Secret Lives of Adjectives

If we divide up the short list of English parts of speech according to status, adjectives are at the top of the B-list. The elites, nouns and verbs, seem to get everyone's attention because without them, sentences wouldn't have a job.

English users, from first-time learners to sophisticated software used in data mining, are all over nouns and verbs. This focused spotlight has meant that until recently, even meticulous observers of language have failed to notice that adjectives possess arcane subtleties — qualities that fly under the radar of the typical language user but that figure importantly in their meaning and usage. Linguists investigating adjectives today discover that there may be a lot more going on than meets the eye when you plop an adjective into the predicate slot of an innocent-looking sentence.

The conventional wisdom about adjectives — and the only thing you're likely to glean about them by looking at dictionary definitions — is that they're gradable (like fat) or not (like multiple), limited to predicate use (like alive), limited to attributive use (like elder), or show somewhat different meanings depending on whether they're used attributively or predicatively (like ready). Is there any more you need to know? Well, that depends on how fluent you want to be. Gradable adjectives — the ones that have comparative and superlative forms with suffixes -er and -est or with more and most — have other semantic potentials that may come into play at the moment of use.

The first clue that an adjective may be up to something that you don't know about is when using it results in a sentence being infelicitous. Here we don't mean infelicitous "unhappy," but infelicitous "not appropriate in application." Linguists use this term to characterize sentences that are not well-formed semantically (despite violating no obvious grammatical rule); in other words, sentences that don't sound right. Unless they are making jokes or having fun with words, fluent native speakers avoid infelicitous sentences instinctively and can spot them a mile away. Learners — including children, speakers of other languages, and computers that would aspire to literacy — are prone to these kinds of linguistic infelicities. The infelicities often turn up in modified adjectives in predicates. Take for example,

a. Sarah's very talkative for a two-year-old.
b. Sarah's very awake for a two-year-old.

a. The table is small for a dining-room table.
b. The table is wet for a dining-room table.

Though the adjectives in all of these sentences are gradable, the b. sentences are things we wouldn't say and that we would correct (or balk at) if we heard or read them. Why? Linguists Chris Kennedy and Louise McNally have looked behind the curtain of adjective behavior and developed an analysis of some gradable adjectives based on this aspect their variable qualities. Their explanation is that some adjectives (like awake, wet, and full, for example) normally get their meaning with reference to an absolute scale, independent of the context in which they're used. Other adjectives (like talkative, small, wide, and in fact most gradable adjectives) mean what they do with reference to a context. The presence of a "for" prepositional phrase supplies a context that may make the absolute kind of adjective infelicitous.

We say "may make" because different contexts call up slightly different senses of some adjectives, and an adjective's meaning may suggest an absolute scale in one context but a relative one in another. We don't object to "The restaurant was full for a Tuesday night" but "The glass was full for a milk glass" sounds a bit off.

It would be a neat and handy trick if all adjectives did not transgress beyond either of these two categories into which most of them naturally fall; but language being what it is, they leap out from time to time to make an appearance in the other category. And that's only the beginning of the story. If you read the following sentences,

a. This stick is not quite straight.
b. *This stick is not quite bent.

a. The door was slightly open.
b. *The door was slightly closed.

a. This water is absolutely pure.
b. *This water is absolutely impure.

Again, you'll probably like the a. sentences; and feel that the b. sentences all don't get it quite right. What's the difference, when the adjectives (all gradable) that constitute the only difference between the pairs are generally considered to be opposites?

Kennedy and McNally's explanation is that many adjectives have reference to scales in which the end points may be either open or closed. Where do these scales reside? Mainly in the minds of native speakers. In the illustration below, a filled circle indicates a scalar endpoint; an open circle indicates unboundedness or open-endedness.

lower limit closed	●――――――○	examples: wet, bent, bumpy
upper limit closed	○――――――●	examples: pure, straight, accurate
both ends closed	●――――――●	examples: opaque, open, necessary

Adjectives that have a scale with a closed end characterize a quality that we think of as a kind of phase transition — a Rubicon that, once crossed, does not admit of scalable or comparative adjustment any more. But there are other adjectives that are open at both ends and do not have a point at which the difference between zero and nonzero reference is fixed by anything other than context.

open at both ends

○――――――○

examples: short, deep, expensive, likely

Open-scale adjectives like these allow a wider range of modification by adverbials, but don't do so well with, e.g., absolutely.

If you play around with adjectives in your lexicon, you may be surprised to discover that despite their gradability, there are restrictions on the kinds of modifiers that they can take. If you're an adult native speaker, you know what these restrictions are, without necessarily being able to articulate the reasons for them. You know that something is unlikely to be called almost completely fluffy, or only slightly vertical. If you're a child, a learner, or a computer, on the other hand, you may only gape in slack-jawed wonder at your mistakes and be at a loss for how to correct them.

Would we all be better served if language somehow encoded these distinctions by way of morphology, syntax, or phonology? Probably not, because language hasn't evolved that way. Our brains seem ready to devote a lot more resources to the software side than the hardware side when it comes to getting meaning right: we interpret things in the moment, rather than storing an unlimited number and variety of forms that do the disambiguating for us. Linguist Kennedy's idea of this is called Interpretive Economy: language favors flexibility of application and tends to let a small set of lexemes do as much work as possible.

You can dive deeper into the subject in the following papers and slideshow:

http://semantics.uchicago.edu/kennedy/docs/vg-epub.pdf

http://semanticsarchive.net/Archive/zU5MjNiN/km-scales.pdf

http://web.uni-frankfurt.de/fb10/grad_koll/mcnally-slides.pdf