Figure 1. Deese's list of 39 contrasting pairs
Egan begins by noting that the term antonym was coined as an opposite for synonym by C.J. Smith in 1867, who explained it in this way: "Words which agree in expressing one or more characteristic ideas in common he has called Synonyms, those which negative one or more ideas he has called Antonyms."1 Egan points out that "negative" is quite vague, making it hard to know exactly what Smith had in mind, so she goes on to consider modern dictionary definitions of antonym. She finds that all of them contain the word opposite, but she points out that opposite itself has a wide range of meanings. She says, for example:
Opposition is a relation involved when two things are placed that: (1) they may be connected in a straight line...(as, opposite windows); (2) they lie at either ends of an axis, diameter, or the like (as, opposite points of the globe) [Egan lists 5 more senses, ending with] (8) they represent the obverse and the reverse (as, the opposite sides of a coin). (Egan 1968, 26a)
Egan says of this wide range of things which can be considered opposite in some way, "One can go no further than to say that opposite represents a setting of one thing against another so as to sharpen their differences or reveal their differences."
Looking at the kinds of words that are generally listed as antonyms in dictionaries and thesauri, Egan identifies several types of opposites, and they generally correspond to the types of opposites described above in sections 1.2.1 and 1.2.2. However, rather than focusing on the features which distinguish the various types, Egan looks for what is common to all of them by looking at what distinguishes all of these typical opposites from the near-opposites (contrasted terms). She finds that near-opposites "differ sharply in some part, but not all parts, of their meaning. They do not clash full force." More specifically, she says:
Put more simply, they [contrasted terms, i.e., near-opposites] differ (1) in their range of application or applicability, one being general, the other specific, or one being more inclusive or less inclusive than the other, and (2) in their depth of meaning--that is, in the number and quality of implications contained in the terms. (Egan 1968, 29a)
She gives a few examples of this. One is the pair rich/destitute. She says that the meaning of rich is much broader and vaguer than that of the much more explicit destitute. She says "rich suggests more possessions than one needs," and thus the best opposite for rich is poor, which, she says, "suggests fewer possessions than one requires and so negates the full meaning of the other [that is, rich]. Unlike poor, though, destitute has some further implications, and these implications are not matched by contrasting implications in rich. She says:
Opulent could be opposed to destitute in narrowness and explicitness of meaning, for destitute suggests the miserable condition where one is deprived of all that is needed for bare existence and opulent the felicitous condition where everything that is desired is possessed in abundance. Though rich and poor come close together (the dividing line being marked by such a word as comfortable) and destitute and opulent are very far apart, being in fact "diametrical opposites," each represents the negation of the other. (Egan 1968, 29a)Egan's observations are similar to Cruse's comments on purity of opposition and Lehrer and Lehrer's ideas of similarity of distribution, but unlike them, she takes these observations as central to understanding the nature of antonymy and bases her definition of antonymy on them. Her definition is:
An antonym is a word so opposed in meaning to another word, its equal in breadth or range of application, that it negates or nullifies every single one of its implications. (Egan 1968, 30a)This definition gives a clear answer to the question of what makes two words antonyms--they are equal in range of application and yet are opposed in meaning--and at the same time provides a natural explanation of why two words can contrast in meaning yet still not be antonyms--they may be different in their range of application. Thus, Egan's definition may be easier to apply to actual cases than the theories of antonymy described in section 1.2, and I have tried to do so in the three case studies that make up the main part of this dissertation.
In the case studies, I take up the notion of antonyms as words which are "equal in breadth or range of application", looking carefully to see if there are in fact differences between pairs of antonyms and pairs of near-opposites in terms of semantic range. For example, in the first case study of big, little, large and small, if Egan is right, it would be expected that the antonyms large and small are equal (which apparently means "identical" ) in their semantic range, while the pair of near-opposites large and little are quite different in their semantic ranges. But when applying Egan's definition in these terms, however, it immediately becomes clear that the definition as it stands is unlikely to be entirely correct.
First of all, given the fact that there are no perfect synonyms (words that absolutely identical in terms of register, range of application and implications), it would seem unlikely that there are any antonyms that contrast perfectly. In fact, the case of happy and its two antonyms, sad and unhappy (the subject of the third case study), demonstrates this point. Unhappy and sad are near, but not perfect synonyms, so they are not exactly the same in terms of range of application or implications. Happy cannot be "equal in breadth or range of application" to both unhappy and sad simultaneously when these two words themselves are not exactly the same. However, I think that Egan has identified some very important factors which, with some modification, can go a long way toward explaining the phenomena of antonymy. In the case studies, I show how "range of application" can be used to explain many of the phenomena associated with antonymy; specifically, I show that antonyms, while not completely identical in their range of application, share a great deal of semantic range and have more shared range in common than near-opposites do.
The clang phenomenon (which she calls "clash") also may require some modification of Egan's definition. She says that two words "clash satisfactorily" when they are equal in range of application and all of their implications contrast, and she gives a few examples of this, including the pairs rich/poor and destitute/opulent in the quotation above. Although it seems that for Egan, both of these pairs are equally contrasting, I believe that most native speakers (including myself) feel a strong sense of "clash" with the first pair, which seem to be prototypical antonyms, but only a weak sense with the second. Destitute and opulent seem more like near-opposites to me.2 Before the case studies then, it will be useful to look at the research into antonymy done by psychologists and psycholinguists since they have looked more deeply into the clang phenomenon.
The distribution of responses evoked by a particular word as stimulus defines the meaning of that word. The only ways in which such meaning can be specified are (1) by the nature of the distribution itself and (2) by the relation that distribution has to distributions of responses to other linguistic forms as stimuli. (Deese 1965, 43)In other words, Deese investigates the meaning of a word by asking many different respondents to give a response (a free association) to it. He then looks at the range of responses and compares it to the range of responses given for other target words. Deese feels that he can measure the similarity of two words by comparing the sets of responses, so that "...two stimuli may be said to have the same associative meaning when the distributions are identical, and they may be said to share meaning to the extent that these distributions agree" (Deese 1965, 45-46)
Since Deese is examining the "associational" meaning of words, his definitions are somewhat different than the kinds of definitions usually given by linguists and lexicographers. Consider, for example, some of the definitions from the Associative Dictionary at the end of his book (Deese 1965): "Rabbit: Animal, Dog" and "Bronze: Metal. Gold, Copper." While rabbit may cause people to think of dogs because dogs are often used in hunting rabbits or because dogs like to chase rabbits, it is unlikely that a dictionary definition of rabbit would include this information. Likewise, gold is not likely to be listed in a dictionary definition of bronze.
However, Deese does uncover some of the same kinds of semantic relations linguists describe, including antonymy. A few nouns and verbs produce antonyms as responses, but he notes that many adjectives, especially the common ones, are strongly associated with their antonyms. This fact is so striking to Deese that he proposes that the entire adjectival vocabulary of English might be organized in the mind around the relation of antonymy. Frequently occurring adjectives are associated with their opposites, and the remaining adjectives can be characterized in terms of their associations with the common adjectives. As an example of the latter, in the Associative Dictionary, famous is said to share the meaning of good and rich, which are in turn defined in terms of their opposites bad and poor.
Given this proposal, Deese says that it is important to find a way of determining whether or not two words are antonyms which does not rely on subjective judgments. He makes this suggestion:
The notion of contrast implies that one member of a pair of words should have its associative meaning most strongly determined by the other member of the pair and that the meaning should be reciprocal. Contrasting words, in brief, are words that are defined in terms of each other... (Deese 1965, 122)
In other words, Deese suggests, because opposites are so strongly associated, if a person is given one member of an antonym pair and asked to give another word in response, he/she is highly likely respond with the antonym. This kind of behavioral response can serve as an identification procedure for antonyms.
Deese used a simple word association test to identify a set of basic antonyms which he could then use to define the meanings of the less frequent adjectives which do not necessarily have antonyms. He says that in this kind of association test, "[A]ll pairs of words in which the stimulus to one is the most frequently occurring response to the other are the best candidates for being reciprocal pairs." That is, if most people respond with hot when they are given cold and if most respond with cold when they are given hot, then hot and cold would be identified as a reciprocal, contrasting pair, i.e., as antonyms.3 Starting with a set of 278 commonly occurring adjectives,4 Deese used a free association test given to undergraduate students to arrive at a set of 39 basic contrasting pairs. These are listed below in Figure 1.
Most of the pairs on this list conform to the standard described above; that is, each word was the most frequent response to the other, but in a few cases (those marked with an asterisk), one of the words was the second most common response; this occurred when one word had two opposites, e.g., easy/hard and hard/soft. Hard was the most common response for both easy and soft; soft was the most frequent response for hard and easy was the second most frequent.
Figure 1. Deese's list of 39 contrasting pairs
There is nothing surprising in the list of opposites produced by this association test; that is, all of the pairs seem like very good, prototypical, opposites.5 Indeed, as Deese says,
The most important aspect of this table [of the 39 contrasting pairs] is the extent to which the principle which generated it produces pairs which do not violate intuitive expectation. (Deese 1965, 122)
Deese notes that his method of identifying antonyms may fail to produce all the English pairs which seem to be opposites, however. He gives the example of brave and cowardly; both of these words were in the original sample, but the undergraduates did not respond to brave with cowardly or vice-versa. Deese gives this explanation:
It is very possible that those antonyms not appearing by the criterion--allowing for the possibility of sampling error--are not in the cognitive sense defined by each other. We do not think of brave as being the opposite of cowardly; we think of brave as being strong and cowardly as being afraid. Cowardly, brave, and afraid are all related in a basic way to the pair strong-weak. (Deese 1965, 123)
Brave and cowardly seem to me to be examples of near-opposites, words which certainly contrast in meaning but lack the "clang" effect; this suggests that the strong association between antonyms which Deese uncovered may be behind the clang phenomenon. All of the adjectives on Deese's list of 39 "sound" like antonyms, and most native speakers would readily identify them as such.
In this work, Deese's concern is simply associations between words. He does not look any deeper into the semantics of the kinds of associations he discovers, and so although his work goes a long way toward answering one of my initial questions, the question about the basis of the clang phenomenon, it does not really touch on the others. Deese does not try to explain why two particular words are associated as antonyms, or why some words have more than one antonym while others have none. But some psycholinguists, in particular, George Miller and his colleagues working on the WordNet project at Princeton, have tried to deal with these questions. They take Deese's idea that all adjectival meanings can be organized around a set of basic antonyms and develop it into a more complete model of antonymy. This work is presented in the next section.
They begin by pointing out that although speakers can easily recognize antonyms, it is hard to define exactly what an antonym is. They say that although lexicographers such as Egan identify many different types of opposites, in general, they can account for the various types of antonyms with just two assumptions:
(1) the semantic role of adjectives is to express values of attributes (properties, qualities, features, or dimensions); and (2) attributes are bipolar in nature. (Gross, Fischer and Miller 1988, 2)
It seems from these assumptions, then, that Gross, Fischer, and Miller believe that antonymy in a sense reflects the world--the bipolarity found in antonymy reflects the fact that many attributes in the world are bipolar. They give as examples the attribute of size (with the bipolar values large and small), the attribute of sex (which has two values, female and male) and what they call the "evaluative" attribute, which ranges from good to bad. They note that there are some exceptional kinds of attributes, e.g., color, which has many values (red, yellow, green, etc.), but they say that by and large, most attributes are bipolar. They note that several different types of opposites, e.g., Egan's complementaries, contradictories and reverse terms, all fit these two assumptions. In other words, all types of opposites involve bipolar attributes.
Gross, Fischer, and Miller go on to say that in building a model which will organize the meanings of most adjectives on the basis of antonymy, they will need to account for the fact that many adjectives do not seem to have antonyms. This includes the kinds of adjectives which Egan called "contrasted" or "loosely contrasted." As an example, they say that the adjective musty contrasts in meaning with fresh, but that the antonym of fresh is stale, not musty. However, they note, musty is similar in meaning to stale, so they propose a third assumption to account for this. They say:
(3) Any adjective with no direct antonym will be similar in meaning to some adjective with a direct antonym. (Gross, Fischer and Miller 1988, 3)
Thus, in theory, all adjectives can be accounted for. Many adjectives have clear examples of antonyms--these are called direct antonyms--and the adjectives which do not have a direct antonym are synonyms of adjectives which do, and thus can participate in antonymy indirectly. Adjective pairs which are thus "mediated" by a pair of direct antonyms are called indirect antonyms. As another example, Gross, Fischer, and Miller say, "The incompatible pair vigilant/careless is mediated, presumably, by the direct antonym careful." That is, in their model, the antonyms careful and careless are directly linked as antonyms, and vigilant is linked to careful as a synonym, thereby being indirectly linked to careless.7
Although synonymy and antonymy are both ways of linking words together in the lexicon, Gross, Fischer, and Miller argue that there is a crucial difference between them. They say that while synonymy is "a relation between lexical concepts," antonymy is "a relation between words, not concepts." This point is somewhat difficult to understand, but I think it can be made clearer with one of their examples, one which involves the basic antonym pair wet and dry and the several adjectives which are related to these antonyms. These include damp, moist, and waterlogged, which they consider to be synonyms of wet, and baked, arid, and parched, which they consider to be synonyms of dry. Gross, Fischer, and Miller say that the synonyms of wet form a conceptual "cluster" in semantic memory, and likewise the synonyms of dry. These two semantic clusters are held together by the fact that they share a bipolar attribute. Although these two conceptual clusters contrast in meaning, they argue that antonymy is a relationship between particular words, so that it cannot be said that the clusters themselves are antonyms. Instead, wet and dry are somehow selected as the "labels" for the two poles of the attribute, and the fact that they are chosen as labels makes them direct antonyms.
As further evidence that antonymy is a relationship between words, not concepts, Gross, Fischer, and Miller note that most antonyms in English are formed by prefixes such as un- and in-, saying, "When an antonym is formed by affixation, the direct bond between the two adjectives is apparent from their shared root."8
Next Gross, Fischer, and Miller presented these various pairs to a new set of subjects, telling them it was a "color study." First, without using the words antonym or opposite, they trained the subjects to recognize an unnamed "relation" by showing them two lists. One list had pairs of opposites--including direct pairs and both types of indirect pairs--and the other list had the pairs of unrelated adjectives and color terms. They asked the subjects if they could identify the relation that held between all the things on the one list but not the other. When the subjects said that they recognized the relation, they were presented with pairs of words on a computer screen. As each pair appeared, the subject was asked to press one key if the pair showed the relation they had been trained to recognize, and another key if it did not. The reaction time was measured.
This experiment found that subjects were able to identify pairs of direct antonyms faster than pairs of indirect antonyms, and that they identified the near indirect antonyms faster than the far ones. Gross, Fischer, and Miller conclude that these results are compatible with their model--if direct antonyms are immediately linked in the mind, it would make sense that they can be identified faster than indirect antonyms which are not so linked.
Gross, Fischer, and Miller then conducted a second experiment to see if subjects could distinguish between direct and indirect antonyms. It was much like the first one, except that this time, they were asked to identify the relation that held among the pairs of direct antonyms only. Indirect antonyms in this case were grouped with the non-antonyms. They found that subjects were able to identify the direct antonyms quickly and to distinguish them from the indirect antonyms, although they had some difficulty in making decisions about the indirect antonym pairs that contained near synonyms (e.g, alive/lifeless). Gross, Fischer, and Miller admit that the results of these two experiments do not prove the validity of their model, but they say, "the conservative statement is that the results of experiments 1 and 2 do not refute the hypothesis that they were designed to test" (Gross, Fischer and Miller 1988, 12).
Gross, Fischer, and Miller consider another possible explanation of their findings, a hypothesis that they call the "co-occurrence hypothesis." They note that some words are used together more often than expected by mere chance, and that this is probably especially true in the case of antonyms since antonyms label the two poles of a single attribute. In other words, the co-occurrence hypothesis is that an adjective such as cold is likely to be used together (in the same sentence, paragraph, or stretch of speech) with hot, and it is more likely to occur with hot than with any of the synonyms of hot such as scorching or ardent (the near and far synonyms of hot in this experiment). They go on to say that the co-occurrence hypothesis, if true, could probably explain why subjects identified direct antonyms faster than indirect antonyms, but they add, "We would object, however, that the co-occurrence hypothesis does not go far enough to explain all that we know about antonymous English adjectives" (Gross, Fischer and Miller 1988, 13). In particular, they feel that the large number of morphologically marked antonyms in English cannot be explained by co-occurrence alone; instead, the prefixes show a particularly tight bond between two antonyms that they feel is better explained through a model which directly links antonyms than through the co-occurrence hypothesis.
This model does not answer all the questions, however. How exactly is one adjective (out of a set of synonyms) chosen to represent the pole of an attribute? And what is the explanation for the fact that some words have more than one antonym, for example, happy, which has two antonyms unhappy and sad?9 Are both unhappy and sad somehow chosen as labels for the same attribute? Or is happy chosen as the label for two different (though related) attributes? Is the distinction between direct and indirect links enough to account for the clang phenomenon, or is that better accounted for simply through frequency of association? What about the fact that subjects had a lot of trouble categorizing the indirect pairs that used near synonyms (e.g, alive/lifeless and living/dead)? The fact that the subjects had difficulty deciding whether such pairs were direct or indirect antonyms argues against a structural difference between these two types of antonyms. And most importantly, even though the model does explain a great deal, is there any evidence that the mind is actually organized in this way? Or could other models explain the antonymy equally well or better? Gross, Fischer, and Miller themselves acknowledge that the results of their experiments did not provide strong support for their model, although they did not contradict it either. However, their model did lead to further study by researchers seeking to answer these questions. Some of this research is discussed in the rest of this chapter.
1Egan quotes this from Synonyms and Antonyms by C.J. Smith, 1867.
2Actually, it would be possible to explain the difference in intensity of "clash" in Egan's own terms by saying that she just has not considered the range of application and implications of these two words carefully enough; for example, destitute is used to describe people, but opulent can be (and I think more often is) applied to people's clothes, homes, lifestyles and so on rather than to people themselves. (Cf. destitute man, ?opulent man; *destitute home, opulent home) Secondly, it does not seem that opulent exactly negates all the implications of destitute. If destitute means "lacking the basic essentials", shouldn't its antonym mean "having more than is needed of the basic essentials?" Instead, opulent implies the possession of luxuries in addition to the basic essentials.
3Pairs which contrast in meaning in some way but which do not produce this kind of reciprocal response are not considered to be antonyms in Deese's model.
4The list contained adjectives that occurred more than 50 times per million words in the Thorndike-Lorge vocabulary, with some participles weeded out.
5Although all of the reciprocal pairs of adjectives identified by Deese's method are antonyms, this is not true of the reciprocal pairs found in other parts of speech. Deese notes that some nouns form reciprocal pairs (that is, the two nouns each occur as the most common response to the other), although this is quite rare. Sometimes, these nouns show some kind of semantic contrast as in mother/father and hand/foot, but with other pairs, the reciprocal association does not involve much contrast at all, as in house/home and clock/time.
6Gross, Fischer, and Miller (1988) use synonym rather widely, both to describe words that are quite close in meaning (e.g., lifeless and dead) and to describe words which are not so similar in meaning but which could be considered synonymous in a particular context (e.g., alive and bouncy).
7Pairs in which both adjectives are linked to antonyms (e.g., parched/soggy, in which parched is linked to dry and soggy is linked to wet) are also considered indirect antonyms in this model.
8They do not mention, however, that these prefixes do not always form antonyms (consider easy and uneasy) or that the most frequently occurring antonyms are not formed by affixation (for example, none of the 39 pairs on Deese's list in Figure 1 are formed this way).
9Gross, Fischer, and Miller note the existence of words that have more than one antonym, and they claim (without any justification that I can see) that this is additional evidence that antonymy is a relationship between particular words, not between concepts. However, they do not explain how their model accounts for one word with two antonyms.