomalous relation does not, as in (21). A relevant analogy is found 
by matching the relevant cell from the source sense-frame with one of the cells from 
the target sense-frame. If the match of cells is composed of a set of sister network 
paths between corresponding word senses in those cells, then this is interpreted as 
analogical and hence indicative of a metaphorical relation. Any other match of ceils is 
interpreted as not analogical and thus an anomalous semantic relation is recognized 
(see Fass 1986; 1987). 
The third component of CS is the semantic vector which is a form of representation, 
like the sense-frame; but sense-frames represent lexical knowledge, whereas semantic 
vectors represent coherence. Semantic vectors are therefore described as a kind of coher- 
ence representation. A semantic vector is a data structure that contains nested labels and 
ordered arrays structured by a simple dependency syntax. The labels form into sets. 
The outer sets of labels indicate the application of the three kinds of lexical semantic 
constraints. The outermost set of labels is 'preference' and 'assertion.' The middle set is 
'relevant' and 'nonrelevant.' The innermost set is the kind of mapping used: 'network 
path' and 'cell matches.' The nesting of labels shows the order in which each source of 
knowledge was introduced. The ordered arrays represent the subkinds of each kind 
of mapping. Five-column arrays are for the five network paths; seven-column arrays 
are for the seven types of cell match. Each column contains a positive number that 
shows the number of occurrences of a particular network path or cell match. 
The fourth component of CS is the process of screening. During analysis of a 
sentence constituent, a Semantic vector is created for every pairwise combination of 
word senses. These word sense combinations are called semantic readings or simply 
"readings." Each reading has an associated semantic vector. Screening chooses between 
two semantic vectors and hence their attached semantic readings. Rank orderings 
67 
Computational Linguistics Volume 17, Number 1 
among semantic relations are applied. In the event of a tie, a measure of conceptual 
similarity is used. 
The ranking of semantic relations aims to achieve the most coherent possible in- 
terpretation of a reading. The class of preference-based semantic relations takes prece- 
dence over the class of assertion-based semantic relations for lexical disambiguation. 
The rank order among preference-based semantic relations is 
literal --* metaphorical ~ anomalous. 
If the semantic vectors are still tied then the measure of conceptual similarity is 
employed. This measure was initially developed to test a claim by Tourangeau and 
Sternberg (1982) about the aptness of a metaphor. They contend that aptness is a 
function of the distance between the conceptual domains of the source and target 
involved: the claim is that the more distant the domains, the better the metaphor. This 
is discussed further in Section 5. The conceptual similarity measure is also used for 
lexical ambiguity resolution (see Fass 1988c). 
5. The Meta5 Program 
CS has been implemented in the meta5 natural language program. The meta5 program 
is written in Quintus Prolog and consists of a lexicon holding the sense-frames of just 
over 500 word senses, a small grammar, and semantic routines that embody collation 
and screening, the two processes of CS. The program is syntax-driven, a form of 
control carried over from the structure of earlier programs by Boguraev (1979) and 
Huang (1985), on which meta5 is based. Meta5 analyzes sentences, discriminates the 
seven kinds of semantic relation between pairs of word senses in those sentences (i.e., 
the program recognizes metonymies, metaphors, and so on), and resolves any lexical 
ambiguity in those sentences. Meta5 analyzes all the sentences given in Sections 3 
and 4, plus a couple more metaphorical sentences discussed in Section 7. 
Below are simplified versions of some of the metonymic inference rules used in 
meta5. The metonymic concepts used in CS contain three key elements: the conceptual 
relationship involved, the direction of inference, and a replacement of the source or 
target. The metonymic inference rules in meta5 contain all three key elements. The 
rules, though written in a prolog-like format, assume no knowledge of Prolog on the 
part of the reader and fit with the role of metonymy shown in Figures 1 and 4. 
Each metonymic inference rule has a left-hand side and a right-hand side. The left- 
hand side is the topmost statement and is of the form metonymic_inference_rule(Source, 
Target). The right-hand side consists of the remaining statements. These statements 
represent the conceptual relationship and the direction of inference, except for the 
bottom most one, which controls the substitution of the discovered metonym for either 
the source or target: this statement is always a call to find a new sense-network path. 
Rule 1 
PROPERTY FOR WHOLE: source-driven. 
metonymic_inference_rule (Source, Target):- 
find_cell (Source, \[Whole, have1, it1\]), \[1\] 
find_sense_network_path (Whole, Target). \[2\] 
This rule represents PROPERTY FOR WHOLE, which is source-driven. State- 
ment \[1\] represents the conceptual relationship and direction of inference. The concep- 
tual relationship is that the source is a property possessed by the whole in a property- 
whole relation. The inference is driven from the source: find_cell searches through the 
68 
Fass Discriminating Metonymy 
source's list of cells for one referring to a "whole" of which the source is a "part." 
Statement \[2\] controls the substitution of the discovered metonym: the "whole" is 
the substitute metonym that replaces the source, and the next sense-network path is 
sought between the whole and the target. 
Rule 2 
CONTAINER FOR CONTENTS: target-driven. 
metonymic_inference_rule (Source, Target):- 
find_cell (Target, \[it1, contain1, Contents\]), \[1\] 
find~sense_network_path (Source, Contents). \[2\] 
This metonymic concept is target-driven. The target is the "container"in a container- 
contents relation (\[1\]). The "contents" is the substitute metonym that replaces the tar- 
get. The next sense-network path is sought between the source and the contents (\[2\]). 
Rule 3 
ARTIST FOR ART FORM: target-driven. 
metonymicJnference_rule (Source, Target):- 
find_genus (Target, Occupation), \[1\] 
find_cell (Occupation, lit1, Make, Art form\]), \[2\] 
confirm_type (create1, Make), \[3\] 
confirm_type (artdorml, Art form), \[4\] 
find~ense_network_path (Source, Art form). \[5\] 
Again, the inference in ARTIST FOR ART FORM is from the target. The target 
is a person who is an "artist" in an artist-art form relation. The occupation of the 
person is found by searching up the sense-network (\[1\]). The list of ceils associated 
with the occupation are searched for a cell describing the main activity involved in 
the occupation (\[2\]), e.g., a cook cooks food and an artist makes art forms. Checks are 
done to confirm that any activity found is indeed making an art form, i.e., that the 
"making" involved is a type of creating (\[3\]) and that the "art form" is a type of art 
form1 (\[4\]). The "art form" is the substitute metonym that replaces the target. A new 
sense-network path is computed between the source and the art form (\[5\]). I will now 
describe how meta5 recognizes some metonymies and metaphors. 
Example 19 
"Ted played Bach" (= the music of Bach). 
In (19), between 'Bach' and the twelfth sense of 'play' in meta5's lexicon (meaning "to 
play music"), there is a chain of metonymies plus a literal relation. The chain consists 
of ARTIST FOR ART FORM and CONTAINER FOR CONTENTS metonymies. Both 
metonymic concepts are target-driven. In ARTIST FOR ART FORM the inference is 
from the ARTIST (the target) to the ART FORM (the source), so the substitute metonym 
replaces the target (the ARTIST) if the inference is successful. 
The sense-frames of the verb sense play12 and the noun senses music1 and jo- 
hann_sebastian_bach are shown in Figure 5. The semantic relation results from match- 
ing the object preference of play12, which is for music, against the surface object, 
which is 'Bach,' short for 'Johann Sebastian Bach.' The preference is the source and 
the surface object is the target. 
We will follow what happens using the flow chart of Figure 4. (Enter diamond 1 
of the chart.) The sense-network path between the source (music1) and the target 
69 
Computational Linguistics Volume 17, Number 1 
sf(play12, 
\[\[arcs, 
\[\[supertype, per forrnl\]\] 1, 
\[node2, 
\[\[agent, 
\[preference, human_being1\]\], 
\[object, 
\[preference,music1\]\]\]\]\]). 
sf(musicl, sf (johann_sebastian_bach, 
\[\[arcs, \[\[arcs, 
\[\[supertype, \[sound1, art_fermi\]\]I\], \[\[supertype, composerl\]l \] 
\[nodeO, \[nodeO, 
\[\[musician1, play12, itt\]\]\]\]). \[\[animacyl, dead1\], 
\[sex1, male1\], 
\[bornt, 1685\], 
\[died1, 17501111). 
\]Figure 5 
Sense-frames for play12 (verb sense), music1 and johann_sebastian_bach (noun senses) 
(johann_sebastian_bach) is sought. The path is not inclusive because johann_sebastian_ 
bach is not a type of music1. 
(Enter diamond 2 of the chart.) Metonymic inference rules are applied. The rules 
for PART FOR WHOLE, PROPERTY FOR WHOLE, CONTAINER FOR CONTENTS, 
CO-AGENT FOR ACTIVITY are tried in turn, but all fail. The rule for ARTIST FOR 
ART FORM, however, succeeds. The discovered metonymic inference is that johann_ 
sebastian_bach (the ARTIST) composes musical pieces (the ART FORM). The metonymic 
inference is driven from the target (the ARTIST), which is johann_sebastian_bach. The 
successful metonymic inference, using the ARTIST FOR ART FORM inference rule 
above, is as follows: \[1\] johann_sebastian_bach (the ARTIST) is a composer1, \[2\] com- 
posers compose1 musical pieces (the ART FORM). Additional tests confirm \[2\], which 
are that \[3\] composing is a type of creating, and \[4\] a musical_piece1 is a type of 
art_form1. 
(Enter the leftmost statement box -- also step \[5\] of the ARTIST FOR ART FORM 
inference rule above.) The original target (johann_sebastian_bach) is replaced by the 
substitute metonym (musical_piece1). 
(Enter diamond 1 for a second time.) The sense-network path between the source 
(music1) and the new target (musical_piece1) is sought. The path is not inclusive. 
(Enter diamond 2 for a second time.) Metonymic inference rules are applied. The 
rules for PART FOR WHOLE and PROPERTY FOR WHOLE fail, but the rule for 
CONTAINER FOR CONTENTS succeeds. The successful inference, using the descrip- 
tion of the CONTAINER-CONTENTS inference rule given previously, is that \[1\] a 
musical_piece1 (the CONTAINER) contains music1 (the CONTENTS). 
(Enter the leftmost statement box for a second time.) The direction of inference in 
the CONTAINER FOR CONTENTS metonymic concept is from the target (the CON- 
TAINER) towards the source (the CONTENTS), so \[2\] the target (the CONTAINER) is 
replaced by the substitute metonym when an inference is successful. Hence in our ex- 
ample, the target (musical_piece1) is again replaced by a substitute metonym (music1). 
The source, which is music1, the object preference of play12, remains unchanged. 
(Enter diamond 1 for a third time.) The sense-network path between the source 
(music1) and the latest target (music1) is sought. The path is inclusive, that music1 is 
a type of music1, so a literal relation is found. 
(Exit the chart.) The processing of the preference-based semantic relation(s) be- 
tween play12, and its preference for music1, and johann_sebastian_bach is completed. 
70 
Fass Discriminating Metonymy 
After an initial preference violation (Johann Sebastian Bach is not a kind of music), the 
semantic relation found was an ARTIST FOR ART FORM metonymic relation (that 
johann_sebastian_bach composes musical pieces) followed by a CONTAINER FOR 
CONTENTS metonymic relation (that musical pieces contain music) followed by a 
literal relation (that music is music). 
Example 20 
"The car drank gasoline." 
There is a metaphorical relation between carl and the verb sense drink1 in (20). The 
source is drink1, whose agent preference is animal1, and the target is carl (see Fig- 
ure 6). 
A metaphorical relation is sought after failing to find an inclusive network path 
or a metonymic inference between animal1 and carl, hence the network path between 
animal1 and carl must be exclusive. The network path found is an estranged one. 
The second stage is the match between the relevant cell of animal1 and the cells of 
carl. In the present example, drinkl is relevant. The list of cells for animal1 is searched 
for one referring to drinking. The relevant cell in the list is \[animal1, drink1, drink1\], 
which is matched against the inherited cells of carl (see Figure 7). A sister match is 
found between \[animal1, drink1, drink1\] and \[carl, use2, gasoline1\] from carl. 
The sister match is composed of two sister paths found in the sense-network. The 
first sister path is between the verb senses drink1 and use2, which are both types 
of expending (Figure 8). The second path is between the noun senses drink1 and 
gasoline1, which are both types of liquid (Figure 9). The effect of the network paths is to 
establish correspondences between the two cells such that an analogy is "discovered" 
that animals drink potable liquids as cars use gasoline. Note that, like Gentner's (1983) 
systematicity principle, the correspondences found are structural and independent of 
the content of the word senses they connect. Note also that the two cells have an 
underlying similarity or "ground" (Richards 1936) in that both refer to the expenditure 
of liquids. This second stage of finding a relevant analogy seems the crucial one in 
metaphor recognition. 
Figure 10 shows the match of the nonrelevant cells from animal1 and carl. The cell 
\[carl, use2, gasoline1\] has been removed. There are three inclusive cell matches as an- 
imals and cars share physical objectlike properties of boundedness, three dimensions, 
sf(drinkl, 
\[\[arcs, 
\[\[supertype, \[ingest1, expendl\]\]J\], 
\[node2, 
\[\[agent, 
\[preference, anlrnall\] \], 
\[object, 
\[preference, drinkl\]\]\]\]\]). 
sf(animatl, sf(carl, 
\[\[arcs, \[\[arcs, 
\[\[supertype, organism1\]\]\], \[\[supertype, motor_vehicle1\]\]\], 
\[nodeO, \[nodeO, . 
\[\[biology1, animal1\], \[lit1, carry1, passenger1\]\]\]\]). 
lit1, drink1, drink1\], 
lit1, eat1, food1\]\]\]\]). 
Figure 6 
Sense-frames for drink1 (verb sense), animal1 and carl (noun senses) 
71 
Computational Linguistics Volume 17, Number 1 
Relevant cell of animal1 
(SOURCE) 
\[animal1, drink1, drink1\] 
Cetls of carl 
(TARGET) 
\[\[bounds1, distinct1\], 
\[extent1, three_dimensionatl 
\[behaviourl, solid1\], 
\[animacyl, nonliving1\], 
\[carl, roll1, \[on3, land1\]\], 
\[composition1, steel1\], 
\[driver1, drivel, carl\], 
\[carl, have1, \[4, wheel1\]\], 
\[carl, have1, engine1\], 
\[carl, use2, gasoline1\], 
\[caN, carry1, passenger1\]\] 
Figure 7 
Match of relevant cell from animal1 against cells of carl 
~ enter1 ~ contract1 
supertype supertype super~l~ertype 
~drinkl suP~I~ use2 
Figure 8 
Sister sense-network path between drink1 and use2 (verb senses) 
~ gy source1 
| supertype supertype supertype J 
Idrinkl ~J gasoline1 Ico.ll I |oodl 
Figure 9 
Sister sense-network path between gasoline1 and drink1 (noun senses) 
72 
Fass Discriminating Metonymy 
Non-relevant cells of animal1 Non-relevant c~lJ~ of carl Cell matches (SOURCE) (TARGET} 
\[\[bour=dsl, distinct1 \] \]\]bounds I distinct1 \] i sam \]extent1, three dimensional1|, \[extent , three d mens ona \],13 e 
\[behaviourt, solid1|, \[behaviourt, sollidl\], ICell matches 
lanimacyl, livingt\], \[animacyl, nonlivingl\], ~2 sister 
\]composition1, flesh1|, \]composition1, steel1|, ~ cell matches 
\]animal1, eat1, food1|, J2 distinctive 
\[biol0gyt, animalt\]\] I source celts (of animalt) 
\]carl, roHt, \]on3, land1||, | 
\]driver1, drivel, carl|, 15 distinctive \]carl, hayer, \[4. wheelt\]\], target cells 
\[cart, hayer, enginet\], (of carl) 
\[cart, carryt, passenger1|| 
Figure 10 
Matches of non-relevant cells from animal1 and carl 
\[preference, 
\[\[network~oath, 
\[0, 0, 0, 0, 1\]\], 
\[cellmatch, 
\[\[relevant, 
\[0, 0, 1, 0, 0, 0, 10\]\], 
\]non_relevant, 
\[0, 3, 2, o, o, 2, 5\]\]\]\]\]\] 
I First array: 
preference violation 
(estranged sense-network path) 
I Second array: 
relevant analogy 
(sister match of relevant cell) I Third array: 
distance betw. conceptual domains 
(matches of non-relevant cells) 
Figure 11 
Semantic vector for a metaphorical semantic relation 
and solidity. Two cell matches are exclusive. Animals are composed of flesh, whereas 
cars are composed of steel. Animals are living, whereas cars are nonliving. There are 
two distinctive cells of animal1 and five distinctive cells of carl. Tourangeau and Stern- 
berg's (1982) hypothesis predicts that the greater the distance between the conceptual 
domains of the terms involved in a metaphor, the more apt the metaphor. The pro- 
portion of similarities (inclusive cell matches) to differences (exclusive cell matches) is 
3 to 2, which is a middling distance suggesting, tentatively, an unimposing metaphor. 
All of these matches made by collation are recorded in the semantic vector shown 
in Figure 11. The crucial elements of the metaphorical relation in (20) are the preference 
violation and the relevant analogy. In Figure 11, the preference violation has been 
recorded as the 1 in the first array and the relevant analogy is the 1 in the second 
array. Information about the distance between conceptual domains is recorded in the 
third array. 
The 'preference' label indicates that a preference has been matched (rather than 
an assertion). The five columns of the first array record the presence of ancestor, 
same, sister, descendant and estranged network paths respectively. When a preference 
is evaluated, only one network path is found, hence the single 1 in the fifth column, 
which indicates that an estranged network path was found between animal1 and carl. 
Cell matches are recorded in the second and third arrays, which each contain seven 
columns. Those columns record the presence of ancestor, same, sister, descendant, 
estranged, distinctive source, and distinctive target cell matches respectively. The 1 in 
the third column of the second array is the relevant analogy -- a sister match of the 
73 
Computational Linguistics Volume 17, Number 1 
relevant cell \[animal1, drink1, drink1\] and the cell \[carl, use2, gasoline1\]. The 10 is 
the ten distinctive cells of carl that did not match \[animal1, drink1, drink1\]. This is 
the match of 12 cells, 1 from the source and 11 from the target (see Figure 7). The sum 
of array columns is: 
((0+0+1+0+0) x2) 4-((0+10) xl)=(1 x2)+(10x1)=12. 
The 3 similarities, 2 differences, 2 distinctive cells of animal1 and 5 distinctive cells 
of carl are the nonzero numbers of the final array. The 3 similarities are all same cell 
matches; the 2 differences are both sister cell matches. A total of 17 cells are matched, 
7 from the source and 10 from the target (see Figure 10). The total of array columns is: 
((0+3+2+0+0) x 2) + ((2+5) x 1) = (5 x 2) + (7 x 1) = 17. 
Example 15 
"The ship ploughed the waves." 
In (15), there is a metaphorical relation between a sense of the noun 'ship' and the 
second sense of the verb 'plough' :in meta5's lexicon. Note that 'plough,' like 'drink,' 
belongs to several parts of speech. Figure 12 shows the sense-frames for the verb sense 
plough2, the noun sense plough1, which is the instrument preference of plough2, and 
the noun sense ship1. 
In (15), meta5 matches senses of 'ship' against senses of 'plough.' When meta5 
pairs ship1 with plough2, it calls upon collation to match ship1 against the noun 
sense plough1, the instrument preference of plough2. 
First, the graph search algorithm searches the sense-network for a path between 
plough1 (which is the preference) and ship1 and finds an estranged network path 
between them, i.e., a ship is not a kind of plough, so plough2's instrument preference 
is violated. 
Next, collation inherits down lists of cells for plough1 and ship1 from their super- 
ordinates in the sense-network. What is relevant in the present context is the action of 
ploughing because (15) is about a ship ploughing waves. Collation then runs through 
the list of inherited cells for the noun sense plough1 searching for a cell that refers to 
the action of ploughing in the sense currently under examination by meta5, plough2. 
sf(plough2, 
\[\[arcs 
\[\[supertype, transfer1\]\]\], 
\[node2, 
\[\[instrument, 
\[preference, plough1\]\], 
\[object, 
\[preference, soil1\]\]\]\]\]). 
sf(ploughl, sf(shipl, 
\[\[arcs, \[\[arcs, 
\[\[supertype, tool1\]\]\], \[\[supertype, watercraft1\]\]\], 
\[nodeO, \[nodeO, 
\[\[farmer1, plough1, it1\], \[\[it1, carry1, shipment1\], 
\[it1, plough2, soil1\]\]\]\]). \[it1, have1, engine1\]\]\]\]). 
Figure 12 
Sense-frames for plough2 (verb sense), plough1 and ship1 (noun senses) 
74 
Fass Discriminating Metonymy 
Relevant cell of DIouQhl 
(SOURCE) 
\[plough1, plough2, soil1\] 
Cells of shiol 
(TARGET) 
\[\[bounds1, distinct1\], 
\[extent1, three dimensional1\], 
\[behaviourl, solid1\], 
\[animacyl, nonliving1\], 
\[composition 1, metal1\], 
\[ship1, use2, energy source1\], 
\[boatman1, sail1, ship1\], 
\[ship1, sail2, water2\], 
\[ship1, carry1, shipment1\], 
\[ship1, have1, enginelJ\] 
Figure 13 
Match of relevant cell from plough1 against cells from ship1 
I Non-relevant cells of olouahl Non-relevant cells of shin1 Cell matches 
(SOURCE) (TARGET) I 1 ancestor 
\[\[composition1, matter1\], \[\[composition1, metal1\], Icell match 
\[bounds1, distirtctl\], \[bounds1, distinct1\], 
\[extent1, three dimensional1\], lextentt, three_dimensionall\],14 same 
\[behaviourl, solid1\], \[behaviourl, solid1\], Icell matches \[animacyl, nonliving1\], \[animacyl, nonliving1\], | 
I 1 sister \[farmer1, plough1, plough1\]\] \[boatman1, sail1, ship1\], Icell match 
use2, energy_source1\], 13 distinctive \[ship1, \[ship1, carry1, shipment1\], Itarget cells 
|ship1, ha,/el, engine1\]\] I(of ship1) 
Figure 14 
Matches of non-relevant cells from plough1 and ship1 
Collation finds a relevant cell \[plough1, plough2, soil1\] and uses its frame-matching 
algorithm to seek a match for the cell against the list of inherited cells for ship1, shown 
in Figure 13 (for ease of reading, it1 has again been replaced by the word senses being 
defined). The algorithm finds a match with \[ship1, sail2, water2\] (highlighted in Figure 
13), and hence collation "discovers" a relevant analogy that both ships and ploughs 
move through a medium, i.e., that ploughs plough through soil as ships sail through 
water. 
Finally, collation employs the frame matching algorithm a second time to match 
together the remaining nonrelevant cells of plough1 and ship1 (see Figure 14). The cell 
\[ship1, sail2, water2\] is removed to prevent it from being used a second time. 
Figure 15 shows the semantic vector produced. As with Figure 11, it shows a 
metaphorical relation. There is a preference violation, an estranged network path in- 
dicated by the 1 in the fifth column of the first array. There is also a relevant analogy, 
shown by the 1 in the third column of the second array: the analogical match of the 
cells \[plough1, plough2, soil1\] and \[ship1, sail2, water2\]. The second array shows that 
11 cells are matched, I from the source and 10 from the target (check against Figure 13). 
The sum of the array's columns is: 
((0 +0 + 1 +0+0) x 2) + ((0+9) x 1) = (1 x 2) + (9 x 1) = 11. 
75 
Computational Linguistics Volume 17, Number 1 
\[preference, 
\[\[network oath, 
\[0, O, O, O, 1\]\], 
\[cell_match, 
\[\[relevant, 
\[o, o, 1, o, o, o, 9\]\], 
\[non_relevant, 
\[1,4, 1, O, O, O, 311\]\]11 
I First array: preference violation 
(estranged sense-network path) 
I Second array: 
relevant analogy 
(sister match of relevant cell) I Third array: 
distance betw. conceptual domains 
(matches of non-relevant cells) 
Figure 15 
Semantic vector for another metaphorical semantic relation 
In the third array, the match of nonrelevant cells, there is 1 ancestor match, 4 same 
matches, 1 sister match, and 3 distinctive cells of ship1. Fifteen cells are matched, 6 
from the source and 9 from the target (see Figure 14). The totals are: 
((1+4+1+0+0) x2)+((0+3)x1)=(6x2)+(3x1)=15. 
Semantic vectors can represent all the semantic relations except metonymic ones. The 
reason is that metonymic relations, unlike the others, are not discriminated by CS in 
terms of only five kinds of network path and seven kinds of cell matches. Instead, 
they consist of combinations of network paths and specialized matches of cells that 
have not fallen into a regular enough pattern to be represented systematically. 
6. Extensions 
Even for those semantic dependencies investigated, the interpretation of semantic re- 
lations seems to require more complexity than has been described so far in this paper. 
Consider the differences between the following sentences: 
Example 20 
"The car drank gasoline." 
Example 27 
"The car drank coffee." 
Intuitively, sentence (20) is metaphorical while (27) is metaphorical/anomalous. 
In (20), the semantic relation between 'car' and 'drink' is thought to be metaphorical, 
and the isolated semantic relation between just 'drink' and 'gasoline' is anomalous, 
but the sentence as a whole is metaphorical because it is metaphorical that cars should 
use up gasoline. 
In (27), the semantic relation between 'car' and 'drink' is metaphorical; the seman- 
tic relation between just 'drink' and 'coffee' is literal; yet the effect of (27) as a whole is 
metaphorical/anomalous. The object preference of 'drink' is for a drink, i.e., a potable 
liquid. It seems that it is metaphorical for cars to "drink" a liquid commonly used up 
by cars, e.g., gasoline, but anomalous if the liquid has nothing to do with cars, e.g., 
coffee, as in (27). 
The problem of understanding the differences between sentences (20) and (27) 
requires some further observations about the nature of semantic relations, principally 
76 
Fass Discriminating Metonymy 
that the differences are caused by the combinations of semantic relations found in the 
sentences and the relationships between those relations. Below is a suggestion as to 
how deeper semantic processing might discriminate the differences between the two 
sentences. 
Before getting to the deeper processing, we need a better semantic vector notation. 
The better semantic vector notation, which developed from a discussion with Afzal 
Ballim, is a modification of the notation shown in Section 5. The key differences are 
reformulation by rewriting the five and seven column arrays in terms of the predicate- 
argument notation used in the rest of semantic vectors, and extension by adding the 
domain knowledge connected by every network path and cell match. 
Figure 16 shows the semantic vector in Figure 11 reformulated and extended. The 
advantage of vectors like the one in Figure 16 is that they record both how the sense- 
frames of two word senses are matched (i.e., as various kinds of network path and cell 
match) and what information in the sense-frames is matched (i.e., all the cells). For 
example, the part of Figure 16 that begins "\[relevant, ..." contains all the information 
found in Figure 7, the match of the relevant cell from animal1 against the cells of carl, 
both the types of cell matches and the cells matched. The equivalent part of Figure 11 
only records the types of cell matches. Recording the contents of the matched cells is 
useful because it enables a deepened analysis of semantic relations. Such an analysis 
is needed to detect the differences between (20) and (27). 
In the description of CS in Section 4, collation discriminates the one or more 
semantic relations in each semantic dependency, but treats the semantic relations in 
one dependency as isolated from and unaffected by the semantic relations in another 
dependency. What is needed is extra processing that interprets the semantic relation(s) 
in a later dependency with respect to the semantic relation(s) established in an earlier 
\[preference, 
\[\[network_path, 
\[estranged, 
\[1, \[animal1, carl\]I\]\], 
\[cell_match, 
\[\[relevant, 
\[\[sister, 
\[1, \[\[\[animal1, drink1, drink1\], \[carl, use2, gasoline1\]\]\]\], 
\[distinctive_target, 
\[10, \[\[bounds1, distinct1\], \[extent1, three_dimensional1\], 
\[behaviourl, solid1\], \[composition1, metal1\], 
\[animacyl, nonliving1\], \[carl, rollf, \[on3, land1\]\], 
Idriverl, drivel, carl\], \[carl, hayer, \[4, wheel1\]\], 
\[carl, have1, engine1\], \[carl, caCryl, passenger1\]\]\]\]\]\], 
\[non_relevant, 
\[\[same, 
\[3, \[\[\[bounds1, distinct1\], \[boundst, distinct1\]\], 
\[\[extent 1, three dimensional1\], \[extentt, three dimensional1 \]\], 
\[\[behaviourl, solid1\], \[behaviourl, solid1\]\]\]\], 
\[sister, 
\[2, \[\[\[composition1, flesh1\], \[composition1, rnetall\]\], 
\[\[animacyl, living1\], \[animacyl, nenlMngl\]\]\]J, 
\[distinctive_source, 
\[2, \[\[animal1, eat1, food1\], \[biology1, animal1\]\]\]\], 
\[distinctive_target, 
\[5, \[\[carl, roll1, \[on3, landt\]\], \[driver1, ddvel, cart\], 
\[carl, hayer, \[4, wheel1\]\], \[carl, have1, engine1\], 
\[carl, carry1, passenger1\]\]\]\]\]\]\]\]\]\] 
Figure 16 
Reformulated and extended version of Figure 11 
77 
Computational Linguistics Volume 17, Number 1 
\[preference, 
\[cell_match, 
\[relevant, 
\[sister, 
\[1, \[\[animal1, drink1, drink1\], \[carl, use2, gasoline1\]}\]\]\]\]\] 
Figure 17 
Vector statement of match of relevant cell from animal1 against cells of carl 
\[preference, 
\[cellmatch, 
\[relevant, 
\[sister, 
\[1, \[\[animal1, drink1, drink1\], \[vehicle1, use2, gasoline1\]\]\]\]\]\]\] 
Figure 18 
Vector statement of match of relevant cell from drink1 against cells of gasoline1 (noun senses) 
one. This processing matches the domain knowledge in semantic vectors, i.e., this 
processing is a comparison of coherence representations. 
In sentences such as (20) and (27) there are two key semantic dependencies. The 
first one is between the subject noun and the verb; the second is between the verb 
and object noun. In each dependency, the source is the verb (through its agent and 
object preferences) and the targets are the nouns. Semantic relations are found for each 
dependency. One way to detect the difference between metaphorical sentences such 
as (20) and metaphorical/anomalous ones such as (27) is in each sentence to consult 
the semantic vectors produced in its two main semantic dependencies and compare 
the matches of the relevant cells that are found by collation. 
Let us go through such an analysis using CS, starting with the first semantic 
dependency between subject noun and verb. In this semantic dependency in both (20) 
and (27), a relevant analogy is discovered as part of a metaphorical relation between 
the target carl and animal1, the agent preference of the source drink1. The semantic 
vector in Figure 16 records the two cells that figure in that relevant analogy. Figure 17 
shows the same information from the semantic vector but written as a statement. 
When the second semantic dependency is analyzed in (20), the target is gasoline1 
and is matched against the noun sense drink1, the object preference of the source 
drink1 (the verb sense). A semantic vector is produced. The relevant cell found in 
the noun sense drink1 is \[animal1, drink1, drink1\]. Its match against \[vehicle1, use2, 
gasoline1\], a cell from gasoline1, is shown in the vector statement in Figure 18. The 
match is a sister match, indicating a relevant analogy. 
Now this is peculiar because "drinking gasoline" is anomalous, yet a relevant 
analogy has been found and this paper has argued that relevant analogies are special 
to metaphorical relations. One possible explanation is that differences exist between the 
recognition of metaphorical relations that concern agents and metaphorical relations 
that concern objects and other case roles. It may be that metaphorical relations are 
indicated by a relevant analogy, but only in selected circumstances. This needs further 
investigation. 
78 
Fass Discriminating Metonymy 
\[preference, 
\[cell_match, 
\[relevant, 
\[ancestor, 
\[1, \[\[animal1, drink1, drink1\], \[human_being1, drink1, coffee1\]\]\]\]\]\]\] 
Figure 19 
Vector statement of match of relevant cell from drink1 against cells from coffee1 (noun senses) 
To return to the analysis of (20), what appears to be important in determining that 
(20) is a metaphorical sentence is the comparison of the two pairs of matched relevant 
cells: 
\[\[animal1, drink1, drink1\], \[carl, use2, gasoline1\]\] 
\[\[animal1, drink1, drink1\], \[vehicle1, use2, gasoline1\]\] 
The two source cells are the same and the two target cells, \[carl, use2, gasoline1\] 
and \[vehicle1, use2, gasoline1\], are almost identical, indicating that the same basic 
analogy runs through the whole of (20), hence the sentence as a whole is metaphorical. 
Now let us analyze the second semantic dependency in (27). The target is coffee1 
and is again matched against drink1, the object preference of the verb sense drink1, the 
source. The relevant cell from the noun sense drink1 is again \[animal1, drink1, drink1\], 
which matches against \[human_being1, drink1, coffee1\] from the target coffee1. This 
time, the match is an ancestor match and hence not a relevant analogy. Figure 19 
shows this match of the relevant cell as a vector statement. Let us compare the two 
pairs of matched relevant cells for (27): 
\[\[animal1, drink1, drink1\], \[carl, use2, gasoline1\]\] 
\[\[animal1, drink1, drink1\], \[human_being1, drink1, coffee1\]\] 
The two source cells are the same but the two target cells, \[carl, use2, gasoline1\] 
and \[human_being1, drink1, coffee1\], are very different. The reason that the sentence as 
a whole is metaphorical/anomalous is because of the clash between these target cells. 
The basic analogy of a car ingesting a liquid does not carry over from the first semantic 
dependency into the second. The anomalous flavor of (27) could not be detected by 
looking at the semantic relations in the dependencies in isolation because one semantic 
relation is metaphorical and the other is literal. Neither relation is anomalous -- the 
anomaly comes from the interaction between the two relations. 
Figure 20 is a proposed representation for sentence (20). The left side of Figure 
20 shows the knowledge representation part of the sentence representation: a simple 
case-frame based representation of (20). The right side of Figure 20, within the grey 
partition, is the coherence representation component of the sentence representation: 
abridged semantic vectors for the two main semantic dependencies in (20). The upper 
semantic vector is the match of the target carl against the source animal1. The lower 
semantic vector is the match of the target gasoline1 against the source drink1, the 
noun sense. The upper abridged semantic vector indicates a metaphorical relation. 
The lower semantic vector also indicates a metaphorical relation though, as was noted 
earlier, "drinking gasoline" when interpreted in isolation is surely anomalous. 
The underlines in Figure 20 denote pointers linking the semantic vectors to the 
case frame. The grey vertical arrows show that the two semantic vectors are also linked 
79 
Computational Linguistics Volume 17, Number 1 
~rlnkl, \[a~t. 
¢ar~l. \[object, 
gasollnet \]\] 
Ii urce' \[\[explicit, ddnkl\], 
\[implicit, animal1\[\[\[, 
target, carl\[l, 
reference, 
network_path, 
\[estranged, \[1, \[anirnalt, car 1\]\]\]\], 
{cell_match, 
\[\[relevant, \[{sister, 
\[1, \[\[\[animall, drink1, drink1\[, \[carl, use2, gasolinet|\]\]\]\]\]\]\]\[\]\] 
\[\[\[source, l \[\[explicit, drink1\[, 
\[Implicit, drink1 \]\]\[, 
\[target, ~\]esollnel\]\], 
\[preference, 8,4 
\[\[network_path, \[dster, 
\[1, \[drink1, gasolinet \]\]\] \], 
(cell_match. 
\[{relevant, {{sister, '~ 
\[1, \[\[\[animal1. drink1, drink1\[, \[vehicle1. use2, gasolinel\]\]\]\]\]\]\]\]\]\]\] 
Figure 20 
Sentence representation for "The car drank gasoline" 
I ~£\[\[\[sou roe, {I \[\[explicit, drink1\[, 
It \[implicit, ~11\]\]\], t~ \[age, carl\[l, 
/ \[p ref erenc'~'. 
| \[\[network path, I| 
\[estranged, ,~ \[1, \[anirneH, carl\]\]\]\], 
II \[cell match, l~ \[\[relevant, 
\[drink1, 
{agent, c.~J \], 
\[object. 
coffee. 1\]\] \[\[\[source. 
{{explicit, driNkl\], \[implicit, drinkt\]\]\], 
\[target, coffee. 1\]\], 
\[preference \[\[network_path. 
\[ancestor, \[1, \[drinkt, 
coffeet\]m, \[cell_match, 
\[\[relevant, \[\[ancestor, 
{(sister, \[1, \[\[\[animal1. drink1, drink1\[. \[carl, use2, gasolinel\]\]\]\]\]\]\]|\]\]\] 
\[1, \[{\[anirnall, drink1, drink1\[, \[human_being1, drink1, coffeel\]\]\]\[\]\]\]\]\]l 
Figure 21 
Sentence representation for "The car drank coffee" 
together via the matches of their relevant cells. In those matches, the arrows are sense- 
network paths found between the elements of the two target cells. The network paths 
indicated in grey, that connect the two abridged semantic vectors, show processing of 
coherence representations. The particular network paths found (indicated in italics), a 
descendant path and two same "paths," show that the same relevant analogy is used 
in both semantic relations -- that both semantic relations involve a match between 
animals drinking potable liquids and vehicles (including cars) using gasoline -- hence 
sentence (20) as a whole is metaphorical. Figure 20 is therefore unlike any of the 
coherence representations shown previously, because it shows a representation of a 
metaphorical sentence, not just two isolated metaphorical relations. 
80 
Fass Discriminating Metonymy 
Compare Figure 20 with Figure 21, a sentence representation for (27). The upper 
semantic vector again indicates a metaphorical relation between carl and drink1. The 
lower semantic vector indicates a literal relation between drink1 and coffee1. What is 
important here is the match of relevant information discovered in the two semantic 
relations, as indicated by the three network paths. The paths found are two estranged 
paths and a. sister path, indicating that the relevant information found during the two 
semantic relations is different: in one semantic relation, information about animals 
drinking potable liquids is matched against cars using gasoline; in the other, the same 
information is matched against human beings drinking coffee; but cars using gasoline and 
human beings drinking coffee are quite different, hence sentence (27) is anomalous 
overall. 
Note that in Figures 20 and 21, the coherence representation part of the sentence 
representation is much larger than the knowledge representation part. The detailed 
"world knowledge" about carl, the verb sense drink1, gasoline1, and coffee1 are all on 
the right side. It is interesting to contrast the figures with early Conceptual Dependency 
(CD) diagrams such as those in Schank (1973) because, rather than the large and 
seemingly unlimited amounts of world knowledge that appear in CD diagrams, the 
two figures present only the world knowledge needed to discriminate the semantic 
relations in (20) and (27). 
7. Discussion and Conclusions 
This section reviews the material on metonymy and metaphor in Section 2 in light of 
the explanation of the met* method given in Sections 3-6. When compared with the AI 
work described in Section 2, the met* method has three main advantages. First, it con- 
tains a detailed treatment of metonymy. Second, it shows the interrelationship between 
metonymy, metaphor, literalness, and anomaly. Third, it has been programmed. 
Preference Semantics addresses the recognition of literal, metaphorical, and anoma- 
lous relations, but does not have a treatment of metonymy. In the case of Preference 
Semantics, the theory described in Wilks (1978) has not been implemented, though 
the projection algorithm was implemented (Modiano 1986) using some parts of CS to 
supply detail missing from Wilks' original specification. 
Gentner's (1983) Structure-Mapping Theory has no treatment of metonymy. The 
theory has been implemented in the Structure-Mapping Engine (Falkenhainer, Forbus 
and Gentner 1989) and some examples analyzed by it but not, to my knowledge, 
examples of metaphor or anomaly. 
Indurkhya's (1988) Constrained Semantic Transference theory of metaphor has no 
treatment of metonymy, anomaly or literalness. It has also not been implemented: see 
Indurkhya (1987) for reasons why. 
Hobbs and Martin (1987) offer a relatively shallow treatment of metonymy with- 
out, for instance, acknowledgement that metonymies can be driven from either the 
source or the target. Hobbs' "selective inferencing" approach to text interpretation has 
been applied to problems including lexical ambiguity (Hobbs 1977; 1982b; Hobbs and 
Martin 1987), metaphor (Hobbs 1977; 1983a; 1983b) and the "local pragmatics" phe- 
nomena of metonymy (Hobbs and Martin 1987), but not anomaly. To my knowledge, 
Hobbs has yet to produce a unified description of selective inferencing that shows 
in detail how lexical ambiguity is resolved or how the differences between metaphor, 
metonymy, and so on can be recognized. Hobbs" earlier papers include a series of 
programs -- SATE, DIANA, and DIANA-2 -- but the papers are not clear about what 
the programs can do. It is not clear, for example, whether any of the programs actually 
analyze any metaphors. 
81 
Computational Linguistics Volume 17, Number 1 
Martin's (1990) work is the only other computational approach to metaphor that 
has been implemented. However, the work does not have a treatment of metonymy. 
Martin's metaphor-maps, which are used to represent conventional metaphors and 
the conceptual information they contain, seem to complement semantic vectors of 
the extended kind described in Section 6. In Section 6, I argued that vectors need to 
record the conceptual information involved when finding mappings between a source 
and target. What metaphor-maps do is freeze (some of) the conceptual information 
involved in particular metaphorical relations. There is some theoretical convergence 
here between our approaches; it would be interesting to explore this further. 
Moreover, the metaphors studied so far in CS seem linked to certain conventional 
metaphors because certain types of ground have recurred, types which resemble Lakoff 
and Johnson's (1980) structural metaphors. Two types of ground have cropped up so 
far. 
Example 28 
"Time flies." 
The first is a use-up-a-resource metaphor which occurs in (20) and in (28) when 
viewed as noun-verb sentence. Both sentences are analyzed by meta5. Use-up-a-re- 
source resembles structural metaphors like TIME IS A RESOURCE and LABOR IS A 
RESOURCE which, according to Lakoff and Johnson (1980, p. 66), both employ the 
simple ontological metaphors of TIME IS A SUBSTANCE and AN ACTIVITY IS A 
SUBSTANCE: 
These two substance metaphors permit labor and time to be quantified -- that is, 
measured, conceived of as being progressively "used up," and assigned 
monetary values; they allow us to view time and labor as things that can be 
"used" for various ends. 
Example 29 
"The horse flew." 
The second type of ground is motion-through-a-medium, a type of ground dis- 
cussed by Russell (1976). This appears in (15) and (29), again both analyzed by meta5. 
Incidentally, it is worth noting that structural metaphors have proven more amena- 
ble to the met* method than other kinds tried. I assumed initially that orientational and 
ontological metaphors would be easier to analyze than structural metaphors because 
they were less complex. However, structural metaphors have proved easier to analyze, 
probably because structural metaphors contain more specific concepts such as "drink" 
and "plough," which are more simple to represent in a network structure (like the 
sense-network of CS) so that analogies can be found between those concepts. 
7.1 Relationship between Literalness and Nonliteralness 
We return here to Gibbs' point concerning the traditional notion of literal meaning that 
\[1\] all sentences have literal meanings that are entirely determined by the meanings 
of their component words and that \[2\] the literal meaning of a sentence is its meaning 
independent of context. Although \[1\] and \[2\] are both presently true of CS, there are 
means by which context can be introduced more actively into sentence interpretation. 
At present, the meaning of a sentence in CS -- whether literal or nonliteral -- 
is not derived entirely independently of context; however, the only context used is a 
82 
Fass Discriminating Metonymy 
limited notion of relevance which is generated by collation from within the sentence 
being analyzed: what is relevant is given by the sense of the main sentence verb. 
Nevertheless, because of this notion of relevance, contextual influence is present in 
semantic interpretation in CS. Moreover, the notion of relevance is recorded in semantic 
vectors (Figures 11 and 15) and the extended coherence representations discussed in 
Section 6. Hence, the processes and representations of CS possess basic equipment for 
handling further kinds of context. 
7.2 Relationship between Metonymy and Metaphor 
The met* method is consistent with the view that metaphor is based on similarity, 
whereas metonymy is based on contiguity (cf. Jakobsen and Halle 1956). Contiguity, 
readers may recall, refers to being connected or touching whereas similarity refers to 
being alike in essentials or having characteristics in common. The difference comes 
from what and how the conceptual information is related. 
Example 1 
"My car drinks gasoline." 
Let us consider what is related first. In metaphor, an aspect of one concept is similar 
to an aspect of another concept; e.g., in (1), an aspect of the concept for animal, that 
animals drink potable liquids, is similar to an aspect of another concept, that cars use 
gasoline. 
Example 2 
"The ham sandwich is waiting for his check." 
However, in metonymy, a whole concept is related to an aspect of another concept. 
For example, in (2) the metonymy is that the concept for ham sandwich is related to 
an aspect of another concept, for "the man who ate a ham sandwich." 
Regarding how that conceptual information is related: in the case of metaphor, 
the met* method assigns a central role to finding an analogy, and an analogy be- 
tween two terms is due to some underlying similarity between them (the ground), 
e.g., in the analogy that animals drinking potable liquids is like cars using gasoline, 
the underlying similarity is that both animals and cars ingest liquids. In an analogy, 
the relationship between aspects of two concepts is purely structural. In metonymies, 
however, the relationships are "knowledge-laden" connections, e.g., PART-WHOLE 
and CONTAINER-CONTENTS. 
So in summary, "similarity" in metaphor is understood to be based on struc- 
tural relationships between aspects of concepts, whereas "contiguity" in metonymy 
is based on knowledge-specific relationships between a concept and an aspect of an- 
other concept. These observations, I would argue, support the view that metonymy 
has primarily a referential function, allowing something to stand for something else -- 
a connection between a concept and an aspect of another concept. The observations 
also support the view that metaphor's primary function is understanding, allowing 
something to be conceived of in terms of something else: the role of analogy is espe- 
cially crucial to this function. 
7.3 Metonymy 
The treatment of metonymy permits chains of metonymies (Reddy 1979), and allows 
metonymies to co-occur with instances of either literalness, metaphor, or anomaly. 
83 
Computational Linguistics Volume 17, Number 1 
The kinds of inferences sought resemble the kinds of inferences that Yamanashi (1987) 
notes link sentences. An obvious direction in which to extend the present work is 
toward across-sentence inferences. 
Example 30 
"John drank from the faucet" (Lehnert 1978, p. 221). 
Example 31 
"John filled his canteen at the spring" (Ibid.). 
Metonymy seems closely related to the work on non-logical inferencing done by 
Schank (Schank 1973) and the Yale Group (Schank 1975; Schank and Abelson 1977; 
Schank and Riesbeck 1981). For example, Lehnert (1978) observes that just one infer- 
ence is required for understanding both (30) and (31). The inference, that water comes 
from the faucet in (30) and the spring in (31), is an instance of PRODUCER FOR 
PRODUCT in which the faucet and spring are PRODUCERs and water is the PROD- 
UCT. However, the inference is not a metonymy because it is from unused cases of the 
verbs 'drink' and 'fill' whereas metonymy only occurs in the presence of a violated 
selection restriction, that neither (30) nor (31) contain. 
7,4 Metaphor 
Metaphor recognition in the met* method is related to all four views of metaphor 
described in Section 2, consisting of: 
1. 
. 
. 
4. 
a contextual constraint violation, such as a preference violation -- as in 
the selection restrictions view; 
a set of "correspondences"-- rather like the system of commonplaces in 
the interaction view; 
a relevant analogy -- cf. the comparison and interaction views; with 
analogies that fall into patterns not unlike conceptual metaphors found 
in the conventional view. 
In CS, the presence of metaphor has been investigated in violations of preferences, 
a kind of lexical contextual constraint. Though clearly this is a small part of the picture, 
it seems worth establishing an extensive picture of preference violation and metaphor 
before moving on to other contextual constraints. 
Collation and the met* method have certain similarities with the comparison view 
of metaphor, especially in the cell matching process. The relevant analogies discovered 
in CS are indeed, to quote Tourangeau and Sternberg, "a comparison in which one 
term.., is asserted to bear a partial resemblance to something else." 
The collation process gives quite a clear picture of the ground and tension in a 
metaphor. The ground is the most specific statement that subsumes both statements 
that figure in the analogy, e.g., \[it1, ingest1, liquid1\] is the ground for the analogy 
involving \[animal1, drink1, drink1\] and \[carl, use2, gasoline1\] (see Figures 8 and 9). 
Moreover, the details of the process match well Aristotle's two basic principles for 
finding the ground of a metaphor in that both terms in a metaphorical relation belong 
84 
Fass Discriminating Metonymy 
to a common category (in the example above, the common categories are it1, ingest1, 
and liquid1) and an analogy is found between them. 
The collation process also takes care of many of the problems Tourangeau and 
Sternberg (1980) note with the comparison view. Regarding the problem that "every- 
thing shares some feature or category.., with everything else," CS is in agreement: the 
only significant combination of features in a metaphor are those involved in a relevant 
analogy. The problem that "the most obvious shared features are often irrelevant," i.e., 
that the most obvious shared features are irrelevant to a metaphor, is borne out by 
experience with CS -- for example, animals and cars share some basic physical object- 
like properties, but these have a minor role in understanding cars drinking. The met* 
method bears out another problem that, "even when a feature is relevant, it is often 
shared only metaphorically." Finally, with the problem that novel metaphors cannot 
be based on "extant similarities," -- the relevant analogies found in the met* method 
are not "extant" but have to be actively discovered. 
In Section 2, two main differences were noted between the interaction and com- 
parison views: first, that similarities are "created" in the interaction view, whereas only 
pre-existing similarities are found in the comparison view, and second, that a whole 
system of similarities are evoked in the interactions view, unlike the comparisons view, 
which focuses upon finding a single similarity. Regarding the first difference, I would 
argue that the difference is a mistaken one and that interaction theorists are simply us- 
ing a sophisticated form of comparison. This is quite evident when one examines, for 
example, the methods Tourangeau and Sternberg propose for relating features across 
domains in their theory. The second of Aristotle's basic principles is finding an anal- 
ogy, yet Tourangeau and Sternberg (1982, p. 218) themselves say that, "in a sense, we 
are proposing that metaphors are analogies that include both tenor and vehicle and 
their different domains as terms." 
And, of course, finding an analogy is central to the met* method on CS. 
Regarding the second difference, I would agree that finding a system of common- 
places is distinctive. However, the extensions to CS described in Section 6 move toward 
the direction of finding a system of commonplaces in that the deeper semantic vec- 
tors, and sentence representations shown in Figures 20 and 21 contain the information 
crucial to finding a system of commonplaces. Having identified the crucial analogy in 
(20), the deeper semantic vector contains the two pairs of matched relevant cells that 
provide the core analogy on which the metaphorical interpretation of (20) is built: 
\[\[animal1, drink1, drink1\], \[carl, use2, gasoline1\]\] 
\[\[animal1, drink1, drink1\], \[vehicle1, use2, gasoline1\]\] 
With this information at hand, the sense-frames for word senses in analogical corre- 
spondence -- the verb senses drink1 and use2, the noun senses animal1 and carl, 
animal1 and vehicle1, and drink1 and gasoline1 -- can be systematically expanded to 
uncover deeper commonplaces between animals and cars. 
In conclusion, the view of metonymy and metaphor in the met* method is consis- 
tent with much of the literature on these phenomena. The met* method is consistent 
with the view that the primary function of metaphor is understanding while that of 
metonymy is referential, like anaphora. Nevertheless, metonymy and metaphor do 
have much in common: both might be described as forms of "conceptual ellipsis," a 
shorthand way of expressing ideas. 
85 
Computational Linguistics Volume 17, Number 1 
The met* method in its present serial form recognizes literalness, metonymy, 
metaphor, and anomaly in the following order and by the following characteristics. 
• Literalness -- a satisfied preference. 
• Metonymy -- a successful conceptual inference. 
• Metaphor -- an underlying relevant analogy. 
• Anomaly -- none of the above. 
The above analysis also illustrates, I hope, why metonymy and metaphor are easily 
confused: both are nonliteral and are found through the discovery of some aspect (a 
property) shared by the source, a preference, and the target, in the above case a surface 
noun. The differences are (a) how that aspect is selected, (b) the operations that follow, 
(c) the effect those operations produce, and (d) subsequent processing. 
In the case of metonymy, (a) the selected aspect forms a regular semantic relation- 
ship with a property from the target; (b) there is substitution, i.e., replacement of one 
concept with another; (c) hence the apparent referential function of metonymy; and 
(d) is unclear at present. 
In the case of metaphor, (a) the selected aspect is relevant; (b) forms an analogy 
with another aspect from the target; and (c) the effect is of surprise discovery of 
similarity between the two concepts; and (d) the discovered analogy is used to unearth 
further similarities between the two concepts (i.e, to deepen the analogy) and to guide 
subsequent sentence interpretation. Moreover, the view of metaphor in CS contains 
elements of the selection restrictions view, the comparisons view, and the interactions 
view of metaphor. 
It should be emphasized that the met* method has only been applied to a small 
set of English sentences. Metonymy interpretation has been investigated only for 
adjective-noun and subject-verb-object constructions; metaphor interpretation, only for 
the latter. The best avenue for progress with the met* method appears to be the exten- 
sions to metaphor interpretation described in Section 6. In the meantime I am looking 
for sentences that contain semantic relations consisting of a metonymy (or chain of 
metonymies) followed by a metaphor. 
Example 32 
"America believes in democracy" (Hobbs 1983b, p. 134). 
On a related point, some sentences are interesting in this respect because they 
have either a metaphorical or metonymic interpretation. In (32), for example, "Are 
we viewing America metaphorically as something which can believe, or are we using 
it metonymically to refer to the typical inhabitant, or the majority of inhabitants, of 
America?" (Ibid., p. 135). 
Example 33 
"Prussia invaded France in 1870." 
Sentence (33), which was discussed in a group working on beliefs at the CRL (see 
Acknowledgments), also has separate metonymic and metaphorical interpretations. 
The key semantic relation is between 'Prussia' and 'invade.' The relation is nonliteral 
because 'army' is the expected agent of 'invade' and 'Prussia' is a country, not an army. 
What, then, is the semantic relation between 'Prussia' and 'army'? One possibility is 
86 
Fass Discriminating Metonymy 
that a chain of metonymies is involved, that the army is controlled by the govern- 
ment which also controls Prussia. A second possibility is that Prussia is understood 
metaphorically as being an animate thing that extends itself into France. 
Acknowledgments 
I would like to thank the many people at 
the Cognitive Studies Centre, University of 
Essex; the Computing Research Laboratory, 
New Mexico State University; and the 
Centre for Systems Science, Simon Fraser 
University, with whom I have had fruitful 
discussions over the years, especially those 
in the beliefs group at the CRL (Afzal 
Ballim, John Barnden, Sylvia Candelaria de 
Ram, and Yorick Wilks); others at the CRL 
(including Xiuming Huang, David Farwell, 
and Eric Dietrich); and colleagues in the 
CSS who made helpful comments on earlier 
drafts of this paper (Chris Groeneboer, Gary 
Hall, and Carl Vogel). A special word of 
thanks for the help given by Yorick Wilks, 
the director of the CRL, and Nick Cercone, 
the director of the CSS. I also gratefully 
acknowledge the financial support provided 
by SERC Project GR/C/68828 while at 
Essex, by the New Mexico State Legislature 
while at NMSU, and by the Advanced 
Systems Institute and the Centre for 
Systems Science while at SFU. 
References 
Amsler, Robert A. (1980). The structure of the 
Merriam- Webster Pocket Dictionary. 
Doctoral dissertation TR-164, University 
of Texas, Austin, Texas. 
Basso, K.H. (1976). " 'Wise words' of the 
Western Apache: metaphor and semantic 
theory." In Meaning in anthropology, edited 
by K.H. Basso and H. Selby, New Mexico 
Press. 
Beardsley, M. (1962). "The metaphorical 
twist." Philosophy and Phenomenological 
Research, 22:293-307. 
Bickerton, Derek (1969). "Prolegomena to a 
linguistic theory of metaphor." 
Foundations of Language, 5:36-51. 
Black, Max (1979). "More about metaphor." 
In Metaphor and thought, edited by 
A. Ortony, 19-43. Cambridge University 
Press. 
Black, Max (1962). Models and Metaphors. 
Cornell University Press. 
Boguraev, Branimir K. (1979). "Automatic 
resolution of linguistic ambiguities." 
Technical Report No. 11, University of 
Cambridge Computer Laboratory, 
Cambridge, England. 
Campbell, P. (1975). "Metaphor and 
linguistic theory." Quarterly Journal of 
Speech, 61:1-12. 
Chomsky, Noam (1964). "Degrees of 
grammaticalness." In The Structure of 
Language: Readings in the Philosophy of 
Language, edited by J.A. Fodor and 
J.J. Katz, 384-389, Prentice-Hall. 
Van Dijk, T. (1975). "Formal semantics of 
metaphorical discourse." Poetics, 
4:173-198. 
Dubois, J.; Edeline, E; Klinkenberg, J.-M.; 
Minguet, P.; Pire, E; and Trinon, H. (1970). 
Rhdtorique Gdndrale. Larousse. Published 
as: A General Rhetoric, translated by 
P.B. Burrell and E.M. Slotkin, 1981, The 
John Hopkins University Press. 
Falkenhainer, Brian; Forbus, Kenneth D.; 
and Gentner, Dedre (1986). "The 
structure-mapping engine: algorithm and 
examples." Artificial Intelligence, 41:1-63. 
Fass, Dan C. (1986). "Collative semantics: an 
approach to coherence." Technical Report 
MCCS-86-56, Computing Research 
Laboratory, June 1986. 
Fass, Dan C. (1987). "Semantic relations, 
metonymy, and lexical ambiguity 
resolution: a coherence-based account." In 
Proceedings, 9th Annual Cognitive Science 
Society Conference, University of 
Washington, Seattle, Washington: 575-586. 
Fass, Dan C. (1988a). Collative semantics: a 
semantics for natural language processing. 
Doctoral dissertation, New Mexico State 
University, New Mexico. 
Fass, Dan C. (1988b). "An account of 
coherence, semantic relations, metonymy, 
and lexical ambiguity resolution." In 
Lexical Ambiguity Resolution: Perspectives 
from Psycholinguistics, Neuropsychology, and 
Artificial Intelligence, edited by Steven 
L. Small, Garrison W. Cottrell, and 
Michael K. Tanenhaus, Morgan 
Kaufmann, 151-178. 
Fass, Dan C. (1988c). "Metonymy and 
metaphor: what's the difference?" In 
Proceedings, 12th International Conference on 
Computational Linguistics (COLING-88), 
Budapest, Hungary, 177-181. 
Fass, Dan C. (1988d). "Collative semantics: a 
study in the discrimination of meaning." 
Technical Report CSS/LCCR TR 88-24, 
Centre for Systems Science, Simon Fraser 
University. 
Fass, Dan C. (1989a). "Lexical semantic 
constraints." Technical Report CSS/LCCR 
TR 89-11, Centre for Systems Science, 
Simon Fraser University. (To appear in 
87 
Computational Linguistics Volume 17, Number 1 
Semantics and the Lexicon, edited by James 
Pustejovsky. Kluwer Academic 
Publishers.) 
Fass, Dan C. (1989b). "Four general 
representations and processes for use in 
problem solving." In Knowledge Based 
Computer Systems (Proceedings of KBCS "89, 
Bombay, India), edited by S. Ramani, 
R. Chandrasekar, and K.S.R. Anjeyulu, 
Narosa Publishing House, 169-178. 
Genette, Gerard (1970). "La rhetorique 
restreinte." Communications, 16:158-171. 
Gentner, Dedre (1983). "Structure mapping: 
a theoretical framework for analogy." 
Cognitive Science, 7:155-170. 
Gibbs, Raymond W. Jr. (1984). "Literal 
meaning and psychological theory." 
Cognitive Science, 8:275-304. 
Grosz, Barbara J.; Appelt, Douglas E.; 
Martin, Paul; and Pereira, Fernando C.N. 
(1987). "TEAM: An experiment in the 
design of transportable natural-language 
interfaces." Artificial Intelligence, 
32(2):173-243. 
Harris, R. (1976). "Comprehension of 
metaphor: a test of a two-stage processing 
model." Bulletin of the Psychonomic Society, 
8:321-324. 
Hesse, M. (1966). Models and Analogies in 
Science. Notre Dame University Press. 
Hobbs, Jerry R. (1977). "Coherence and 
interpretation in English texts." In 
Proceedings, 5th International Joint 
Conference on Artificial Intelligence 
(IJCAI-77), Cambridge, Massachusetts, 
110-116. 
Hobbs, Jerry R. (1979). "Coherence and 
coreference." Cognitive Science, 3:67-90. 
Hobbs, Jerry R. (1980). "Selective 
inferencing." In Proceedings, 3rd National 
Conference of the Canadian Society for 
Computational Studies of Intelligence 
(CSCSI-3), Victoria, British Columbia, 
Canada, 101-114. 
Hobbs, Jerry R. (1982a). "Towards an 
understanding of coherence in discourse." 
In Strategies for Natural Language 
Processing, edited by Wendy G. Lehnert 
and Martin H. Ringle, Erlbaum 
Associates, 223-243. 
Hobbs, Jerry R. (1982b). "Representing 
ambiguity." In Proceedings, 1st West Coast 
Conference on Formal Linguistics, Stanford, 
California, 15-28. 
Hobbs, Jerry R. (1983a). "Metaphor 
interpretation as selective inferencing: 
cognitive processes in understanding 
metaphor (Part 1)." Empirical Studies of the 
Arts, 1:17-33. 
Hobbs, Jerry R. (1983b). "Metaphor 
interpretation as selective inferencing: 
cognitive processes in understanding 
metaphor (Part 2)." Empirical Studies of the 
Arts, 1:125-141. 
Hobbs, Jerry R. (1985). "On the coherence 
and structure of discourse." Technical 
Report No. CSLI-85-37, Center for the 
Study of Language and Information 
(CSLI), Stanford University. 
Hobbs, Jerry R. and Martin, Paul (1987). 
"Local Pragmatics." In Proceedings, lOth 
International Joint Conference on Artificial 
Intelligence (IJCAI-87), Milan, Italy, 
520-523. 
Huang, Xiuming (1985). "Machine 
translation in the SDCG (Semantic 
Definite Clause Grammars) formalism." 
In Proceedings, Conference on Theoretical and 
Methodological Issues in Machine Translation 
of Natural Languages, Colgate University, 
New York, New York, 135-144. 
Indurkhya, Bipin (1985). "A computational 
theory of metaphor comprehension and 
analogical reasoning." Technical Report 
#85/001, Boston University. 
Indurkhya, Bipin (1987). "Approximate 
semantic transference: a computational 
theory of metaphors and analogies." 
Cognitive Science, 11(4):445-480. 
Indurkhya, Bipin (1988). "Constrained 
semantic transference: a formal theory of 
metaphors." In Analogica: Proceedings of 
the First Workshop on Analogical Reasoning, 
edited by Armand Prieditis, Morgan 
Kaufmann (Pitman Research Notes in 
Artificial Intelligence), Los Altos, 
California, 129-157. 
Jacobs, Paul S. (1985). "A knowledge-based 
approach to language production." 
Technical Report No. UCB/CSD 86/254, 
Computer Science Division (EECS), 
University of California, Berkeley. 
Jakobsen, Roman and Halle, Morris (1956). 
Fundamentals of Language. Mouton. 
Johnson, Mark (1980). "A philosophical 
perspective on the problems of 
metaphor." In Cognition and Figurative 
Language, edited by Richard P. Honeck 
and Robert R. Hoffman, 47-68, Erlbaum 
Associates. 
Katz, Jerrold J. (1964). "Analyticity and 
contradiction in natural language." In The 
Structure of Language: Readings in the Phil- 
osophy of Language, edited by J.A. Fodor 
and J.J. Katz, 519-543, Prentice-Hall. 
Lakoff, George and Johnson, Mark (1980). 
Metaphors We Live By. Chicago University 
Press. 
Lees, R.B. (1960). "The grammar of English 
nominalizations." International Journal of 
American Linguistics, 26(3). 
88 
Fass Discriminating Metonymy 
Lenat, Douglas B. and Guha, R. V. (1990). 
Building Large Knowledge-Based Systems: 
Representation and Inference in the CYC 
Project. Addison-Wesley. 
Levin, Samuel R. (1977). The Semantics of 
Metaphor. John Hopkins University Press. 
Loewenberg, Ina (1975). "Identifying 
metaphors." Foundations of Language, 
12:315-338. 
Malgady, R. and Johnson, M. (1976). 
"Modifiers in metaphors: effects of 
constituent phrase similarity on the 
interpretation of figurative sentences." 
Journal of Psycholinguistic Research, 5:43-52. 
Martin, James H. (1990). A Computational 
Model of Metaphor Interpretation. Academic 
Press. 
Martin, James H. (1988). "A computational 
theory of metaphor." Technical Report 
No. UCB/CSD 88/465, Computer Science 
Division (EECS), University of California, 
Berkeley. 
Martin, James H. (1985). "Knowledge 
acquisition through natural language 
dialogue." In Proceedings, 2nd Annual 
Conference on Artificial Intelligence 
Applications, Miami, Florida. 
Matthews, R. J. (1971). "Concerning a 
'linguistic theory' of metaphor." 
Foundations of Language, 7:413-425. 
Miles, J. (1967). Style and Proportion. Little, 
Brown and Co. 
Mish, Frederick C. (1986). Webster's Ninth 
New Collegiate Dictionary. Merriam- 
Webster Inc. 
Modiano, Nicole (1986). "Two procedures 
for sense projection." Unpublished ms., 
Computing Research Laboratory, New 
Mexico State University. 
Van Noppen, Jean-Pierre; De Knop, S.; and 
Jongen, R. (1985). Metaphor: A Bibliography 
of post-1970 Publications. Amsterdam 
Studies in the Theory and History of 
Linguistic Science (Series V: Library and 
Information Sources in Linguistics), 
Volume 17, John Benjamins Publishing 
Company. 
Ortony, Andrew (1980). "Some 
psycholinguistic aspects of metaphor." In 
Cognition and Figurative Language, edited 
by Richard P. Honeck and Robert 
R. Hoffman, 69-83, Erlbaum Associates. 
Ortony, Andrew (1979). "Beyond literal 
similarity." Psychological Review, 
86:161-180. 
Percy, Walker (1954). The Message in the 
Bottle. Farrar, Strauss, and Giroux. 
Quillian, M. Ross (1968). "Semantic 
memory." In Semantic Information 
Processing, edited by Marvin Minsky, 
216-270, MIT Press. 
Perrine, Lawrence (1971). "Psychological 
forms of metaphor." College English, 
33:125-138. 
Reddy, Michael J. (1979). "The conduit 
metaphor -- a case of frame conflict in 
our language about language." In 
Metaphor and Thought, edited by Andrew 
Ortony, 284-324, Cambridge University 
Press. 
Reddy, Michael J. (1969). "A semantic 
approach to metaphor." In Chicago 
Linguistic Society Collected Papers, Chicago 
University Press, 240-251. 
Richards, Ivor A. (1936). The Philosophy of 
Rhetoric. Oxford University Press. 
Russell, Sylvia Weber (1976). "Computer 
understanding of metaphorically used 
verbs." American J~ournal of Computational 
Linguistics, Microfiche 44. 
Schank, Roger C. (1975). "The structure of 
episodes in memory." In Representation 
and Understanding, edited by Daniel 
G. Bobrow and Allan Collins, 237-272. 
Academic Press. 
Schank, Roger C. (1973). "Identification of 
conceptualizations underlying natural 
language." In Computer Models of Thought 
and Language, edited by Roger C. Schank 
and Kenneth M. Colby, 187-247. 
W.H. Freeman. 
Shibles, Warren (1971). Metaphor: An 
Annotated Bibliography and History. The 
Language Press. 
Stern, Gustaf (1968). (first published in 
Sweden 1931) Meaning and Changes of 
Meaning. Indiana University Press. 
Tourangeau, Roger and Sternberg, Robert J. 
(1982). "Understanding and appreciating 
metaphors." Cognition, 11:203-244. 
Tversky, Amos (1977). "Features of 
similarity." Psychological Review, 
84:327-352. 
Ullmann, Stephen (1962). Semantics: An 
Introduction to the Science of Meaning. Basil 
Blackwell & Mott Ltd. 
Waldron, Ronald A. (1967). Sense and Sense 
Development. Andre Deutsch. 
Weiner, E. Judith (1985). "Solving the 
containment problem for figurative 
language." International Journal of 
Man-Machine Studies, 23:527-537. 
Weiner, E. Judith (1984). "A knowledge 
representation approach to understanding 
metaphors." American Journal of 
Computational Linguistics, 10:1-14. 
Wheelwright, P. (1962). Metaphor and Reality. 
Indiana University Press. 
Wilks, Yorick A. (1978). "Making 
preferences more active." Artificial 
Intelligence, 11:197-223. 
89 
Computational Linguistics Volume 17, Number 1 
Wilks, Yorick A. (1975b). "An intelligent 
analyzer and understander for English." 
Communications of the ACM, 18:264-274. 
Wilks,Yorick A. (1975a). "A preferential 
pattern-seeking semantics for natural 
language inference." Artificial Intelligence, 
6:53-74. 
Wilks, Yorick A. (1973). "An artificial 
intelligence approach to machine 
translation." In Computer Models of 
Thought and Language, edited by Roger 
C. Schank and Kenneth M. Colby, 
114-151, W.H. Freeman. 
Wilks, Yorick A. and Fass, Dan C. (In press). 
"The preference semantics family." To 
appear in Computers and Mathematics with 
Applications, Special Issue on Semantic 
Networks in Artificial Intelligence. 
Winston, Patrick H. (1980). "Learning by 
creatifying transfer frames." In Artificial 
Intelligence: An MIT Perspective, Volume 1, 
edited by Patrick H. Winston and Richard 
H. Brown, 347-376, MIT Press. 
Yamanashi, Masa-aki (1987). "Metonymic 
interpretation and associative processes in 
natural language." In Language and 
Artificial Intelligence (Proceedings of an 
International Symposium on Language and 
Artificial Intelligence held in Kyoto, Japan, 
16-21 March 1986), edited by Makoto 
Nagao, 77-86, Elsevier Science Publishers. 
Glossary of Main Terms 
Anomalous relation: a semantic relation 
indicated by a violated preference and the 
absence of a relevant analogy \[see 
Section 3\]. 
Assertion: a word sense-based contextual 
constraint in which semantic information is 
imposed onto the local context of a word 
sense \[Section 3\]. 
Collation: a process that discriminates the 
semantic relation(s) between two word 
senses by matching the sense-frames for the 
word senses \[Section 4\]. 
Literal relation: a semantic relation indicated 
by a satisfied preference \[Section 3\]. 
Metaphor: a trope in which one entity is 
used to view another entity to which it 
bears a partial resemblance \[Sections 2-7\]. 
Metaphorical relation: a semantic relation 
indicated by a violated preference and the 
presence of a relevant analogy \[Section 3\]. 
Metonymy: a trope in which one entity is 
used to refer to another that is related to it 
\[Sections 2-7\]. 
Metonymic relation: a semantic relation 
indicated by failure to satisfy a preference 
and the presence of one or more conceptual 
relationships like PART-WHOLE \[Section 3\]. 
Preference: a word sense-based contextual 
constraint in which semantic information 
restricts the local context of a word sense 
\[Section 3\]. 
Screening: a process that resolves lexical 
ambiguity by choosing among semantic 
vectors on the basis of rank orderings 
among semantic relations and a measure of 
conceptual similarity \[Section 4\]. 
Semantic relation: the basis of literalness, 
metonymy, metaphor, etc.; found by 
evaluating lexical semantic constraints in 
sentences \[Section 3\]. 
Semantic vector: a data structure that 
represents semantic relations by recording 
the matches produced by collation 
\[Section 4\]. 
Sense-frame: a framelike data structure that 
represents a word sense and is composed of 
other word senses having their own 
sense-frames \[Section 4\]. 
Sense-network: a semantic network of word 
senses formed from information in 
sense-frames \[Section 4\]. 
Source: the lexical item in a semantic 
relation whose contextual constraint(s) are 
enforced \[Section 3\]. 
Target: the lexical item in a semantic relation 
on which contextual constraint(s) are 
applied \[Section 3\]. 
Trope: the technical term for a nonliteral 
figure of speech, e.g., metaphor, metonymy, 
simile, understatement (litotes), 
overstatement (hyperbole), and irony 
\[Section 1\]. 
90 
