... distinctions.indd 106-107 Should vs Ought to 107 —(only for ought to) do-support vs absence of do-support —(only for ought to) infinitive marker to vs reduced (oughta) vs absent The quantitative... of ought to (cf Kennedy (2002)), which amounts to a should/ ought to ratio of 19:1. Similarly, in Collins’s (2009) study (involving different corpora), the ratio of should to ought to appears to. .. observed from the table that should is much more frequent than ought to in the corpora, with a 9:1 should/ ought to ratio in the spoken corpus and a 36:1 should/ ought to ratio in the written corpus. These
Should vs. Ought to [Modals] Should vs. Ought to* Bert Cappelle and Gert De Sutter University College Ghent / Ghent University 0. Preface The seeds of this text were sown within Renaat Declerck’s ambitious but unfinished project on mood and modality. At one project work meeting in 2006, one of us (B.C.) presented some quantitative findings about significant factors in the choice between should and ought to, based on a fairly restricted data set (150 examples of each modal verb). A mutual comparison of the contributions of these factors was lacking, and this paper is an attempt to provide this. The idea of using an extensive set of parameters to investigate the distribution of modal auxiliaries lay at the heart of the modality project as formulated in its proposal. In this sense, the present paper is in the spirit of Declerck’s research aims. * Many thanks go to Renaat Declerck for allowing the research within his final project, on mood and modality, to be conducted from different theoretical and descriptive angles. This text, as well as the text by Ilse Depraetere in this volume, should prove that such space for intellectual freedom can bear fruit which we hope will carry Renaat’s approval. For the implicit views on modality, this text has undoubtedly benefited from the many e-mail exchanges and meetings with Ilse Depraetere, Susan Reed and An Verhulst which took place within that project. We are also especially grateful to Marie Baert and Fanny Lauwaers, who coded the corpus examples of should and ought to in written discourse as part of their course work for the 2006-2007 master’s course Descriptive English Linguistics: Selected Topics at K.U. Leuven. Finally, we would like to thank Naoaki Wada for his helpful suggestions. Any remaining shortcomings are our own. 93 At the time, however, Declerck did not think that a multifactorial statistical analysis of the sort presented here would advance the project. Instead, he argued that larger sets of examples should instead be examined with an eye to discovering rare or undescribed usage types. For this paper a set of examples more than three times the original size has been investigated, but our primary aim is not to report special types. Rather, our intention is to show that systematic analysis and coding of what is still a relatively restricted data set (approx. 500 authentic examples of each modal verb) can lead to confident conclusions about rarely described or hitherto undetected usage trends which distinguish these two semantically close lexical items. 1. Should and Ought to: The Challenge As the following sentences demonstrate, should and ought (to) can be used as what appear to be stylistic variants of each other (1a, b), as mutual paraphrases to strengthen each other’s meaning (2a, b), and, most remarkably, in tag questions as proforms for verb phrases containing the other verb, a possibility pointed out by Harris (1986: 353), Palmer (1990: 122), Perkins (1983: 55) and Swan (1995: 496)—though note that, for reasons that will become clear, a tag with shouldn’t is more frequent than one with oughtn’t (3a, b): (1) a. Tina had no moral sense about this question, no feeling that children ought to know who their fathers were or should be fathered by the men their mothers lived with or were married to. (British National Corpus (henceforth BNC), written discourse) b. Check the quality of the paper. It should not be limp, shiny or waxy and the heavily printed areas ought to feel crisp and slightly rough. (Cobuild corpus, Today Newspaper) (2) a. I’m not all that I should be and all that I ought to be, but by this time next year I’m going to be a bit better than I am just now, in spiritual terms. (BNC, spoken discourse) 92 distinctions.indd 92-93 10.7.26 11:52:15 AM 94 Bert Cappelle and Gert De Sutter b. It is all very well setting goals—but what if the very idea causes psychological reversals, or the athlete’s logical mind says they ought, should, could do it while their emotional mind says it doesn’t want to? (Cobuild corpus, UK books) (3) a. Suppose I ought to tell him that shouldn’t I? (BNC, spoken discourse) b. And yes, we should be mindful of Muslims’ sensitivities. But such mindfulness really should run both ways, oughtn’t it? (http://ace.mu.nu/archives/061590.php) Such examples appear to confirm the received opinion about the modals should and ought to, namely that they are so similar in meaning that they can typically substitute for one another. For instance, Palmer (1990: 122) writes, “It is not at all clear that (…) English makes any distinction between should and ought to. They seem to be largely interchangeable.” In Collins’s (2009: 52-53) summary of the literature, similar sentiments are expressed by Coates (1983: 69), Quirk et al. (1985: 227) and Huddleston and Pullum (2002: 186). Both should and ought to can also be used as epistemic modals. Unlike in their deontic use, their function is then not to speak about obligations or to offer advice but rather to express that the fulfilment of a situation is quite likely or can be reasonably expected given some (perceived) facts. For example: (4) a. Michael Saunders, UK economist at Salomon Brothers, said that with output prices already falling sharply, underlying inflation should drop below 2.5 per cent by the middle of this year. (Cobuild corpus, The Times newspaper) b. After all that, some morale-boosting planetary aspects on the 16th, 18th and 22nd ought to make you glow again. (Cobuild corpus, UK magazines) In their epistemic use, should and ought to convey a weaker version of logical necessity as expressed by must. That means that they do not refer to an inescapable conclusion. Indeed, sometimes they are even used distinctions.indd 94-95 Should vs. Ought to 95 when the likelihood of a situation is zero (in counterfactual situations); they then merely express that there was a reasonableness of expectation, which the context makes clear is not borne out, as in the following examples: (5) a. “You stupid fool,” he said aloud. “It should be obvious what’s happening to the Lands. They’re dying.” (Cobuild corpus, UK books) b. “Why doesn’t the Thing know where to go?” said Gurder. “It could find Floridia, so somewhere important like Blackbury ought to be no trouble.” (BNC, written discourse) Clearly, these examples show that should and ought to need not indicate the speaker’s/writer’s supposition that the statement he makes is true. Quite often, an epistemic interpretation is mingled with a deontic one. Consider: (6) a. Unless you are going somewhere subtropical, a wet suit is essential, as is a “foamie”, a fat foam board specially designed for beginners. Any good surfing school should have plenty of both. (Cobuild corpus, The Times newspaper) b. Who’d like to tell me what twenty-six will be? And then the next number? Jody. Come on nearly everybody ought to be able to work this one out now. Eleanor. (…) That’s lovely. Can you all tell me together what twenty-eight will be? (BNC, spoken discourse) In (6a), there is both an epistemic sense (‘go to any good surfing school and you’re likely to find that they have plenty of both’) and a deontic sense (‘any good surfing school owes it to itself to have plenty of both’). Similarly, in (6b), the modal should in the teacher’s utterance is epistemic (‘Since all these exercises are similar, I expect that this particular exercise will not pose any difficulty’) but it is undeniably tinged with a deontic reading (‘You’d better be able to work this one out, otherwise it’s clear to me you haven’t paid any attention’). Note, crucially, that in these in- 10.7.26 11:52:15 AM 96 97 Bert Cappelle and Gert De Sutter Should vs. Ought to stances of purely epistemic use or ‘merger,’ should and ought to are again interchangeable. Given the examples above and the generally accepted view that should and ought to are semantically equivalent, the challenge is to establish on which grounds, if any, should and ought to can nonetheless be distinguished. 2.1. Subjectivity vs. Objectivity Echoing Swan (1980: 550), Declerck (1991: 377, fn. 21) claims: “Although should and ought to are often interchangeable, there is a slight difference of meaning between them. When using should the speaker expresses his own subjective view; ought to is more objective and is used when the speaker wants to represent something as a law, duty or regulation. For this reason ought to may sound more emphatic than should.” Declerck offers the following contrast and accompanying comment, again loosely based on Swan (1980: 550): 2. Differences Proposed in the Literature We are not the first to rise to the challenge of finding differences between should and ought to. Some linguists have been dissatisfied with the generally acknowledged view that these modals are semantically equivalent and have put forward claims about subtle semantic differences. That there are differences in meaning between should and ought to is only to be expected. Despite their different ancestry (should originally being the past tense of shall and ought to the past tense of owe), the close synonymy of should and ought to goes back to at least Old English (i.e. to before the mid-twelfth century), as appears from attested examples adopted in the Oxford English Dictionary (Simpson (2000)). The fact that should and ought to have all this time been allowed to co-exist as distinct forms occupying the same semantic patch “would seem to justify the conclusion,” according to Visser (1963: 1637), “that they never had exactly the same meaning, however closely synonymous they always have been.” Yet, exactly what distinguishes these two modals semantically is hard to pin down. In the following subsections, we will consider three proposals made in the literature: (i) subjectivity vs. objectivity, (ii) absence vs. presence of an implication of non-fulfilment and (iii) relative frequency vs. infrequency of epistemic reading. In a final subsection, we will consider some other, non-semantic, differences proposed in the literature. distinctions.indd 96-97 (7) a. You should / ought to congratulate her. b. I ought to congratulate her, but I don’t think I will. (Should would sound odd here: it would be strange to give yourself advice and then add that you were not going to follow it.) (Declerck (1991: 377, fn. 21)) Likewise, Collins (2009: 54) makes the claim (though without backing it up with concrete figures) that while should and ought to are both “more commonly subjective than objective (…) the proportion of objective cases is higher with ought to.” He calls a deontic modal “subjective” when it “indicates what the speaker considers desirable, appropriate or right” (Collins (2009: 45)) or when “the speaker is giving advice authoritatively to the addressee” (Collins (2009: 54)) and “objective” when “the appropriateness or desirability of the course of action described stands independ ently of the speaker’s endorsement” (Collins (2009: 45)), i.e., when “generally accepted standards of appropriate behaviour are being invoked” (Collins (2009: 54)). A very similar distinction can be found in Myhill’s (1996) study, who argues in detail that should expresses an individual opinion while ought to stresses that an opinion is shared by a group. This reference to shared opinion, associated with objectivity, can be further linked to the claim that ought to, more strongly than should, suggests that the obligation is a duty leading to the greater public good (cf. Gailor (1983: 348); Aarts and Wekker (1987: 193)). It is hard to test the impact of subjectivity objectively: in individual examples, it’s not always clear whether the source of modality is grounded 10.7.26 11:52:15 AM 98 Bert Cappelle and Gert De Sutter in the speaker/writer or in some speaker-external source. We have used a somewhat different notion of subjectivity in our corpus investigation: whenever the modality (whether grounded in the speaker/writer or in some other source) emanates from a personal viewpoint, it is subjective; when the modal statement is issued without the speaker taking a personal stance whatsoever, it is objective. One way of testing subjectivity indirectly but fairly reliably might be to count the number of sentences in a corpus in which should and ought to occur in a clause which is the complement of an expression of cognition (assume, believe, feel, reckon, suppose, think, take the view, …). It can be hypothesised that if should is relatively more subjective and ought to relatively more objective, sentences like I think you should leave are more frequent than I think you ought to leave. 2.2. Absence vs. Presence of a Non-fulfilment Implication Ought to has been claimed to suggest that the actualisation of the situation referred to “is overdue or may be delayed” (Close (1981: 121)) or that it might not take place at all (Gailor (1983: 348-349)), implications which are thought to be absent with should. As Westney (1995: 170) remarks, it could also be the reason why As you should know… is less aggressive than As you ought to know … as an opener to give (superfluous) information or advice, since the latter would impolitely suggest that the hearer might not yet know. Support for Westney’s observation can be found in the relative normality of titles of websites or articles starting with “Ten things you ought to know about…”: here the writer has a good motivation to assume or suggest that the reader does not yet have the knowledge about the relevant topic, since otherwise writing the text would be rather pointless. However, Westney correctly points out that the suggestion of nonfulfilment is by no means always present with ought to, as is witnessed by the following example: (8) Whenever he got a chance, Malek broke into a canter, and one of those bursts turned into a long twilight gallop; he may have distinctions.indd 98-99 Should vs. Ought to 99 thought we were far from home and ought to get a move on… (Leigh Fermer, Patrick (1986) Between the Woods and the Water, Murray, London, quoted by Westney (1995: 168)) Non-fulfilment is therefore at best a (pragmatic) implicature rather than a (semantic) implication (i.e. entailment) in the strict sense. Sentences like (8), where actualisation of the situation can be deduced from the context, are rather exceptional, however. Whether or not the situation actually comes to fulfilment is not coded in the modal but is usually something to be derived extra-linguistically. This neutrality with respect to actualisation especially holds when the proposition refers to a future situation (Palmer (1990: 123)). In Degani’s (2009: 338) corpus-based study of ought to, the majority of sentences are indeed “non-factual” (rather than “actualised” or “counterfactual”). In such sentences, ought to simply conveys the idea that there is a situation whose actualisation is considered desirable or expected. It is hard to verify via corpus research whether ought to in such cases, on top of this notion of desirability or expectation, commonly suggests a lesser likelihood of actualisation than if should had been used. In fact, it would not be a reliable method to check whether the likelihood of fulfilment increases or decreases when replacing ought to by should in individual examples, since such judgments could only be made on a subjective basis. A more valid operationalisation would be to count the number of instances of should and ought to followed by a perfect infinitive (e.g. You {should/ought to} have asked me first) or a present progressive (e.g. I {should/ought to} be revising now), since these verb forms implicate counterfactuality (but see section 5 for some complications). 2.3. Relative Frequency vs. Infrequency of Epistemic Use A number of linguists have claimed or reported that should conveys epistemicity more commonly than ought to does. Coates (1983) reports that the ratio of epistemic versus deontic uses is roughly 1:4 in the case of should but only roughly 1:8 in the case of ought to. This difference is 10.7.26 11:52:16 AM 100 101 Bert Cappelle and Gert De Sutter Should vs. Ought to even more pronounced in Collins’s (2009) recent corpus counts, from which we can infer an epistemic/deontic ratio of about 1:6 for should and of about 1:32 for ought to. According to Palmer (1987: 134), it is only “theoretically possible to imagine ought to being used epistemically but that seems very rarely to occur. In general ought to is interpreted deontically.” In Palmer (1990: 60) he even reports not to have found a single example of epistemic ought to, although he acknowledges the existence of cases of merger. In sharp contrast, Degani’s (2009: 333) corpus counts of ought to in the FLOB and Frown corpora (representing 1990s English in the UK and the US, respectively) reveal percentages as high as 31% and 36% for the epistemic use. Expressed as epistemic/deontic ratios (disregarding cases of merger), this amounts to roughly 1:1.5 for UK English and almost 1:1 for US English. Interestingly, Degani also shows that the epistemic use of ought to seems to have increased in frequency since the early 1960s, from 21% in the LOB corpus (UK English) and 24% in the Brown corpus (US English), or from epistemic/deontic ratios of roughly 1:3 and roughly 1:5, respectively. We must conclude that the assessment of epistemic ought to differs widely across studies, ranging from being only a theoretical possibility to an interpretation which is as frequent as deontic ought to. Degani’s (2009) study only reports recent changes in the use of ought to, so our study will complement her study by comparing the use of ought to with the use of should in 1990s UK English. 19:1. Second, Celce-Murcia and Larsen-Freeman (1983: 89) claim that deontic should and deontic ought to contrast with each other in that the latter is more informal (most markedly so if it is phonologically reduced to oughta). Not differentiating between deontic and epistemic uses, a similar register-based difference for ought to has been noted by Collins (2009: 46): “Despite its small numbers in the present study, ought to was found to be considerably more robust in British and American speech than writing,” with the speech/writing ratios being roughly 3:1 and 4:1, respectively. (The number of epistemic uses of ought to was too low in Collins’s study to allow a comparison between speech and writing.) Collins (2009: 52) also notes that should is much more evenly distributed across speech and writing, but that “deontic should shows a stronger tendency to be associated with the written word (…) than does epistemic should.” This might be somewhat surprising, given that ought to has been shown to be in serious decline (e.g. Leech (2003))—one usually expects to encounter archaic linguistic items in the written rather than the spoken mode. But note that many remnants of older language phases which have long disappeared from the standard language can be preserved in dialects, which are by their nature also informal. Finally, it has been stated that “unlike should, ought to occurs mostly in positive statements, not in negative and interrogative sentences” (Harris (1986), Aarts and Wekker (1987: 193)). 2.4. Other Differences Proposed in the Literature Apart from the semantic differences mentioned in the previous subsections, it has been noted that should and ought to differ in a number of other respects. First, and most conspicuously, should is much more frequent in use than ought to. In the BNC, there are 111,237 occurrences of should but only 5,979 occurrences of ought to (cf. Kennedy (2002)), which amounts to a should/ought to ratio of 19:1. Similarly, in Collins’s (2009) study (involving different corpora), the ratio of should to ought to appears to be distinctions.indd 100-101 3. The Present Study: Data, Parameters, Methodology 3.1. Data For both modals, should and ought to, we made sure that at least 500 occurrences in context were extracted from corpora of British English. Half of the occurrences are from the spoken part of the BNC, which was searched with the accompanying SARA software. Oddly, the software did not allow us to select all the written texts in the BNC in any obvious way, which led to the decision to use another corpus, the Cobuild corpus, for which we selected all the written subcorpora with British English lan- 10.7.26 11:52:16 AM 102 Should vs. Ought to Bert Cappelle and Gert De Sutter guage material. A difference between the BNC and the Cobuild Corpus is that a search for should and ought in the BNC also yielded instances of shouldn’t, shoulda, oughtn’t and oughta, whereas these forms had to be looked up separately in Cobuild. We made sure that the number of examples of these forms extracted from Cobuild matched the proportion of the number of base forms to the total occurrences in the corpus (about 1% for should, about a third for ought). Table 1 provides an overview of the total number of hits for each search term and the number of hits actually extracted. Obviously, all hits we extracted for our investigation were chosen randomly. Should Shouldn’t Shoulda Ought Oughtn’t Oughta Spoken (BNC) Total Hits Extracted 12113 250 (included in above) (included in above) 1278 250 (included in above) (included in above) Written (Cobuild corpus) Total Hits Extracted 25816 250 896 9 7 0 719 250 11 4 7 2 Table 1: C orpus Finds and Number of Examples Extracted for All Search Terms It can be observed from the table that should is much more frequent than ought to in the corpora, with a 9:1 should/ought to ratio in the spoken corpus and a 36:1 should/ought to ratio in the written corpus. These frequencies confirm the general predominance of should and the relative informality of ought to mentioned in section 2.4. Removed from the data were sentences which were severely anacoluthic and non-interpretable, as well as sentences such as those in (9)-(13), containing should in a sense in which it simply could not be replaced by ought to: (9) a. I should like to ask Mick regarding the long service ambulance personnel. (BNC, spoken discourse) b. I should say it’s convenient for neither I should imagine. distinctions.indd 102-103 103 (BNC, spoken discourse) c. When’s dad coming back?—I don’t know yet, I should think it’ll be by the end of the month. (BNC, spoken discourse) d. I should sincerely hope not. (Cobuild corpus, Times newspaper) (10) a. Let me know if you need anything, or if someone close to you should suddenly die. (Cobuild corpus, UK books) b. Always try and make up a working prototype of your ideas should an inspection be requested. (Cobuild corpus, UK magazines) (11) … and I think it is absolutely critical that we should be devoting our attention er, to policing that catches criminals and prevents crime … (BNC, spoken discourse) (12) a. Yes, interesting you should say that, because the definition I had was something quite the reverse. (BNC, spoken discourse) b. It is so cruel that such a thing should happen to such a nice, normal family. (Cobuild corpus, Sun newspaper) c. It may be coincidence that you should choose to write to me now when what you dearly want is within reach. (Cobuild corpus, Sun newspaper) d. That Oxford and Cambridge should be capable of sustaining such a profile in the professional era is to the credit of those who run the university clubs. (Cobuild corpus, Times newspaper) (13) So the point of conclusion is if the myth is different in the Bible then the likely explanation is that it was tampered with, but that the scribes and the people who wrote the Bible altered the … and they changed it round why should they change this (BNC, spoken discourse) myth? The reason why such sentences were excluded is that we only wanted to find factors which favour the choice for should or ought to in sentences in which they both could in principle occur. In (9a-d), should is used as an 10.7.26 11:52:16 AM 104 Should vs. Ought to Bert Cappelle and Gert De Sutter alternative to would after the subject I. In (10a, b), should is used in a conditional clause to add a sense of tentativeness, i.e. to suggest that the speaker doesn’t think the actualisation of the situation is highly likely. In (11), should is used in a that-clause depending on an expression of necessity. In this sentence should alternates with the present (‘mandative’) subjunctive. However, we did include most of the sentences in which the modal occurs in a that-clause complementing an expression of suggestion, advice and even occasionally necessity, since we can observe that ought to is not wholly excluded from such contexts (although grammars do not record this usage): (14) a. Indeed by this time Mao is specifically advocating that the Party ought to take a more cautious and less radical approach to land reform so as to not antagonize the interests of the middle and rich peasants. (BNC, spoken discourse) b. It almost seems unfair to require that the computer ought to be able to do things that the programmer has not foreseen. (http://www.independent.co.uk/news/business/analysis-andfeatures/computers-with-a-mind-of-their-own-709785.html) In (12) above, should is used in content clauses occurring with expressions indicating a (typically emotion-fuelled) personal opinion. In such sentences an indicative form could generally be used instead of should, but by using should, the speaker indicates that he considers the mere idea of something as surprising or remarkable. A somewhat related use concerns should after why to suggest that the speaker can’t see any obvious reason for actualisation of the situation; cf. (13), where ought to would have been odd. Yet, most sentences with why were retained since, again, it can be found that ought to is not completely impossible in this context, in so far as failing to see a reason for something is often tantamount to failing to see a (deontic or epistemic) need or necessity for it: (15) a. There is absolutely no reason why film technicians ought to be expected to work for less than, say, a shelf stacker in a supermarket… (shootingpeople.org/poll/minimumwage/) distinctions.indd 104-105 105 b. I don’t see why it ought to be a high turnout election. (Corpus of Contemporary American English) The purification of our corpus examples resulted in an actual data set of 461 examples with should and 494 examples with ought to. As can be inferred from the above discussion, deciding whether an example should be expunged or retained was not always an easy matter. We have been as conservative as possible, thus leaving in perhaps more examples than other researchers would have done. All in all, we do not think this jeopardises the replicability of our study. 3.2. Parameters Investigated in This Study To the best of our knowledge, this paper investigates the potential impact of the largest number of parameters (i.e. variables) ever considered in the comparison of two modal verbs. These are listed below, grouped for convenience by a few headings (but different groupings are of course possible). Semantic parameters: —deontic vs. epistemic vs. merger —subjective (stance-taking) vs. objective (no stance-taking) —past-time reference (modality is situated in the past) vs. no pasttime reference Properties of the subject: —1st vs. 2nd vs. 3rd person —animate vs. inanimate —agentive (subject control over action) vs. non-agentive (lack of subject control) Properties of the verb following the modal: —dynamic vs. stative —present tense vs. perfect tense perfect: contracted (’ve, sometimes spelled of) vs. noncontracted —simple vs. progressive —active vs. passive 10.7.26 11:52:16 AM 106 Bert Cappelle and Gert De Sutter Syntactic properties of the clause: —code (use of modal to substitute entire VP, e.g. You should!) vs. no code —extraposition of the subject (e.g. It should be noted that…) vs. no extraposition —position of an adverb (if there is one) before (e.g. You really ought to…) vs. after the modal (You should really…) —clause type: embedded by think or a similar expression of cognition (assume, believe, feel, it seems to me, …) —clause type: embedded by suggest or a related item (advocate, advise, argue, consult, insist, legislate, recommend, request, require, persuade, stipulate, etc.; also nouns like suggestion/ proposal (that) and periphrastic constructions like think it wise (that) and it is appropriate/essential (that)) —polarity: positive vs. negative —negation: external (e.g. You shouldn’t be frightened = ‘It’s not necessary to be frightened’) vs. internal (e.g. This privilege should not be abused = ‘It is necessary not to abuse this privilege’) —negation: contracted (-n’t) vs. non-contracted —negation: negative raising (e.g. I don’t think you ought to say that) vs. otherwise —negation: negative subject (e.g. No marriage ought to be contracted on this day) vs. otherwise —negation: near-negative (e.g. We should only…) vs. otherwise —declarative vs. interrogative (including dependent questions, e.g. [boys who worry about] what moisturiser they should be using) —introduced by why (or a covert form, as in There’s no reason _ teenagers shouldn’t earn pocket money) vs. otherwise —subject-operator inversion vs. no subject-operator inversion —coordination of modals (e.g. What the trial judge can and should do…) vs. no coordination of modals distinctions.indd 106-107 Should vs. Ought to 107 —(only for ought to) do-support vs. absence of do-support —(only for ought to) infinitive marker to vs. reduced (oughta) vs. absent The quantitative results for these individual parameters are provided in the tables presented as addenda. Remember, however, that the primary aim of this study is to discover which factors have a unique contribution to the selection of should vs. ought to. Perhaps paradoxically, this can only be achieved by considering all the parameters together: only then can we know whether a parameter has a unique impact or whether it has to be disregarded because its effect correlates strongly to that of another parameter whose impact is stronger (and which should therefore be retained, unless it in turn correlates with a yet stronger parameter). Exactly how such a multivariate analysis proceeds is explained in the next subsection. One additional semantic parameter (‘counterfactuality’) was later on assembled on the basis of two of the above parameters: if the verb following the modal was either a full perfect auxiliary or the progressive auxiliary be, the modality was considered to convey counterfactuality (nonfulfilment); otherwise it was not. For a motivation of this operationalisation, see section 2.2, but see also section 5 for further discussion. The reason why only full perfect auxiliaries but not contracted ones were assigned the value ‘counterfactual’ was that we knew, at the time we assembled this variable, that contracted perfect auxiliaries occur significantly more frequently after should than after ought to, while we suspected, on the basis of the linguistic literature, that a suggestion of non-fulfilment occurs especially frequently with ought to. If we had included contracted auxiliaries in the set of ‘counterfactual’ examples, a possible impact of counterfactuality on the choice of should vs. ought to would therefore have been harder to detect, because it might have been cancelled out by the impact of contracted auxiliaries. The parameter ‘aspect’ (simple vs. progressive infinitive after the modal auxiliary) was removed from the analysis after assembling the parameter ‘counterfactuality,’ and the parameter ‘tense’ then acquired a purely morpho-phonological interpretation (presence or absence of a reduced perfect auxiliary). 10.7.26 11:52:16 AM 108 3.3. Methodology In order to find out which of the parameters mentioned above contributes significantly to the should/ought to variation in English, we fitted a so-called logistic regression model. This technique, which is very popular in sociolinguistic circles (cf. Goldvarb; Sankoff, Tagliamonte and Smith (2005)), is a multivariate analysis method that measures the simultaneous effect of one or more parameters, which may be categorical and continuous in nature, on a binary outcome (in this case should vs. ought to). By means of such a multivariate technique, significant variables can be distinguished from non-significant ones and the relative impact of each of the significant variables is estimated (i.e. the impact of parameter X while controlling for the effects of all other parameters included in the model). Finally, such techniques can yield insight into the descriptive, explanatory and predictive power of the resulting model (i.e., to what extent can the observed variation be described, explained and predicted by the explanatory variables). The statistical package R 2.7.0 (R Development Core Team (2008)) was used to conduct this statistical analysis. Before presenting the results of the logistic regression model, some remarks have to be made. First, for technical reasons, we reorganised parameters with more than two values, as well as parameters where one of the two values led to one or more further distinctions, to single-level, binary parameters. In doing so, we grouped those values together which had a similar effect on the should/ought to variation, or to put it somewhat differently, which maximised the difference between the values of the non-binary parameters. Thus, for the parameter polarity, we decided to regroup the values such that each sentence could be assigned either of two values: (i) negative polarity with the negation marked on the modal auxiliary (i.e. should not / ought not to / shouldn’t / oughtn’t to) or (ii) positive polarity or negative polarity which is not marked on the modal auxiliary. Second, we did not include any interaction effects in the model, since preparatory analyses pointed out that such a model is corrupted by multicollinearity. More particularly, the estimates of the interaction terms correlated heavily with the main terms, as a consequence of which the estimates became unstable. Third, some of the parameters mentioned in the distinctions.indd 108-109 Should vs. Ought to Bert Cappelle and Gert De Sutter 109 previous section had to be removed from the model, as preparatory analyses revealed that their effect correlated heavily with the effect of other parameters (yielding a new multicollinearity problem). In particular, the parameters animacy, agentivity, dynamicity, sentence type (declarative vs. interrogative), voice (active vs. passive) and presence of why were excluded for that reason. For instance, the parameter sentence type correlated with inversion, and the parameter voice (active vs. passive) correlated with agentivity, which correlated with meaning (deontic vs. epistemic or merger) and with animacy, which in turn also correlated with person. In choosing which of the correlating parameters should be retained, we opted for the parameters with the strongest effect on the variation at hand. This way, we ensured that the resulting model of the should/ought to variation only contained parameters with a unique impact on the variation and no parameters whose effect is redundant to that of another parameter. 4. Results Our statistical model, made up of 14 non-correlating parameters, fits the data well (p = 0.14). This means that there is no reason to assume that there is a discrepancy between the variation accounted for by the model and the actual (observed) variation. The model’s predictive power, however, is rather low (c = 0.69, a value closer to 0.5 (no predictive accuracy) than to 1 (full predictive accuracy)). Table 2 provides an overview of the retained parameters, ordered by individual strength (as expressed by the odds ratio). Starting from the base-line expectation that should and ought to have an a priori equal chance of being selected (i.e., controlling for the overall higher frequency of should compared to ought to noted in section 3.1), we can read the results in Table 2 as follows. First, in sentences with inversion between the subject and the operator (e.g. {Should we / Ought we to} do that?), should is more than twelve times as likely to be selected compared to when there is no inversion. This effect is highly significant. Second, when the word following the modal is a contracted perfect 10.7.26 11:52:16 AM 110 Should vs. Ought to Bert Cappelle and Gert De Sutter Parameter and Its Setting for Selection of Should Subject-operator inversion: yes Following word: ’ve Adverb: none or right after the auxiliary Polarity: verb-marked negative Modal proposition embedded by an item like suggest: yes Past-time reference: no Modal proposition embedded by a cognition item: no Person of the subject: third Meaning: epistemic or merger Counterfactual: no Subjective: yes Code: yes Extraposition: yes Coordination of modals: yes Odds Ratio Significance 12.6 10 3.9 2.9 2.7 2.2 2.1 1.9 n.a. n.a. n.a. n.a. n.a. n.a. *** *** * *** ** *** *** *** n.s. n.s. n.s. n.s. n.s. n.s. Table 2: S tatistical Model of the Variation between Should and Ought to, with Predictors for should, Odds Ratios and Significance Levels (‘***’: p < 0.001; ‘**’: p < 0.01; ‘*’: p < 0.05; ‘n.s.’: not significant); ‘n.a.’ = not applicable auxiliary (’ve, sometimes spelled of, as in You {should / ought to} {’ve / of} told me), the likelihood of that modal being should is ten times higher than when the following word is not a contracted perfect auxiliary. This effect is highly significant. Third, when there is no adverb or when it is positioned between the (verbal part of the) modal and the to-infinitive (as in You {should probably / ought probably to} ignore this), that modal is almost four times as likely to be should compared to when there is an adverb before the modal (as in You probably {should / ought to} ignore this), in which case of course the likelihood of that modal being ought to is more than four times as likely (compared to when there is no adverb or an adverb right before the toinfinitive). This effect is significant at the 0.05 level. Fourth, when the polarity is negative and expressed by means of not or -n’t following the (verbal part of the) modal (as in We {shouldn’t / oughtn’t to} do this), that modal is almost three times as likely to be distinctions.indd 110-111 111 should compared to when the modality is positive (as in We {should / ought to} do that) or negative without the negation being marked directly on the modal (as in I don’t think we {should / ought to} do that). This effect is highly significant. Fifth, when the modal proposition is a complement of suggest or a similar expression of suggestion, advisability, etc., should is almost three times as likely to be selected compared to when the modal proposition is not embedded by such an item. This effect is significant at the 0.01 level. Sixth, when there is no past time reference (as in Listen to me, you {should / ought to} get some tests done), should is more than twice as likely to be selected compared to when there is past-time reference, as is the case in the corpus sentence Then my father told me I ought to get some tests done and the polyps were diagnosed (Cobuild corpus, Today newspaper). This effect is highly significant. Seventh, when the modal proposition is not a complement of think or a similar expression of cognition, should is more than twice as likely to be selected compared to when the proposition is embedded by such an item (as in I think you {should / ought to} give it a try). This effect is highly significant. Eighth, when the subject of the modal proposition is a third person (as in He {should / ought to} apologise), should is almost twice as likely to be selected compared to when the subject is a first or a second person (as in {I / you / we} {should / ought to} apologise). This effect is highly significant. All other parameters are not significant—notably, this is the case for meaning (epistemicity/merger vs. deonticity) and for counterfactuality—or were left out of the model because they correlated with one or more of the other parameters. 5. Discussion Our model of the unique factors governing the choice between should and ought to allows us to describe this variation fairly parsimoniously in terms of eight significant factors and six more factors whose impact turns 10.7.26 11:52:17 AM 112 Bert Cappelle and Gert De Sutter out not to be significant. We have managed to rank the relevant factors according to their relative strength. This is a major descriptive advance with respect to previous studies. Our statistical model also allows us to explain some of the variation in terms of the intrinsic formal difference between should and ought to. The four strongest factors can all be linked to the fact that should is a standard modal auxiliary verb whereas ought to is, in fact, only a semi-modal verb, as appears from the fact that it (standardly) cannot take a bare infinitive. Ought to has therefore been described as a blend between a true modal and a full, lexical verb. As a concomitant of its hybrid status, we encounter it least frequently in structures where we do not normally find full verbs: inversion (e.g. Ought you to go?, cp. *Like you to go?), addition of a reduced auxiliary (e.g. We ought to’ve gone, cp. *We would have liked to’ve gone), insertion of an adverb (e.g. You ought probably to go, cp. *You like probably to go) and verb-marked negation (e.g. We ought not to go, cp. *We like not to go). That a contracted perfect auxiliary (’ve instead of have) is avoided strongly with ought to, as with other verbs taking a to-infinitive, is because to is itself phonologically weak and in that respect a dispreferred attachment site for another reduced item (compare also I won’t / I haven’t, where the reduced negator attaches to phonologically strong material, and *I’lln’t / *I’ven’t, where it does not). At first sight, there may be not much reason for being euphoric about the predictive accuracy of our statistical model. The C-value of our model is 0.69, but this is a correlation value, so one should not interpret this in terms of a fraction (i.e., it doesn’t mean that the model would manage to correctly guess the identity of ‘masked’ modals in our set in over two thirds of the cases). Note, indeed, that the C-value always gives a score between 0.5 and 1. A value of 0.5 would mean that there is no correlation at all between the actual choice between should and ought to in our set of examples and the choice of the modal auxiliary that the model would ‘predict’ on the basis of the specific values each example has for the factors included in the model. A value of 1 would mean that there is full correlation between actual and predicted choice of modal auxiliary. Clearly, while the C-value of our model is removed from the 0.5 bottom distinctions.indd 112-113 Should vs. Ought to 113 score, there is much room for improvement, and some claim that only from a 0.8 score onwards can one conclude that a statistical model has sufficient stability. The fact that our model has rather low predictive power can be interpreted in two different ways. One way is to refer to the received opinion about should and ought to, which is that they are highly synonymous (cf. section 1). Any model which yielded a C-value which is rather higher than 0.5 should therefore be considered as a triumph over the view that there is no clear distinction between should and ought to and that there is free variation between them. Perhaps a higher score can’t be reached, no matter how hard one looked for extra determining factors. Language allows for some variation, which means speakers can in some usage situations choose between multiple options. Even if one could find factors which heighten the likelihood of a particular option in given circum stances, there might always be cases where speakers act against the odds, for no apparent reason (cf. Cappelle (2009)). Another view is that the fairly low C-value suggests that not all relevant factors have been found and/or that the factors may not have been optimally operationalised. For instance, we might refine our operationalisation of the factor non-fulfilment (cf. section 2.2) since a perfect infinitive only suggests nonactualisation if the use of the modal is deontic. As Declerck (1991: 411) rightly points out, a sentence like (16) is ambiguous: (16) It’s two o’clock. The train should have crossed the border by now. (after Declerck (1991: 411)) This sentence could be interpreted as ‘Given the time, I expect that the train has crossed the border,’ with the perfect infinitive expressing anteriority to the time of speech. However, it could also be read as ‘Given the time, it’s not normal that the train hasn’t crossed the border yet (damn this unreliable railway company).’ Even with deontic uses, there need not be an implication of nonactualisation (Declerck (1991: 379); cf. also Huddleston and Pullum (2002: 204)): (17) To be eligible for the post you should have worked for the firm 10.7.26 11:52:17 AM 114 for at least ten years. (Declerck (1991: 379)) Similarly, a progressive infinitive does not suggest nonactualisation if the use of the modal is epistemic. Consider (17): (18) The students should be studying hard now. If should is epistemic, the interpretation is something like ‘(given the time of year,) we can only conclude that the students are studying hard now,’ with no suggestion that this is not the case (although the speaker is more tentative in his conclusion than if he had used must). Only if should is used deontically do we (automatically) get the interpretation that the students are not, in fact, studying hard now. In the light of these complications, we might have to admit that our operationalisation of the potential factor ‘suggestion of non-fulfilment’ was not very adequate and that a better operationalisation might have yielded better results. After all, there is some evidence that ought might indeed be preferred if the speaker does not want to imply that the (deontic) necessity actually has to be fulfilled. Consider an utterance like (19): (19) Whoever perpetuated the bare-belly look ought to be hanged. (BNC, written discourse) Here, there is no suggestion that the subject should actually be prosecuted and punished with death by hanging. At most, the author suggests that in an ‘ideal’ world where one could inflict horrible punishments at will and on a whim, this would be the course of action he’d suggest—all the while acknowledging that this is not something one could realistically ask for in the real world. In the Corpus of Contemporary American English, we searched for the modals should and ought to followed by be and a past participle. From the list of past participles, we selected 10 verbs referring to situations whose likelihood of realisation is low or nonexistent and 10 verbs referring to situations that are more likely to be realised. The results are given in Table 3. The distribution is highly significant (χ2 = 44.4, d.f. = 1, p < 0.001). distinctions.indd 114-115 115 Should vs. Ought to Bert Cappelle and Gert De Sutter Unlikely Situations (fired / shot / killed / castrated / euthanized / exiled / horsewhipped / hanged / bombed / crucified) Likely Situations (considered / treated / called / encouraged / regarded / pointed [incl. pointed out] / interpreted / recognized / emphasized / avoided) Should Be… Ought to Be… 213 32 3318 135 Table 3: D istribution of Should Be and Ought to Be before Past Participles Referring to Likely and Unlikely Situations in the Corpus of Contemporary American English This remarkable difference between should and ought to might also be due in part by a difference in formality between the verbs involved—remember that ought to has a higher likelihood of occurrence in informal than in formal discourse. Furthermore, it may be suspected that the first group of verbs would come up more naturally in discourse in which the speaker expressed his personal views than in discourse where the speaker is less personally (and emotionally) involved. This brings us to the issue of subjectivity vs. objectivity, which Declerck (1991) and others have related the choice of should and ought to, the latter being claimed to be more objective than the former (cf. section 2.1). In the remainder of this section, we will discuss this parameter in some detail. We are not convinced that should is more subjective than ought to. In fact, we see three counter-indications, dealt with in the next three subsections. A more fundamental problem of the subjectivity/objectivity distinction is then discussed in 5.4. 5.1. Corpus Evidence against Ought to as Being “More Objective” Our corpus contained few instances in which the speaker expresses an ‘objective’ opinion (as in (1a) above)—and besides, whether an opinion counts as truly objective is always hard to verify—or in which the speaker just plainly states which requirements ‘objectively’ have to be fulfilled for 10.7.26 11:52:17 AM 116 Should vs. Ought to Bert Cappelle and Gert De Sutter some other situation (as in (1b) above). At any rate, we did not find corroboration that ought to occurs more frequently than should in such cases. However, if ought to were really more objective than should, then it should be avoided in contexts containing hedges like I think or If you ask me, but sentences like the following do occur, seem perfectly natural and have been shown in this study to be indeed more frequent than similar sentences with should (cf. the seventh result stated in section 4). (20) a. If you ask me, though, it ought to be twice that size. (www.deadline.com/hollywood/mr-rogers-gone-but-notforgotten/) b. “I think this woman ought to be replaced immediately by myself,” he said. (Cobuild corpus, The Times newspaper) c. …what’s been er been going through my head recently is, is er the, looking at the pattern of the meetings and the way the meetings are arranged and, and how, erm, at the last meeting we had a speaker er and that I think, we all found that quite interesting and the one, one from Central America that things and I feel we ought to have that much more frequently than we do have er, a, either a speaker or a focus of some sort of meetings erm, so I think that’s something I’d like to raise and get the A G M at the next meeting I think [a] similar thing we ought to consider there… (BNC, spoken discourse) Moreover, our study also reveals that ought to (vs. should) is chosen twice as likely with first and second person subjects, which refer to the author and his addressee, as with third person subjects (cf. our eighth result). This suggests that ought to is more (inter)subjective than should. 5.2. Should is Not Necessarily Speaker-grounded It is not hard to find examples similar to those which Declerck (1991) claims should sound odd: (21) a. I should probably apologise… But I won’t. distinctions.indd 116-117 117 (losingit.me.uk/2009/08/21/i-should-probably-apologise) b. I know I should shut up now, but I won’t. (https://lists.ubuntu.com/archives/ubuntu.../000324.html) One might try to save the subjective value of should by proposing that should more generally refers to someone’s subjective view, whether this view can be attributed to the speaker/writer or to someone else. It could indeed be argued that in (21a, b) should is not used to give advice to oneself but to represent an obligation or advice which arises outside the speaker/writer—these sentences are in this respect not altogether different from cases of ‘distancing indirect speech/thought’; cf. Vandelanotte (this volume and the references presented there). For instance, (21a) might be seen as a shortened version of ‘Some people would probably give me the advice that I should apologise…’ and (21b) might be equivalent to ‘I know that you think I should shut up now….’ However, if the necessity expressed by should happens to be speaker/ writer-external, it still need not be a subjective necessity. Consider (22): (22) I should be fasting for Ramadan but I won’t today—maybe tomorrow. (www.telegraph.co.uk/.../How-money-stopped-mattering-forDragons-Den-mogul-James-Caan.html) Again, the necessity in this sentence isn’t grounded in the writer. The sentence wouldn’t differ noticeably in its interpretation from “According to the Koran, I should be fasting for Ramadan....” But a course of action prescribed by the Koran can hardly be considered as a necessity based on a subjective view. 5.3. When No Stance Is Taken, Ought to Cannot Be Used For ‘objective’ requirements as can be found in official letters, instruction booklets, public notices and so on, it is actually should, but not ought to, which can be used (cf. Close (1981: 120-122), Westney (1995: 172)): (23) a. Applications should be submitted by March 31st at the latest. (Close (1981: 120)) 10.7.26 11:52:17 AM 118 Bert Cappelle and Gert De Sutter b. Entries should be clearly typed. (Gailor (1983: 348)) c. Carbon knives should be wiped clean after use. (from a written instruction) (Westney (1995: 166)) d. This coffee-machine should be turned off after use. (from a printed public notice) (Westney (1995: 172)) It may be the case that degree of formality plays a role here, too (cf. section 4). Yet, the relatively high formality of these utterances cannot be invoked as the sole reason why ought to would be odd here, since this modal is otherwise definitely not excluded from writing. Still another parameter, namely implication of fulfilment vs. non-fulfilment, discussed above and in section 3.2, might be at work here. Take, for instance, (23a): it would be very awkward to state that applications must be submitted by a specified date while implying that this requirement might not be met. Whichever additional factors may or may not play a role, though, the above examples prove wrong Swan’s (1980: 550) and Declerck’s (1991: 377, fn. 21) claim that ought to is the modal to be used for talking about laws, duties and regulations. We believe that ought to is in fact excluded in decrees, official rules, guidelines, etc., which do not involve stance-taking, that is, where the utterances do not express someone’s personal opinions (whether it be the speaker/writer’s or someone else’s) but merely form statements of procedures to be followed. 5.4. Other Issues with Subjectivity vs. Objectivity Both should and ought to are claimed by Declerck (1991: 377-378) to “merely express a moral obligation (rather than an objective necessity)” (our emphasis). But if should and ought to already differ from each other in terms of whether the obligation expressed is based on a subjective opinion or something more objective, how can one then go on to claim that should and ought to as a pair differ from other deontic modals, such as must and have (got) to, in exactly those terms? We are forced to conclude that the distinction ‘subjective’ vs. ‘objective’ is too coarse to differentiate all deontic modals. (We are leaving epistemic modals out of the distinctions.indd 118-119 Should vs. Ought to 119 picture here, since Declerck (1991) doesn’t discuss the difference between epistemic should or ought to on the one hand and must on the other in terms of subjectivity but in terms of tentativeness.) If taken as a cline, subjectivity might be more adequate (as in Degani (2009: 334)). This is, in fact, how Declerck apparently looks at this parameter (cf. “ought to is more objective”; our emphasis). But then subjectivity also becomes rather impractical as a variable when it comes to judging individual examples: on what basis can we (objectively) decide for any two instances whether they are equally subjective/objective or whether one of them is more subjective/objective than the other? This problem does not apply to Collins’s claim quoted in section 2.1 (“the proportion of objective cases is higher with ought to”), from which we can infer that he doesn’t consider subjectivity itself to be scalar. Rather, he allows deontic modals to be ranked according to this binary criterion, on the basis of how many instances of each modal qualify as ‘subjective.’ On the other hand, with a two-way distinction, there may be many cases where there is doubt whether the example should be classified as subjective or objective, and where a middle category might therefore have been helpful. The parameter ‘modality embedded by a cognition expression’ which we used in our study might perhaps be considered the most objective approximation variable (‘proxy’) of subjectivity and it is also in accordance with the desideratum that other modals can be ranked with respect to it. 6. Some Further Observations In this section we will focus on some interesting examples that we came across when coding the corpus data. These examples have no immediate bearing on the main findings of this study but are worth discussing in a study on should and ought to. 6.1. Didn’t as Tag after Ought to The following example is curious in that it shows that ought to cannot only be felt to be more like a full verb than a modal auxiliary, hence the 10.7.26 11:52:17 AM 120 Should vs. Ought to Bert Cappelle and Gert De Sutter use of do-support, but also in that it appears to be thought of as a past tense form—which, historically, it is: (24) Right! Your school ju, your school jumper ought to be cleaned (BNC, spoken discourse) didn’t it? This example might perhaps more naturally be explained as a case of free indirect speech, i.e. with an understood They said or You said, in which case ought to can be interpreted as having past time reference. As our study revealed, past time reference is a frequent option with ought to. 6.2. Should + Have to Unlike ought to, should is occasionally combined with deontic have to, a possibility noted by Harris (1986) (cf. also Kennedy (2002)). We have found a single instance in our corpus: (25) We should only have to look as far back as a season or at the most a couple. (Cobuild corpus, Sun newspaper) This apparently redundant addition of have to is probably most common with Why should…, where the basic deontic interpretation of should is bleached (cf. the discussion of (13) in section 3.1). In the BNC, 27 hits are returned for the sequence “why should * have to.” 6.3. E pistemic Should / Ought to and Situations Expressing a Positive Idea A little-observed property of epistemic should and ought to, which they do not share with epistemic must, is that they can seemingly only be used for situations which are felt to be positive rather than negative (cf. Quirk et al. (1985: 227), Westney (1995: 179)). This property has intrigued Renaat Declerck, to whom the following examples are due: (26) a. It {should / ought to} be {easy / *difficult} to solve that problem.—That problem {shouldn’t / oughtn’t} to be {*easy / difficult} to solve. b. It must be {easy / difficult} to solve that problem.—That distinctions.indd 120-121 121 problem can’t be {easy / difficult} to solve. We cannot go into the hypothesis Declerck has offered in a working document to explain this contrast. Instead, let us consider the following extended corpus extract, which at first sight would seem to constitute a counterexample: (27) “Not surprisingly, against this background, consumer and business confidence is still deteriorating,” the Oxford group said. Oxford said that interest rates in both Europe and America should fall further, predicting 2.75 per cent rates in Germany by the spring and 4.75 per cent in the US by the summer. Most other countries would follow suit. Lower interest rates would bolster growth, it added. (Cobuild corpus, Times newspaper) Note, however, that the last sentence reveals that the situation of falling further is not necessarily bad. 7. Conclusions This corpus-based study on should and ought to in contemporary spoken and written British English usage has shown that these two near-synonymous modals are not identical in use. The data confirmed previous observations that should is used much more frequently than ought to and that the frequency of ought to in spoken language is higher than its frequency in written language. Out of many more investigated parameters, eight were shown to exert a unique and significant impact on the choice of should vs. ought to, in order of decreasing strength: (i) inversion, (ii) a following contracted perfect infinitive, (iii) a following adverb (or none), (iv) negation, (v) embedding by suggest or a similar item, (vi) reference to the non-past, (vii) no embedding by think or a similar cognition expression and (viii) third (vs. first or second) person subject. The four strongest parameters can be linked to the fact that should is a member of the set of core modal auxiliaries while ought to is, morpho-syntactically speaking, only a so-called semi-modal or 10.7.26 11:52:18 AM 122 quasi-modal. While ought to does have the NICE properties (allowing negation, inversion, code and emphasis without do-support), it tends to be avoided precisely in environments where it has to display auxiliary-like properties. We have also provided evidence, pace Declerck (1991) and some other scholars, against statements that should is more subjective than ought to. If anything, the contrary appears to be the case. Meaning Subjectivity Time reference Deontic Epistemic Merger Subjective Objective Non-past Past Ought to Written Spoken 200 (79%) 225 (94%) 35 (14%) 9 (4%) 19 (8%) 6 (3%) 240 (95%) 236 (98%) 14 (6%) 4 (2%) 173 (68%) 224 (93%) 81 (32%) 16 (7%) Table A1: Distribution of Should and Ought to by Semantic Factors Person Animacy Agentivity Should Written Spoken 1st 21 (9%) 58 (26%) 2nd 21 (9%) 42 (19%) 3rd 200 (83%) 114 (52%) Animate 145 (60%) 149 (68%) Inanimate 97 (40%) 70 (32%) Agentive 112 (46%) 103 (47%) Nonagentive 130 (54%) 116 (53%) Ought to Written Spoken 49 (19%) 95 (40%) 23 (9%) 45 (19%) 182 (72%) 96 (40%) 156 (61%) 198 (83%) 98 (39%) 40 (17%) 102 (40%) 148 (62%) 152 (60%) 92 (38%) Table A2: D istribution of Should and Ought to by Properties of the Subject distinctions.indd 122-123 Dynamicity Tense Dynamic Stative Present Perfect Written Should Spoken Voice Full Simple Progressive Active Passive Written Ought to Spoken 166 (69%) 122 (56%) 152 (60%) 158 (66%) 76 (32%) 97 (44%) 102 (40%) 82 (34%) 219 (91%) 197 (90%) 219 (86%) 226 (94%) 23 (10%) 22 (10%) 35 (14%) 14 (6%) Contracted 1 (0%) Aspect Appendices Should Written Spoken 195 (81%) 183 (84%) 34 (14%) 15 (7%) 13 (5%) 21 (10%) 227 (94%) 206 (94%) 15 (6%) 13 (6%) 216 (89%) 197 (90%) 26 (11%) 22 (10%) 123 Should vs. Ought to Bert Cappelle and Gert De Sutter 22 (9%) 8 (4%) 14 (6%) 0 (0%) 2 (1%) 35 (14%) 12 (5%) 232 (97%) 203 (93%) 240 (94%) 220 (92%) 10 (4%) 16 (7%) 14 (6%) 20 (8%) 194 (81%) 197 (90%) 211 (83%) 221 (92%) 48 (20%) 22 (10%) 43 (17%) 19 (8%) Table A3: D istribution of Should and Ought to by Properties of the Infinitive Code Extraposed Subject Position of Adverb Yes No Yes No Before modal After modal Embedded by Yes think-clause No Embedded by Yes suggest-clause No Polarity Positive Written Should Spoken Written Ought to Spoken 3 (1%) 9 (4%) 5 (2%) 8 (3%) 7 (3%) 5 (2%) 5 (2%) 8 (3%) 3 (1%) 2 (1%) 8 (3%) 13 (5%) 239 (100%) 210 (96%) 249 (98%) 232 (97%) 235 (98%) 214 (98%) 249 (98%) 232 (97%) 20 (8%) 5 (2%) 12 (5%) 12 (5%) 3 (1%) 32 (15%) 38 (15%) 54 (23%) 237 (99%) 187 (85%) 216 (85%) 186 (78%) 17 (7%) 16 (7%) 4 (2%) 9 (4%) 225 (94%) 203 (93%) 250 (98%) 231 (96%) 216 (90%) 183 (84%) 236 (93%) 228 (95%) 10.7.26 11:52:18 AM 124 Negative Internal External Contracted Raising Neg. subject Near-neg. Sentence type Declarative Interrogative Introduced by Yes why No Inversion Coordination of modals Do-support Infinitive marker Should vs. Ought to Bert Cappelle and Gert De Sutter Yes No Yes No Yes No to Reduced Absent 26 24 2 8 0 2 2 (11%) (10%) (1%) (3%) (0%) (1%) (1%) 36 31 5 21 5 2 1 (16%) (14%) (2%) (10%) (2%) (1%) (0%) 18 14 4 4 2 4 0 (7%) (6%) (2%) (2%) (1%) (2%) (0%) 12 (5%) 11 (5%) 1 (0%) 1 (0%) 5 (2%) 0 (0%) 0 (0%) 211 (88%) 199 (91%) 8 (3%) 224 (93%) 8 (3%) 7 (3%) 0 (0%) 1 (0%) 12 (5%) 12 (5%) 2 (1%) 0 (0%) 4 (2%) 0 (0%) 4 (2%) 0 (0%) 0 (0%) 1 (0%) 31 (13%) 20 (9%) 246 (97%) 16 (7%) 234 (98%) 212 (97%) 254 (100%) 239 (100%) 230 (96%) 207 (95%) 252 (99%) 240 (100%) 238 (99%) 219 (100%) 250 (98%) 240 (100%) n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 254 (100%) 239 (100%) 249 (98%) 236 (98%) 2 (1%) 3 (1%) 1 (0%) 3 (1%) Table A4: D istribution of Should and Ought to by Syntactic Properties of the Clause (Note: ‘n.a.’ = not applicable) References Aarts, Flor and Herman Wekker (1987) A Contrastive Grammar of English and Dutch, Martinus Nijhoff, Leiden. Cappelle, Bert (2009) “Can We Factor Out Free Choice?” Describing and Modeling Variation in Grammar, ed. by Dufter, Andreas, Jürg Fleischer and Guido Seiler, 183-202, Mouton de Gruyter, Berlin and New York. Celce-Murcia, Marianne and Diane Larsen-Freeman (1983) The Grammar Book: An ESL/EFL Teacher’s Course, Newbury House, Rowley, MA. Close, Reginald A. (1981) English as a Foreign Language: Its Constant distinctions.indd 124-125 125 Grammatical Problems, 3rd ed., George Allen & Unwin, London. Coates, Jennifer (1983) The Semantics of the Modal Auxiliaries, Croom Helm, London and Canberra. Collins, Peter (2009) Modals and Quasi-Modals in English, Rodopi, Amsterdam and New York. Declerck, Renaat (1991) A Comprehensive Descriptive Grammar of English, Kaitakusha, Tokyo. Degani, Marta (2009) “Re-analysing the Semi-Modal Ought to: An Investigation of Its Use in the LOB, FLOB, Brown and Frown Corpora,” Corpus Linguistics: Refinements and Reassessments, ed. by Andrew Kehoe, 327-346, Rodopi, Amsterdam. Gailor, Denis (1983) “Reflections on Should, Ought to, and Must,” English Language Teaching Journal 37, 346-349. Harris, Martin (1986) “English Ought (to),” Linguistics across Historical and Geographical Boundaries: In Honour of Jacek Fisiak on the Occasion of His Fiftieth Birthday. Volume 1. Linguistic Theory and Historical Linguistics, ed. by Dieter Kastovski and Aleksander Szwedek, 345-358, Mouton de Gruyter, Berlin, New York and Amsterdam. Huddleston, Rodney and Geoffrey K. Pullum (2002) The Cambridge Grammar of the English Language, Cambridge University Press, Cambridge. Kennedy, Graeme (2002) “Variation in the Distribution of Modal Verbs in the British National Corpus,” Using Corpora to Explore Linguistic Variation, ed. by Randi Reppen, Susan M. Fitzmaurice and Douglas Biber, 73-90, John Benjamins, Amsterdam and Philadelphia. Leech, Geoffrey (2003) “Modality on the Move: The English Modal Auxiliaries 1961-1992,” Directions in Corpus Linguistics. Proceedings of the Nobel Symposium 82, Stockholm, 4-8 August 1991, ed. by Roberta Facchinetti, Manfred Krug and Frank Palmer, 223-240, Mouton de Gruyter, Berlin and New York. Myhill, John (1996) “Should and Ought: The Rise of Individually Oriented Modality in American English,” English Language and Linguistics 1, 3-23. Palmer, Frank Robert (1987) The English Verb, 2nd ed., Longman, London and New York. Palmer, Frank Robert (1990) Modality and the English Modals, 2nd ed., Longman, London and New York. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik (1985) A Comprehensive Grammar of the English language, Longman, London. Perkins, Michael R. (1983) Modal Expressions in English, Pinter, London. 10.7.26 11:52:18 AM 126 Bert Cappelle and Gert De Sutter R Development Core Team (2008) R: A Language and Environment for Statistical Computing, Version 2.7.0, Vienna, R Foundation for Statistical Computing. Sankoff, David, Sali A. Tagliamonte and Eric Smith (2005) “Goldvarb X: A Variable Rule Application for Macintosh and Windows,” Department of Linguistics, University of Toronto, Toronto. Simpson, John, chief ed. (2000) Oxford English Dictionary Online, Oxford University Press, Oxford. Swan, Michael (1980) Practical English Usage, Oxford University Press, Oxford. Swan, Michael (1995) Practical English Usage, 2nd ed., Oxford University Press, Oxford. Visser, Fredericus Theodorus (1963) An Historical Syntax of the English Language. Part Three. First half. Syntactical Units with Two Verbs, Brill, Leiden. Westney Paul (1995) Modals and Periphrastics in English: An Investigation into the Semantic Correspondence between Certain English Modal Verbs and Their Periphrastic Equivalents, Niemeyer, Tübingen. [Modals] The English Comparative Modals—A Pilot Study* Johan van der Auwera and Astrid De Wit University of Antwerp 1. Introduction English has a number of modal constructions which could be called ‘comparative.’ The core examples of these comparative modals contain an auxiliary and an adverbial of comparison. There are three subtypes, depending on whether the adverb is a superlative (1a), a comparative (1b) or an expression of equality (1c). Within each subtype, one can distinguish further subtypes, depending on the type of auxiliary or the type of adverbial. (1) a. b. c. had best, ’d best had better,’d better, better would rather, ’d rather, had rather, should rather would sooner, ’d sooner, had sooner, should sooner would (just) as soon as may (just) as well might (just) as well * This work was done within the ‘Grammaticalisation and (Inter)Subjectivity’ project (Belgian Federal Government—Interuniversity Attraction Poles P6/44). Special thanks are due to Folker Debusscher, Mariela Gonzalez Gomez, Benita Lašinytė, Adinda Robberechts, Daniel Van Olmen, and Dirk Noël (the latter for making the penultimate version of this paper take account of Collins (2009)). 127 distinctions.indd 126-127 10.7.26 11:52:18 AM [...]... in a study on should and ought to 6.1. Didn’t as Tag after Ought to The following example is curious in that it shows that ought to cannot only be felt to be more like a full verb than a modal auxiliary, hence the 10.7.26 11:52:17 AM 120 Should vs Ought to Bert Cappelle and Gert De Sutter use of do-support, but also in that it appears to be thought of as a past tense form—which, historically, it is:... objective, how can one then go on to claim that should and ought to as a pair differ from other deontic modals, such as must and have (got) to, in exactly those terms? We are forced to conclude that the distinction ‘subjective’ vs ‘objective’ is too coarse to differentiate all deontic modals. (We are leaving epistemic modals out of the distinctions.indd 118-119 Should vs Ought to 119 picture here, since... reduced auxiliary (e.g We ought to ve gone, cp *We would have liked to ve gone), insertion of an adverb (e.g You ought probably to go, cp *You like probably to go) and verb-marked negation (e.g We ought not to go, cp *We like not to go). That a contracted perfect auxiliary (’ve instead of have) is avoided strongly with ought to, as with other verbs taking a to- infinitive, is because to is itself phonologically... Sutter out not to be significant. We have managed to rank the relevant factors according to their relative strength. This is a major descriptive advance with respect to previous studies Our statistical model also allows us to explain some of the variation in terms of the intrinsic formal difference between should and ought to. The four strongest factors can all be linked to the fact that should is a standard... school jumper ought to be cleaned (BNC, spoken discourse) didn’t it? This example might perhaps more naturally be explained as a case of free indirect speech, i.e with an understood They said or You said, in which case ought to can be interpreted as having past time reference. As our study revealed, past time reference is a frequent option with ought to 6.2. Should + Have to Unlike ought to, should is... has intrigued Renaat Declerck, to whom the following examples are due: (26) a It {should / ought to} be {easy / *difficult} to solve that problem.—That problem {shouldn’t / oughtn’t} to be {*easy / difficult} to solve b It must be {easy / difficult} to solve that problem.—That distinctions.indd 120-121 121 problem can’t be {easy / difficult} to solve We cannot go into the hypothesis Declerck has... (losingit.me.uk/2009/08/21/i -should- probably-apologise) b I know I should shut up now, but I won’t (https://lists.ubuntu.com/archives/ubuntu /000324.html) One might try to save the subjective value of should by proposing that should more generally refers to someone’s subjective view, whether this view can be attributed to the speaker/writer or to someone else. It could indeed be argued that in (21a, b) should is not used to. .. distinctions.indd 112-113 Should vs Ought to 113 score, there is much room for improvement, and some claim that only from a 0.8 score onwards can one conclude that a statistical model has sufficient stability The fact that our model has rather low predictive power can be interpreted in two different ways. One way is to refer to the received opinion about should and ought to, which is that they are... speaker/writer’s or someone else’s) but merely form statements of procedures to be followed 5.4. Other Issues with Subjectivity vs Objectivity Both should and ought to are claimed by Declerck (1991: 377-378) to “merely express a moral obligation (rather than an objective necessity)” (our emphasis). But if should and ought to already differ from each other in terms of whether the obligation expressed... emotionally) involved This brings us to the issue of subjectivity vs objectivity, which Declerck (1991) and others have related the choice of should and ought to, the latter being claimed to be more objective than the former (cf section 2.1). In the remainder of this section, we will discuss this parameter in some detail. We are not convinced that should is more subjective than ought to. In fact, we see three