Antonymy has several characteristics which set it apart from the other lexical-semantic relations. First, as Cruse says, "Of all the relations of sense that semanticists propose, that of oppositeness is probably the more readily apprehended by ordinary speakers." (Cruse 1986, 197) In fact, as he points out, most languages have a non-technical term for antonyms, e.g., English opposite, German gegensatz, and Japanese hantaigo.
A second interesting fact about antonymy is that most (probably all) languages have morphological processes which can create antonyms. As Lyons writes:
In many languages, including English, the most commonly used opposites tend to be morphologically unrelated (good:bad, high:low, beautiful:ugly, big:small, old:young). But these are outnumbered in the vocabulary by such morphologically related pairs as married:unmarried, friendly:unfriendly, formal:informal, legitimate:illegitimate, etc. (Lyons 1977, 275)
In addition to un- and in-, English also has the prefixes dis- (like/dislike, honest/dishonest) and de- (colonize/decolonize, emphasize/deemphasize), as well as the suffixes -less and -ful, which together sometimes form pairs of antonyms (harmless/harmful).2 However, English does not have derivational processses involving other kinds of lexical-semantic relations; that is, there are no affixes which create synonyms or hyponyms or form the name of a part from a whole.
Third, there seems to be something "mysterious" or even "mystical" about antonymy because opposites seem as different as they can possibly be, yet they still have something in common. In some measures of semantic similarity, antonyms pattern much like synonyms. For example, in word association tests, people often respond to a word with its opposite or with a synonym (see Deese 1965), yet people clearly feel that antonyms and synonyms are not the same. Cruse describes this mysterious quality of antonyms in his book Lexical Semantics (1986). He writes:
Opposites possess a unique fascination, and exhibit properties which may appear paradoxical. Take, for instance, the simultaneous closeness, and distance from one another of opposites. The meanings of a pair of opposites are felt to be maximally separated. Indeed, there is a widespread idea that the power of uniting or reconciling opposites is a magical one, an attribute of the Deity... The closeness of opposites, on the other hand, manifests itself, for instance, in the fact that members of a pair have almost identical distributions, that is to say, very similar possibilities of normal and abnormal occurrence... Philosophers and others from Heraclitus to Jung have noted the tendency of things to slip into their opposites states; and many have remarked on the thin line between love and hate, genius and madness, etc. (Cruse 1986, 197)
Cruse says that this feeling of simultaneous closeness and distance can be at least partially explained by semantic dimensions--opposites are two words which share a semantic dimension, e.g., the dimension of TEMPERATURE in the case of hot and cold, but which occupy opposite poles along this dimension.
Another interesting thing about antonymy is that language learners expect that there will be pairs of antonyms available to label certain types of concepts. Miller and Fellbaum (1991) remark that students learning a second language "when given only one member of an antonymous or opposed verb pair, will insist upon being taught the other member." And Kagan says:
As the child creates categories, she is disposed to invent their complement. Soon after learning the meaning of up, the child learns the meaning of down; after learning the meaning of high, she learns low; after good, she develops the meaning of bad. (Kagan 1984, 187)
Kagan claims that children do not have to be explicitly taught that there are such things as opposites; instead they seem to expect to find them. This kind of expectation may exist for other kinds for lexical-semantic relations too, but it seems to be especially strong in the case of antonymy.3 Even adult speakers feel the usefulness of antonyms when expanding the vocabulary of their native language. As Egan puts it:
It is good, we feel, to know the exact antonym of a word, for not only will it give us a firmer grasp of the meaning of the word to which it is opposed, but inversely, of itself. (Egan 1968, 28a)
Anthropologists (e.g., Casagrande and Hale 1967) have found that antonymy is one of the semantic relations commonly used in folk definitions; a thing or quality can often be defined in terms of what it is not, for example, big can be defined as 'not small.'
Native speakers have strong intuitions about which pairs of words are good examples of antonyms; for example, probably all English speakers would agree that cold is the opposite of hot, that happy is the opposite of sad, and that down is the opposite of up. This intuition has been called the "clang phenomenon." It occurs strongly with a rather small set of prototypical antonyms, but it is not so strong for less prototypical opposites; for example, some, but not all speakers would agree that vigorous and feeble or loud and faint are opposites.4
The research in this dissertation was prompted by a desire to learn more about the properties of antonyms in order to explain the strong intuitions that people have about them. I began with the initial questions listed in (1) below.
(1)
- What makes two words antonyms? What exactly are the semantic dimensions which antonyms are said to share?
- Why do some words have antonyms while others have none? Why do some words have more than one antonym, e.g., good/bad and good/evil or happy/sad and happy/unhappy?5
- What accounts for native speakers' strong intuitions about which words are antonyms (the "clang phenomenon")? What is the difference between pairs of words which are good, prototypical examples of antonyms, such as hot and cold, wet and dry, and pairs which are seem to contrast in meaning but which many people would not consider antonyms, e.g., hot and chilly or loud and faint? Is there a model of antonymy that can explain why some pairs are "good" antonyms and some are not?
There has been a great deal of research on antonymy in fields ranging from anthropology to lexicography to computational linguistics, but much of this research focuses on a specific aspect of antonymy (i.e., the phenomenon of markedness, described briefly in section 1.2.1 below) rather than on the basic questions listed above. The rest of this section, therefore, reviews only the literature which focuses on the kinds of larger issues I am most interested in.
Most linguists, however, have found this use of antonym too broad to be useful or interesting, and so many linguists, including Palmer (1976), Lyons (1977), Lehrer and Lehrer (1982), and Cruse (1986), have tried to categorize opposites into different types. They then define the term antonym in a more narrow sense, using it only to refer to the set of gradable opposites, a set which has many interesting properties such as markedness and committedness. Although the adjectives I focus on in the case studies all happen to be antonyms in the narrow sense, my model of antonymy is relevant to a wider range of opposites as well, so the next two sections describe some of the different types of opposites which have been discussed in the literature. Section 1.2.1 focuses on antonyms in the narrow sense and section 1.2.2 focuses on other types of opposites (antonyms in the wider sense). Then, since understanding antonymy also necessitates understanding why some pairs of words are not antonyms, section 1.2.3 focuses on pairs that are almost-but-not-quite opposites.
The set of gradable opposites includes many common and prototypical pairs of opposites including big/little, good/bad, high/low, hot/cold, happy/sad, tall/short, and wet/dry. The defining property of this set is that the opposites name qualities which are gradable, that is, qualities which can be conceived of as 'more or less'; therefore the scale (dimension) with which each pair is associated has a neutral mid interval. Take for example hot and cold, which describe opposite ends of the scale of TEMPERATURE. Hot and cold are both gradable; for example, we can say "A is hotter than B," "C is fairly cold," "D is very hot," and so on.6 Between the opposite poles named by hot and cold, there is a mid interval, so that if something is neither hot nor cold, it might be warm, cool, or lukewarm, etc.7
Although there are also nouns (e.g., friend/enemy) and verbs (e.g., love/hate and like/dislike) which show properties of gradability, most attention has been given to the adjectives of this type, perhaps because the adjectives most clearly exhibit other characteristic properties of gradable opposites, such as implicit comparison, committedness, and markedness.
Implicit comparison can easily be seen in examples such as big and little, tall and short, young and old, and hot and cold. Something is described as big or tall or hot in comparison to other things of the same type. This means, for example, that a tall child is tall in comparison to other children of the same age, but may in fact be much shorter than a short adult, and that a hot day describes a hotter than average day, but an overall temperature that is much lower than the one described by a hot oven.
Committedness involves an adjective's behavior in questions. An adjective is said to be committed if it implies a particular value when used in a question, and impartial or uncommitted if it does not have such an implication. For example, tall is uncommitted in a question like "How tall is Pat?" This question is neutral and can be used whether or not the speaker knows Pat's approximate height and whether Pat is tall, short or of average height. In contrast, the adjective short is committed; a speaker would only ask "How short is Pat?" if there is some reason to believe that Pat is shorter than average height. Many pairs of gradable antonyms contain one committed term and one uncommitted, e.g., old/young, heavy/light, fast/slow; many other pairs are made up of two committed terms, e.g., innocent/guilty, beautiful/ugly, happy/sad.
Markedness has been used as cover term for several related phenomena which distinguish the marked member of an antonym pair from the unmarked member. Lehrer (1985) discusses several of the criteria which have been proposed to define markedness. Committedness is one of them: the uncommitted member of an antonym pair is said to be unmarked and the committed member is said to be marked, so old is unmarked, while young is marked. It has also been noted that if the name of the semantic scale is morphologically related to one of the antonyms, it is related to the unmarked member, so for example, the name of the scale of LENGTH is related to the unmarked long rather than the marked short.
Another criterion of markedness is that the unmarked antonym can generally appear in a wider range of syntactic contexts; in particular, unmarked antonyms can occur with measure phrases but marked ones usually cannot, so we can say that something is 3 feet tall but not 3 feet short. Similarly, ratios are usually only possible with the unmarked antonym; we can say that Kim is twice as old as Pat, but we can't say that Pat is twice as young as Kim. Morphology also plays a role: in pairs in which one antonym is derived from the other, the derived member is said to be marked, so happy is unmarked and unhappy is marked.
Most research on antonymy has focused on gradable opposites, antonyms in the narrow sense, perhaps because the properties described above are quite subtle and fascinating, but a few people, including Lyons (1977) and Cruse (1986) have tried to characterize the other sorts of commonly occurring opposites. These other types lack the special properties found with gradable opposites, but like them, they show a "dependence on dichotomization" (Lyons 1977). In other words, like antonyms in the narrow sense, these other types of opposites are also pairs of words which share some kind of semantic dimension.
The type of opposite which is most similar to the gradable opposite is the complementary opposite, sometimes known as the contradictory. 8 Examples of complementaries include adjectives such as true/false, dead/alive, and male/female. Like the gradable adjectives, the complementary adjectives share a semantic dimension, but it is a dimension which has no middle values. As Cruse describes it:
[T]he essence of a pair of complementaries is that between them they exhaustively divide some conceptual domain into two mutually exclusive compartments, so that what does not fall into one of the compartments must necessarily fall into the other" (Cruse 1986, 198).
For example, we know that if a person is not dead, s/he must be alive; if a statement is not true it is false, and so on.
It is sometimes hard to decide whether a pair of opposites belongs in the set of gradable adjectives or in the set of complementaries, as in the case of clean/dirty. Clean and dirty are both gradable adjectives: we can say that something is fairly clean, very clean, extremely dirty, and we can say that X is cleaner/dirtier than Y. However, the scale of clean and dirty does not seem to have a middle term; whenever something is not clean, it can be described as dirty, so as Cruse says, it sounds strange to say "It's neither clean nor even slightly dirty."9 The case of wet and dry (the subject of my third case study) is similar in that there are words such as damp and moist which appear to name midpoints of the scale. However, as I will show, it seems that damp and moist are actually just more specific terms for types of wetness. We can gloss damp as 'slightly wet', but we cannot gloss a true midpoint word in this way (e.g., we can't gloss lukewarm as "slightly hot").
In addition to adjectives, verbs such as pass/fail and obey/disobey, nouns such as day/night, prepositions such as in/out, and adverbs such as backwards/forwards are also sometimes considered examples of complementaries.
Although by definition, complementaries are pairs which allow no logical middle term, in actual use, complementaries are sometimes used like gradable adjectives; for example, we can say that something is almost true, or that someone is barely alive. However, as Lyons (1977) points out, in these cases it may be the "secondary implications" of the words that are being graded rather than the main sense. That is, someone who is barely alive is actually entirely alive, but s/he is not as lively or energetic as most people are.
Directional opposites are another type of opposite, described in Lyons (1977) and in greater detail in Cruse (1986). These are generally adverbs or prepositions and include pairs such as up/down, in/out, and clockwise/anticlockwise.
Reversive opposites, described in Lehrer and Lehrer (1982) and Egan (1968), are yet another type of opposite. Egan describes reversive opposites in this way:
[T]hese comprise adjectives or adverbs which signify a quality or verbs or nouns which signify an act or state that reverse or undo the quality, act, or state of the other. Although they are neither contradictory nor contrary terms, they present a clear opposition. (Egan 1968, 27a)
This class contains many verbs, for example, tie/untie, marry/divorce, enter/leave, appear/disappear. Cruse and Lyons consider the reversive verbs to be a subtype of directional opposites, because they all describe activities which result in an object undergoing a change from one state to another; the two members of the reversive pair involve the same two states, but the direction of change is different in each case; for example, the verb tie means roughly 'to cause something to go from the state of being untied to the state of being tied,' while untie means ' to cause something to go from the state of being tied to the state of being untied.' Thus Cruse says the opposition seen in pairs of reversive verbs is similar to the kind of opposition in pairs of directional prepositions such as to/from.
Relational opposites (Cruse 1986) (also called relative terms (Egan 1968) and conversive terms (Lyons 1977)) include pairs such as above/below, predecessor/successor, parent/child and teacher/student. Egan describes these as "pairs of words which indicate such a relationship that one of them cannot be used without suggesting the other." Cruse considers this class to also be a subclass of the directional opposites. He says that these pairs "express a relationship between two entities by specifying the direction of one relative to the other along some axis." In examples such as above/below, this axis is spatial, but other examples (e.g., ancestor/descendant) involve "an analogical or metaphorical extension of spatial dimensions" (Cruse 1986, 231). Lyons points out that many opposites of this type involve social roles (teacher/student, doctor/patient) or kinship relations (father/mother), and these types of reciprocal relations have been well documented in many languages in the anthropological literature.
The various types of opposites discussed so far--antonyms, complementaries, directional opposites, and so on--all illustrate the essential properties of antonymy: they are pairs of words which simultaneously seem close and yet far apart in meaning, words which share some kind of semantic dimension but denote contrasting values along that dimension. But simply looking at the different types of opposites does not go very far in explaining what makes two particular words antonyms and why some words have no antonyms. To address these questions, it is useful to look at near-opposites, pairs of words which contrast in some way but which do not seem to be "real" opposites. Understanding what prevents some words from being opposites will lead to a deeper understanding of the essential properties of antonymy. In the next section, some examples of near-opposites are given, along with the suggestions that have been given to explain why these pairs are just near-opposites rather than antonyms.
One of the types of near-opposites described by Cruse is what he calls "impure" opposites. He says that these are opposites "which encapsulate, or include within their meaning, a more elementary opposition." For example, he says:
[G]iant:dwarf can be said to encapsulate the opposition between large and small (but this opposition does not exhaust their meaning); likewise, shout and whisper encapsulate loud and soft, criticize and praise encapsulate good and bad..." (Cruse 1986,198)
Cruse's discussion of this example is brief but suggestive. For example, if the semantic opposition found in impure opposites "does not exhaust their meaning", then is it the case that the opposition does somehow exhaust the meaning of a pair of "pure" (prototypical) opposites? Cruse's examples of pure opposites are adjectives and his impure opposites are basically nouns or verbs, so the "impurity" may lie in the nominal or verbal elements of meaning associated with these words. It certainly does seem that with adjective pairs such as large/small or loud/quiet, the shared semantic dimension (SPEED or TEMPERATURE) makes up the largest part of their meaning, while nominal or verbal pairs have additional semantic elements; shout and whisper, for example, are manner-of-speaking verbs, while dwarf and giant describe people. This idea of "purity of opposition" is vague, yet it may contain the seeds of an explanation of why there are more antonyms among adjectives than nouns or verbs.
Another type of near-opposite described by Cruse includes pairs that are only weakly contrasted because of "the difficulty of establishing what the relevant dimension or axis is" (Cruse 1986, 262). He gives work/play and town/country as examples. Cruse does not analyze these pairs at all, but it seems likely that they seem somewhat like opposites because in a particular context, there are two alternatives that provide an either/or choice. At the same time, we know that the two alternatives do not exhaust the logical possibilities (in addition to being at play or at work, a person could be asleep, for example), so they are not felt to be true opposites. It is not clear, however, how this type of near-opposite is different from prototypical pairs such as black/white (there are also other color terms) or happy/sad (there are other emotions, such as awe, frustration or boredom).
Finally, Cruse says that "nonpropositional meaning" is important to antonymy. He says that:
[A] good pair of opposites must be closely matched in respect of their nonpropositional meaning: that is why, for instance, tubby and emaciated are not fully satisfactory opposites, although they incorporate a binary directional opposition. (Cruse 1986, 262)
He does not explain this further, but apparently “nonpropositional meaning” refers to the connotations associated with the words. Tubby is an adjective that has rather positive connotations in that it is used to affectionately describe someone who is somewhat overweight but not dangerously so, while emaciated is more impersonal (it might be used as part of a medical description while tubby would not be) and also conveys the idea that the thinness is possibly life-threatening.
This idea of nonpropositional meaning seems potentially useful in explaining the behavior of many pairs of near-opposites, but there are still some questions to be answered. First, what kinds of nonpropositional meaning are relevant? Is it enough simply to speak of some kind of general positive or negative connotation (for example to say that tubby is somewhat positive, while emaciated is quite negative)? Is register important? (Tubby is an informal word used in conversation more than in writing, while emaciated is a more formal word, likely to occur in medical literature). In the first two case studies of this dissertation (in Chapters Two and Three) I will return to this question. To anticipate, the synonyms little and small appear to differ in terms of register and associated connotations--for example, little is often used to convey a sense of "cuteness," which is not found so often with small. Likewise, the synonyms of wet described in the second case study have strong connotations--unpleasant ones in the case of damp, dank, humid, and positive in the case of moist.
Lehrer and Lehrer (1982) also describe a few different types of near-opposites, but these types and their explanations of why these pairs are near-opposites rather than antonyms are quite different from Cruse's. One type includes pairs such as excellent/bad and hot/freezing, in which both words are associated with the same semantic dimension, but one term describes a state that is more "extreme" than the other. They account for these cases by saying, for example, that excellent, good, bad, and terrible all lie along the same semantic dimension, but that excellent and terrible name the extreme ends of the dimension, farthest from the midpoint, while good and bad name points nearer to the midpoint.
Based on these examples, Lehrer and Lehrer say that the definition of antonymy also needs the qualification that two antonyms should be the same distance from the midpoint, which would insure that good/bad and excellent/terrible are antonyms, while excellent/bad and good/terrible are only near-opposites. Similarly, in the case of temperature words, this qualification would rule out antonym pairs such as freezing/warm and cold/boiling. Although "distance from a midpoint" does seem to work well in explaining the behavior of these two sets of words, I cannot think of other cases in which this criterion is necessary, and Lehrer and Lehrer (1982) do not provide any others, so I wonder whether it is really necessary or whether the behavior of these two sets of words can be explained in some other way. Lehrer and Lehrer themselves acknowledge that distance from a midpoint cannot explain all cases of near-opposites, and they mention the case of large and little as one example.
Lehrer and Lehrer point out that big and large are near synonyms, associated with the same pole of the dimension of SIZE, with neither one more "extreme" than the other; the same is true of small and little. Big and little are antonyms, of course, as are large and small, but big/small and large/little might also be expected to form pairs of antonyms. Many speakers do in fact regard big/small as antonyms (I saw this pair in many of the lists of antonyms in the papers I read while doing this study, for example), but it seems that native speakers do not feel that large/little are antonyms. Lehrer and Lehrer do not analyze this case in any detail, but they do make a useful suggestion. They note, "their distribution is not identical, since big and little but not large and small overlap with older and younger. My big sister is not equivalent to my large sister on one interpretation" (Lehrer and Lehrer 1982, 496). In the first case study, I look at the distribution patterns of these four adjectives by looking at the types of nouns each adjective typically modifies. I show that distribution patterns suggest an explanation of why little and large are not antonyms. This notion of difference in distribution may also explain the behavior of many other pairs of near-opposites, for example, pairs examined in the second case study such as wet/arid and dry/damp.
Lehrer and Lehrer have one more thought about near-opposites, suggesting that some near-opposites may be a result of multi-dimensional scales. They give an example involving words related to intelligence, saying that dictionaries of antonyms list many different words as possible antonyms of clever, including dull, stupid, unintelligent, unskillful, slow, dense, simple, and foolish. These "antonyms" of clever are all partial synonyms of each other, and clever itself has many partial synonyms, including smart, able, intelligent, skillful, ingenious. All of the synonyms of clever contrast at least loosely with all of the synonyms of stupid, but only a few of the possible pairs form good examples of antonyms, e.g., skillful/unskillful and intelligent/unintelligent. Most of the combinations seem only to form near-opposites, e.g., clever/unskillful or ingenious/dense. Lehrer and Lehrer suggest that while all these words are related to a scale of intelligence, this scale is not a simple one. They say:
Each of these words may be connected to the various ways of being clever. Smart and intelligent might be more appropriate for mathematical skills than for interpersonal relationships, whereas skillful may be more applicable to interpersonal relationships than to mathematical abilities. However, we do not have very strong intuitions on that point. (Lehrer and Lehrer 1982, 496)
The third case study, on happy, unhappy and sad, looks at a similar case, showing that each of the antonyms sad and unhappy corresponds to a different way of being happy.
A few linguists, namely Cruse (1986) and Lehrer and Lehrer (1982) have tried to identify the factors which determine whether two words are antonyms or just near-opposites. Each of these factors, in identifying what keeps two words from being antonyms, also makes an implication about the nature of antonymy, and together these factors seem likely to be useful in answering some of the "big questions" about antonymy, in particular, the questions of what makes two words antonyms and why some words have no antonym. These factors are listed in (a) to (f) of (2) below, and they will be examined in more detail in the case studies.
(2) Factors which affect the "goodness" of a pair of opposites:Linguists are not the only ones who have worked on antonymy. The next section presents the work of a lexicographer who takes a somewhat different approach to the nature of antonyms and near-opposite pairs.
- The purity of the semantic opposition: In some pairs of near-opposites (e.g. whisper/shout) the semantic opposition does not exhaust the meaning of the words. The implication is that in prototypical pairs of opposites, the semantic contrast does in some sense exhaust the meanings of the words.
- The ease with which a semantic dimension can be imagined: With near-opposite pairs such as town/country, it is hard to determine what the relevant semantic dimension could be. The implication is that for prototypical opposites, the semantic dimension can be easily identified.
- Correspondence of nonpropositional meaning: Some near-opposites (e.g., tubby/emaciated) have very different connotations. The implication is that prototypical opposites are very similar in terms of non-propositional meaning.
- Distance from the midpoint of a semantic dimension: In some cases, one member of a pair of near-opposites (e.g., terrible) seems to name a more extreme value than the other member (good). The implication is that prototypical opposites lie at equally distant points from a midpoint.
- Similarity in distribution: For example, both big and its antonym little can refer to relative age rather than physical size in constructions such as big sister and little brother, but the near-opposite of little, that is, large, can only refer to physical size with the nouns sister and brother. The implication is that prototypical opposites are similar in distribution.
- Whether the semantic contrast involves a single dimension or multiple dimensions: Some near-opposites seem to be located along different dimensions, although the dimensions seem to be related to the same general concept. The implication is that prototypical opposites share a single dimension, even if that dimension is related to a concept associated with multiple dimensions.
1I am following Evens, et al. (1980) in using the term lexical-semantic relation as a kind of neutral term, rather than either lexical relation or semantic relation, terms which have theoretical connotations for many people.
2Of course, these affixes do not always create antonyms; it is easy to find examples which, due to semantic drift or some other cause, look morphologically as if they should be opposites but which actually are not, e.g., integrate/disintegrate and easy/uneasy.
3In the case of synonymy, in contrast, research suggests that children do not expect to find synonyms. In order to explain children's early word learning, a "mutual exclusivity" principle has been proposed (see Markman 1994, for example). According to this principle, when young children hear a new word, they assume that it is distinct in meaning from the words they already know.
4Mettinger (1994), for one, considers both of these pairs to be examples of opposites.
5I am not so interested here in cases in which it is clear that one of the words has two very different senses, as in the case of right, which has two antonyms, left and wrong. The sense of right which is paired with left is quite distinct in meaning from the one which is paired with wrong and presumably this explains why right has two antonyms.
6Compare this to a typical pair of non-gradable adjectives such as alive/dead. Except in figurative uses, people do not say “A is extremely dead ,” or “B is fairly alive.”
7The temperature scale is more lexicalized than most of the scales in that it has words to name the mid intervals; most gradable scales do not have names for mid intervals, but that does not mean that they are not there. Consider the SIZE scale for example. We know that it is possible for something to be neither big nor little but some size in between, even though we don't have a word to describe this. (We could use a phrase such as of average size though.)
8Contradictories and contraries are terms from logic which largely overlap with complementaries and antonyms as used by linguists. For a discussion of the differences between the logical terms and the linguistic terms, see Lyons (1977, 272).
9As Cruse points out, it sounds fine to say "It's neither clean nor dirty," because dirty is more often used for things that are distinctly dirty.