1. Introduction
Geoff Pullum’s career has been characterized by meticulous reevaluations of established linguistic conventions on a remarkably broad range of topics, from the philosophical basis of linguistics to the rigorous categorization of individual words (e.g. Huddleston & Pullum Reference Huddleston and Pullum2002, Payne, Huddleston & Pullum Reference Payne, Huddleston and Pullum2007, Reference Payne, Huddleston and Pullum2010). In a similar spirit of reevaluation, I argue that more and less are always determinatives, contrary to the categorization in The Cambridge grammar of the English language (CGEL) (Huddleston & Pullum Reference Huddleston and Pullum2002).
CGEL holds that while more and less are generally determinatives, they are adverbs exceptionally in the context of analytic comparatives, such as in more interesting or less quickly. The justification for this categorization is that analytic more ‘does not enter into any [degree modifier] contrast with much: we can say This is more porous than that, but not *is this much porous?’ (Huddleston & Pullum Reference Huddleston and Pullum2002: 1123).
This paper challenges that claim, arguing that such contrasts do in fact exist. I propose that the distributional facts can largely be explained by the semantics of -er/more and much. Specifically, following Zhang & Ling (Reference Zhang and Ling2021), I adopt the position that -er/more establishes a salient minimum value in the discourse (Zhang & Ling Reference Zhang and Ling2021) where none might otherwise exist. And following Kennedy & McNally (Reference Kennedy and McNally2005), I argue that much requires such a value in order to function as a modifier. This semantic interplay between -er/more and much provides a pragmasemantic explanation for their distribution, without needing to posit a category distinction.
Furthermore, I argue that a mere lack of contrast should not be relied upon for making categorial determinations in any case, following the arguments by Payne et al. (Reference Payne, Huddleston and Pullum2010). For these reasons, positing two different lexemes for more and less is neither necessary nor parsimonious. The determinative analysis can account for all of the relevant data.
The paper is structured as follows: Section 2 lays out CGEL’s analysis of more and less and the notion of analytic comparatives. Section 3 presents contrasts between more/less and much/little in various syntactic contexts, challenging CGEL’s empirical claim. Section 4 examines the complex factors influencing the distribution of degree modifiers and argues that more and less do not pattern distinctly from other determinatives. Section 5 proposes a semantic explanation based on scale structure for the distributional patterns observed. Finally, Section 6 argues against relying on distributional complementarity for making categorial distinctions, before the paper concludes in favor of a unified determinative analysis of more and less.
2. CGEL analysis
This section introduces the CGEL framework that are essential for understanding the subsequent analysis.
2.1. Terminology
CGEL (24) distinguishes between lexical and phrasal categories on one hand and syntactic functions on the other. The lexical category of a word is what you would find in the dictionary – its part of speech. Phrasal categories are derived from these lexical categories. Syntactic functions, in contrast, are relational notions. For example, happy is always and only a member of the adjective category, but the adjective phrase (AdjP) very happy can function as a modifier in a very happy child or as a predicative complement in the child is very happy . I argue that the lexemes more and less belong to the lexical category of determinative. In CGEL, determinatives include articles, demonstratives, cardinal numbers, universals such as all and both, and a few other items. These elements head determinative phrases (DPs), which should not be confused with the notion of a ‘determiner phrase’ in the DP hypothesis. Perhaps surprisingly, CGEL and Chomsky (Reference Chomsky2020) agree on this point: the widely accepted DP hypothesis is mistaken, despite its prevalence in generative syntax. In this paper, a phrase like far fewer is analyzed as a DP, while a phrase like this word is considered a noun phrase (NP).
The main syntactic function that DPs perform is the determiner function in NPs (e.g. every change). But this is a many-to-many relationship: DPs serve other functions as well, such as modifier in AdjPs (e.g., this happy). Conversely, other constituents besides DPs – primarily NPs – can also function as determiners in NPs (e.g. the group ’ s output).
2.2. CGEL’s analysis of more and less as adverbs
CGEL (539) analyzes more and most as inflected forms of the determinative much, and similarly less and least as inflected forms of little, with one exception: in analytic comparatives, they are considered adverbs (more
$ {}_{\mathrm{a}} $
, less
$ {}_{\mathrm{a}} $
). CGEL states: ‘For the comparative category, analytic marking is by means of the adverb more’ (1123), and extends this analysis to most
$ {}_{\mathrm{a}} $
, less
$ {}_{\mathrm{a}} $
, and least
$ {}_{\mathrm{a}} $
(64). This adverb classification is not due to their role as degree modifiers (395, 459, 549), since CGEL notes that ‘apart from the interrogatives and relatives, virtually all determinatives that can occur in NP structure with a non-count singular head can also function as modifier to verbs and/or adjectives and adverbs’ (565). Rather, it hinges specifically on the claim that more fails to enter into any degree of modifier contrast with much in this environment.
Much and little (all forms) occur as degree adjunct in clause structure: Jill little realised what they were planning; It didn’t hurt as much as last time. The plain forms much and little modify comparative expressions: much better, little different, much more cheese, little less intrusive. Very much modifies a wider range of expressions: very much in control, very much an intellectual. (More and less modify adjectives, adverbs, etc., but we take these to be degree adverbs, rather than comparative forms of much and little: see Ch. 13, §4.1.1.). (CGEL 395)
Notably, more and less are the only determinatives in CGEL that are argued to be homonymous with adverbs. In categorizing them as adverbs in analytic comparatives, CGEL aligns with the long tradition in English grammars and dictionaries. For example, this is the position taken by Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 463) and the Oxford English Dictionary. Payne et al. (Reference Payne, Huddleston and Pullum2010: 37) define the ‘distributional core’ of adverbs as follows:
Any item which can appear after a subject and before a verb (and does not by other distributional criteria belong to another category) will be adjudged to belong to the adverb distributional core.
Although rarely, more and less do meet this criterion, as in Example (1).
However, CGEL analyzes such cases as containing determinatives, diverging from the traditional adverb analysis (534). The adverb categorization is applied specifically to analytic comparative constructions. In the following sections, I argue that this restricted dual categorization is unmotivated, and that a unified determinative analysis is preferable.
2.3. Analytic comparatives
The term analytic comparative refers to a construction where ‘separate words realize grammatical distinctions that in other languages [or in other contexts in the same language] may be realized by inflections’ (Matthews Reference Matthews2003). While CGEL restricts its discussion of analytic comparatives to AdjPs and AdvPs as in (2; 533), it’s worth considering other possible analytic comparative constructions.
There are clearly gradable preposition phrases (PPs) which allow comparative forms, either inflectionally as in closer to home or analytically, as in Example (3a). These would seem to qualify as analytic comparatives. Verb phrases (VPs) and NPs like Examples (3b and 3c) appear structurally quite similar, but lack inflectionally comparative counterparts in English, so it may be inappropriate to consider them true analytic comparatives.
Interestingly, while CGEL explicitly analyzes more and less as determinatives in PPs like more out of sorts (395, 533), it does not extend the adverb analysis to these clear cases of PP-based analytic comparatives. This seems like an inconsistency, especially since the lack of contrast with much that CGEL uses to justify the adverb categorization applies here as well.
2.4. Summary of the CGEL analysis
In summary, while CGEL generally analyzes determinatives as being able to modify a wide range of phrase types without exhibiting homonymy with adverbs, it makes an exception for more and less in analytic comparative AdjPs and AdvPs, though not PPs. The sole basis given for this dual categorization is the claim that more fails to contrast with much in these environments.
In the following sections, I demonstrate that such contrasts do in fact occur, even in AdjPs and AdvPs. For the remainder of this paper, I focus primarily on much and more, but the general argument applies equally to less and little.
3. Contrasts in Various Contexts
This section presents evidence of contrasts between much and more in various syntactic environments, challenging CGEL’s claim that such contrasts never occur in analytic comparatives.
3.1. Contrasts in AdjPs with plain-form heads
To identify which adjectives head AdjPs that allow modification by much, I searched the Corpus of Contemporary American English (COCA, from which all examples are taken unless otherwise specified; Davies Reference Davies2023).Footnote 1 The query [be] much JJ returns hits for all forms of be (e.g. is, were, being) followed by much plus an adjective, excluding negated forms like isn’t. As expected, the results are dominated by comparative adjectives, but some plain-form adjectives also appear.Footnote 2 Similar searches with [seem] much JJ and [become] much JJ produce the set of adjectives in Table 1. I then searched for each of these adjectives preceded by be/seem/become + more (e.g. [be] more akin). The resulting frequencies are shown in Table 1.
Table 1 The counts of all words tagged as adjectives in COCA heading AdjPs modified by much and more in complement function for any form of be, seem, or become, along with the ratio of much:more tokens.
a These counts unfortunately conflate adjective and preposition uses due to tagging errors.
b Four of these are determiner uses of much, and the rest appear to be errors (e.g. *It’s much hard to anticipate).
Strikingly, much not only modifies plain-form adjective heads in these AdjPs, but sometimes does so more frequently than more (e.g. with different). Moreover, despite much being a negative polarity item (NPI) in many contexts, it exhibits little polarity sensitivity with the adjectives in Table 1. The distribution of much is even broader when it is itself modified (e.g. very/as/pretty much equal; Huddleston & Pullum Reference Huddleston and Pullum2002).Footnote 3
The existence of these contrasts undermines CGEL’s motivation for a dual-category analysis of more by demonstrating that much and more are not in strictly complementary distribution. The adjectives in Table 1 appear to fall into four main groups: comparative governors, past participials, a- adjectives (see §3.1.3), and others.Footnote 4
3.1.1. Comparative governors
Much–more contrasts occur with what CGEL (1,104) calls comparative governors, items which license comparative complements: different, superior, inferior.Footnote 5 Additional comparative governor adjectives occurring in the relevant contexts in COCA, though tagged as other categories, are given in Example (4).
The comparative governors in Example (4a) co-occur with much, more, most, little, less, and least in COCA. Those in Example (4b) are unattested with little, and there are no instances of least akin/superior, though all seem possible. Little is generally rare as an AdjP modifier. Some other comparative governor adjectives like other, such, and else tend to resist degree modification altogether (see Section 5 for discussion).
3.1.2. Past-participial adjectives
Jespersen (Reference Jespersen1956: 399) observes that with past participles, ‘much, which was required on account of the verbal character’, was increasingly being replaced by very.Footnote 6 Some of the lower-frequency items in Table 1 seem marginal to me, but all except unsinewed, diseased, and well-versed qualify as adjectives by CGEL’s criteria (Ch. 16, §10.1.3).
Other cases such as much broken and much frightened do not occur in the corpus with non-modified much (see Footnote footnote 3) and seem less acceptable. Many of these adjectives also exhibit little/less contrasts (e.g. little/less concerned).
3.1.3. A adjectives
The adjectives in Example (6) start with a, resist attributive use, and are not participles. Those in Example (6a) contain the originally prepositional prefix a- meaning ‘on, in, into’. Both alike and alive seem to be comparative governors in that they require two semantic arguments: the subject of alike must be plural or conjoined (X and Y are alike, we/my brothers are alike), while akin explicitly licenses a PP complement. The a in the adjectives in Example (6b) is not the same prefix, but they may pattern similarly by analogy.
Expanding the search beyond the be/seem/become frames, there are two instances of much averse versus 23 of more averse in COCA, averse being another a-prefixed adjective from CGEL’s list (559).
3.2. Less and least with comparatives and superlatives
Less rarely occurs as a pre-head modifier of comparative adjectives, but Jespersen (Reference Jespersen1956: 368) mentions less happier, and COCA contains 22 instances of less worse, four of less happier, and three of less riskier, along with a few other sporadic examples. These clearly contrast with little worse/happier/riskier. Similarly, least occurs with worst and other superlatives. COCA also has three instances of less modifying analytic comparatives: less more likely/efficient/vulnerable, contrasting with little more likely/efficient/vulnerable.
3.3. Contrasts in other phrases
Having shown in detail in Section 3.1 that contrasts exist in analytic comparative AdjPs, I use this section to briefly illustrate similar patterns in other phrasal categories.
3.3.1. Contrasts in AdvPs
Contrasts in AdvPs are even rarer than in AdjPs. Much differently (181 instances) contrasts with the analytic comparative more differently (16 instances), and would more rather contrasts with would much rather. Much too (e.g. much too good) is a rare case of much with a plain adverb head, but curiously, *more too good is impossible, so there is no contrast in this case. Nevertheless, it is an interesting construction, to which I return in Section 5.1.1.
3.4. Contrasts in PPs
In PPs, CGEL analyzes more and less as determinatives, but given the existence of synthetic comparatives in PPs like closer to home, it is reasonable to consider examples like Example (7) as involving contrasts in analytic comparatives (see Section 2.3).
Many of the hits for like in Table 1, and perhaps alike as well, are actually in preposition phrases. As is also common in these PPs, as in Example (8).
3.5. Contrasts in VPs
Although VPs modified by more are not considered analytic comparatives by CGEL, contrasts with much do occur (mostly limited to negative polarity contexts), as shown in Example (9).
Relevant contrasts in post-head modifiers are also common, as in Example (10).
3.6. Summary of contrasts
The data presented in this section challenge CGEL’s claim that more and much never contrast in analytic comparatives. While such contrasts are not especially frequent, they do occur across a range of AdjPs, AdvPs, and PPs. The next section investigates whether the observed distributional differences between more and much, even if not strictly complementary, could still justify a category distinction.
4. The Vagaries of Modification
In this section, I examine the differences in the distribution of modifiers within a given category across different phrase types. I conclude that, even if the contrasts identified in Section 3 could somehow be explained away, the distributional differences between more and much fall within the expected range of variation for determinative intensifiers and do not by themselves justify a dual-category analysis for more.
Intensifier distribution in adjective phrases is notoriously idiosyncratic. Bolinger (Reference Bolinger1972) devotes an entire 30-page chapter to ‘Some restrictions on intensifiers primarily with adjectives’, yet describes it as ‘perhaps better than a sampling, but … far from complete’. Restrictions may be dialectal (e.g. right pleased), register-specific (e.g. We were little affected by what we saw), or positional (enough only occurs post-head). Semantic and prosodic factors also play a role: highly frightful is odd because ‘frightful is already stronger than highly … In addition to the semantic restriction there is a tendency to avoid mono-syllabic adjectives’ (Bolinger Reference Bolinger1972: 52).
This idiosyncrasy extends to the choice between analytic and synthetic comparatives. Jespersen (Reference Jespersen1956: 359), noting that ‘it is not always easy to see why writers prefer one or the other method of comparing adjs’, dedicates nine pages to apparently unpredictable cases of periphrasis versus inflection in AdjPs, plus two more pages on AdvPs. He observes, ‘the periphrastic comparatives and superlatives with preposed more and most are found not only in those cases in which the endings -er and -est cannot be used for phonetic reasons, but also extensively in other cases” (Jespersen Reference Jespersen1956: 382).
4.1. Determinatives as modifiers
If more and less in analytic comparatives patterned distinctly from other determinatives in AdjPs, this could potentially support an adverb analysis. But the distribution of determinative modifiers in AdjPs is quite varied, as shown in Table 2.Footnote 7
Table 2 Determinatives as modifiers in AdjPs.
a Post-head
The old cluster includes old, fast, small, hard, young, strong, and high. The recent group includes intelligent, dangerous, fortunate, generous, embarrassing, expensive, and comfortable. The comparative group includes all words tagged as comparative adjectives (e.g. much JJR). The different group comprises different, separate, preferable, superior, unequal, and inferior. The afraid group includes afraid, alike, afraid, alive, alone, and aware. And the improved group contains improved, pleased, delighted, concerned, beloved, interested, and surprised.
Enough and that pattern similarly to more in their ability to modify a wide range of AdjPs, but not synthetic comparatives like bigger. In contrast, much is more restricted, while no and any are the most limited. Viewed in this broader context, the distributional data do not single out more as exceptional.
For AdvPs (Table 3), I’ve included a little, which patterns like much in modifying too, in place of little which does not occur in this context. The determinative all (not shown) also modifies too. Once again, no distributional pattern clearly identifies more as belonging to a distinct category.
Table 3 Determinatives as modifiers in AdvPs.
a Just for the words given + all tagged as JJR for Comparative
b Post-head
Table 4 shows determinative modifiers in PPs. It is difficult to identify relevant preposition classes, so the items included are more opportunistic and reflect strings rather than constituents. For example, up to covers up to date, up to the individual, and similar cases. Still, no clear pattern emerges suggesting that more alone is an adverb.
Table 4 Determinatives as modifiers in PPs.
a Pre- or post-head
What this highly selective comparison illustrates is considerable variation in the distribution of determinatives as modifiers. This undermines the argument for assigning more and less in analytic comparatives to a distinct category based solely on distributional facts. The differences between their behavior and that of much or little appear to fall within the expected range of variation for determinatives, potentially explainable by semantic factors.
4.2. Adverbs as modifiers
If more and less in analytic comparatives patterned similarly to adverbs in their distribution as pre-head modifiers in phrases with gradable heads, this could provide evidence for an adverb analysis. However, as shown in Table 5, more exhibits a distribution quite distinct from other adverbs. In fact, there is no clear pattern; each of the eight adverbs considered has its own unique distribution.
Table 5 Selection of more, less, and adverbs as modifiers.
Clearly, there is no typical pattern for adverbs functioning as modifiers in these contexts. Even so, more appears unusually restricted in its distribution, with less only slightly less so. At minimum, Table 5 provides no positive evidence for CGEL’s claim that more and less in analytic comparatives are adverbs.
5. The Explanatory Force of Scales
Although distributional evidence may not justify categorizing more and less as adverbs, their differences from much and little still call for explanation. Building on the distributional analysis in the previous sections, this section appeals to the scale structure of adjectives (and other categories) to account for the differing distributions of more and less versus much and little.
Stevens (Reference Stevens1946) proposed a classification system for attribute data, divided into four levels, as shown in Figure 1 (adapted from Zhang & Ling Reference Zhang and Ling2021: 250). These scales provide a framework for understanding the differing distributions of intensifiers and modifiers based on the scale structure of the adjectives they modify.
-
• Nominal Scale: Categorizes data into distinct groups based on qualitative properties.
-
• Ordinal Scale: Categorizes and orders data, but does not provide information about the intervals between categories.
-
• Interval Scale: Categorizes, orders, and establishes equal intervals between categories, but lacks a true zero point.
-
• Ratio Scale: Categorizes, orders, establishes equal intervals, and includes a true zero point.
Figure 1 Venn diagram illustrating Stevens (Reference Stevens1946)’s scales hierarchy.
CGEL distinguishes between gradable (
$ \hskip-0.35em \approx \hskip-0.35em $
ordinal, e.g. important) and non-gradable (
$ \hskip-0.35em \approx \hskip-0.35em $
nominal, e.g. mutual) adjectives, hinting at scalar constraints on possible modifiers in AdjPs. But further distinctions can be made based on the other scale levels (Table 6). As more semantic scale levels are added, more constraints are imposed. For example, one sixth as dense is possible because density is a ratio scale, but ?one sixth as kind seems anomalous, presumably because kindness is not a ratio scale.
Table 6 Syntactic compatibility of intensifiers with plain-form adjectives of different semantic scales.
a Half as good or twice as nice are pseudo-ratio modifiers merely meaning ‘much worse’ or ‘much better’ without a genuine multiple.
b The ordinal adjectives assign an ordinal rank, but each rank has only a nominal scale (something is or is not first).
c The sense of these adjectives ‘corresponds to their interpretation when they are associated with units’ (Sassoon Reference Sassoon2007: 243), in contrast to the general sense in He retired before he grew old.
Scale structure can impose further constraints on possible modifiers. Scales can be open or closed at either end (Kennedy & McNally Reference Kennedy and McNally2005). Totalizing and approximating modifiers work best with an inherent upper bound (e.g. completely/almost straight), but are odd with open-ended scales (e.g. ?completely/almost bent). Conversely, minimizing modifiers prefer a lower bound (e.g. slightly bent), and are degraded without one (e.g. ?slightly straight). Scales can also have contextually determined reference points based on expected norms, as with tall meaning ‘of a height noticeably greater than normal for the relevant reference group’ (vs. e.g. 1.7 m tall).
While the existence of the contrasts in Section 3 undermines CGEL’s motivation for a dual-category analysis, more and much (and less and little) do exhibit different distributions. I propose that these differences can largely be explained by their semantics and the scale structure of the expressions they modify, without positing distinct syntactic categories. Specifically, I argue that the limited distribution of much and little as modifiers follows from their requirement for a discourse-salient reference point on the relevant scale, while the broader distribution of more/-er and less stems from their ability to establish such a reference point. This semantic account captures the observed distributional patterns while maintaining a unified categorial treatment of these items as determinatives.
5.1. Scale-structure limitations on much and related modifiers
As shown in Section 4, the semantic properties of adjectives, particularly their scale structure, play a significant role in determining which modifiers they allow. For example, adjectives that inherently encode a limit or boundary on a scale, such as straight or pure, are more compatible with modifiers like completely or almost that pick out the endpoints of a scale, while adjectives that lack such boundaries, like big or interesting, are not. This is because the semantics of the modifier must be compatible with the scale structure of the modified expression.
Given this general principle, it should not be surprising that specific restrictions might apply to the modifier much. As a modifier in an AdjP, much requires the presence of a discourse-salient, scale-internal reference point for comparison (Kennedy & McNally Reference Kennedy and McNally2005). Crucially, this reference point need not coincide with a norm or average value on the relevant scale, although it may do so coincidentally. Rather, the reference point is typically established explicitly in the discourse context, often by a comparative complement. For instance, in the AdjP much taller than me, the reference point is the height of the speaker, not the average height for a person. Thus, while much requires a reference point, it is not necessarily a norm or average, but rather a contextually salient scale value.
Establishing such a point is one function of -er/more (Zhang & Ling Reference Zhang and Ling2021), an idea with precedents in earlier work on the semantics of comparatives.Footnote 8 The point may be implicit, but it can typically be made explicit in a comparative complement like a than-PP. For instance, It happened much more recently than yesterday picks out yesterday as the relevant point of reference. Once this point is established, much and related modifiers become possible, as shown in Figure 2. Without such a point, much is anomalous because the comparison size is indeterminate.Footnote 9
Figure 2 Illustration of the interaction between more/-er’s than-complement and much in It happened much more recently than yesterday.
At the same time, other modifiers are blocked, including most, too, very, and extremely. *Much most/-est is likely ruled out because most/-est has a strictly ordinal semantics, while much – like slightly, somewhat, significantly, greatly, and other similar degree adverbs (?) – requires an interval semantics. Very and extremely may be incompatible because they target an absolutely high degree on the scale, without establishing the relevant reference point. In comparative constructions, it is the difference in degrees, not the absolute degree, that matters.
5.1.1. Too
Too presents an interesting case. As noted in Section 3.3.1, plain-form adverbs generally disallow modification by much, but too is an exception, as in much too big. Notably, too resembles more/-er in its semantics and syntax. Jespersen (Reference Jespersen1956: 391) calls it a ‘latent comparative’, and von Stechow (Reference von Stechow1984) analyzes it as a degree operator involving universal quantification over degrees, similar to comparative morphemes. Both too and more/-er function primarily as modifiers in phrases headed by plain ordinal adjectives, adverbs, or prepositions; both license a complement (a to-infinitival for too); and in both cases, the complement’s semantic value establishes the scale-internal, lower reference point. The key difference is that, without an overt complement, too sets its reference level to the maximum acceptable level based on pragmatics and social norms (Meier Reference Meier2003). It is too’s lower reference level that allows much to occur in AdjPs like It’s much too big (to fit), as shown in Figure 3.
Figure 3 Illustration of the interaction between too’s to-complement and much in It is much too big to fit.
Conversely, even though more can often modify plain-form adverbs, *more too big (to fit) (than that) is impossible because more and too establish competing points of comparison, making it unclear whether the comparison is with the size of another object or with the fit size.
It’s worth noting, however, that the unacceptability of these constructions could potentially be explained by factors other than the semantic incompatibility proposed here. As pointed out by an anonymous reviewer, the difficulty in processing sentences like Kim is less happier than Lee than Pat might contribute to their perceived unacceptability. The complex structure of such sentences, with multiple embedded comparisons, poses a significant cognitive challenge. This raises the possibility that the unacceptability is due more to grammatical or processing constraints than to any fundamental cognitive limitation on combining these degree modifiers.
In sum, an adjective head’s semantic scale structure significantly impacts the modifiers it allows, and these restrictions offer a compelling explanation for why much and more generally appear in different contexts. However, as shown in Section 3, there are contexts where both more and much may alternate. I now turn to explaining these.
5.2. Comparative governors
In Section 3.1.1, I showed that comparative governors like different allow modification by both more and much. In Section 5.1, following Kennedy & McNally (Reference Kennedy and McNally2005), I claimed that much-type modifiers require an established scale-internal reference point. And following Zhang & Ling (Reference Zhang and Ling2021), I claimed that more/-er establish such a point. It’s unsurprising that comparative governors allow more/-er modification, but it remains to be seen why they allow much without more/-er to establish a reference point.
I propose that, like more/-er, comparative governors allow much modification because their semantics establish a reference point, which may be expressed in a than or from PP complement. In Example (11a), the reference point is a contextually salient price made explicit in the from-PP.
But if that’s the case, why is more also possible? The answer is that it establishes (implicitly or explicitly) a distinct reference point on a second-order scale – a price-difference scale. Metaphorically, more different is to different as acceleration is to speed. In Example (11b), the reference point is not another price, but rather the normal or expected difference between prices. In Example (11b), the reference point is not another price but rather the normal or expected difference between prices.
5.3. Participial adjectives
In Section 3.1.2, I showed that non-comparative participial adjectives like improved, refreshed, recovered, and diminished may head AdjPs in which much is a modifier. The explanation is the same as above: these adjectives tend to have a semantics that establishes a scale-internal reference point, which can sometimes be expressed in a PP complement, as in It’s much improved from the first draft and the examples in Example (12), as illustrated in Figure 4.
Figure 4 Illustration of the interaction between improved’s from-complement and much in It’s much improved from the first draft.
In contrast, participial adjectives like broken lack this reference point, which disallows much as a modifier in Example(13).
However, this explanation does not cover all of the possible participial adjectives. For instance, inclined and impressed do not seem to establish the expected reference point. Perhaps, then, it is ‘on account of the verbal character’ (Jespersen Reference Jespersen1956: 399), and/or other factors.
An anonymous reviewer suggested Kennedy & McNally (Reference Kennedy and McNally2005)’s analysis of participial adjectives as a different explanation for why these words resist modification by much. Under this analysis, rather than lacking a reference point entirely, these adjectives may have a fixed reference point at the minimal element of their scale. For example, broken always refers to the state of an object just as it becomes broken, not to intermediate states of partial brokenness. This fixed minimal reference point may be incompatible with the semantics of much, which requires a contextually salient reference point that can be shifted.
5.4. The A- adjectives
As with the participial adjectives, some of the a- adjectives seem to have a natural reference point (e.g. akin & alike), while others do not (e.g. alive & afraid). I don’t have a good story, though, for why examples like I was much aware seem much better than ?Something was much amiss.
5.5. Other plain-form adjectives
I have found no other plain adjectives that accept simple much as a modifier in Modern English (e.g. *that is much true),Footnote 10 but there are rare examples like as much true of China …. In such cases, the comparative-governor adverb as may license the relevant reference point. The question then becomes why this is not more broadly applicable. In fact, modified much (e.g. so much) is more flexible than much alone (see Footnote footnote 3) for reasons that are unclear to me. A syntactically identical construction with different semantics does seem to apply more broadly, an example of which is Sanctions are as much psychological as they are punitive.
5.6. Problems with this account
Though the explanation accounts for much of the data, it does not account for all of it. First, as mentioned above, there are past-participial adjectives and a- adjectives that may head AdjPs with modifier much, without establishing any obvious reference point.
Second, most adjectives allow a post-head for PP introducing a comparison class, as in Example (14a), but as Example (14b) shows, this does not satisfy much’s need for a discourse-salient reference point.
An anonymous reviewer suggested an explanation, argued for by Fara (Reference Fara2000) and Kennedy (Reference Kennedy2007): for-phrases specify comparison classes relative to which standard values are computed, but they do not determine those values directly. The actual reference points remain indeterminate, and can vary across contexts, even within the same comparison class (Qing Reference Qing2020). This indeterminacy may explain why for-phrases do not license much modification: they do not provide the definite reference point that much requires.
Third, there are two senses of adjectives such as tall, an ordinal one taking the standard expected tallness for the reference class as its base, as in Example (15a), and a ratio one taking 0 cm as its base, as in Example (15b). Again, neither seems to satisfy much’s need for such a reference value, as illustrated by Example (15c).
The same anonymous reviewer observes that the distribution of measure phrases with non-comparative adjectives is highly idiosyncratic, with somewhat arbitrary variation both within and across languages (Schwarzschild Reference Schwarzschild2005, Grano Reference Grano2012). For example, while tall accepts measure phrases in English, its equivalent in some other languages may not. This idiosyncrasy poses a potential problem for the proposed analysis of much: if the acceptance of measure phrases is taken as a diagnostic for the presence of a reference point, then the variable acceptability of measure phrases across adjectives and languages suggests that the availability of a reference point is similarly variable.
However, a more regular pattern emerges if we focus on comparative constructions. As noted by Schwarzschild (Reference Schwarzschild2005) and Grano (Reference Grano2012), comparatives consistently accept both measure phrases and much-equivalents across languages. This suggests that whatever factor underlies the idiosyncratic distribution of measure phrases with non-comparative adjectives, it is distinct from the factor that licenses much. The consistent acceptability of much in comparatives, which all provide a reference point, supports the proposed analysis of much as requiring a definite reference point.
Nevertheless, the idiosyncratic distribution of measure phrases with non-comparative adjectives remains a puzzle.
Fourth, the story about too may not be the whole story. If it were, then exceedingly might be expected to behave similarly. On the other hand, too generally has a negative affective orientation to it, while exceedingly can have a positive affect, so perhaps their semantics simply aren’t similar enough in the right ways.
5.7. Payne’s counter-proposal
John Payne (personal communication, September 4, 2022), speculated that a difference exists between the adverb more, which deals with degrees, and the determinative more, which deals with quantities.Footnote 11 If a case had already been made for the dual categorization analysis, and I hope I have shown that it has not, then this would be an interesting observation, but it doesn’t seem sufficient to motivate distinct categories.
Moreover, it doesn’t hold consistently across the CGEL analysis. For example, Example (1) is clearly a degree difference, as opposed to I dance more, which would be a quantity difference.Footnote 12 But there are also not so much in control and Kim isn’t much of an actor, which are explicitly determinative uses of much (CGEL 395) and yet clearly require a degree interpretation (CGEL 415). Furthermore, as Payne acknowledges, it isn’t the case that determinatives in general do not deal with degrees. For instance, ‘that seems to be more flexible. As a demonstrative, it can point either to a particular degree for a degree concept or a particular amount for a quantity concept’ (personal communication, September 4, 2022). Nor does this proposal clear up any of the difficulties with the explanation advanced in Section 5.6.
Payne has also suggested (personal communication, October 30, 2023) that perhaps there are two much items, one being an NPI, and the other not. Indeed, in the modifier cases discussed in Section 3, much shows little polarity sensitivity, while elsewhere much is polarity sensitive. But Israel (Reference Israel2011: 41) has observed that ‘even the most robustly polarity sensitive forms tend to have usages which belie their status as polarity items.’ In fact, the overwhelming majority of examples in the OED before 1950 show no polarity sensitivity, while only about half since do (Oxford English Dictionary 2002). Given the above, it wouldn’t do to base a category distinction on polarity sensitivity, nor does such a move have any precedence in CGEL.
5.8. Summary of the explanatory force of scales
Overall, then, scales and the scale structures of individual adjectives have a significant impact over the selection of modifiers allowed by various heads in AdjPs. In other words, semantic and pragmatic factors play an important role in (dis)allowing modifiers. This extends to most of the observed difference in distribution between much and many as modifiers in AdjPs. This does not preclude a categorial difference between much and more (and by extension, little and less), but it certainly undermines the motivation for it.
6. The Complementarity Argument
In arguing that adverbs are a distinct category from adjectives, Payne et al. (Reference Payne, Huddleston and Pullum2010: 61) conclude that,
clearly, in these cases where the same forms are complementary in some environments and contrastive in others, it is not the distribution per se which leads us to think of a derivational relation between wood and wooden, and an inflectional relation between [the Russian nouns] soldat and soldatom. And even if, as a thought experiment, wood and wooden on the one hand and soldat and soldatom on the other always stood in complementary distribution, would this alter our decision? We think not: it seems that factors other than simple distribution are the crucial ones.
Although this quote pertains to the entire lexical category of adverbs, the principle should apply equally to individual words. The question of whether the relations between much and more and between little and less should be considered inflectional in all cases or only sometimes should not hinge on whether or not they stand in complementary distribution. And yet, the fundamental assumption underlying the categorization of more as an adverb relies on precisely this reasoning. Table 7 demonstrates the lack of a consistent relationship between distributional contrast and category assignment in CGEL’s analyses.
Table 7 Distribution and category contrast.
6.1. Little and less?
The arguments in this paper have focused primarily on the much/more distinction, but they apply equally to little and less. This contrasts with CGEL’s original claim about the lack of contrast. While AdjPs like less worse and less happier are not common, an example like Example (16) seems perfectly acceptable to me. In this case, there was a deal which was worse than the current situation, and now there’s a third deal which is still worse, but slightly less so.
If only half of the analytic comparative pair (more/less) truly lacked contrast – and to be clear, I argue that neither lacks contrast – then in which direction should a complementarity argument pull, assuming such arguments had any force to begin with? It’s not at all clear that it would necessarily group less with more into a different category from much and little.
6.2. Categorization options
Even if all the evidence and arguments presented here were set aside, and it could be shown conclusively that more and much belonged to different categories, the question of which specific categories they belong to would still need to be addressed. While determinative and adverb are plausible options, CGEL offers no principled reason for assigning more to the adverb category as opposed to much. The choice seems arbitrary.
7. Conclusion
In this paper, I have argued that the words more and less are determinatives in all contexts, contrary to their categorization in CGEL. I have shown that CGEL’s conception of analytic comparatives overlooks PPs such as closer to home and more like home. Because CGEL analyzes more as a determinative in these cases, its analysis is internally inconsistent. I have demonstrated that contrasts between more and much exist in various contexts, including with comparative governors (e.g. more/much different) and certain participial adjectives (e.g. more/much improved), contradicting CGEL’s claim that more and much never contrast in analytic comparatives. Although I have focused on more and much, the arguments extend to most, less, little, and least.
I have proposed an explanation for the distributional patterns in AdjPs based on the pragmasemantics of more/-er and much. Specifically, more/-er establishes a salient minimum value in the discourse where none might otherwise exist, while much requires such a value. This explains not only why much tends to be limited to comparative contexts, but also why it appears with comparative governors, certain participial adjectives, and too. It also explains why more and too cannot modify each other. However, some puzzles remain, such as the variable applicability of much among the a- adjectives.
Finally, I have argued that CGEL’s reliance on the lack of contrast is theoretically unsound.
I conclude that more and less are most parsimoniously categorized as determinatives, and that their categorization as adverbs is not justified in any context.
Beyond the specifics of more and less, this study highlights the importance of empirical evidence in grammatical analysis. By rigorously examining corpus data, we can test existing grammatical descriptions and propose refinements where necessary. Even widely accepted analyses should be subject to ongoing scrutiny and revision in light of new evidence, as Geoff Pullum has demonstrated throughout his career.
Moreover, the categorization of more and less serves as a case study in the contribution of various linguistic sub-fields to the resolution of grammatical categorization. While traditional approaches have often prioritized distributional criteria for category assignment, this study suggests that semantic and pragmatic factors can also play a crucial role. The fact that the distribution of more and less can be largely explained by their meaning and pragmatic context raises questions about the primacy of syntactic criteria. More broadly, it suggests that a comprehensive understanding of linguistic categories requires an approach that integrates insights from multiple aspects of linguistic analysis.
Supplementary material
To view supplementary material for this article, please visit http://doi.org/10.1017/S0022226724000239.
Acknowledgments
Many thanks to John Payne, Chris Potts, Nathan Schneider, Peter Evans, Linmin Zhang, Chandra Hodgson, Michael Israel, Dan Lassiter, and Geoff Pullum for extremely helpful comments! Thanks to Humber College for travel support and research time, and to my friend, Bert Remijsen, for lodging. Thank you, Geoff, for so much more!
In compliance with Humber College’s policy on use of AI, I acknowledge using GPT-4 (OpenAI 2023) to draft R and LATEX code. I also acknowledge using Claude 3 Opus (Anthropic 2024) to draft responses to reviewer comments.
Appendix A. Statistical analysis of adjectives in COCA
In the main body of the paper (Section 3.1), I discussed the patterns of adjectives that head AdjPs modified by much based on the Corpus of Contemporary American English (COCA). This appendix provides a more detailed statistical analysis of these adjectives, along with a visual representation, to support the arguments made in the paper.
Principal component analysis (PCA) was performed on the pointwise mutual information scores of the adjectives as modified by the determinatives (e.g. much bigger, details follow) to reduce the dimensionality of the data. With the elbow method, I determined that three clusters were optimal. Then I applied the k-means clustering algorithm, with the resulting clusters shown in Figure A1. The first two principal components (PC1 and PC2) are used to visualize the data in two dimensions. The R code (R Core Team 2023) used to generate the figure is as follows:
# Install and load required packages
if (!requireNamespace("ggplot2", quietly = TRUE)) install.packages("ggplot2")
if (!requireNamespace("ggrepel", quietly = TRUE)) install.packages("ggrepel")
library(ggplot2)
library(ggrepel)
# Load the data
data <- read.csv("adjnpmi.csv", row.names = 1)
# Keep only numeric columns
data_numeric <- data[sapply(data, is.numeric)]
# Scale the numeric data
data_scaled <- scale(data_numeric)
# Perform k-means clustering with 3 clusters
set.seed(123)
km_result <- kmeans(data_scaled, centers=3)
# Add cluster assignments to the data
data$cluster <- as.factor(km_result$cluster)
# Perform PCA for visualization
pca_result <- prcomp(data_scaled)
pca_data <- as.data.frame(pca_result$x[,1:2])
colnames(pca_data) <- c("PC1", "PC2")
pca_data$cluster <- data$cluster
pca_data$label <- row.names(data)
# Define a custom color-blind friendly palette
cb_friendly_palette <- c("#D81B60", "#1E88E5", "#004D40")
# Create the plot
ggplot(pca_data, aes(x=PC1, y=PC2, color=cluster, shape=cluster)) +
geom_point(size = 4, alpha = 0.7) +
geom_label_repel(aes(label=label),
box.padding = 1, point.padding = 1,
segment.color = ’grey50’, max.overlaps = 1000) +
scale_color_manual(values = cb_friendly_palette) +
scale_shape_manual(values = c(16, 17, 18)) +
theme_bw() +
theme(axis.title = element_blank(),
axis.text = element_blank(),
axis.ticks = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
legend.position = "bottom",
plot.background = element_rect(fill = "lightgrey", color = NA)) +
labs(title = "PCA Visualization of Clustered Data",
color = "Cluster", shape = "Cluster")
# Save the plot
ggsave("pca_cluster_plot.png", width = 10, height = 8, dpi = 300)
Figure A1 Visualization of adjective types based on pointwise mutual information scores. Points represent individual adjectives positioned by principal component analysis. Colors differentiate three k-means clusters: PINK circles: basic adjectives + separate, GREEN diamonds: comparatives + the outlier different, and BLUE triangles: other adjective forms (the comparative governors, the a- adjectives, and the past-participial adjectives) + expensive.
The PCA visualization (Figure A1) shows that comparatives (in green diamonds) form a tight cluster, indicating shared linguistic characteristics, with different as an outlier. The pink circle cluster includes the most typical adjectives that participate in synthetic and analytic comparatives. Expensive, though in the blue triangle cluster, could fit into the pink circle one. Separate, in the pink circle cluster, was included as a quasi-comparative governor (governing a from PP). The blue triangle cluster contains the other comparative governors (except different), the a- adjectives, and the past participial adjectives, which are kept distinct in the paper to explain their compatibility with both much and more.
Overall, the clustering aligns well with the observations in the main text, reinforcing the distinctions between adjective types in terms of determinative modification. This supports the paper’s main arguments about the distributions of much and more. The adjnpmi.csv file is available from supplementary material, along with the Excel file from which it is derived. The csv includes normalized pointwise mutual information scores (NPMI, ranging from
$ -1 $
for no occurrences to
$ +1 $
for
$ N-1 $
occurrences; see Equation 1; Bouma Reference Bouma2009) for adjectives modified by determinatives from the set {more, less, enough, that, little, much, no, any} in two contexts: Mod + Head + punctuation and Mod + Head + preposition (enough only post-modified). The ‘DQ’ (disqualified) column counts the number of irrelevant or false positive examples that were manually identified and removed from the analysis. For modifier-head pairs with over 100 hits, a random sample of 100 was checked, and the proportion of disqualified examples was extrapolated to the full count. After this manual verification process, I’m confident the NPMI scores in the csv file are not distorted by false positives, though some false negatives may exist.
$$ \hskip0.1em \mathrm{NPMI}\hskip0.1em \left(x;y\right)=\frac{\log_2\left(\frac{n_{xy}}{N}\times \frac{N}{n_x\times {n}_y}\right)}{-{\log}_2\left(\frac{n_{xy}}{N}\right)}. $$
Here,
$ N $
is the COCA size: 1,002,889,754 words (personal communication, Mark Davies, October 21, 2023).
Appendix B. NPMI Scores for modifier–head pairs
This section examines specific adjective, adverb, determinative, and preposition heads, showing their pairings with different modifiers. NPMI scores from COCA assess the grammaticality of these pairings, empirically evaluating the judgments in Tables 2–5 in the main text. The results are visualized as heatmaps.
While there is no widely accepted statistical measure of grammaticality, NPMI serves as a reasonable proxy when investigating modifier constraints in AdjPs. For instance, much strongly correlates with comparative adjectives, yielding an NPMI of 0.38, whereas the ungrammatical *any old has no relevant instances, resulting in an NPMI of –1.Footnote 13
Figure B1 was produced using the following R script, with minor variations for subsequent figures:
# Load necessary libraries
library(ggplot2)
library(reshape2)
# Load data
data <- read.csv(’/path/to/your/d-adj-table.csv’)
# Melt data for ggplot2 using the corrected column
name
data_melted <- melt(data, id.vars = "Adjective.group")
# Set the factor levels for "Adjective.group" to
retain the original order
# Reverse the order so that it matches the original
table
data_melted$Adjective.group <- factor(data_melted$
Adjective.group, levels = rev(unique(data$
Adjective.group)))
# Define colors for heatmap
midpoint <- (-1 + 0.28) / 2
color_scale <- scale_fill_gradient2(low = "#F26161",
mid = "white", high = "#5DB56E",
midpoint =
midpoint,
limits =c(-1,
0.28))
# Plot heatmap
ggplot(data_melted, aes(x = variable, y = ‘Adjective.
group‘, fill = value)) +
geom_tile() +
color_scale +
theme_minimal() +
labs(x = "Modifier", y = "Head")
Appendix B1. Determinative modifiers + adjective heads
Figure B1 displays NPMI scores for determinative modifiers and adjective heads, grouped as in Figure A1. The old group contains basic adjectives; the recent group includes other plain adjectives; comparative adjectives are grouped together; the different group has comparative governors; the afraid and improved groups contain a- and participial adjectives, respectively.
Figure B1 Heatmap of normalized pointwise mutual information (NPMI) scores between adjective Heads and determinative Modifiers in COCA. Scores range from –1 (red) to 0.38 (blue), with neutral associations centered in white.
Appendix B2. Determinative modifiers + adverb heads
Figure B2 shows NPMI scores for determinative modifiers with individual adverb heads.
Figure B2 Heatmap of normalized pointwise mutual information (NPMI) scores between adverb Heads and determinative Modifiers in COCA. Scores range from –1 (red) to 0.34 (blue), with neutral associations centered in white.
Appendix B3. Determinative modifiers + preposition heads
Figure B3 displays NPMI scores for determinative modifiers with preposition heads.
Figure B3 Heatmap of normalized pointwise mutual information (NPMI) scores between preposition Heads and determinative Modifiers in COCA. Scores range from –1 (red) to 0.22 (blue), with neutral associations centered in white.
Appendix B4. Adverb modifiers + various heads
Finally, Figure B4 presents NPMI scores for adverb modifiers with various heads.
Figure B4 Heatmap of normalized pointwise mutual information (NPMI) scores between various Heads and adverb Modifiers in COCA. Scores range from –1 (red) to 0.46 (blue), with neutral associations centered in white.
Open access
