English comparative modals and their complements

English comparative modals are combinations of the adverbs rather, sooner and better with an auxiliary. There is recent consensus that the comparative modals rather and sooner have over time developed a different syntax and semantics than better. However, potential differences in the syntax of rather and sooner with respect to patterns of complementation haven’t been explored. This article reports the results of a corpus study of these two modals and finds that rather patterns like object-raising verbs, allowing a range of complements that are unavailable for sooner. Our analysis of these patterns draws on recent work in the Construction Grammar framework, with forays into its formal implementation, Sign-Based Construction Grammar, and we propose that rather differs from sooner in that it constitutes a micro-construction whose features are licensed by both the Modal Construction and the Object-Raising Construction, the latter a subtype of the Transitive Construction.


Introduction
This article addresses the question of whether English comparative modals can be analyzed as a coherent class, or as a single subschema in the terminology of Construction Grammar (CxG).The term comparative modals refers to combinations of the adverbs rather, sooner and better with an auxiliary.All three are depicted in the examples in (1)-(3), harvested from the Corpus of Contemporary American English (COCA, Davies 2008-).
(1) I am concerned, but I would rather them have a full investigation, because an investigation plays out based on the evidence.(SPOK: Connect the World, 2013) (2) Hebert would sooner lose $49 million in new funding for roads, schools and police than provide an accounting of how the money is spent.(ACAD: American Indian Quarterly, 2017) (3) Obama had better win this election and save an already conservative court from getting far worse.(BLOG: www.queerty.com/what-would-bescarier-than-a-romney-presidency-a-romneysupreme-court-20121030/,2012) Recent research on the history of these modals shows that RATHER and SOONER have followed a different path than BETTER, not only with respect to their semantics, but also their syntax (Denison & Cort 2010;van der Auwera & De Wit 2010;van der Auwera et al. 2013;Van linden 2015;Traugott 2016Traugott , 2019)). 2 It has been argued on the basis of these findings that RATHER and SOONER belong in a separate category from BETTER (Traugott 2016(Traugott , 2019)).We offer novel corpus evidence in general support of this argument, but we also demonstrate that the set consisting of RATHER and SOONER requires further differentiation due to the former's distinct ability to license raising to object (note the accusative-marked NP them followed by an infinitival complement in (1)), which is a feature of lexical verbs rather than modal auxiliaries.Our evidence is based on patterns of complementation that characterize RATHER and SOONER in present-day US English, an area still largely unexplored.We follow Traugott (2016Traugott ( , 2019) ) in adopting a CxG perspective on comparative modals, such that they are defined as constructions and (changes in) their semantics and syntax are considered in parallel.
The version of CxG we adopt here is Goldberg (1995Goldberg ( , 2006)).Its basic assumption is that linguistic knowledge can be represented as a network of constructions, which range from morphemes to clausal elements.Constructions of any size and complexity are understood as pairings of form and meaning.Generalizations over constructions are captured by means of inheritance hierarchies, where more schematic constructions license more specific ones and where features can be shared among several constructions in the network.In the remainder of this article, inheritance hierarchies will play a role in capturing the range of complementation patterns available for RATHER and SOONER, and in particular links that RATHER has to object-raising verbs.Our focus here is on (i) RATHER and SOONER, whose complementation possibilities have not remained stable over time unlike those of BETTER, and (ii) differences in the syntax of the comparative modals, given that their divergent semantics is uncontroversial.When describing the properties of the comparative modals we make occasional forays into Signed-Based Construction Grammar, a formal implementation of CxG (see Boas & Sag 2012).
The rest of this article proceeds as follows.In section 2, we briefly sketch Traugott's (2019) constructionist analysis of the three comparative modals, which we then expand in section 5 in light of the corpus data to be presented here.Section 3 describes the data, which were harvested from COCA, while section 4 walks the reader through the results, i.e. complementation patterns observed in RATHER and SOONER, also drawing comparisons with equatives like would ( just) as soon.In section 5, we address, in addition to a theoretical analysis of the comparative modals, a related question, namely whether the syntactic category of the adverb rather itself is shifting (or has shifted) from adverb to verb.Section 6 concludes.
2 Constructionist analysis of RATHER, SOONER and BETTER Traugott's (2019) analysis of the comparative modals is based on changes to their semantics and syntax from the Old English period (650-1100) onwards.We begin by briefly summarizing these changes.

Historical development of RATHER, SOONER and BETTER
The predecessors of the adverbs rather and sooner in Old English were temporal adverbs meaning 'sooner in time, instead' that had the ability to co-occur with the verb would.The first indications of the modern preference meaning are found in them in Middle English (1100-1500), and there the preference is associated with the sentential subject, that is, it represents participant-internal rather than participant-external modality. 3The adverbs co-occur with the verbs would, will, should and had, and there is an explicit or implied standard of comparison.By the sixteenth century, strings consisting of an auxiliary (predominantly would) and rather/sooner have clear preference meanings and can be considered modal verbs, as illustrated in (4) and ( 5).
(4) I would rather be torn with wild Horses, than forsake my Religion.(1571, CED: DiTNORFO; Traugott 2016: 115) (5) Nay I meane to follow yee: I will sooner leese my life, then fight of you till this dinner be done.(1595, CED: DiCWARNE;Traugott 2019: 116) Subsequent developments include a strong association with the auxiliary would and an expansion in complementation patterns.While RATHER and SOONER take only bare infinitival complements until the sixteenth century (see ( 4) and ( 5)), the eighteenth century sees the addition of finite clauses as complements, with or without the complementizer that.The subject of the embedded clause may differ from the subject of the main clause, and a standard of comparison may be optionally expressed, as depicted in (6).
(6) I would rather that the House should fall down, and knock his Brains out, than he should be hanged.(t17431012-37;Traugott 2019: 122) While the availability of these complementation patterns has been acknowledged in previous research (Van linden 2015;Traugott 2016Traugott , 2019)), it hasn't been explored in any detail what other complements are permitted by these modals.We demonstrate in section 4 that present-day RATHER and SOONER don't pattern exactly the same: RATHER behaves like an object-raising verb, licensing combinations of NP complements with predicative complements and of NP complements with infinitival complements in addition to infinitival and finite-clause complements.This suggests that rather could be analyzed as a lexical verb, rather than an adverb, co-occuring with an auxiliary.In fact, Juge (2002) and Klippenstein (2012) articulate the idea that rather has changed its lexical class from adverb to verb.Klippenstein (2012) cites the earliest verbal uses of rather going back to the 1500s, initially without verbal morphology, as in (7).She concludes that rather must be a lexical verb here because there is no other finite verb in this clause.
(7) I rather be borne Fortunate, then to be very wise.(Edwards 1576: 25) As of the 1800s, rather is attested with morphological specifications, as in ( 8). ( 8) What care and what pains and hazard I have been in since coming here, I will remit to others to declare; but if the matter had not proceeded as it has done, I should have rathered I had been buried here.(Sinclair 1899: 119) Klippenstein's (2012) data reveal that the historical record of rather includes both adverbial and verbal uses.She argues on this basis that rather has been reanalyzed from adverb to verb, following Juge (2002), who points to the morphological nontransparency of the comparative form rather compared to sooner as one of the potential reasons for the reanalysis.It's interesting to note at this point that if rather indeed is a verb (we return to its potential lexical-verb status in sections 4.1, 4.3 and 5), its history tracks that of better in the sense that both exhibit verbal properties, although rather does so much earlier.As we will see shortly, however, verbal properties of better are aligned with auxiliary verbs, while rather clearly behaves like a lexical verb.We now briefly turn to the history of the comparative modal BETTER.
BETTER originates from a type of Middle English impersonal copula construction hosting the adjective betere (Denison & Cort 2010).This construction is illustrated in (9).
(9) A mon were betere for is sunne be sori and vnssriue flanne issriue a man would-be better for his son be sorry and unshriven than shriven wifloute sorinesse.without sorriness 'It would be better for a man if his son were sorry and not receive the sacrament than to receive the sacrament without being sorry.' (a1325 SLeg.(Corp-C 145) 131/91 [MED unshriven ppl.a; Denison & Cort 2010: 352) The syntax of the construction that hosts better changes in several ways after the loss of impersonal constructions in Middle English.The changes are: the subject position in a new copula construction is first filled by expletive it (i.e.It were better), and then by referential NPs in place of expletive it, which is followed by better being associated with the auxiliary had and used as an adverb in this context (Denison & Cort 2010).These syntactic changes can be seen as the first steps toward the formation of a new modal.According to Traugott (2019), BETTER has preference semantics through the seventeenth century.This is plausible, given that, even if the meaning of BETTER comes closer to advisability, it can still be seen as intersecting with participant-external preference.That is, preference is dictated by circumstances external to both the sentential subject and the speaker, which stands in contrast to participant-internal preference conveyed by RATHER and SOONER, as we have seen above.
By the early eighteenth century there are reasons to see BETTER as a full-fledged modal (Traugott 2019).Its semantics has shifted away from preference and toward advice offered by the speaker to others or themselves (if a first-person pronoun serves as subject).Syntactically, BETTER licenses only infinitival complements and is rarely used with a standard of comparison.While BETTER is slower to develop into a modal than RATHER and SOONER are, there is a rapid increase in its frequency compared to that of RATHER and SOONER from the eighteenth century on.Furthermore, there are indications that bare better (i.e.used without the auxiliary had) has recently been moving in the direction of modalhood, as argued by Denison & Cort (2010) on the basis of semantic and syntactic evidence including contracted forms shown in (10).
(10) lol well you better start staying in then betternt you!!!!!! lol (http://profile.myspace.com/index.cfm?fuseaction=user.viewprofile&friendid=81374563,2007; cited in Denison & Cort 2010: 380) In sum, although originally an adjective, better later patterns like the adverbs rather and sooner in terms of its lexical class and co-occurrence with an auxiliary, developments that are subsequently followed by a divergent semantics, complementation patterns and, partially, lexical class.

Traugott's (2019) analysis
Traugott (2019) asks whether there is enough evidence that we can group the three comparative modals into separate subschemas.She operates with distinctions that reflect the generality of particular constructions organized into an inheritance hierarchy.The most general nodes are macro-schemas, followed by schemas, within which several coherent sets can be identified.Schemas are further differentiated into subschemas grouping similar members, and the two lowest levels are microconstructions (i.e.individual types like RATHER, SOONER and BETTER) and constructs (i.e.individual tokens).The macro-schema that licenses the network of English modals is the Modal Construction (ModalCxn), and a number of daughter nodes can be differentiated within it, as depicted in figure 1. 4, 5  The criterion for placing comparative modals within this network is the degree to which each of them shows evidence of being a conventionalized pairing of form and meaning.Given the historical changes detailed in the previous section, Traugott (2019) argues there is reason to believe that comparative modals have undergone constructionalization.Traugott (2022) defines constructionalization as in (11). 611) Constructionalization: The establishment of a new symbolic link between form and meaning which has been replicated across a network of language users, and which involves an addition to the constructicon.7 (Traugott 2022: 49) On this definition of constructionalization, syntactic changes must be accompanied by semantic changes, and if this isn't the case, we can only speak of constructional changes that can appear before or after constructionalization itself.It is clear that all three comparative modals have developed a new syntax.They are adverbs forming a string together with a preceding auxiliary and minimally taking infinitival complements.It's equally clear that all three have developed a new semantics.RATHER and SOONER express preference after previously expressing temporal meanings, while BETTER expresses advice, and more recently hope, after previously expressing preference.We are thus justified in placing the comparative modals in the network shown in figure 1.Given the semantics of the comparative modals, the subschemas in figure 1 that are of interest to us are the AdviceCxn and PreferenceCxn.Traugott (2019) argues that RATHER and SOONER are micro-constructions falling under the PreferenceCxn, while BETTER instantiates the AdviceCxn, and thus they aren't subsumed under a single subschema, but constitute a family of constructions at the higher levels.This analysis aligns with Van linden (2015), whose diachronic investigation of the comparative modals in US English reveals semantic and syntactic differences between BETTER and the two preference modals RATHER and SOONER.But there is also a grammaticalization path all three comparative modals have followed with respect to phonetic reduction of the auxiliary, although BETTER has progressed farther along this path than RATHER and SOONER and is closer to auxiliarihood.

Data
The research reported here is based on data harvested from COCA, the current size of which exceeds one billion words representing various registers of spoken and written US English.The time span is twenty-nine years from 1990 to 2019.Because we are interested here specifically in the range of material that can follow RATHER and SOONER (BETTER was excluded, given its inability to license complements other than infinitives), the search strings were designed to include all complements, contracted as well as full forms of the auxiliaries would and had, and null auxiliaries (this was the case only with RATHER, as there were no instances of SOONER with null auxiliaries and relevant preference semantics in our data). 8As will become clear in the next section, RATHER has progressed farther than SOONER in developing a new syntax in US English.The novelty is especially its ability to appear in environments characteristic of object-raising verbs, i.e. with NP complements followed by infinitival complements and with NP complements followed by predicative complements.In this respect, RATHER also distributes differently than equatives (i.e.strings like would ( just) as soon).Equatives are defined as members of a larger set of comparative modals discussed in van der Auwera & De Wit (2010: 127), a set which ranges from superlatives (e.g. had best) to equatives.We include data on equatives in table 1 (and in table 2 in section 4.3) in an attempt to locate the syntax of RATHER in a broader landscape, although equatives are not the focus of this article.

Results
We begin by summarizing the range of complements available for RATHER and SOONER in table 1.These counts include both declarative and interrogative clauses hosting rather and sooner along with contracted and full forms of the auxiliaries would and had, as well as declarative and interrogative clauses hosting bare rather (i.e.without any auxiliary).The counts for finite-clause complements include those with an overt complementizer or without it, and all finite verb forms.Among the instances of finite-clause complements to RATHER and would ( just) as soon are clauses with subjunctive verb forms (as in (15)), a pattern that is not attested with SOONER (we briefly return to subjunctives in section 4.1).Table 1 shows that the most frequent complements are infinitives, and these are followed by finite clauses across the board.Columns 4-6 show the lesser-known kinds of complement, which the following sections focus on.All five kinds of complements in table 1   As can be seen from table 1, the characteristic that RATHER and SOONER share with each other and with equatives is the possibility of taking infinitival and finite-clause complements.There is no evidence in COCA that SOONER can take any of the remaining complements that RATHER can, that is, either an NP and an infinitival complement or an NP and a predicative complement in addition to NP complements, as in ( 18)-( 22).9Because these are complements that normally are licensed by object-raising verbs (e.g.consider, expect, make, want, have), they demonstrate that RATHER has object-raising-verb properties and has thus syntactically moved further away from both SOONER and the core modals.We turn to raising verbs next.

Subject-and object-raising verbs
Object-raising verbs are a type of raising verbs.The term raising verbs goes back to Postal (1974) and refers to verbs taking infinitival complements where the unexpressed subject of the embedded verb instead occupies the position of subject or object in the higher clause without being selected by the higher verb.Depending on the lower subject's grammatical function in the higher clause, two kinds of raising verbs can be distinguished: subject-raising verbs and object-raising verbs.The former are illustrated by seem in (23), and the latter by consider in (24) and the second instance of make in (25).
(23) This book seems to be targeted to ordinary Joe Sixpacks who want to learn to make good home movies with their camcorders.(BLOG: http://newsvideographer.com/2007/06/02/ book-review-three-the-completedigital-video-guide/, 2012) (24) I willfully continue to say hi to them anyway, and I just consider them to be rude and obnoxious human beings when they don't reply.(WEB: www.apartmenttherapy.com/howfriendly-are-your-neighbors-andwhat-do-you-do-if-the-answer-is-not-at-all-174677,2012) (25) Republicans are just mad that he made a really good decision at a good time and made them look bad in the process.(BLOG: http://politicalirony.com/2012/06/16/damned-if-you-do,2012) The characteristic of all raising verbs is that they don't assign any semantic role to their subject or object arguments.That is, this book in (23) receives its semantic role not from the subject-raising verb seem but from the lower verb, and, similarly, the pronoun them in ( 24) and ( 25) receives its semantic role from the lower verbs and not from consider or make.These facts are supported by the ability of raising verbs to take expletive subjects or objects (e.g.There seem to be lots of rules, We consider there to be a breach of the employment contract) and to leave the meaning of idiom chunks unchanged when co-occuring with them (e.g.The cat seems to be out of the bag) (for recent overviews of theoretical analyses of raising verbs, see e.g.Polinsky 2013 and Abeillé 2021).
Although the term raising may suggest that movement operations are involved in theoretical analyses of raising verbs, this is not always so.This term is used in constraint-based frameworks like Head-Driven Phrase Structure Grammar and in constructionist frameworks like Sign-Based Construction Grammar, neither of which employ movement operations.Instead, raising is accounted for by allowing the unexpressed subject in the lower clause to share its syntactic and semantic features with the subject (under raising to subject) or object (under raising to object) in the higher clause at the same time that the subject or object is not included among the semantic arguments of the higher verb (see Abeillé 2021 for a recent overview).We are assuming here that an analysis along these lines is appropriate for raising verbs.
English auxiliaries, including the core modals (e.g.should, may, must), are considered to be subject-raising verbs.This view is shared by both transformational (Wurmbrand 1999(Wurmbrand , 2001) and non-transformational frameworks (Pollard & Sag 1994), including again Sign-Based Construction Grammar (Sag 2012;Sag et al. 2020). 10s subject-raising verbs, auxiliaries license VP complements and do not assign any semantic role to their subject arguments.The three comparative modals fall into the subject-raising category as well when they appear in their default subcategorization frames, i.e. with bare infinitives as complements.11However, our data demonstrate that RATHER is also an object-raising verb, much like consider and make.Examples ( 24)-( 25) illustrate raising to object very clearly with consider and make: the NPs showing up in object position in the higher clauses have accusative case specifications, as they are headed by pronouns, and the embedded infinitival clause may host either a bare or a to-infinitive, depending on the higher verb.12In all the items in table 1 that follow the NP + infinitival complement pattern, the relevant NPs also have accusative case specifications (as in examples ( 21)-( 22)).We follow the essential insights of Head-Driven Phrase Structure Grammar and Sign-Based Construction Grammar in assuming that object-raising verbs take two complements: an NP and a VP, the first of which is not assigned any semantic role by the higher verb (see Sag 2012 and Abeillé 2021 for more detail).
Note that NPs that lack accusative case specifications and appear with bare infinitives would make the structure ambiguous between raising to object and finite-clause complements, specifically finite clauses hosting subjunctive verb forms, one example of which we saw in (15) (but example (15) is itself not ambiguous between raising to object and finite clause, because the NP is headed by a nominative-marked pronoun).Further examples of such ambiguity, which we have classified as subjunctives, appear in ( 26)-( 27) below.These are grouped together with finite-clause complements in table 1.However, a to-infinitive makes the structure unambiguously raising to object, as depicted in ( 28).
(28) I would love to sit down with Goldman.I would rather Goldman and Marcia Clark to be sitting right here interviewing me, but they're too busy making money, you know.(SPOK: ABC World News, 1996) Example ( 28) is one of two with a to-infinitive among our object-raising examples.The ability to take either a to-or bare infinitive varies among the set of verbs that license raising to object in present-day English, which include: allow, assume, believe, cause, consider, declare, deem, estimate, expect, find, get, hold, imagine, intend, judge, know, let, make, mean, need, perceive, prefer, project, repute, rumor, report, say, see, show, suppose, think, want and wish.
Another property RATHER shares with object-raising verbs is that it permits one more subcategorization frame with two complements.We saw examples of NP complements followed by predicative complements (here PPs) in ( 19) and ( 20), repeated here for convenience as ( 29) and ( 30); there is a total of fourteen of them in the current data.Notice the parallels between ( 29)-( 32) and the object-raising verbs we have discussed thus far: the NP complements all appear in object position and with accusative case specifications at the same time that they are the unexpressed subjects of the respective predicative complements. 13The raising analysis assumed for such verbs is that the first complement (i.e. the object of the higher verb) is shared with the unexpressed subject of the predicative complement and receives no semantic role from the higher verb (see Abeillé 2021 and references therein for more detail).
Several object-raising verbs (e.g.expect, prefer, believe, want) permit either infinitival complements or predicative complements as their second complements.This fact speaks in favor of rather being a verb when it appears together with an auxiliary.Wood (2013: 64, 82) argues that would by itself doesn't license raising to object in English, and nor does bare rather, based on examples similar to (33)-( 36).Although few, we did find counterexamples to Wood's claim regarding bare rather.It appears three times with an accusative-marked NP and an infinitival complement as the second complement, as in (37).
(37) I rather him stand by them and be open about it … than do like Mitt and try to pretend like he has no idea what they are doing.(WEB: www.mediaite.com/online/president-obama-was-theonly-candidate-who-could-afford-to-denounce-super-pacs-yet-he-won%E2%80%99t/,2012) Bare rather is otherwise only found with infinitival complements (47 instances) and finite clauses (7 instances) in our data (these counts are included in the data in table 1).In addition to these, we have harvested example (38), which shows a clearly verbal rather used with do-support.
(38) Do you rather they wear a condom or get HIV? (BLOG: http://scienceblogs.com/pharyngula/ 2008/07/24/the-great-desecration/, 2012) Example ( 38) suggests that rather has been developing verbal characteristics regardless of the presence of an auxiliary, but we leave it for future research to explore why the presence of an auxiliary appears to impact the range of complements that are permitted.In any case, our three object-raising examples along with (38) strengthen the case for analyzing rather as a verb even when it's not accompanied by an auxiliary, which we return to in section 5.

NP complements
This section addresses the remaining pattern reported in table 1, namely NP complements, that is found with RATHER.Twenty-six instances of RATHER are followed by an NP complement in our data.This count includes both nominal and pronominal heads within the complement, as shown in ( 39)-( 40) (see also (18) for another example of a pronominal head).
(39) Though that was still bad, he'd rather Jack than Roger.(FIC: We Remain, 2019) (40) Chief did make a few bad plays, but I would rather him than Ariza.(BLOG: www.hornets247.com/2012/11/17/milwaukees-three-point-shooting-sinks-hornets, 2012) This complementation pattern again sets RATHER apart from both SOONER and equatives, which resist NP complements, while further aligning it with the set of object-raising verbs (e.g. with verbs like prefer and want).
We now move beyond the data included in table 1 by discussing the use of the auxiliary have with RATHER when it takes finite-clause complements.

Auxiliary have and RATHER
In this section, we shift our focus to another difference that sets RATHER apart from SOONER and equatives.It's well known that both the two comparative modals and equatives may express counterfactual semantics, and do so by co-occuring with the auxiliary have at the same time that the lower verb bears perfect participle morphology.The auxiliary have typically follows rather, sooner and ( just) as soon, as in ( 41)-( 43).But have may also follow would, preceding rather, sooner or equatives, as in ( 44)-( 46).The order illustrated in ( 44) is interesting due to the ability of rather to take verbal morphology in this environment in some US English dialects, as reported in Wood (2013Wood ( , 2019)).Example (47) illustrates this point: it hosts the participle rathered in place of rather and is one of two examples we have harvested from COCA.14 (47) I would have rathered that this simple-minded President simply vote 'present' and leave the important decisions up to the adults in the room.(WEB: www.politico.com/blogs/bensmith/0909/Polish_PM_wouldnt_take_US_calls.html,2012) Data like these are suggestive of rather having changed its lexical class from adverb to verb and thus having the ability to license the auxiliary have in examples like (47) (recall Klippenstein's (2012) and Juge's (2002) proposals discussed in section 2.1).
A parallel lexical-class change is not observed in sooner.But example (47) reveals that differences between RATHER versus SOONER and would ( just) as soon go even deeper.
A closer examination shows that RATHER, but not SOONER or would ( just) as soon, has the option of taking finite-clause complements when the auxiliary have precedes rather.This option is illustrated in ( 48)-( 49).
(48) Honestly Shoshana, would you have rather the authorities not have inspected you and the other two men and found that you never made it home that day to see your children? (BLOG: http://shebshi.wordpress.com/2011/09/12/some-real-shock-and-awe-racially-profiledand-cuffed-in-detroit,2012) (49) I still feel that they would have rather I kept quiet.(WEB: http://captainawkward.com/2012/08/07/322-323-my-friend-group-has-a-case-of-the-creepy-dude-how-do-we-clear-that-up, 2012) To put this use into perspective, let us consider the data in table 2. While the frequency of finite-clause complements is low, they could be indications of the status of rather as a verb, though lack of overt participial specifications on it is problematic, as is the fact that participial specifications on rather are generally rare in COCA, as we have seen.Wood (2013) argues that RATHER allows raising to object in some dialects of US English when have precedes rather as well, which supports the potential status of rather as a verb.
We have found no instances of raising to object in this environment in our data.However, even finite-clause complements combined with the lack of a perfect participle leave us with the question of whether there is a verb here at all to help license the auxiliary have.
A theoretical analysis articulated by Wood (2013) is that verbal features are borne by the silent verb have, while rather itself remains an adverb.He proposes this solution specifically to explain uses of RATHER with raising to object: the silent have is present in the structure to license accusative case specifications on the object in the higher clause and to receive participial morphology assigned by the auxiliary have.This reasoning is based on clear distributional parallels between the lexical verb have and RATHER, such that both can license raising to object with similar semantics, as shown in (50) from Wood (2013: 78).
(50) Would you have him kill them rather than arrest them?
We don't pursue Wood's analysis here for two reasons.First, silent categories are not permitted in the CxG framework; second, finite-clause complements illustrated in examples ( 48) and ( 49) would make illicit complements for the lexical verb have, introducing a complication to Wood's analysis.One would be forced to propose another silent verb on account of such examples, prefer being the best candidate, as in ( 51).
(51) I still feel that they would have preferred I kept quiet.
But because prefer doesn't license bare infinitives under raising to object, although RATHER does, this would require positing either have or prefer as a silent verb to account for the entire range of complements available for RATHER.We therefore turn to the theoretical possibilities for accounting for the complementation patterns of RATHER offered by CxG in section 5.However, before doing so, we explore another pattern associated with RATHER that offers more insight into the categorial status of rather.

Premodifiers of rather
The data we present in this section center on two adverbs able to premodify rather, which can serve as a diagnostic for its verbal status.The first of them is a lot, which modifies adverbs and adjectives, but not verbs, and the second is much, which may modify all three categories.Thus, if rather is preceded by a lot this forces adverbial interpretation on it.We hypothesized that the string would/had a lot rather doesn't appear in environments associated with a verbal rather, such as raising to object.The string would/had much rather, on the other hand, could well be found with all the patterns we listed in table 1 for RATHER, since much doesn't block verbal use of rather.We add data on the corresponding strings with sooner for comparison.We have confirmed our hypotheses with data also harvested from COCA.First, the string would/had a lot rather appears three times in the corpus, each time with an infinitival complement (all three examples are depicted in ( 52)-( 54) below).The string would a lot sooner appears a single time, also with an infinitival complement (55).
(52) All I know in my state is that people would a lot rather have jobs than they would some empowerment zones or whatever you want to call it.These examples demonstrate that if rather is interpreted as an adverb, the comparative modal RATHER lacks uses that would indicate the presence of a lexical verb.Not surprisingly, the same holds true of the comparative modal SOONER.
Moving on to the strings would/had much rather and would much sooner, the data are given in table 3. Exactly as expected, would/had much rather again permits finite clause complements (56), NP complements (57) and the NP + Infinitival complement pattern associated with raising to object (58), while would much sooner permits only infinitival complements.These data include strings with bare rather, which host only infinitival or finite-clause complements.The picture that emerges from these data is that rather, unlike sooner, does exhibit distinct verbal characteristics when it appears as part of the comparative modal RATHER despite lacking overt verbal morphology.
5 Revised analysis of RATHER and SOONER The current corpus results reveal that the idea that SOONER and RATHER have developed into parallel micro-constructions requires some amendments.We have seen specifically that RATHER permits a much wider range of complements and that rather itself has distinctly verbal characteristics.We therefore take the stance that RATHER, but not SOONER, represents a micro-construction whose properties are licensed by the Raising-to-objectCxn in addition to the ModalCxn.In other words, the set of properties of RATHER consists of those that are passed down to it by more than one schematic construction in the network, a licensing mechanism known in the CxG literature as multiple inheritance (for more discussion of multiple inheritance, see e.g.Goldberg 1995;Hoffmann & Trousdale 2013;Trousdale 2013;Sommerer 2020).We begin by spelling out how the complementation patterns of SOONER and RATHER are licensed and then proceed to address the categorial status of rather.
Recall from section 2.2 that the set of micro-constructions falling under the PreferenceCxn includes the comparative modals SOONER and RATHER and the lexical verb prefer.We leave the semantics of these micro-constructions aside in this section, revisiting only their syntax.The PreferenceCxn, being itself licensed by the macro-schema ModalCxn, licenses the modal-auxiliary properties of SOONER and RATHER, that is, their ability to take bare infinitives.Since both SOONER and RATHER also exhibit transitive-verb properties by permitting finite clause complements, we assume that these are licensed by the macro-schema TransitiveCxn.The TransitiveCxn additionally licenses the object-taking ability of RATHER.There is of course nothing unusual about allowing these comparative modals to be licensed by both the PreferenceCxn and the TransitiveCxn, since English already has verbs (e.g. the semi-modals need and dare) that straddle the boundary between modal auxiliaries and lexical verbs.The remaining subcategorization frames that are permitted only by 89 ENGLISH COMPARATIVE MODALS AND THEIR COMPLEMENTS RATHER -NP+ predicative complement and NP + infinitival complementare licensed by the Raising-to-objectCxn, which we take to be a subschema falling under the TransitiveCxn.The Raising-to-objectCxn licenses constructions where NPs appear in object positions without being assigned any semantic roles in these positions.These links are represented schematically in figure 2. We are assuming here that all transitive-verb properties of RATHER are licensed by virtue of the fact that it must satisfy all constraints imposed on the Raising-to-objectCxn and those imposed on the TransitiveCxn, which licenses the Raising-to-objectCxn. Therefore, RATHER doesn't inherit directly from the TransitiveCxn.
Syntactic changes where a fully formed construction (or a form-meaning pairing being formed) gradually expands its range of syntactic contexts are known as syntactic expansion. 15Because syntactic expansion triggers changes to the form, but not to the meaning, it doesn't affect a form-meaning pairing as a whole and, therefore, can only be seen as constructional change if one adopts either Traugott & Trousdale's (2013) or Traugott's (2022) definitions of constructionalization, as we do here. 16We assume that the expanded range of complementation patterns that has become available to RATHER instantiates constructional change (specifically, syntactic expansion) surrounding the constructionalization of RATHER.According to Traugott & Trousdale (2013), syntactic expansion often follows constructionalization, but may be coextensive with it as well.
With respect to the categorial status of rather, we propose that it functions as a verb when in combination with an auxiliary.We are forced to conclude that this is so, given in particular the ability of RATHER to license raising to object and the availability of the auxiliary have to express counterfactual semantics with no other verbal form to be assigned participial features, as discussed in sections 4.3 and 4.4.Furthermore, it is due to the presence of verbal rather that the combination of would and rather licenses the full range of complementation patterns we have seen in the current data.The auxiliary would, whether on its own or combined with a clearly adverbial rather, licenses only infinitival complements.The specific theoretical possibility we defend here is that the lexical-verb component of the combination of rather with an auxiliary comes from the construction as a whole rather than any of its individual members.That is, this micro-construction has a lexical-verb slot that is filled by rather, which allows the construction to pattern syntactically much like it would if the verbal slot was filled by the verb prefer.That rather is able to fill this slot, or, in other words, that it has a distinct potential to transition into a lexical verb, is not surprising, given that it does have a record of being used as a verb (however scarce the corpus evidence to support this claim).We can thus think of rather as a defective verb for the paucity of examples with overt verbal morphology,17 or as underspecified with respect to its lexical class.

Conclusion
In this article we have presented novel evidence that the comparative modal RATHER has followed a different developmental path than SOONER.RATHER owes its special character to a wider range of complements, which requires a theoretical analysis that goes beyond that offered by Traugott (2019).Toward this purpose, we have offered a CxG analysis where RATHER represents a micro-construction licensed by the ModalCxn and the Raising-to-objectCxn, the latter a subtype of the TransitiveCxn.We have also proposed that rather fills the verbal slot in this micro-construction.

(
26) I mean if we're to have a cookout, I'd rather friends come over.(MOV: The Neighbor, 2019) (27) Besides, I would rather Houston be known as a 'world-class city' on the basis of improvements to a variety of services that don't involve 'corporate welfare'.(NEWS: Houston Chronicle, 2000)

( 29 )
Though I'd rather him with his shirt off, I agree with this handsome piece of work.(Edge of Fear (IMBD, Open subtitles), 2018) (30) She's taken them to the house in Georgetown.I'd rather them at home.(MOV: Jackie, 2016) Verbs taking NP and predicative complements (which may be realized as NPs, APs or PPs) also belong to the class of raising verbs, typical examples of which are elect and prove, illustrated in (31)-(32).

(
33) *I would Todd Akin in the United States Senate than Claire McCaskill.(34) *I would him cook than not cook.(35) *I rather Todd Akin in the United States Senate than Claire McCaskill.(36) *I rather him cook than not cook.

(
41) I think we played pretty well.I would rather have seen Ezeli more in the second half.(BLOG: http://blogs.mercurynews.com/warriors/2012/11/17/promising-how-the-pieces-fit-warriors-106-minnesota-98,2012) 85 ENGLISH COMPARATIVE MODALS AND THEIR COMPLEMENTS (42) Christine, he was quite certain, would sooner have paraded in a potato sack than anything remotely resembling pants.(FIC: Lady of Skye, 2001) (43) For example, his recent writings about the Regional Air Quality Council's meetings exposed shenanigans I'm sure the 'growth elite' would as soon have kept quiet.(NEWS: Denver Post, 1995)

( 44 )
Frankly, I would have rather been backed over by a school bus than sat down to a meal.(MAG: Cosmopolitan, 2006) (45) Yabrith would have sooner believed the sun would voluntarily refuse to rise than those men be lost to the guild.(FIC: The Hollow Queen, 2016) (46) Though a very big part of him would have just as soon kissed her.(FIC: The Bad Boy, 2004) (SPOK: PBS_Newshour, 1993) (53) I'd a lot rather have talked to the puppies.(FIC: Fantasy & Science Fiction, 2007) (54) I'd a lot rather have somebody shooting hoops than shooting bullets.(SPOK:PBS_Newshour, 1994) (55) She is more than competent and qualified, but there are others I would a lot sooner see on the Supreme bench.(BLOG: http://emptywheel.firedoglake.com/2010/04/18/a-concurrence-inthe-case-against-elena-kagan,2012)

Figure 2 .
Figure 2. Simplified constructional networks that license RATHER and SOONER are depicted in the examples below in the order in which they appear in table 1.
(15) So you don't mind that John is poor?No, but I would rather he have a house.(Mov:Little  Women, 1994)

Table 1 .
Complementation patterns for RATHER, SOONER and would ( just) as soon (COCA) Perhaps we should make you ambassador to the emperor.He doesn't deserve her.Would he sooner we didn't get the ships?(Mov:DangerousBeauty, 1998) (17) In fact, the TSAwould just as soon nobody even stood up during the flight at all.(SPOK: NPR She's taken them to the house in Georgetown.I'd rather them at home.(Mov:Jackie, 2016)NP + infinitival complement:(21) To me, I feel like when we see Tribe on-stage we want to believe that there's love.And if it's really not and behind the scenes it's some BS going on I'd rather them not be up there.

Table 2 .
Complementation patterns for RATHER, SOONER and would ( just) as soon when auxiliary have follows would (COCA)

Table 3 .
Complementation patterns for would/had much rather and would much sooner (COCA) We're in a new age of nutrition discovery, and as a dietitian, I'd much rather people focus on what they should be adding to their diets.(NEWS: Atlanta Journal Constitution, 2013) (57) I'd much rather Howard Zinn's paranoid style of American History than have progressives airbrush themselves into Mathew Brady's photographs.(BLOG: http://pjmedia.com/richardfernandez/2009/09/19/to-the-manner-born, 2012) (58) But I'd much rather him ask those questions and get it out there, prove to the American people that nothing has been violated in terms of the Constitution.(SPOK: CNNAM, 2006) Furthermore, we have found an example much like those discussed in section 4.3, where the auxiliary have follows would and a finite clause serves as the complement, given in (59).(59) Just worked for six weeks in spring training to get here, plus the offseason.So would have much rather this popped up in spring training.(MAG: ESPN, 2019)