Facing Two Ways Syntactically: On the Grammar and Use of Promise and Threaten in Three Regional Varieties

ABSTRACT The predicates promise and threaten with a to infinitive complement can be used in two main senses, with each sense representing a different argument structure: subject control, and subject to subject raising. Recent work has shown both uses to have been available for several centuries, with the raising variant emerging later than the control. The present investigation aims to provide insight into the usage patterns of the two predicates in the Englishes of New Zealand, Britain, and North America. The former is compared to the two latter varieties to help further our understanding of the complex system of the grammar of English predicate complementation. The data is examined from quantitative and qualitative perspectives, and the results indicate not only significant variation in usage among the varieties, but also high levels of correlation in some areas. New findings emerge from the study, including on the notion of stance, offering fresh avenues of research into the nature of the complementation patterns examined and their use in three core varieties of English.


Introduction
One of the fundamental insights in the area of sentential complementation to emerge from the generative-transformational model of linguistic analysis concerns the basic dichotomy of two types of to infinitive complements in English. To illustrate the dichotomy, consider sentences (1a-b).
(1) a. John seems to be a millionaire.
b. John wants to be a millionaire.
The matrix verb of (1a) is seem, and that of (1b) is want, with each selecting a to infinitive complement. In each case the sequence of overt constituents is: "NP-Verb-to-be NP". However, in the generative model their derivations and the resulting structures are sharply different. In (1a) the matrix subject is moved from the lower subject position into the matrix subject position by subject to subject raising, which is a type of NP Movement rule, leaving a coindexed NP trace behind. The matrix subject, the NP John, receives its theta role in its original subject position in the lower clause. A convenient way to describe the nature of the matrix verb in (1a) is to say that seem is a subject to subject raising verb. As regards (1b), there is no movement into the higher subject position. Instead, the higher subject is generated in the matrix subject position by phrase structure rules. The to infinitive complement is still sentential, for instance because of the need to satisfy the theta grid of the lower verb. It has its own subject, which is implicit or covert, and represented by the symbol PRO. PRO is a pronominal NP that is not pronounced. The complement structure construction is one of subject control, and the matrix subject controls the reference of PRO and imposes coreference between the two NPs in (1b). In this case, the matrix subject receives its theta role from the higher predicate, with PRO getting its theta role from the lower predicate. Regarding the nature of the matrix verb in (1b), it can be said that want in (1b) is a subject control verb.
The model explains for instance why a sentence of the type of (2a), which is from the News on the Web (NOW) Corpus, is completely natural, whereas a sentence of the type of (2b), freely invented using (2a) as a model, is unlikely.
(2) a. The latest news seems to be encouraging.  b. *The latest news wants to be encouraging.
In (2b) the higher predicate, want, again assigns a theta role to its subject, and it would prefer a human (or at least an animate) subject. By contrast, in (2a) seem does not assign a theta role to the NP in the higher subject position, which explains why even an abstract NP, raised into that position from the lower clause, by subject to subject raising, is fine in that position, as in (2a). The number of matrix verbs of the subject control type is relatively large in English, while the number of matrix verbs of the subject to subject raising type is relatively small. This conclusion is grounded in the list in Alexander and Kunz. 1 Their list is illustrated with the example My father WANTED to see the world, but it comprises both types of verbs. The number of subject to subject raising verbs is relatively small in their list, while verbs of the subject control type is much larger (see also the lists in Rudanko). 2 At the same time, the disparity in the numbers of verb types does not mean that they are infrequent at the level of usage, since for instance seem is very frequent with to infinitive complements.
The list of verbs in Alexander and Kunz is also of interest for another reason. When the verbs listed are considered from the point of view of the subject control versus subject to subject raising dichotomy, it is observed that for the vast majority of the verbs it is true to say that with to infinitive complements they subcategorise either for subject control or for subject to subject raising. In other words, verbs that are compatible with either structure are very rare in the list. When the verbs listed by Rudanko in the subject control pattern with to infinitives are considered, it is likewise observed that very few of them are triggers for subject to subject raising. The purpose of this study is to investigate two matrix verbs that are exceptional in this respect in that they can be found both in 1 Alexander and Kunz, Some Classes of Verbs in English, Vol. 1, 75. 2 Rudanko, Complementation and Case Grammar, 21-3. subject control and subject to subject raising structures. The verbs in question are promise and threaten, the latter the "evil twin" of promise, as Culicover and Jackendoff 3 label threaten.
Initial illustrations of these verbs are given in (3a-b) and (4a-b), from the Great Britain part of the NOW Corpus.
(3) a. The mayor promised to write back to the committee … (20-07-23 GB) b. This promises to be quite an exciting biking year … (10-01-14 GB) (4) a. The accused […] threatened to post clips on social media if the victim went to the police.  b. Ultra-low rates threaten to cause serious distortions;  The sentences in (3a) and (4a) illustrate subject control constructions, with the matrix verbs assigning theta roles to their subjects, and there is no relevant movement rule applying. For their part, the sentences in (3b) and (4b) illustrate subject to subject raising constructions. In them the higher subject is generated in the lower subject position and then raised into the higher subject position, with a coreferential NP trace remaining behind.
The difference in the interpretations of the control and raising variants is also reflected in a difference of the senses of the matrix verbs in the two constructions. To consider promise first, the sense of the verb in (3a) is the one represented under sense 1 of the verb in the Oxford English Dictionary (OED). Part of this sense is "to undertake or commit oneself to do or refrain from (a specified thing or act)". On the other hand, for sentence (3b), where raising is applicable, the sense of the verb is that of sense 5 in the OED. Part of this sense is "to give strong or reasonable grounds for expecting (future achievements or good results)". An analogous difference can be discerned for threaten. In the case of the subject control construction, as in (4a), the sense of threaten is the one represented by sense 3.a in the OED "to hold out or offer (some injury) by way of a threat; to declare one's intention of inflicting". Also relevant to the subject control construction is the OED sense 7, "In weakened use: to express an intention to do something, not necessarily evil", with the example "he threatened to give me money". As for the subject to subject raising construction, illustrated by (4b), sense 4.b in the OED is appropriate, which is "to appear likely to do some evil". It is possible to say that the control constructions involve meanings of the matrix verbs that are more lexical and more specific, and that the subject to subject raising constructions involve meanings of the matrix verbs that are more bleached and more grammaticalised. 4 In this sense the raising constructions have a degree of epistemic modality about them and, as Traugott notes, 5 they are similar to semi-auxiliaries, such as be going to. The OED senses of raising promise and threaten reproduced above include the verb phrases to give … 3 Culicover and Jackendoff, Simpler Syntax, 437. 4 Traugott, "The Conflict Promises/Threatens to Escalate into War", and "Subjectification and the Development of Epistemic Meaning" for discussion of promise and threaten in this context; Verhagen, "Subjectification, Syntax, and Communication"; Cornillie, "On the Pace of Syntactic Elaboration" and "Syntactic Complexity in Standard Average European"; Kissine, "Metaphorical Projection, Subjectification and English Speech Act Verbs". See Hopper and Traugott, Grammaticalization, for background on this topic. 5 Traugott, "Subjectification and the Development of Epistemic Meaning," 190. grounds for expecting and to appear likely to … , which strongly suggest that the evaluation of the likelihood of the enactment of the action denoted by the lower verb rests upon the speaker. Traugott 6 provides what is probably the most detailed discussion of the development of the epistemic uses of promise and threaten within the concept of subjectification. This widely-discussed process, which, in Traugott's conception, places an emphasis on "meanings that express speaker attitude or viewpoint", 7 is important to note here, because in the classification of corpus data much can rest on the extent to which the evaluation of the lower clause proposition can be attributed to a subjective viewpoint.
Earlier work has shown that both the more lexical and the more grammaticalised senses of the matrix verbs have existed in English for several centuries. 8 It also emerges from this body of earlier work that the proportions of the two senses and of the two types of syntactic structures associated with them have sometimes shifted in the course of recent centuries. 9 Regarding earlier studies, Traugott's 1993 study is an important investigation of the evolution of the two types of constructions with the two verbs in British English from Middle English onwards, and the present authors recently examined the patterns in question, again in British English, up to the 1920s on the basis of a systematic study of the Corpus of Late Modern English Texts, version 3.0. 10 However, as far as we are aware, the incidence of the properties of the two types of constructions have not been explored in very recent English on the basis of large electronic corpora. It is one of the aims of the present study to fill this gap in the literature. In view of their earlier work on the patterns in question in BrE up to the 1920s, the present authors naturally chose BrE as one of the varieties to be investigated in this article. This study also investigates the two types of constructions in very recent New Zealand English and American English. The choice of the former is made because the area of sentential complementation has sometimes been neglected in the study of that regional variety, 11 and is further motivated by the historical link between the two varieties of NZE and BrE. 12 AmE is chosen in order to represent another major source of potential influence for a smaller English variety. The three varieties are all inner-circle varieties, facilitating the comparison of the uses of the variants and their properties from that perspective.
A comparison of regional varieties invites discussion of language change under the influence of the dominant varieties of AmE and BrE, or, in the case of the varieties under discussion here, what has been termed "colonial lag". 13 The influence of AmE on other Englishes in general has been the subject of research in recent years, 14 and 6 Ibid. 7 Traugott, "(Inter)Subjectivity and (Inter)Subjectification," 60. 8 Traugott, "The Conflict Promises/Threatens to Escalate into War" and "Subjectification and the Development of Epistemic Meaning"; Cornillie, "On the Pace of Syntactic Elaboration"; Rickman and Rudanko, "Straddling a Syntactic Divide." 9 Cornillie, "On the Pace of Syntactic Elaboration"; Rickman and Rudanko, "Straddling a Syntactic Divide." 10 Traugott, "The Conflict Promises/Threatens to Escalate into War"; Rickman and Rudanko, "Straddling a Syntactic Divide"; on the corpus in question, see Diller, De Smet, and Tyrkkö, "A European Database of Descriptors of English Electronic Texts." 11 The present study also forms part of a larger project of research on predicate complementation in NZE currently being conducted by the first author. 12 Gordon et al., New Zealand English, for in-depth discussion of the development of NZE and its relationship with BrE, though from a largely phonological perspective. 13  the availability of appropriate corpora has no doubt helped to make this research possible. The present study is synchronic, however, and thus we will not inquire into changes of complementation tendencies with promise and threaten. Nonetheless, a picture of the current state of affairs can be pieced together, which will be useful in providing a foothold for future work.

Data and Methodology
The NOW Corpus was selected as the source of data for the present study because it offers a large body of data of very recent English and because it is possible to collect information from it on a country-by-country basis. The NOW Corpus currently comprises 15.3 billion words of data collected from online newspapers and other publications from 20 countries around the English-speaking world. The majority of the data are from American and British-based sources, but NZE is also very well represented, making it possible to obtain large amounts of data on promise and threaten in all three varieties. The search strings used were "[promise]_v* to" and "[threaten]_v* to". These strings were designed to find the verbal uses of promise and threaten followed directly by to. From each variety, a random set of 500 tokens produced by each string was taken using the built-in random sample function of the www.english-corpora.org interface. This retrieves a non-reduplicating set of random tokens from the entire period covered by the corpus, which is 2010 to the present day. (At the time the data were collected, the most recent material available in NOW was from the end of 2021.) This approach was chosen over simply taking the most recent 500 available tokens in order to minimise the presence of duplicate tokens in the datasets. Duplicate tokens were still found in the promise datasets of all three varieties, however, where, in each case, a sentence was reproduced in different publications several times, resulting in the removal of as many as 62 tokens in the case of the US promise dataset, and somewhat fewer in the NZ and GB datasets. 15 The vast majority of tokens retrieved are relevant to the present investigation, but there are some that are not. They include tokens where to, rather than introducing a to infinitive complement, is instead the head of a prepositional phrase in a different complement pattern, as in (5).
… the money promised to him by his aunt in the UAE was still nowhere to be seen. (16-07-28 NZ) The rare instances of the type shown in (6a-b) were also excluded. Sentence (6a) does involve a sentential complement of promise and the surface subject of promise, the NP more stock, does originate in the lower clause. However, the raising rule it has undergone is subject to object raising, after which it is moved by the NP movement rule of Passivisation into the subject position. Subject to subject raising is not involved in its derivation, and the pattern, apparently quite rare with promise, can be set aside in the present investigation. Sentence (6b) also involves a to infinitive clause, but the construction is of the type "NP call [upon NP] [to Verb …]", with the to infinitive sentence being a complement of call. Also set aside in the present investigation are tokens of the type in (7a-b).
(7) a. No school is being threatened to close. We're not doing any of that. (19-09-10 US) b. … five years after the affair she was threatened to keep quiet by a man she did not recognize in a Las Vegas parking lot.  Although the to infinitive clause in (7a-b) is a sentential complement of threaten, there are structural differences between this pattern and the subject to subject raising and subject control types discussed so far which rule it out of consideration in the present study. The clearest difference is the fact that, like (6a), the higher subject, the NPs no school and she, are the surface subjects of passive clauses and were therefore moved into that position via passivisation, instead of being generated in that position and receiving their theta roles as the original subjects of threaten. It seems clear that this type of subject originated as the object of threaten, with the theta role of Patient, and left behind an NP trace as a result of the passivisation process, which acts as the controller of PRO. 16 This is therefore neither a case of subject control nor subject to subject raising. In addition to the examples in (7a-b) there was only one other such example in the dataset, found in the US data, which appears to be a reproduced token stemming from the same news item as (7b).
The search string was intended to retrieve verbs but part-of-speech taggers are not infallible and other word classes also turned up in the datasets: nouns in the case of promise, and adjectives in the case of threaten, as shown in (8a-b). (8) a. … fulfilled the French president's campaign promises to pull no punches with Russia … (17-05-30 NZ) b. None were as divisive or as threatening to the royal family's standing as those surrounding race. (21-03-10 US) The datasets of both verbs were manually scanned for irrelevant tokens. The total numbers of the remaining relevant tokens are given in Section 3.

Results and Discussion
This section presents the results of the corpus data analysis, dealing firstly with the results for promise, followed by the results for threaten, and a discussion of the data taken as a whole. Table 1 gives the numbers of promise tokens categorised as subject control and subject to subject raising in the three varieties.

Promise in NOW GB, US and NZ
The results show that promise is used more frequently as a control predicate than a raising predicate in all three varieties. Furthermore, in our datasets, there is some clear variation in the ratio of control to raising tokens in each variety. In BrE the share of the total number of instances of promise that is occupied by the raising variant is slightly over one-third, and as such it is higher than it is in either of the other varieties. AmE and NZE are at comparable levels, with the raising variant accounting for 27% and 28%, respectively, of all instances of promise. The difference between the BrE and the AmE uses of promise is statistically significant at the 0.5% level (chi-squared = 5.63, p < 0.05). The difference between the BrE and the NZE results is also statistically significant at the 0.5% level (chi-squared = 4.39, p < 0.05). The difference between the AmE and NZE uses of promise, however, is not statistically significant, indicating that NZE and AmE are more closely aligned to one another than they are to BrE in terms of the ratio of control to raising.
(9a-c) give examples of promise in its subject control use, from the datasets of each of the three regional varieties. (9) a. As always happens at such times, politicians promised to go to the ends of the earth to find the perpetrators. (18-08-11 GB) b. That includes 120W HyperCharge technology that promises to juice your phone full in less than 20 minutes. (21-09-16 US) c. … it's the story of four teenage friends who promised to be at each other's weddings no matter what.  These are, at least in the case of (9a) and (9c), standard, uncontroversial examples. The subjects of promise in these two tokens are both animate, human entities and as such are "able to perform an illocutionary act", 17 and make suitable subjects for promise in its performative (subject control) meaning. These are thus prototypical subject control constructions. The subject in (9b), on the other hand, is not human, nor animate, but this exampleand several others in the datasets like itwas classed by the present investigators as subject control. The commitment to undertake the act encoded in the lower clause is interpreted as having been made by the creators and/or marketers of the 120W Hypercharge technology, thus a link between a human agent and the inanimate grammatical subject seems evident, even if such tokens do not represent prototypical subject control constructions. 18 Clues lending support to this interpretation can often be found in the complement clause in these types of tokens, i.e., the detail in the prepositional phrase in less than 20 minutes seems less likely to be the evaluation of an outside observeras it would have to be under a raising interpretationand more likely to be the repetition of assurances given by the creators of the technology. The compatibility with a suitable manner adverb was used as a test to determine the classification of a number of the borderline tokens that turned up in the datasets. For example, the insertion of explicitly into the frame of (9b) is shown below, in (9´).
(9´) That includes 120W HyperCharge technology that explicity promises to juice your phone full in less than 20 minutes.
In all such examples, the insertion of the adverb had to give a plausible-sounding result in order for the token to be classed as subject control.
Turning to the subject to subject raising use of promise, (10a-d) give examples. The examples in (10a-c) are the prototypical raising types. The subject of (10a), this year's route, is [-Animate], and, in particular, the dummy subject there and weather it of (10b-c), are raising subjects par excellence. 19 Alongside such prototypical examples, however, it can also be seen that [+Human] subjects, as in (10d), are also possible with raising promise. Rickman and Rudanko 20 observed in their dataset from Early Modern English that in such cases often "the lower verb describes a human quality that one does not usually have control over, such as physical and mental attributes", as in the girl promises to be a good athlete.
As well as a few of these types, another type is found in the data, one which may be described as the expectation of a certain type of performance, based on the reputation of the referent of the NP subject. (10d) was an example of this, and we supplement it with the illustrations in (11a-c).
(11) a. … they promise to mesmerise you with their voices and invisible instruments … The majority of such types in our datasets describe entertainment or sporting events, and the subjects are the performers or athletes from whom a good performance is expected. Judgements on this type of example are likely to be a matter of interpretation dependent on context: it is theoretically possible, in an appropriate context, to interpret these as examples of subject control, but to the present investigators it seems unlikely that e.g., the lecturer of (10d) made a promise to give an outstanding talk, or the musician of (11c) made a pre-concert pledge to be interesting. Instead, the sentences express the speakers' (or writers') views of what is likely to be the case and involve subject to subject raising. The [+Human] subject type may have been among the first of the epistemic uses of promise with a non-finite complement to emerge in the eighteenth century, 21 and it seems in Present-day English to have found a certain semantic area with which it tends to be linked.
Turning to the complement clause with promise raising structures, the predominant lower verb by far in all three varieties is be. As noted by Traugott,22 this was the earliest verb to emerge with the non-finite complement clause as the epistemic use of promise developed in the eighteenth century. It was initially restricted to descriptions of inchoative events, i.e., the girl promises to be a good athlete, and, although the inchoative type is infrequent in the NOW datasets, be is clearly the most common lower verb in raising structures today. The GB dataset has be in 89 (or 54.6%) out of all 163 examples; US has 41 (36.6%) out of 112; and NZ has 84 (65.6%) out of 128. A wide variety of other lower verbs are of course possible, as evidenced in (10c-d), the more frequent include deliver, bring, make, and change. (Lower clause verbs in promise control structures, on the other hand, are much more uniform in their distribution, with be generally as frequent as activity verbs such as do, make, take, return.) One further pattern, an off-shoot of the raising structure of promise, turned up in the data with noticeable frequency and deserves attention here. Examples of the "what promises to be X" pattern are given in (12a-d).
(12) a. … on Sunday Pittodrie will be the scene of what promises to be a full-blooded collision between Aberdeen and Rangers … (19-03-02 GB) b. We're thrilled to have her here with 2degrees for what promises to be a stunning free show. The pattern appears to be a useful discourse strategy, providing focus for the following NP, and it is worth noting that the what promises to be clause is structurally entirely unnecessary, as the removal of those four words shows in (12´a-b).
(12´) a. Pittodrie will be the scene of […] a full-blooded collision between Aberdeen and Rangers 21 Traugott, "The Conflict Promises/Threatens to Escalate into War," 352-3; Traugott, "Subjectification and the Development of Epistemic Meaning," 188-9. 22 Traugott, "Subjectification and the Development of Epistemic Meaning," 188. b. We're thrilled to have her here with 2degrees for […] a stunning free show.
It is most commonly used in the context of a much-anticipated sporting event or concert performance, as seen in (12a-b), and occasionally some other type of event or product (12c-d) that deserves to be heralded with the type of extravagant and superlative NP modifiers seen in most of these examples. The past tense what promised to be is possible but it is far less common than the present simple, and is found only once in our data; this pattern is used predominantly with a future orientation. There is little noteworthy variation in the context of use across the three varieties in our data. Frequencies of usage though, are a little different: the pattern comprises 26 (or 16%) of the 163 GB raising examples, 12 (10.7%) of the 112 US, and 15 (11.7%) out of the 128 NZ tokens. These ratios are largely in line with the general raising usage frequencies for each variety seen in Table 1, with BrE somewhat higher than the other two varieties. 23 Table 2 provides the numbers of control and subject to subject raising tokens in the threaten data in the three varieties.

Threaten in NOW GB, US and NZ
Like promise, threaten is clearly used more frequently as a control predicate than a raising predicate. Unlike promise though, the ratio of raising to control uses with threaten appears to be much more uniform across the three regional varieties. AmE and NZE are almost exactly the same, at 37.7% and 37.8% raising, respectively, and BrE is only a small margin ahead, with 38.2% raising. Beginning with subject control, (13a-c) below give examples.
(13) a. The court heard in September 2015 she threatened to bite the nose off a security guard at the RVI … (19-12-12 GB) b. The NFL is threatening to force teams to forfeit games if they have COVID-19 outbreaks among unvaccinated players.  c. The letters threatened to contaminate infant formula if New Zealand did not stop using 1080.  Subjects in control structures with threaten are most typically animate and human, as in (13a), or collective bodies such as governments and associations, as in (13b). These are prototypical subject control constructions. For its part, the inanimate subject control type that is fairly frequent with promise (120W HyperCharge technology that promises to … ) is much less commonly found with threaten. One of the few examples is given in (13c). Here, however, the human agentthe writer of the lettersis unmistakable.
23 Given the frequency of promise in this pattern, it is of interest to look into the pattern with threaten. Our dataset contains only two examples of what threatens/threatened to be NP, and these were found in the NZ and US data. Both are from the sports/entertainment context, and, in contrast to promise, both are in the past tense. They are given in (a-b) below.
a. It was an incredible finish but it should not completely gloss over what threatened to be a major controversy with France once again accused of manipulating the Head Injury Assessment protocols … (18-02-04 NZ) b. … thrillingly meshing physical production, lighting, projection, sound and hydraulicslifts what threatened to be a movie retread into a live entertainment triumph. (21-09-13 US) The finding suggests that the pattern is used almost exclusively with promise, compared to threaten.
A handful of similar NP subjects denoting e.g., messages, emails, cards, executive orders, are found in the data. Moving on to threaten in its raising role, (14a-c) give examples.
(14) a. … several major blazes threatened to merge into a single dangerous fire front that could reach outlying districts of Sydney.  b. Even the film's style, characterized by conspicuous crosscutting and stuttering camera pans, is loose-lipped, threatening to let cats out of bags.  c. Heads threatened to roll unless they could find an immediate response and, to coach Michael Maguire's credit, he did just that.  The examples in (14a-c) all have the prototypical inanimate raising subject, and (14c) is particularly noteworthy in this respect, with the subject idiom chunk from the idiom heads will roll. The compatibility of raising predicates with subject idiom chunksonly possible with raising predicatesis among the standard set of diagnostics for distinguishing cases of control from raising. 24 [+Human] subjects are found with the raising predicate threaten, as they are with raising promise, and, like promise, they appear to be used most often in sporting/entertainment contexts. (15a-c) give examples.
(15) a. In Duloc we meet pint-sized baddie Lord Farquaad, who threatens to steal the show with his camp comedy and incredible timing.  b. Wylie was threatening to pull off a sweep with a 22-19 cushion in the third, but the Lady Cats hung tough and won the next six points to force a fourth set.  c. … the free-swinging 21-year-old threatened to cause a huge boilover, only for Kerber to regroup in the nick of time.  Recalling the way in which raising promise with a [ + Human] subject can be used to express an evaluation of a performance based on the subject's reputation, it can be seen that threaten can be used in a comparable way. The difference though, is that the description is often that of a surprisingly good performance, one contrary to expectations, the loser coming up from behind to almost "steal the show". A comparison of this type of use in the GB, US and NZ data shows its frequency to be relatively high in GB and NZ,and in US somewhat lower. 25 Thus, all three varieties use raising promise and threaten with [ + Human] subjects in similar ways to describe topics related to sports/entertainment, and in the case of threaten this may be more widespread in BrE and NZE than it is in AmE.  There are also examples of the more standard type of raising use, as shown in (16).
(16) Rapidly growing numbers of New Zealanders living with dementia threaten to overwhelm our health system unless government acts quickly …  A further noteworthy feature of raising threaten in the NOW data is its appearance in the progressive, as shown in (15b). While quite acceptable with control uses of threaten, the progressive is thought to be possible but somewhat limited when threaten is used as a raising verb. 26 (As noted by Traugott, raising verbs in general are not normally compatible with the progressive, particularly with animate subjects). Our data support the observation of this particular property of threaten, and indicate that raising threaten is in fact used in the progressive form with what might be called reasonable frequency, as shown in Table 3. As for promise, Traugott notes a split between the two verbs in this area, with promising "restricted to adjectival and gerund constructions". 27 This is supported by the NOW data, with no promise raising tokens and only one promise control token in the progressive.
Examples of raising threaten in the progressive are given in (17a-c).
(17) a. … the 17 million visitors they get annually are threatening to swamp the city. (17-11-01 GB) b. … they released water from dams that were threatening to overflow. (21-07-23 US) c. Storms are threatening to spoil Guy Fawkes parties across the north of the North Island.  NZE is known as a variety that favours the use of the progressive, or at least did so in the late twentieth century, 28 and this may go some way towards explaining the slightly higher figure for that variety. AmE, however, is not far behind NZE in this respect, with BrE noticeably further behind. In terms of subject animacy, which is a variable known to be a significant factor affecting not only raising verbs but also progressives in general, 29 the data show that six of the 18 GB threaten progressives contain animate subjects, as do 10 of the 28 NZ progressive tokens, while only three of the 26 US tokens have animate subjects. 30 Such differences point to a need for further investigation of the relationship between the progressive aspect and raising predicates in different regional varieties of English.
As for complement clauses with threaten, in its raising use the two verbs found to be the most frequent in all three varieties are derail and undermine, with others, such as break, destroy, and overwhelm also somewhat common but more variety-specific. (18a-c) give examples. 26 Traugott, "Subjectification and the Development of Epistemic Meaning," 194. 27 Ibid., 204. 28 Hundt, New Zealand English Grammar, 75-7; Collins, "The Progressive Aspect in World Englishes"; Collins, "The Progressive." 29 Hundt and Szmrecsanyi, "Animacy in Early New Zealand English." 30 Like Hundt and Szmrecsanyi, "Animacy in Early New Zealand English", we include collective nounsthe All Blacks, the group etc.among the animate types. These in fact account for a good many of the animate types in the datasets of all three varieties.
(18) a. An increase in Saudi oil flowing to Europe threatens to undermine Russia's principle market.  b. As the Celtics navigated a tumultuous regular season that threatened to derail their title aspirations … (19-05-03 US) c. … cleavers (aka sticky weed) will be climbing up and over many plants, threatening to overwhelm them if left to their own devices.  The frequencies of derail and undermine are very similar in the three varieties: derail -GB: 9, US: 10, NZ: 9; undermine -GB: 9, US: 9, NZ: 8. 31 With a strong collocation with lower predications headed by verbs such as these, it is obvious that threaten is a verb that entails a negative semantic prosody. 32 Indeed, it has been described as the "evil twin" 33 of promise, and, as noted above, the OED sense most applicable to raising threaten is "to appear likely to do some evil". With such descriptions, it is safe enough to say that wherever threaten is involved, bad things are rarely far away. Nevertheless, OED sense 7, also quoted above, is "to express an intention to do something, not necessarily evil", but even so, positive or even neutral events encoded by the lower predications of threaten are unlikely to be frequent. This expectation is indeed borne out by the subject control section of the data, with only a handful of candidates for the weakened "not necessarily evil" sense emerging. Two of these are given below.
(19) a. Kate had threatened to do a naked cartwheel if she wonsomething no one wanted to see. This was the Great British Bake Off after all … (17-10-31 GB) b. The developer of a Mosgiel subdivision has won the right to name his own streets and pathways, but only after threatening to use hard-to-pronounce Scottish village names instead.  The semantic prosodies of the lower predicates themselves in (19a-b) can be seen as neutral or, at best "not necessarily evil", but the selection of the higher predicate threaten lends them the somewhat negative implication that the intended act is undesirable. Still, these examples show that threaten can be used in a light-hearted, non-serious way, consistent with the OED sense 7. It also seems possible to say that the degree of commitment to carry out the act expressed in the lower clause is weaker than it would be had promise Note: Brackets = percentages of the total number of raising tokens. 31 These figures prompted a further search of the NOW corpus as a whole in order to gain a better understanding of the collocational strength of the threaten to derail combination. The results of a search using the string "VERB to derail" shows that not only is derail the preferred lower verb of the raising verb threaten, but threaten is the most common matrix verb selecting derail in NZ, GB and US datasets in the NOW corpus. The NZ and GB results were manageable enough to allow us to use all tokens returned by the search string (274 and 822 tokens, respectively), while a sample of 1000 tokens was taken from the total 3028 US results. The three varieties show a striking similarity in the collocational strength of threaten and derail: the two verbs were found together in 50.7% of the sampled US tokens, in 51.3% of the NZ tokens, and in 52.4% of the GB tokens. 32 Louw, "Irony in the Text or Insincerity in the Writer?"; Partington, "Evaluative Prosody" for comments on semantic prosody in individual lexical items. 33 Culicover and Jackendoff,Simpler Syntax,437. been the matrix verb instead, e.g., Kate promised to do a naked cartwheel if she won seems to entail a stronger commitment on the part of the subject.
The choice of matrix verb can also be useful in expressing stance or "spin" in political contexts. 34 In terms of stance, Gray and Biber 35 list the nouns promise and threat as stance nouns with a to-complement clause, and it seems clear that this can be extended to cover the verbs as well. Our focus here is on "spin", however, which is a term typically associated with US politics. It may be described as "the intentional slanting of ambiguous political events and situations to promote an interpretation favorable to one's own side". 36 A political agent, whether a politician or a newspaper editor committed to a particular party, may engage in positive spin by highlighting the advantages of his or her political agenda and the assumptions that go with it. It is also possible to talk of negative spin ("Frequently in to put a positive (negative, etc). spin on. colloquial (chiefly U.S. Politics", 37 )). In the latter case a political agent generally seeks to attack, and to find fault with, the competing political agenda of his or her enemies. In texts that are meant to argue for a certain point of view, where reality can be constructed in a particular way in support of that point of view, the choice of the higher predicate is an important one. In the context of the two verbs promise and threaten, the former is the clear choice for a writer seeking a positive spin, and the latter verb for a negative spin.
The influence that the verb choice can have on the framing of a speaker's message is seen clearly in (20a-b).
(20) a. Montgomery threatened to fight the lawsuit all the way to get Voice Tech Corp's patents invalidated.  b. In two tweets, Trump threatened to bring in the National Guard to control the situation.  Here, both the referents of the higher NP subjects and the events described in the lower clauses are invariably affected by the negative connotations of threaten, and are presented to the reader as undesirable. Replacing threaten with promise, however, would lend a positive spin to the story, and would help to encode a positive assessment of the referent of the higher subjects, and the events of the lower clauses. We would be more inclined to view the idea of fighting the lawsuit and bringing in the National Guard as good things. Consider the example of promise in (21). (21) Since his days on the campaign trail, President Donald Trump has promised to roll back environmental regulations, boost the use of coal and pull out of the Paris climate agreement … (20-02-01 US) Even though many people would no doubt think that they are undesirable and detrimental, the events described in the lower clausesas well as the subject NP referentare given a decidedly positive spin, situated in the benevolent shade of promise, the good twin of threaten. Such devices can be powerful tools of persuasion, but they can also 34 Gray and Biber, "Stance Markers"; Rudanko, "Representations of the Baltimore Riots of July 1812." 35  be used to simply raise reader interest in a story or to add humour, as (19a-b) appear to do.
Turning to the raising tokens, however, more obviously neutral or positive lower clause propositions are more likely to be found. They appear to be limited to the [+Human] subjects in descriptions of sports/entertainment, as in (22a-c) below.
(22) a. … made it feel a bit like one minute France were threatening to win, the next the game was all over … (18-06-09 NZ) b. Kuate threatened to become a cult hero with his stunning goal against Morton … (17-05-25 GB) c. As the song's energy begins to rise, she threatens to get up, but maintains her post.  The lower predications here, win, become a cult hero and get up are undoubtedly positive/ neutral events, but due to the choice of the matrix verb, the hypothetical events of the lower clause can be framed as events that are undesirable. This is clearly the case in (22a-b) but (22c) appears to be an exception, with threaten being used in a truly weakened sense. The example in question continues as follows, in an undeniably positive review of the performance: Then … she can't control the emotions of the song and rises up to face the crowd and belt the incredible chorus as the song peaks and closes. In light of this example, it seems that the weakened OED sense, viewed as a control construction in Section 1 above and seemingly presented as such in the OED, can also be associated with the raising use of threaten.

Conclusion
The verb pair promise and threaten are rare cases among the English system of predicates selecting non-finite complements, in straddling the syntactic divide between the subject control and subject to subject raising categories. Few other verbs lend themselves to use in quite such a variety of ways, with a colourful and arresting range of subject types and contexts. This study set out to shed light on the uses and distribution of these verbs in three regional varieties of English, based on the evidence of very recent newspaper data. Both prototypical and less prototypical cases of each type have been discussed, with the implication being that a gradient exists between the two syntactic categories, and the establishment of clear boundaries is not always a straightforward matter. Indeed, it is often the ambiguity stemming from the lack of clear boundaries that makes these verbs so intriguing and worth investigating.
In terms of the findings to emerge from this study, we have shown that there are significant differences between the three regional varieties in terms of how promise is used, with more scope for the raising variant in BrE, and less in both NZE and AmE, the latter two varieties appearing to be relatively similar in this respect. A synchronic snapshot is all we have given here, but the finding invites further diachronic work to uncover any evidence of regional influence. We have also shown that raising promise, in all three varieties, lends itself well to descriptions of entertainment and sporting events when used with a [+Human] subject. The sports/entertainment domain is also seen to be a significant factor behind the "what promises to be X" pattern, a structurally superfluous but pragmatically useful device for providing focus for an upcoming NP.
On the other hand, the data on threaten has shown some unexpected similarities among the three varieties, with raising and control ratios running almost in parallel in the three datasets. As well as the correlation in the overall frequencies, there appear to be clear similarities in the collocational strength between threaten and certain groups of lower predicates. Support was found in our data for previous observations on the compatibility of threaten with the progressive, with progressive forms reaching their highest levels in NZE. In short, the findings on threaten all deserve to be investigated in more detail in future work.
Taking the notion of semantic prosody as a starting point, the study has pointed to the use of the two verbs in the language of politics and of advocacy more generally. Political agents typically promote particular agendas involving sets of assumptions about the nature and status of controversial issues and each of the verbs can be used by a speaker as a vehicle of spin in an attempt to frame the perceptions of such issues, either positively (promise) or negatively (threaten), with a view to influencing voting behaviour. The present article illustrates such uses, providing a stimulus for further work combining grammatical analysis with the study of pragmatic function.
Going beyond the datasets examined here, the present article also invites follow-up studies on the uses of the two verbs and their complementation patterns in other regional varieties. Later work can for instance focus on the patterns in question in non-core or non-native varieties of English. In investigations of those varieties, attention would obviously need to be paid to the potential influence of the native languages of the speakers in question on their use of the patterns, but the analytic framework presented here may still serve as an inspiration for, and as a suitable theoretical underpinning of, such follow-up studies.

Disclosure statement
No potential conflict of interest was reported by the author(s).