2.1 More on intensification

‘Intensification’ is an umbrella term for a functional category or for a field of study, sometimes labelled degree (Bolinger Reference Bolinger1972) or graduation (Martin & White Reference Martin and White2005), including ‘boosters’ or ‘maximizers’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985). Although the present study is not solely about intensification, some background is necessary here. ‘Intensification’ generally refers to the ability of some modifiers to place the modified word on a relative scale of either gradability (‘more or less’) or appropriateness (‘truly or not’). Martin & White (Reference Martin and White2005) termed these ‘force’ and ‘focus’, respectively; we will term them Degree and Focus. Intensification is a big tent, however, which can encompass not only Degree and Focus, but also intense reaction (incredibly good) and emphatic epistemic meaning (clearly dangerous).

Intensification has been widely studied: as Méndez-Naya (Reference Méndez-Naya2008) points out, some extensive studies date back to the beginning of the twentieth century (Stoffel Reference Stoffel1901; Borst Reference Borst1902; Fettig Reference Fettig1934), and it is still an active area of research (Paradis Reference Paradis1997; Kennedy & McNally Reference Kennedy and McNally2005; Nevalainen Reference Nevalainen2008; D'Arcy Reference D'Arcy2015; Schweinberger Reference Schweinberger, Lucek and Amador-Moreno2021; Nouwen Reference Nouwen2024). Bolinger (Reference Bolinger1972) devoted an entire book to open-class words (nouns, verbs, adjectives, adverbs) and their scalability. In relation to adjectives, he relied on lists from Borst (Reference Borst1902) and Kirchner (Reference Kirchner1955) which contain many combinations still prominent in present-day corpora, such as absolutely necessary, bitterly cold, incredibly short or strikingly handsome.

In the realm of intensification, Bolinger states, perhaps more than anywhere else in the grammar, processes of grammaticalization seem to always be at work, with each ‘newcomer’ intensifier elbowing others aside (Bolinger Reference Bolinger1972: 18). To illustrate this point, Bolinger (Reference Bolinger1972) categorized very as fully grammaticized, but really as less so. Fast forwarding a few decades, we see that really is now grammaticalized, as Tagliamonte (Reference Tagliamonte2008) found in the language of several generations of Toronto speakers. All four of the Degree adverbs studied by Tagliamonte (very, really, so, pretty) were undergoing change in preference at the time of her study, with older speakers (over 50) preferring very, 30- to 49-year-olds using both really and very, and younger speakers preferring really, but also being the most frequent users of so. Tagliamonte describes this as ‘intensifier recycling’, a process whereby speakers recycle existing intensifiers and use them preferentially (see also Peters Reference Peters, Kellermann and Morrissey1992, Reference Peters1993; Partington Reference Partington, Baker, Tognini-Bonelli and Francis1993; Stenström Reference Stenström, Hasselgård and Okesfjell1999; Lorenz Reference Lorenz, Wischer and Diewald2002; Ito & Tagliamonte Reference Ito and Tagliamonte2003; Nevalainen Reference Nevalainen2008; Paradis Reference Paradis2008; D'Arcy Reference D'Arcy2015). Nouwen (Reference Nouwen2024) argues that intensifiers tend to be deadjectival adverbs and are thus an open class, just like adjectives are.

Some of the less delexicalized intensifiers have strict collocational restrictions: bitterly cold, but not *bitterly hot. This kind of collocation restriction has been theorized in Melʹcuk's Meaning-Text Model (Melʹcuk Reference Melʹcuk2012), where MAGN, magnification, is postulated as a basic lexical function that selects specific lexical modifiers depending on the head word. For instance, in English MAGN(rain) is realized as heavy rain, whereas MAGN(wind) is realized as strong wind. Similar restrictions are also explored by Morzycki (Reference Morzycki2012), who finds that some intensifiers such as flat-out or positively modify only extreme adjectives (gigantic, fantastic). Thus, positively gigantic sounds right, but positively big less so.

It is combinatorial aspects such as these that we explore in our article. Thus, we study not so much the use, recycling and replacement of specific intensifiers, but rather how combinations of adverbs and adjectives in a grammatical construction may convey intensification and/or evaluation, broadly construed (cf. Lewis Reference Lewis2020), along with a variety of other meanings – and seek possible discourse-functional motivations for their differing distributional profiles in different registers.

2.2 Previous classification schemes in review

Existing classifications of adjectives and adverbs are relevant in that the broad semantic categories of ‘adverb-ly adjective’ constructions are largely predictable from those of the adjectives and adverbs individually. We here review semantic classifications of adjectives and adverbs in the main reference grammars of English.

The semantic classification of Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 434ff.) is high level, recognizing three classes of oppositions, as shown in (1).

1. (1) Stative/dynamic. Most adjectives are stative (tall), but some gradable adjectives can be dynamic (abusive, helpful, irritable).

   Gradable/nongradable. Most adjectives are gradable, but some denominal adjectives are not (atomic, hydrochloric).

   Inherent/noninherent. Inherent adjectives characterize the referent of the noun directly (a wooden cross). Noninherent modification (a wooden actor) is ‘an extension of the basic sense of the noun’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 435), which we would describe as metaphorical meaning.

Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 445ff.) propose a classification of adverbs that touches upon their role as modifiers of adjectives. They distinguish amplifiers (absurdly, extremely, sharply, entirely, too) and downtoners (fairly, a bit, nearly, relatively). These are treated separately from emphasizers, which, they say, add to the force, as distinct from the degree, of the adjective. In this class are all, certainly, indeed, just, really.

Huddleston & Pullum et al. (Reference Huddleston and Pullum2002: 555ff.) divide attributive-only adjectives into the semantic classes shown in (2).

1. (2) Degree and quantifying – complete fool, extreme end, outright lie, perfect stranger, total disarray, blithering idiot, thumping majority

   Temporal and locational – current girlfriend, former wife, present manager, lower lip, southern states

   Associative – clerical duties, criminal law, foreign affairs, mathematical genius, medical journal

   Process-oriented – big eater, fast worker, heavy smoker, rapid reader, strong advocate

   Modal – actual cause, certain winner, likely benefits, true course of events, mock trial

   Particularizing – certain house, particular area, chief reason, principal factor

   Expressive – dear mother, poor father, bloody tax inspector

   Hypallage (transferred) – discreet cigarette, insane cackle, nude photo of the mayor, quiet cup of tea, stupid fault

Most of the categories in (2) are transparent from their names. We will later turn our attention to the Associative category because it is closely related to our Topical class. Huddleston & Pullum et al. (Reference Huddleston and Pullum2002) characterize Associative adjectives as those which do not directly denote a property of the head noun, but rather a property associated with it; thus, criminal law is not a law that is criminal, but a branch of law that concerns crime. Raskin & Nirenburg (Reference Raskin and Nirenburg1995) label these as nonpredicating adjectives (e.g. criminal lawyer), as opposed to predicating examples such as a lawyer who is criminal. Hypallage adjectives are those that denote a transferred epithet. In a phrase like a discreet cigarette, it is not the cigarette itself which is discreet; rather, the adjective describes the way it was smoked.

Huddleston & Pullum et al.'s section on adverb classification (Reference Huddleston and Pullum2002: chapter 6, §7.2) proposes that many adverbs, especially those ending in -ly, can be paraphrased in one of two ways: ‘in an Adj manner/way’ and ‘to an Adj degree’. In agreement with this, we adopt the terms Manner and Degree, respectively.

The final comprehensive grammar classification we will discuss comes from the Longman grammar of spoken and written English (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999). The semantic domains for adjectives proposed therein include color, size, time, evaluative and topical, arranged under two large classes (descriptors and classifiers), as set out in table 1.

Table 1. Semantic domains of attributive adjectives from Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan1999)

Biber et al.'s classification of adverbs is summarized in table 2. We exclude the Linking class, which contains sentence adverbs (e.g. additionally, overall, namely, however, incidentally).

Table 2. Semantic domains of adverbs from Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan1999)

Besides these three reference grammars, other researchers have also proposed classifications of adjectives. One notable scheme is Dixon's (Reference Dixon1977): dimension (big, small), physical property (hard, sweet), colour (black, red), human propensity (jealous, generous), age (new, old), value (good, atrocious) and speed (fast, slow). Dixon also studied the properties of adverbs derived from adjectives, pointing out that not all adjectives can form derived adverbs, and that not all derived adverbs carry the semantics of the original adjective. Dixon noted that when a deadjectival adverb is possible, it tends to be derived from the metaphorical meaning of the adjective. For instance, warmly recommended corresponds to the metaphorical meaning of warm. This metaphorical extension is often present in the meaning of the subtypes we discuss in the next section.

3.1 Degree

We first distinguish between what we term Degree (this subsection) and Focus (next subsection). Degree combinations involve scalar gradability. They can be paraphrased as ‘to an Adj1 degree’ (cf. Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 565). The Adj1-ly is usually replaceable by very, very much or not very, very being the prototypical delexicalized intensifier (e.g. Lorenz Reference Lorenz, Wischer and Diewald2002; Tagliamonte Reference Tagliamonte2008). However, if the adverb does not already imply ‘very’, it can itself be modified by very, e.g. very slightly different.

1. (3) Examples of Degree

   1. (a) extremely difficult/important; highly unlikely/skilled/specialized/intelligent; deeply ashamed/concerned/disturbing; bitterly cold; infinitely better

   2. (b) enormously positive; hugely entertaining; exceptionally attractive; especially important/true; particularly important/difficult; slightly different; equally important; mildly amusing

   3. (c) fairly large/simple; relatively small/new/high

   4. (d) barely visible/able; clearly legible/audible

   5. (e) fully aware/capable/operational

   6. (f) extraordinarily daring; exceptionally different/acceptable

As mentioned in section 2.1, some of the Adj1-ly adverbs are subject to very tight collocational restrictions because of their idiomatic nature. As a modifier of adjectives, bitterly mainly collocates with cold and disappointing (though with verbs it can go with complain and oppose), heightening negative evaluative prosody (Partington Reference Partington, Aijmer and Rühleman2014). Other examples of highly specific collocations include eminently, as in eminently qualified (deserving, suitable); brightly, as in brightly coloured; heavily, as in heavily armed; and acutely/keenly, as in acutely/keenly aware.

Degree is the most frequent and prototypical semantic category of adjective-modifying adverbs. Martin & White (Reference Martin and White2005) had termed this category ‘force’ and contrasted it with ‘focus’, which applies to non-scalar, non-gradable concepts, as we will see in section 3.2. Adverbs other than those ending in -ly also participate in this category, most notably very, quite and really (a lexicalized -ly adverb); see section 3.2.1 for more on really.

3.2 Focus

Focus applies to non-scalar, non-gradable concepts. Whereas Degree corresponds to intensity or amount, Focus concerns prototypicality: ‘the preciseness by which category boundaries are drawn’ (Martin & White Reference Martin and White2005: 137). The intuition behind Focus is that this kind of modification is metalinguistic in character. The diagnostic paraphrase is ‘Adj2 is (or isn't) a very appropriate word.’

1. (5) Examples of Focus

   1. (a) Main adverbs:

      • – really

      • – utterly: absolutely, entirely, perfectly, simply, essentially, genuinely, entirely, truly, wholly, utterly ridiculous; genuinely terrific; perfectly legal/fine; simply brilliant; entirely inappropriate/different; essentially Finnish; truly sad/democratic/amazing/pathetic; decidedly average/unimpressive

   2. (b) Some adverbs, such as nearly and virtually force a reading of the Adj2 as ‘extreme’ and non-scalar. The paraphrase to check the extreme nature is ‘one can almost say that it is Adj2’: nearly impossible (‘one can nearly say it is impossible’), nearly complete; virtually impossible/extinct/identical

Some Adj2 are open to being modified both in terms of Degree and from the point of view of prototypicality (Focus); for example, very red, truly red; very upset, truly upset (Martin & White Reference Martin and White2005: 138).

The Longman grammar of spoken and written English (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999: 521) uses the term ‘emphatic adverbs’ for some of the adverbs that convey Focus, noting that they can modify non-scalar adjectives, e.g. quite motionless, absolutely continuous, really tremendous. This brings us to a more detailed discussion of really.

3.3 Manner

In Manner combinations, Adj2 is typically deverbal: a passive or present participle. The diagnostic paraphrase is: ‘when she Verbed it, she Verbed it Adj1-ly’ (i.e. ‘in an Adj1 way’), where Verb stands for the verbal root of Adj2. For example, randomly selected: ‘When they selected them, they selected them randomly’; exquisitely performed: ‘When she performed it, she performed it exquisitely’. Depending on the semantics of Adj1, such combinations can be strongly evaluational.

1. (6) Examples of Manner

   1. (a) randomly selected; patiently assembled

   2. (b) exquisitely/delightfully performed; brilliantly executed

3.4 Reaction

The Reaction category contains the kind of examples that first drew our attention to the ‘adverb-ly adjective’ construction. They are combinations that are frequently used in film and book reviews to describe how the reviewer felt about the work of art. We borrow the label Reaction from the Appraisal framework (Martin & White Reference Martin and White2005: 56), where it is used to describe a type of Appreciation. Our diagnostic paraphrase for the Reaction class is: ‘It's Adj1 that it's so Adj2’, e.g. strangely compelling => ‘it's strange that it's so compelling’. The intensifier so in this use implies very, and that is why there is a strong substrate of Degree in this class. Adj1 implies a ‘feeling’ component (cf. Goddard et al. Reference Goddard, Taboada and Trnavac2019). If Adj2 is open to scalar modification, the construction implies ‘very Adj2’, even if Adj1 does not carry this implication, e.g. strangely compelling implies very compelling.

1. (10) Examples of Reaction

   • (a) incredibly good/bad; stunningly original; devastatingly effective; strangely compelling; breathtakingly simple; wonderfully moving; satisfyingly tense; surprisingly thought-provoking; impressively big; unbelievably good/bad; astonishingly good/bad; delightfully good; eerily similar; outrageously expensive; suspiciously high; breathtakingly presumptuous

   • (b) staggeringly incompetent; bravely melancholy, ridiculously complex; disappointingly simplistic; jaw-droppingly simplistic/misguided; comically predictable; frightfully short; astonishingly slipshod; woefully inadequate/underfunded; disarmingly minimalist

   • (c) fabulously titled; gorgeously written. Roughly: ‘it is fabulous that it is titled in this way’ (i.e. that it has been given a certain title). Interestingly, the deverbal Adj2 (titled, written) seems to always relate to speech, writing or words.

3.5 Topical

In the Topical combinations, the adverb sets the topic or frame of reference, and Adj2 passes comment. There are two groups, the first one being very common in technical, bureaucraticand academic registers (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999). For both groups, Adj1 itself derives from, or is semantically very close to, a noun, e.g. politics –> political; finance –> financial. Huddleston & Pullum et al. (Reference Huddleston and Pullum2002: 556) classify comparable Adj+N constructions as ‘associative’, where the property expressed by the adjective applies to some entity associated with the noun. For instance, in clerical duties, the duties are not clerical; they are simply associated with being a clerk. Similarly, in politically incorrect statement, the modified noun (statement) is not intrinsically or necessarily incorrect; it is only so if one associates it with the context of politics and ideology. In this sense, Topical can perhaps be considered an extension of the Focus meaning, whereby the Adj1 narrows or focuses the semantic space to which Adj2 applies (see section 3.9 for discussion).

We have found two different subtypes of Topical constructions, which we have simply named Topical 1 and Topical 2.

3.6 Time

The Time category includes several further subtypes such as frequency, duration, relative timing or pace. The Adj1 specifies or restricts the type of time or time span that pertains to Adj2.

1. (13) Examples of Time

   Frequency:

   • frequently cited/mispronounced; occasionally offensive; regularly scheduled

   Duration:

   • (a) briefly famous; momentarily suspenseful; consistently brilliant/believable; permanently disabled/scarce; perpetually angry

   • (b) ?relentlessly erudite

   Relative timing:

   • (a) currently involved

   • (b) recently released/promoted/retired/deceased; newly released/independent/married/developed/minted/ emerging; previously unknown; immediately obvious/apparent

   • (c) freshly ground/grated/baked (i.e. ground, grated or baked a short time before)

   • (d) historically Black (neighbourhood)

   Pace (often implies ‘very’):

   • increasingly popular/difficult; rapidly growing/expanding; steadily mounting

3.7 Epistemic

The Epistemic subtype relates to source of knowledge or certainty and can include evidentiality. In general, this category includes combinations where Adj1 indicates how or with what certainty one knows that Adj2. A diagnostic paraphrase is ‘Adj1-ly, Adj2 is the case’, as in potentially, this is fatal or presumably, this is deliberate, where Adj1-ly can act as a sentence adverb.

1. (14) Examples of Epistemic

   • (a) probably true/illegal/plagiarized; potentially fatal/lethal/dangerous/harmful/deadly/important/serious; potentially useful/life-threatening/lucrative/damaging

   • (b) intuitively obvious; clearly incapable; seemingly effortless/endless; glaringly obvious

   • (c) presumably deliberate

   • (d) demonstrably false; patently false

Such examples have been discussed in previous literature about adverbs. For instance, Lorenz (Reference Lorenz, Wischer and Diewald2002) proposes a Modal class, where modal adverbs like apparently, certainly, possibly or supposedly are used to ‘express the extent to which a speaker is willing to attest to the truth of a proposition’. Partington (Reference Partington, Baker, Tognini-Bonelli and Francis1993) discusses a shift from modal to intensifier in words such as very, originally meaning truly or genuinely (from Old English verray/verray, ‘truth’), and utterly, entirely, absolutely, certainly, positively. In Partington's words, ‘it is a short step from averring truth to being emphatic about it’ (Partington Reference Partington, Baker, Tognini-Bonelli and Francis1993: 181).

3.8 Minor subtypes

There are several minor subtypes, with few members each, which could not be classified elsewhere. They do not appear to share any common denominator. We group them together here for the sake of completeness, but do not expand on them in any detail.

• (a) Relational

• (15) mutually exclusive/beneficial; mutually assured (destruction); inversely proportional; diametrically opposed

  ?internally consistent/inconsistent

• (b) Agentive

In these combinations, Adj2 is deverbal (past participle form) and the Adj1 is usually related to an abstract noun. Often a paraphrase with a by-phrase is available, e.g. federally funded = ‘funded by the federal government’; legally required = ‘required by law’.

1. (16) federally/privately funded; constitutionally mandated; legally required (abstract agent or source)

   parentally subsidized/forbidden/arranged

   sexually transmitted

• (c) Path-outcome

1. (17) terminally ill (cf. terminal illness, terminal cancer); fatally ill (cf. fatal illness); mortally injured (cf. mortal injury)

   upwardly mobile

• (d) Unnamed but recognized minor subtypes

• (18) (a) happily married/retired; happily employed/self employed

  (b) publicly/commercially/readily/widely/freely available

  (c) sweetly scented; sharply pointed

3.9 Some additional thoughts on the typology

All typologies are imperfect. Most suffer from an excess of either granularity or abstraction, sometimes from both. Having examined many examples of the ‘adverb-ly adjective’ construction, we are confident that the types discussed in this section capture most of its meanings and uses. One possible addition to the typology would be to endow it with more abstraction, i.e. to find a higher level of classification that captures these seven or eight types. One could argue that the typology can be largely subsumed into two enlarged classes, Degree and Focus. The enlarged Degree conveys intensification or evaluation (Degree, Reaction, Epistemic), whereas the enlarged Focus narrows the semantic space that the adjective applies to (Focus, Manner, Topical, Time).

The confluence of intensification and evaluation in modifiers that would be found in the enlarged Degree class has been discussed before in the literature. According to Bolinger, ‘investigation will probably reveal that virtually any adverb modifying an adjective tends to have or to develop an intensifying meaning’ (Bolinger Reference Bolinger1972: 23). He cites the example He was deliberately mean and comments: ‘deliberately has become a virtual intensifier’. He also writes of ‘less grammaticalized intensifiers’ (Reference Bolinger1972: 22), but insists that ‘the mere existence of a modifier tends to be taken in an intensive sense’ (Reference Bolinger1972: 22–3).

Nouwen (Reference Nouwen2024) hypothesizes that that the intensifying function of deadjectival adverbs is linked to the evaluation expressed by the adjectival base of those adverbs. He also advances two generalizations regarding the phenomenon: evaluative adjectives with a negative evaluative meaning tend to turn into deadjectival intensifiers expressing high degree, while adjectives with a positive meaning make intensifiers of medium degree, perhaps because extreme values on a scale are evaluative negatively, as they involve excess. The bleaching away, or delexicalization, of the positive and negative meaning may be the result of a diachronic process, whereby the deadjectival adverb initially expresses positive or negative evaluation, typically associated to medium or high regions, respectively, of the adjective's scale. Once the evaluative meaning is bleached away, the association to medium or high degree remains and becomes the functional semantics of the intensifier.

Guimier, likewise, proposes that adverbs ending in -ingly followed by an adjective tend to become intensifiers (Guimier Reference Guimier1986: 222). As for the evaluative aspect, Lewis (Reference Lewis2020) suggests that to characterize the adverb in an Adv+Adj combination simply as a ‘degree modifier’ fails to account for the development and use of -ly adverbs expressing speaker evaluation that cannot be reduced to intensification or indeed to degree. Lewis (2000) argues that there is a difference between, on the one hand, grammaticalized intensifiers, such as really, and, on the other hand, evaluative Adv+Adj combinations whose main function is to convey speaker evaluation. In the oxymoronic ‘adverb-ly adjective’ construction strangely ordinary, the negative element strangely does not carry the semantics of ‘degree modifier’ but evaluative semantics. The ‘adverb-ly adjective’ construction, then, can be seen as conveying intensification or evaluation, and sometimes both. This broader classification would subsume accounts for the Degree, Reaction and Epistemic subclasses in our typology.

The other subclasses (Focus, Manner, Topical, Time) do not convey either an intensifying or an evaluative meaning. They can, however, all be seen as types of focus in an enlarged sense, i.e. as narrowing the semantic space to which the adjective applies.

This more abstract and macro classification into Degree and Focus is akin to the distinction between force and focus in the Appraisal framework (Martin & White Reference Martin and White2005). Whether this higher-level two-way division will be useful in future work is speculation on our part. For present purposes, the finer-grained classification is needed because it offers us the opportunity to explore the frequency distribution of the subtypes in different registers, as we see in the next section.

4.1 Frequency by corpus

The five corpora differ not only in size but also in composition. In order to get a sense of the relative frequency of the ‘adverb-ly adjective’ combination, we took account of the total number of adjectives in each corpus. In addition to traditional normalized frequencies per million words, we calculated normalized frequencies of ‘adverb-ly adjective’ instances over total number of adjectives (in both cases calculating over token counts, not types).

As shown in the middle columns of table 3, we find that the ‘adverb-ly adjective’ construction varies in frequency across corpora, with the Movie$ corpus having a dramatically higher frequency, both per million words and per 1,000 adjectives. The next highest frequency per 1,000 adjectives is found in the TIME corpus, followed at a distance by the other corpora. We speculate that the high frequency found in Movie$ is due to its evaluative nature, as it contains reviews of movies, the register that is more likely to include novel combinations of the Reaction type, a theory that we explore further in sections 4.4 and 4.5.

Table 3. Frequency of ‘adverb-ly adjective’ instances in five corpora

4.2 Frequency over time

The first question we wanted to answer was whether the ‘adverb-ly adjective’ construction has been increasing in frequency over time. For this purpose, we turned to the corpus that is more clearly diachronic, TIME. This corpus contains one hundred years’ worth of data from Time magazine. The answer seems to be that the construction has indeed been increasing over time, having perhaps peaked around the 1990s. Figure 1 displays the frequency of the ‘adverb-ly adjective’ construction (a total of 205,678 instances across the entire period) with respect to the total number of all adjectives in the TIME corpus (over 4.3 million). We see that the construction was relatively popular in the 1920s–50s period, ranging from 3.7 per cent in the 1940s to 4.4 per cent in the 1920s, where we see examples such as largely responsible, entirely new or seriously ill. In the 1960s, there is a notable step-up, peaking at 6.3 per cent in the 1990s, where politically correct, politically incorrect and mentally ill were among the most frequent combinations.

Figure 1. ‘Adverb-ly adjective’ construction as a percentage of all adjectives, TIME corpus

In general, the trend is upwards, as shown with a different perspective in figure 2, this time with frequency per million words. There are peaks in the 1970s and 1990s, with a clear upward trend line (the solid blue line, with its dispersion in grey). In summary, in the TIME corpus, ‘adverb-ly adjective’ is a frequent construction and has been increasing in frequency over time.

Figure 2. ‘Adverb-ly adjective’ constructions in TIME corpus, frequency per million words

4.3 Frequency across registers

As mentioned, our initial intuition was that ‘adverb-ly adjective’ combinations were more frequent in evaluative registers, such as in the book and movie reviews we studied for another project (Goddard et al. Reference Goddard, Taboada and Trnavac2019). We also had the intuition that there were differences in the relative frequency of subtypes, with some subtypes appearing more frequently in certain registers. To examine this further, we counted the number of ‘adverb-ly adjective’ combinations in the corpora for which we have register categories, i.e. COCA and CORE.

COCA was subdivided, at the time of data collection, into five main registers: Fiction, Spoken, News, Academic and Magazine.Footnote ⁷ We categorized the frequency of the construction per million words, as shown in figure 3. We found that Magazine and Academic top the list, with News and Spoken next, and Fiction last. This makes sense, as much writing in magazines is stylistically similar to movie and film reviews: it piles on modifiers and tries to be creative. As in reviews of creative products, the need is for descriptions that are vivid and memorable, with examples like intensely personal, aberrantly aggressive, unquestionably gorgeous and winsomely generous. In addition, magazine writing includes recipes, which explains the top five constructions in Magazine: freshly ground, really good, finely chopped, thinly sliced and relatively small.

Figure 3. ‘Adverb-ly adjective’ constructions in COCA, by register

The high frequency of ‘adverb-ly adjective’ combinations in academic writing is largely due to the Topical subtypes described in section 3.5, though Degree and Focus subtypes are also much used. Top combinations in the Academic section include statistically significant, relatively small, culturally diverse and commonly used. In general, academic writing uses more specialized vocabulary and denser noun phrases. Some of the constructions in academic writing are of the type that become acronyms, such as genetically modified organisms (GMOs). Some of the combinations are also related to the language of bureaucracy, such as federally funded, often found in academic writing.Footnote ⁸

In CORE (figure 4), Information Persuasion is the register that shows the highest frequency of ‘adverb-ly adjective’ constructions. Information Persuasion includes information blogs, which may be opinionated but not necessarily creative, with examples such as really good, really nice, completely different and completely new at the top of the distribution for this register. Next are Spoken and Opinion. The Spoken register features highly evaluative language in interviews and speeches, in addition to broadcast transcripts. The top examples in Spoken often contain really (really good, really cool, really important). This seems to confirm the intuition that the construction is more frequent in opinionated language. At the same time, the data seem to show that in informal registers, such as those in CORE, creativity is not the top priority, with instances coming predominately from the Degree and Focus subtypes.Footnote ⁹

Figure 4. ‘Adverb-ly adjective’ constructions in CORE, by register

4.4 Distribution by subtype, at the top

We have thus far examined the distribution of the ‘adverb-ly adjective’ construction in several corpora, finding that it seems to be increasing in relative frequency over time and that it is more frequent in evaluative language. We next wanted to explore whether some of the semantic subtypes are more frequent than others. To do so, the only available procedure consists of painstaking annotation of each instance. This is because the typology is very much meaning-driven and thus relies on human judgement. Given the time-consuming nature of such annotation, we considered only a small subset of the examples, starting with the most frequent cases, i.e. the distribution at the top of the frequency scale.

We extracted the top 200 combinations in COCA and coded them into the subtypes identified in section 3. COCA was chosen as the most varied and representative corpus. Figure 5 shows that Degree and Focus clearly dominate; of the two, Degree is much more frequent, at 39 per cent of the total. The instances are common cases with absolutely, barely or especially (absolutely amazing, barely able, especially difficult). The next most frequent subtype is Topical, with examples like academically talented, clinically significant and environmentally sound. The Time category includes the examples currently available and freshly baked. The other subtypes contain fewer instances, perhaps notable among them actively involved (Manner), deceptively simple (Reaction) and certainly possible (Epistemic).

Figure 5. ‘Adverb-ly adjective’ subtypes in COCA (200 most frequent combinations)

We also coded the most frequent 200 combinations in Movie$ corpus, and here we see a picture which is similar in some respects but notably different in others (see figure 6). Degree/Focus still tops the list, as do Topical and Time. Reaction, however, increases significantly, with instances like perilously close, explosively funny and oddly endearing – precisely the type of more creative combinations that we are interested in. Epistemic examples also go up in frequency as compared to the COCA distribution, including instances such as immediately clear, predictably disastrous and undeniably funny.

Figure 6. ‘Adverb-ly adjective’ subtypes in Movie$ (200 most frequent combinations)

This comparison, between a more general corpus like COCA and a specialized dataset like the Movie$ corpus (albeit with a small amount of data), is consistent with the initial intuition that Reaction is more frequent in creative and evaluative registers than in the general language.

4.5 Distribution by subtype, at the bottom

To truly test the creativity of the ‘adverbly-adjective’ construction, however, the best approach is to examine not the most frequent combinations, but the least frequent; that is, those combinations at the bottom or tail of the distribution. To that end, we collected all the hapax legomena in three of our corpora. Hapax legomena, nonce words, or hapaxes, are instances that occur only once in the corpus under study (Manning & Schütze Reference Manning and Schütze2003: 199). Examining hapax ‘adverb-ly adjective’ combinations may give us a sense of which combinations are unique and novel, and in what proportion relative to the semantic subtypes we have identified.

We began by counting the relative frequency of hapaxes in three different corpora: SOCC, CORE and Movie$. SOCC is a corpus of online comments, containing comments on very many different topics from the news, and thus highly evaluative. CORE is a corpus of online registers. As we have seen, it contains more instances of the ‘adverb-ly adjective’ combination per million words than COCA or TIME, perhaps because it contains more opinion, a defining feature of online registers (Biber & Egber Reference Biber and Egbert2018). Finally, Movie$ is all opinion, as it consists entirely of film reviews. Table 4 shows the hapax frequencies in these three corpora. As shown in the second and third rows, ‘adverb-ly adjective’ hapaxes increase across the three corpora, consistent with our view of how opinionated each corpus is and how much creativity is valued in each register. The proportionate frequency in Movie$ is strikingly higher.

Table 4. Relative frequency of ‘adverb-ly adjective’ hapaxes in three corpora

We then turned to the question of semantic subtype. We collected a random sample of 200 hapaxes from the Movie$ corpus and annotated them, revealing the distribution of subtypes shown in figure 7. Compared with figures 5 and 6 (in section 4.4), which presented the distribution at the top, there are dramatic differences. Manner and Reaction have increased sharply, indicating that these subtypes are more likely to include unique and novel combinations, i.e. hapaxes. This increase comes at the expense of the Degree/Focus type, which seems to contain the more frequent combinations.

Figure 7. Hapax legomena subtypes in Movie$ (200 random examples)

4.6 Oxymoronic ‘adverb-ly adjective’ combinations

A particularly interesting subset of ‘adverb-ly adjective’ combinations are those in which the adverb and the adjective seem to have opposing polarities, from the point of view of sentiment or affect (Martin & White Reference Martin and White2005; Taboada Reference Taboada2016). Take, for example, the phrase wonderfully imperfect, in which the word wonderfully has strongly negative sentiment, whereas imperfect has a negative sentiment. This is a classic example of an oxymoron, an apparent contradiction in terms. We previously found, in the context of work in sentiment analysis (Taboada et al. Reference Taboada, Brooke, Tofiloski, Voll and Stede2011; Taboada Reference Taboada2016; Benamara et al. Reference Benamara, Taboada and Mathieu2017; Taboada et al. Reference Taboada, Trnavac and Goddard2017), that many such combinations could be classified as oxymorons, presumably used for their creativity and impact.

We therefore extracted oxymorons and attempted to classify them into our semantic subtypes. The process we followed started with splitting the ‘adverb-ly adjective’ pair into separate adverb and adjective units, then consulting their polarity and strength in standard sentiment dictionaries. If the adverb was not found in the dictionary, we converted it into an adjective by stripping the -ly, assuming that the polarity would be the same for the adjective and its derived adverb. Once we had the polarity for each member of the pair, we could compute the absolute difference in polarities, to identify those that were oxymorons. Part of the decision in this respect had to do with what threshold constituted enough of a difference to call something an oxymoron. We used SentiWordNet 3.0 (SWN), a standard dictionary for sentiment analysis (Baccianella et al. Reference Baccianella, Esuli and Sebastiani2010) with a high coverage for most of the words we were looking up (it was missing only about 4 per cent of the words).Footnote ¹⁰

We found that, in all corpora, the combinations with the highest difference between the terms had to do with a positive adverb modifying a negative adjective, creating a novel and surprising combination, often in the Reaction class. COCA contained instances such as comically nefarious, genuinely disgusting and wonderfully imperfect. In CORE, we found amusingly sardonic, thoroughly ruthless and beautifully heartbreaking. TIME provided deplorably high-minded, amusingly mangled and zealously moderate.

In some cases, we find a negative adverb modifying either positive or negative adjectives, in a process of delexicalization (Partington Reference Partington, Baker, Tognini-Bonelli and Francis1993; Tagliamonte Reference Tagliamonte2008; Schweinberger Reference Schweinberger, Lucek and Amador-Moreno2021; Nouwen Reference Nouwen2024), whereby intensifiers such as awfully, terribly or frightfully are mostly delexicalized when combined with positive adjectives. Many of the instances we found as oxymorons were also made up of a Degree or Focus adverb, typically positive, followed by a negative adjective. Examples included combinations with the adverbs absolutely, barely, certainly, entirely, fundamentally, genuinely, totally or truly. In these cases, we would argue that the adverb does not contribute positive meaning, as per SentiWordNet 3.0, but only an intensifying meaning, whether up or down.