2.1. Modality as a functional domain and a research paradigm

Due to their being considered ‘a paradigm case of grammaticization’ (Plank Reference Plank1984: 308), modal verbs have been and remain a chief testing ground for the study of language change.Footnote ¹ Attention has for instance been given to the auxiliation of modal verbs (see critical overview in Fischer Reference Fischer2007), the semantic developments of these verbs into and out of the modal domain (e.g. Goossens Reference Goossens, Kellerman and Morrissey1992; van der Auwera & Plungian Reference van der Auwera and Plungian1998; Traugott & Dasher Reference Traugott and Dasher2001: 118–47; Narrog Reference Narrog2012; Traugott Reference Traugott2016) and especially the relation between root and epistemic values from the perspective of both diachrony and language acquisition (e.g. Sweetser Reference Sweetser1990; Bybee, Perkins & Pagliuca Reference Bybee, Perkins and Pagliuca1994; Cournane Reference Cournane2015: 51). These discussions have informed the reflections on whether modals are monosemous or polysemous (Depraetere et al. Reference Depraetere, Cappelle, Hilpert, De Cuypere, Dehouck, Denis, Flach, Grabar, Grandin, Hamon, Hufeld, Leclercq and Schmid2023: 35–51) and have contributed to creating and comparing taxonomies of modal meaning (Depraetere & Reed Reference Depraetere, Reed, Aarts and Hinrichs2021). With the advent of digital corpora, English linguistics ‘has witnessed a remarkable quantitative turn over the last two decades’ (Kortmann Reference Kortmann2021: 1207), and this has impacted research practice in the analysis of the development of the English modals. Key publications include seminal work by Krug (Reference Krug2000) on the grammaticalisation of ‘emerging’ modals and a debate between Leech (Reference Leech2003, Reference Leech2011, Reference Leech, Arerese, Carretero, Hita and van der Auwera2013) and Millar (Reference Millar2009) on whether modals are declining, all of which spearheaded a series of studies (e.g. Smith Reference Smith2003; Biber Reference Biber, Curzan and Emmons2004; Leech et al. Reference Leech, Hundt, Mair and Smith2009: 71–117; Nesselhauf Reference Nesselhauf2012; Seggewiß Reference Seggewiß2012; Smith & Leech Reference Smith, Leech, Aarts, Close, Leech and Wallis2013: 75–84; Kranich & Gast Reference Kranich, Gast, Zamorano-Mansilla, Maíz, Domínguez and de la Rosa2015; Mair Reference Mair and Collins2015; Mair & Leech Reference Mair, Leech, Aarts and Hinrichs2021). With a few exceptions, these studies were largely based on the Brown family of corpora (especially the Brown/LOB/Frown/F-LOB quartet) which, though well designed, exhibit notable limitations in both quantitative aspects (their size) and qualitative aspects (their temporal scope). By comparison, the development of larger corpora such as the COHA (Davies Reference Davies2010) has allowed the study of more data from a longer time range. This, together with ever finer corpus methods, has led to the emergence of new studies (e.g. Lorenz Reference Lorenz2013; Hilpert Reference Hilpert2016; Daugs Reference Daugs, Hiltunen, McVeigh and Säily2017, Reference Daugs, Hohaus and Schulze2020, Reference Daugs, Hilpert, Cappelle and Depraetere2021, Reference Daugs2022, Reference Daugs2023; Jäger Reference Jäger2018; Kranich Reference Kranich, Kranich and Brenan2021; Mair Reference Mair2021), which served to fine-tune and nuance some of the findings from previous research.

The quantitative turn has, however, been shown to correlate with a decrease in qualitative observations in more recent publications (Larsson, Egbert & Biber Reference Larsson, Egbert and Biber2022), a trend which we regret. We share Kortmann’s (Reference Kortmann2021: 1219) view that linguists should ‘interpret [their] quantitative findings against previous linguistic research and theory building’, and we sympathise with Fillmore (Reference Fillmore and Svartvik1992) when he claimed the following:

Research in the field of modality, as witness the references mentioned so far, fares rather well in this respect. Yet it is not fully immune to the qualitative limitations associated with big data either. It is for instance well established now that some of the core modal auxiliaries are declining (Daugs Reference Daugs, Hiltunen, McVeigh and Säily2017). While it should be an empirical question what underpins this trend, Leech (Reference Leech, Arerese, Carretero, Hita and van der Auwera2013: 114) argues for instance that ‘the most plausible explanation … is that grammaticalization of the emergent modals in speech has been associated with increasing frequency, progressively leading to competition with the core modals, which consequently have been undergoing decline in recent English.’ More recently, Love & Curry (Reference Love and Curry2021) also discuss these changes in the modal domain, but only focus on individual quantitative differences across modal verbs and on the (re)distribution of ‘dynamic’, ‘deontic’ and ‘epistemic’ values through time. As we see it, this fails to deliver on the promise to provide ‘a comprehensive perspective on both formal and functional change in modality’ (Love & Curry Reference Love and Curry2021: 538). For indeed, changes within the modal domain show greater formal and functional complexity. At the formal level, recent studies have shown the need to focus not only on modal verbs but also on the specific constructions they appear in. This is in keeping with usage-based Construction Grammar, according to which it is clear that prefabs (i.e. multi-word units) ‘are important to the understanding of the fabric of grammaticization’ (Bybee & Torres Cacoullos Reference Bybee, Cacoullos, Corrigan, Moravcsik, Ouali and Wheatley2009: 212). So quantitative patterns of change are best tracked not only at the level of the verbs but also at the level of the individual constructions they are part of. Daugs (Reference Daugs, Hohaus and Schulze2020: 33) for instance nicely shows that while the auxiliary should is generally declining since the 1950s, the form shouldn’t is actually increasing. Likewise, Kranich (Reference Kranich, Kranich and Brenan2021: 275) shows that the pattern ‘SUBJ must be V-EN’ is decreasing much sharply than the modal must overall. Specific modal patterns have received increased attention in the recent research (see e.g. Adolphs Reference Adolphs and Skandera2007; de Haan Reference de Haan2012; Cappelle, Depraetere & Lesuisse Reference Cappelle, Depraetere and Lesuisse2019; Daugs Reference Daugs, Hohaus and Schulze2020, Reference Daugs, Hilpert, Cappelle and Depraetere2021; Kranich Reference Kranich, Kranich and Brenan2021; Leclercq Reference Leclercq2022, Reference Leclercq2024; Schneider Reference Schneider2023; Daugs & Lorenz Reference Daugs and Lorenz2024; Lorenz Reference Lorenz2024), but still too few examine English modality from a diachronic perspective. This focus on lower-level constructions is all the more important since they have been shown to display (and develop) unique functional profiles, which is the other main issue we seek to address in this article. Attention has too often been limited to the classical categories of root and epistemic modal meanings. However, the function of modal expressions largely extends past these two domains. For one, these two notions only concern the (truth-conditional) semantics of modal verbs, yet they are also rich in (non-truth-conditional) pragmatic information (Depraetere Reference Depraetere2019). Depraetere et al. (Reference Depraetere, Cappelle, Hilpert, De Cuypere, Dehouck, Denis, Flach, Grabar, Grandin, Hamon, Hufeld, Leclercq and Schmid2023: ch. 4) for instance have shown that the choice of verb to express root necessity is to a large extent determined by a pragmatic factor, namely the source of the modality (i.e. the driving force or factor behind a necessary state of affairs, such as specific circumstances or rules and regulations). So in addition to tracing the semantic profile of individual modal verbs and modal constructions, it appears equally crucial to identify how they (come to) differ pragmatically, both synchronically and diachronically. We suggest that this is an approach where qualitative analysis of variation is just as important as the identification of quantitative trends in big data. In addition, even if one were to concentrate on the semantic level only, it is clear since at least van der Auwera & Plungian (Reference van der Auwera and Plungian1998) that modal verbs do not simply express root or epistemic possibility/necessity; these verbs can also acquire post-modal meanings (i.e. meanings which are no longer associated with root or epistemic possibility/necessity, but instead, for instance, with the marking of concession or conditionality). We are therefore also interested in whether modal constructions (here, collocations with the adverb well) might grow to express meanings other than root/epistemic modality, such as was shown in the case of may and might in factual concessives such as in (1) discussed in Leclercq (Reference Leclercq2024: 147):

All of these observations show that, given the trend towards analyses of data from bigger and better corpora (as Fillmore suggested), it is also important that research in the modal domain take a closer look at the local properties of the forms involved in the observed changes and the exact meanings which are involved in these changes. For instance, it appears that post-modal meanings are especially intersubjective and speech-act oriented (Sweetser Reference Sweetser1990; Narrog Reference Narrog2012; Traugott Reference Traugott2016), as was indeed argued in the case of concessive may/might. In our view, it remains an empirical question in English whether these are the only or the primary functional features involved. In keeping with Traugott (Reference Traugott and Hickey2003: 128), Leclercq (Reference Leclercq2024) argues that the increase in intersubjectivity in the modal domain must be viewed as correlating with an increase in politeness and attention to face needs. This would entail a tendency towards speech-act softening rather than speech-act boosting, a claim that needs to be empirically verified. This argument is plausible, though, in view of language-external (social) factors that have been identified as crucial in the development of certain semantic categories including modality, namely a trend towards ‘democratisation’ (see e.g. Myhill Reference Myhill1995; Kotze & Van Rooy Reference Kotze and Van Rooy2020; Kranich, Hampel & Bruns Reference Kranich, Hampel and Bruns2020), which in language ‘refers to the speakers’ tendency to avoid unequal and face-threatening modes of interaction’ (Farrelly & Seoane Reference Farrelly, Seoane, Nevalainen and Traugott2012: 393). In the field of modality, this would for instance explain the decline of strong deontic meanings such as that expressed by must, or the increasing use of modal verbs to serve as polite hedging markers of specific speech acts such as requests (Kranich, Hampel & Bruns Reference Kranich, Hampel and Bruns2020: 3).

These observations show that, besides the broad quantitative trends of each verb and their semantic (root/epistemic) distribution, a number of other formal and functional factors now deserve much attention in the modal domain. Our focus on the patterns may well, might well, well may and well might is one step in this direction.

2.2. Modal collocations with well: previous research

In this subsection, we identify and discuss some of the major themes that have emerged in research on collocations between the modal verbs may and might and the adverb well, in the history of English.

First, it must be noted that there are rather few studies that directly focus on these particular collocations. They have recently been looked at from a quantitative perspective by Flach (Reference Flach2021), and previously they had been approached from a more qualitative perspective mostly by Hoye (Reference Hoye1997) and Shibasaki (Reference Shibasaki2009) but also briefly by Visser (Reference Visser1963) as well as Defour (Reference Defour, Davidse, Vandelanotte and Cuyckens2010: 185), who suggests that may well ‘has turned into a fixed idiom in present-day English’. The relative absence of research on these collocations is particularly striking given that may well and might well have been found to feature among the thirty most strongly associated ‘modal + adverb’ combinations (Flach Reference Flach2021: 754), in the Corpus of Contemporary American English (COCA), with may well even being in fourth position. Flach (Reference Flach2021: 760) also shows that, in this corpus, well is in fact ‘strongly repelled by all other modal auxiliaries’. This quantitative observation highlights the unique relationship between may, might and the adverb, and it adds to the need for a careful study of these collocates. Indeed, we concur with Flach (Reference Flach2021: 745) that ‘probabilistic information about combinatorial patterns is part of speakers’ knowledge’, and that they reveal a degree of idiomaticity that deserves to be studied further.

The main question to address then is what the functional contribution of these collocations is, from both a synchronic and a diachronic outlook. In this respect, it is primarily qualitative approaches that have addressed the question, and chiefly from a synchronic perspective. It is interesting to note for instance that Huddleston & Pullum et al. (Reference Huddleston and K. Pullum2002: 181) argue that well serves to strengthen the epistemic value of may in sentences like He may well have left it downstairs (‘It is quite likely’). This reveals a somewhat compositional view of the collocation, with the adverb simply modifying the force of the modal expression. More detailed qualitative analyses are somewhat less inclined to favour a fully compositional analysis though. Hoye (Reference Hoye1997: 233) discusses the difference in this regard between may well and well may. He considers the first to be more clearly idiomatic (it has a ‘unity of meaning’) which has the function of indicating a strengthening of the speaker’s certainty regarding what is asserted in the rest of the clause (i.e. it suggestsepistemic probability, rather than mere possibility). By contrast, the second is more compositional, with a meaning ‘indeed possible’. Hoye suggests that the unity of may and well (in that order) is not just restricted to meaning, noting that it is ‘invariable, and resistant to any form of modification’ (Reference Hoye1997: 234), which is as much of a formal property as it is a functional one. When the adverb well precedes the modal, the expression serves ‘to both reinforce the truth-value expressed in the sentence and additionally reflect the strength of the speaker’s conviction’ (Hoye Reference Hoye1997: 172). For Hoye, the use of well therefore serves ‘to transform rather than heighten or reinforce’ the epistemic modal meaning (Reference Hoye1997: 165). From a historical perspective, Jucker (Reference Jucker1997: 99–100) observes that there are three meanings associated with well that were established by the Middle English period which may be connected to the more grammaticalised uses of well in later periods. Of most relevance to the present discussion is the one which is concerned with greater certainty and a lack of doubt. Defour (Reference Defour, Davidse, Vandelanotte and Cuyckens2010: 168) suggests that this meaning is associated with a ‘more subjective’ use of the adverb. The connection between epistemic modality and initial position of well (as in well may) is noted as far back as the Old English period (Traugott & Dasher Reference Traugott and Dasher2001: 175; Defour Reference Defour, Davidse, Vandelanotte and Cuyckens2010: 184). Concerning the relationship between well may and may well, Defour (Reference Defour, Davidse, Vandelanotte and Cuyckens2010: 187) provides a quantitative analysis of patterns in the Helsinki Corpus (looking at data from 850CE to 1710CE), noting a ‘general – although not absolute – tendency’ towards a medial rather than initial position for well suggesting that well may have become a modal particle associated with speaker stance and the marking of politeness (Defour Reference Defour, Davidse, Vandelanotte and Cuyckens2010: 188; see also Hoye Reference Hoye1997: 209). We return to these issues in the analysis of our data below.

The specific properties of may/might well can be further understood in connection with Hoye’s general approach to the alignment of epistemic modal and adverbial meaning (see table 5.1 in Hoye Reference Hoye1997: 240). Certain combinations are ‘harmonic’ (Lyons Reference Lyons1977: 807f.) in the sense that they both express similar degrees of likelihood (might + perhaps both express possibility, must + surely both express certainty), while others are non-harmonic (should + perhaps combines a ‘probability’ modal with a ‘possibility’ adverb). May well and might well are both cases of non-harmonic combinations (a ‘possibility’ modal with a ‘probability’ adverb), but as Hoye (Reference Hoye1997: 240, footnote to table 5.2) observes, the ‘probability’ sense of well is only attested in combination with may and might (and could, which we do not discuss further). Thus it would be reasonable to suggest that the probability sense is a property of the modal construction (understood here as a conventional combination of modal and adverb) rather than the individual parts.

In both Huddleston & Pullum et al. (Reference Huddleston and K. Pullum2002) and Hoye (Reference Hoye1997), the use of well with may and might has been discussed in relation to the epistemic sense of the two verbs, and the deontic value of the verbs is not mentioned (nor any other values, modal or post-modal, for that matter). This issue regarding the relationship between deontic and epistemic uses of the collocations has been addressed, again from a more qualitative perspective, mostly in their historical development. Shibasaki (Reference Shibasaki2009: 65) suggests that the collocation of may and might with well ‘enhanced’ the diachronic emergence of epistemic senses of the modals. That suggests that in the earlier history of the language, both ‘sets’ of collocations (i.e. both where the adverb precedes the modal and where it follows) had a more compositional meaning (somewhat like Hoye’s paraphrase ‘indeed possible’), and that the compositionality of the meaning has decreased in the case of may/might well over time, certainly to a greater extent than is the case with well may/might. At the same time, there are good reasons to believe that this pattern of change (from a compositional to a non-compositional epistemic interpretation), which has received most attention, is not the only idiomatic functional feature at play in the development of the collocates. For indeed, discussing the particular collocation may well, Visser (Reference Visser1963: §1667) notes that in Middle English, the expression was used to ‘stress the reasonableness of the statement’. If anything, this particular value cannot be considered compositional. So the development of the strong epistemic sense identified earlier is not the only feature of non-compositional meaning that can be found. It is interesting that in the more recent data that will be presented in the rest of this article, this particular use can still be found in Late Modern English. It thus remains an empirical question what exactly constituted the functional profile of may and might in combination with well, how it has evolved and where it is headed.

There is for instance the issue of the relative timing of the development of (inter)subjective meanings associated with the collocations, especially with regard to concessive use, as exemplified in (2).

Traugott & Dasher (Reference Traugott and Dasher2001: 115–16) do not refer to particular collocations of modals with other lexical categories, but observe more generally that concessive uses, given their metatextual function, will by hypothesis develop rather later semasiologically (see also Bybee, Perkins & Pagliuca Reference Bybee, Perkins and Pagliuca1994: 120). This was recently confirmed by Leclercq (Reference Leclercq2024) in the case of may and might. The question we will explore is whether what holds true for individual modals as a category also holds true for modal idioms.