Not only apples but also oranges: Contrastive negation and register

Olli O. Silvennoinen
University of Helsinki

Abstract

1. Introduction

Big data has in recent years become a buzzword in fields ranging from marketing and public policy to academic research. Along with complementary terms such as thick data and rich data, it is invoked with increasing regularity even in fields that have traditionally relied on qualitative or small-scale quantitative data, such as syntax and semantics. However, what is big depends on the context. For linguistic features that are easily retrievable from corpora with high precision and recall, a big data set may reach a size similar to that of big data in the natural sciences (e.g., Grieve et al. 2015), although this naturally raises the issue of whether such a dataset can accurately represent the target population in an ecologically meaningful way (cf. Biber 1993): traditionally, big correlates with messy in corpus design (see Mair 2006 for discussion). On the other hand, for features that require a lot of work by the researcher that is both time-consuming and error-prone, the question may not be “what is big?” but “what is big enough?” This is the case of constructions that consist of polysemous items that are put together in an emergent fashion (see, e.g., Hopper 2011).

This paper considers a case of the latter type, i.e., contrastive negation in English. To do this, I use an approximately 3.3-million-word subset of the British National Corpus (BNC) consisting of newspaper discourse as well as conversation. Contrastive negation refers to expressions in which one element is negated and another one is presented as its alternative. The family of negative-contrastive constructions is large. One basic syntactic distinction cuts between negative-first and negative-second variants (cf. Tottie 1991), exemplified in (1) and (2) respectively: [1]

The basic syntactic facts about these constructions have been discussed by Gates & Seright (1967), McCawley (1991) and Huddleston & Pullum (2002: 811–812, 1313–1314). Syntactically, the constructions are bipartite and the two parts are parallel. Semantically, the two elements that are contrasted are part of the same domain and replaceable by one another. Gates & Seright have argued that the constructions are headed by the affirmative element, to which the negated one is subordinate. [3] The negated element is redundant and therefore it can be omitted without a change in the truth-conditional meaning:

The semantic redundancy means that the constructions are pragmatically motivated. The constructions may correct an idea that is active in the common ground of discourse participants, setting up an alternative mental space in the discourse (see Dancygier 2012).

The best-known account of the pragmatics of contrastive negation is given in Horn’s (1985; 1989) work on metalinguistic negation. Although McCawley (1991) has shown that only some tokens of contrastive negation are metalinguistic, the term has dominated much of the discussion around negative-contrastive constructions. In addition, the inventory of negative-contrastive constructions is seldom considered as a whole, which may stem from the fact that few studies have attempted to describe the phenomenon using corpus data, most of the research being based on intuitive examples or convenience samples, i.e., tokens collected unsystematically by the authors. This is in conflict with the results of corpus linguistics, which suggest that grammatical choices are sensitive to or even dependent on register (e.g., Kerz & Wiechmann 2015; Biber 2012; for an overview, see Biber et al. 1999). The corpus studies that do exist (Tottie 1991; Anderwald 2002; Pitts 2011) target negation as a whole rather than focusing on contrastive negation in particular. These will be discussed in section 2.

My study asks (i) what the prevalence and distribution of the various negative-contrastive constructions are, and (ii) how they can be explained. The prevalence and distribution are investigated in terms of register, which I define, following Biber (2009: 823), as “a cover term for any language variety defined by its situational characteristics, including the speaker’s purpose, the relationship between speaker and hearer, and the production circumstances”. My study is couched in a tradition in which grammatical choices have been shown to vary according to register in ways that are systematic, meaningful and non-trivial. Linguistic choices are made to conform to register expectations, and at the same time they become entrenched so that a genre may become instantly recognisable through a marked linguistic choice (e.g., Günthner 2010). My study examines what Dorgeloh & Wanner (2010: 4) call “constructional variation”, i.e., the prevalence and distribution of competing, functionally equivalent constructions, which I define as entrenched pairings of form and function (Goldberg 2006). At the same time, I assume that there are probably subtle functional differences between the competing forms (e.g., Bolinger 1977; Goldberg 1995; see also Tottie 1991: 90–96 for discussion). In fact, I believe that studying the variation of these forms is essential in explaining their functional differentiation.

Comparison is an inherent part of any variationist study. However, it is far from obvious which registers should be compared to one another. On the one hand, there are studies that describe linguistic variation on the level of macro-registers. An example of this is the Longman Grammar of Spoken and Written English (Biber et al. 1999), which has four register categories: conversation, fiction, newspapers and academic prose. On the other hand, some studies opt for a more micro approach, comparing closely related registers. For instance, Hiltunen (2010) studied a set of grammatical constructions in research articles in four fields: medicine, physics, law and literary criticism. The first approach compares apples with oranges: the objects of comparison are maximally different, which means that the variation is large, and the internal variability of the registers is seldom considered. The second approach compares apples with apples: the objects of comparison are similar, which means that the variation is likely to be more constrained, but the internal variability of a macro-register comes to the fore.

This study aims to combine the two approaches making both apples-to-oranges and apples-to-apples comparisons. To do this, I employ both corpus linguistics and interactional linguistics. I also tackle contrastive negation considering a larger set of cases than previous research. In addition, I use a more fine-grained syntactic classification of the category than has been used in previous research.

The structure of the paper is as follows: In section 2, I present previous research on negative constructions in English. Section 3 explains the data and methods used. In section 4, I give the results of the corpus analysis. Section 5 concludes the paper.

2. Contrastive negation

There have been a number of large-scale corpus analyses of negation in English as well as other languages. However, contrastive negation receives only sporadic attention in them. Anderwald’s (2002) focus, for instance, is on sentence negation in regional dialects of spoken British English, and contrastive negation is not treated as a separate topic. By contrast, Tottie (1991: 161–170) does treat what she terms “contrastive constructions” as a topic in its own right. Her study focuses on the difference between not-negation and no-negation. The distinction is exemplified in the following (Tottie 1991: 87–88):

According to Tottie’s statistical analysis, contrastive constructions are a reliable trigger of not-negation. She found that contrastive negation is rather rare, accounting for 5% of the negations in her spoken corpus (14 cases) and 4% in her written corpus (21 cases). Since negation is overall twice more common in speech, the numbers would indicate that contrastive negation is more prevalent in speech, too. In her data, negative-first constructions are the more common option in writing while in speech negative-second variants predominate. However, since her data set is much smaller than the one employed here, the number of cases is rather low. [4] In addition, Tottie’s definition of contrastive negation is broader than mine, including also tokens that do not fall under the definition of contrast employed here.

Another study worth mentioning is by Pitts (2011), who studied the forms not and n’t in the spoken part of the Great Britain component of the International Corpus of English (ICE-GB), which contains over 600,000 words. However, Pitts’s focus is pragmatic rather than syntactic: she considers the target of the negation and does not analyse her data in terms of the syntactic constructions that are present in it. In addition, even though her corpus is stratified by register, she only presents raw frequencies of her findings. All in all, we know surprisingly little about the variation of negative-contrastive constructions.

To some, the term “contrastive negation” might seem tautological. The general view is that all negative statements evoke their positive counterparts (Givón 1978; Horn 1989; Verhagen 2005; Dancygier 2012). If we accept this view, all negation that is not derived morphologically (e.g., unhappy, disallow) could be regarded as contrastive with a concomitant positive. [5] In this study, however, contrast is defined as a relation between explicitly mentioned discourse elements. The reasons for adopting this view is that there is no one-to-one correspondence between positive and negative propositions. If my t-shirt is green, it is not yellow, but it also is not red, black or white, and it might not be made of linen in any of these cases. For a statement like My t-shirt is not yellow to be informative, it needs to have strong contextual support (see Horn 1989 for discussion).

Furthermore, this study does not consider just any negative that comes with an attendant positive. My aim is to find grammaticalised ways of combining negation and explicit contrast – negative-contrastive constructions. Contrast means that there are two alternatives to fill the same space in a semantic frame, be that space a proposition, a referent or a designation. Therefore, elements are contrastive if and only if they are about the same thing: they are foci in the same topical background (see Lambrecht 1994). Consider:

In this example, the negation that is highlighted sets up a discourse space: not really complaining is in an oppositional relationship to the ensuing list of events. However, they are not contrastive under the definition adopted here since they are not properly speaking alternatives to one another. The focus of the negation is the verb complain, which does not receive a replacement in the ensuing discourse: there is no focus that would be contrastive with it (Lambrecht 1994). Rather, the following text accounts for the negation by providing a reason for expressing it (Ford 2002).

In many works, the not X but Y construction is taken to be the archetypal form of contrastive negation. According to McCawley, the not X but Y construction “contrasts two ways of filling a syntactic position, one that (according to what the sentence says) results in a false proposition and one that results in a true proposition” (McCawley 1991: 189). This is a syntax-driven definition that leads him to enumerate the following as “forms of contrastive negation” (McCawley 1991: 190):

Forms from (a) to (c) are termed “short” while (d) and (e) are “expanded”. There is a broad synonymy between the forms, as evidenced by (8) and (9), which are both drawn from the corpus and are likely to report the same speech act but exhibit slightly different constructional forms:

Example (8) exhibits the canonical not X but Y construction, while (9) merely juxtaposes two main clauses. That these two clauses are seen as one unit is probably the result of inferencing based on cues such as prosody, discourse context and world knowledge (see Laury & Ono 2014) in addition to the formal and functional parallelism.

In spite of its tidiness – or perhaps because of it – McCawley’s classification leaks: there are forms which readily suggest themselves as synonyms with his list but which he does not mention, such as those in (10) and (11):

Semantically, the contrasts effected by negative-contrastive constructions have three main types. In replacive contrasts, one element is rejected and another is offered in its stead. This represents the purest form of contrastive negation.

In additive contrasts, no element is rejected; what is rejected is the construal in which only one of them is offered. Additive constructions are recognised on the basis of a restrictive element such as only in the scope of the negation:

In restrictive contrasts, no element is rejected in full, just as in additive contrasts. They differ in that in restrictive contrasts the speaker offers something less instead of the rejected element. There are two main ways of expressing restrictive contrast. One is a negative-first construction that employs a focusing expression in the affirmative part (Dik et al. 1981) while the other is a negative-second construction that includes an adversative but (Huddleston & Pullum 2002: 1313). [6]

Replacive contrast can be seen as the basic type on which additive and restrictive contrasts are based, for several reasons. First, it is structurally the simplest, literally “unmarked” by such additions as only or adversative but. Second, it is occasionally polysemous with the other construction types, while they are not polysemous with one another. For instance in (16), a replacive construction expresses a meaning that is available to restrictively contrastive constructions:

Indeed, in many previous studies (e.g., McCawley 1991), only replacive constructions have been counted as contrastive negation, but some have made allowances for additives (Quirk et al. 1985; Tottie 1991), some for restrictives (Gates Jr. & Seright 1967; Huddleston & Pullum 2002) or even both (Dik et al. 1981). I have chosen to work with a wider notion of contrastive negation because of the formal and functional parallels of the different types.

3. Data and methods

What kind of data is needed for a usage-based study of contrastive negation? Data can be big or small, thick or thin, rich or poor, good or bad. The discussion on the merits of big and thick data in particular seems to recycle familiar arguments about quantitative and qualitative methods. I aim to overcome these distinctions by bringing together two research traditions: corpus linguistics and interactional linguistics. In my view, both are needed because of the nature of the data sets as well as the phenomenon being investigated.

To see how contrastive negation is distributed I explored the use of negation in the British National Corpus, using the BNCweb interface hosted by Lancaster University. [7] My aim was to keep the procedure as open-ended as possible to maximise recall. For this reason, I only searched for the analytic as well as synthetic absolute negators in English (Huddleston & Pullum 2002: 788). In other words, I searched for the words not (including n’t and the first n in dunno), no, none, nobody, no one (including no-one and noone), nothing, nowhere. [8] The search was later supplemented with the colloquial forms nope, nah, nay, nowt, naught and nae (subsumed under not, as it only appears in the sequence cannae in the spoken data). [9] Since this obviously returns too many hits for feasible analysis, the search was restricted to two subsets of the corpus, one representing written language and the other spoken language.

For written language, I relied on the national broadsheet newspaper component of the corpus. This component is based on Lee’s (2001) genre classification and consists of circa 3,000,000 words, divided into eight categories that roughly correspond to the usual sections in a newspaper. Newspaper discourse was selected as data for several reasons. First, many corpus-linguistic studies use newspaper discourse as a proxy for written language in general, making it possible and meaningful to compare my results to them. Second, the subdivision of newspaper discourse reflects many of the parameters identified in multi-dimensional register analysis (Biber 1989; 2009); for instance the category Editorial includes argumentative texts while Sports is mostly narrative. Therefore, the structure of the subcorpus allows me to investigate the variation caused by communicative function while keeping other factors such as the level of formality constant. Third, despite their internal register variation, broadsheet newspapers as a whole make for an ecologically valid category: a newspaper is (or was at the time that the BNC represents, the early 1990s) generally published as a whole comprising all the sections that form the subcorpus. The details of the broadsheet newspapers subcorpus are presented in Table 1. [10]

For spoken language, I made a search similar to the one made on the newspaper data in the demographically sampled component of the spoken BNC. To make it feasible in terms of time spent on data collection, I only looked at a small part of this data set, i.e., those files that have been published as the sampler of Audio BNC (Coleman et al. 2012), which comprises around 270,000 words and over 25 hours of conversation. [11] The data set is meant to be as ecologically valid as possible, and contextual information is taken into account in the analysis of every token. The details of the spoken corpus files are in Table 2.

The spoken data is fairly diverse in terms of speakers’ sociolinguistic properties such as age, gender, social class and domicile. However, the data set is too restricted for these categories to be used in the analysis, so this is not attempted here. The diversity of the data is meant to reduce the effect of potential bias caused by regional and social dialects as well as idiolectal preferences. To my knowledge, however, contrastive negation has never been identified as a dialect feature anywhere on the British Isles. Rather, dialectal variation in English negation concerns such issues as negative concord and the regularisation of verbal morphology (see Anderwald 2002).

To identify and annotate the relevant tokens, the search results were exported to a spreadsheet programme with a four-sentence window for context. This was necessary because sometimes the constructions extend over several sentences, as in (17):

The written results were analysed by checking the concordance lines, identifying the relevant cases and annotating them. By contrast, the spoken material was listened to in its entirety using version 2.0.6 of Audacity® (Audacity Team 2014) to not only see but hear the whole contexts of the uses of the negators. This was necessary since it was expected that the constructions are emergent (e.g., Hopper 2011) and therefore cannot be identified using the written transcription only. Ultimately this stems from seeing spoken language on a par with – or even having primacy over – written language, which means that it needs to be studied on its own terms (see Couper-Kuhlen & Selting 2001). From this perspective, the transcription is secondary – the actual data is the recordings. From a more practical point of view, contrastive negation in spoken English is almost impossible to study using only concordance lines. An illustrative example is (18), which is the concordance line for part of extract (19). Only by listening to the extract are we in a position to identify the negative-contrastive constructions in this segment:

The extract contains three negative-contrastive constructs, labelled (a), (b) and (c). They are uttered in partial overlap as the interlocutors try to find what car model is the correct one. In line [184], L denies that the car in question is a Fiat Uno using an expanded negative-first construction, which prompts C to contest this in line [186]. L initiates another such construction in line [187] but this one is completed by R in line [189]. Starting in line [191], C then utters a third expanded negative-first construction, which overlaps with L’s turn in line [192]. The extract demonstrates that in order to produce quantitative generalisations on this topic, both thick and rich data are needed. In methodological terms, corpus linguistics needs to be complemented with interactional linguistics and its holistic focus on communicative situations.

What kind of data is this? Big data has been defined as data sets that are too large for traditional software to handle (Manyika et al. 2011: 1). While my study obviously does not reach this threshold, the number of tokens studied is larger than has been customary in studies on grammar, particularly previous studies on contrastive negation. Thick data refers to the ethnographic detail behind the numbers crunched in big data studies (Wang 2013). The term is most applicable to my study when it comes to the analysis of speech, which often required paying close attention to the surrounding co-text as well as prosody. Rich data, referring to a data set that aims to combine breadth and depth, comes to the fore in the analysis of the tokens found. My data is rich: each token comes with an extensive set of metadata as well as annotations added by the researcher. I would argue that only such a combination of data sets and methodologies makes it possible to study contrastive negation across modes and registers.

4. Results

This section reviews the results of the corpus analysis. Section 4.1 covers the overall patterns in the data: the prevalence of contrastive negation in each of the subcorpora, and the proportion of not-negation and no-negation in negative-contrastive constructions. Section 4.2 presents a more fine-grained analysis of the different construction types. [12]

4.1 Contrastive negation in the data

The search for negators yielded 33,689 tokens. [13] From this raw data set, a total of 2,886 negative-contrastive constructs were extracted manually. These constructs are the data set proper in this study. In these 2,886 constructs, a total of 2,996 negators were used. The discrepancy is due to contexts in which several negators participate in the creation of one contrast. This may happen on purpose (as in example 20):

The co-occurrence of multiple negators may also be due to a false start or other performance phenomena (as in 21):

When there were two or more negators, one of them was chosen as primary. Since my main focus is on constructions rather than negators, I shall mostly use the lower total number, i.e., the number of negative-contrastive constructs that is 2,886. Table 3 displays the prevalence of contrastive negation in the two modalities, divided according to semantic type, i.e., whether the tokens are replacive, additive or restrictive. The table includes first the raw frequencies and then frequencies normalised to 1,000 words (in parentheses); this is the number to which frequencies will be standardised in the remainder of this paper as well. Normalised frequencies are not reported for the two data sets combined since they are uneven in size and because they do not form an ecologically valid whole as such.

Following Tottie’s (1991) numbers, it was predicted that contrastive negation should be more frequent in speech than in writing. The table shows that this is indeed the case: contrastive negation is more than one and a half times as frequent in speech as in writing. The difference between the overall prevalence of contrastive negation in speech and writing is very highly statistically significant (χ² ≈ 55.6, d.f. = 1, p < .001). This means that the null hypothesis that there is no difference in the prevalence of contrastive negation in speech and writing is rejected. However, this is almost exclusively due to replacives, which make up the majority of negative-contrastive constructs in the spoken data. In writing, the situation is not as dramatic although even there replacives clearly outnumber both additives and restrictives. This supports the idea that replacives are the central case around which the category of contrastive negation is built.

Table 4 presents the written language data divided into the eight sub-registers. Again, the table gives both raw and normalised frequencies.

The table shows that broadsheet newspapers are not homogenous when it comes to the semantic types of contrastive negation: the differences among the eight sub-registers are highly statistically significant (χ² ≈ 34.11, d.f. = 14; p < .01). In other words, we can reject the null hypothesis according to which there is no difference in the prevalence of the three semantic types of contrastive negation in sub-registers of the broadsheet newspapers sub-corpus. The Editorial sub-corpus in particular stands out: when normalised frequencies are considered, the prevalence of contrastive negation in Editorial exceeds even the level found in conversation. At the other end of the spectrum we find Commerce and Sports, which have low levels of all kinds of contrast. The range of values is largest in replacives: the difference between the register with the highest and the lowest normalised frequencies of replacives is 1.13 per 1,000 words, compared with 0.22 for additives and 0.17 for restrictives.

I now move on to the status of contrastive negation in the overall grammar of English negation. Since Tottie has claimed that contrastive negation is associated with not-negation, I looked into this. Table 5 shows the proportion of primary contrastive instances of not versus the other negators, which together form Tottie’s category of no-negation. Note that the table includes all negators that participate in negative-contrastive constructions, not only the primary ones.

According to the table, contrastive negation is slightly more likely to be formed using not-negation in speech than in writing. Although statistically significant (χ² ≈ 10.35, d.f. = 1, p < .01), the difference is slight and some sub-registers of writing come close to the proportion of contrastive not-negation that is found in spoken language. The numbers for not-negation exceed the overall proportion of not-negation in these data sets, suggesting that Tottie’s generalisation holds true. The distribution of not-negation and no-negation in negative-contrastive constructions was found to be different from the overall pattern of all negative constructions at a level that is very highly statistically significant in both speech (χ² ≈ 89.2, d.f. = 1, p < .001) and writing (χ² ≈ 486.4, d.f. = 1, p < .001). Therefore, the hypothesis that contrastive negation is associated with not-negation is supported by the data.

The table also shows that registers differ in how large a proportion of all negations are contrastive. Conversation has a particularly low percentage of contrastive negation among all negations even though negative-contrastive constructions are more common in speech than in writing. In practice, this shows just how much more common negation is overall in spoken language. In writing, around one tenth of negations are contrastive. This is a marked difference to Tottie’s study, in which only 4% of the negations in her written data were contrastive. This may be due to the differences in the data sets of the studies: in addition to newspaper writing, Tottie’s data includes, for instance, popular lore, biographies and scientific writings (see note 4), which are narrative or expository registers and therefore likely to contain fewer instances of contrastive negation.

4.2 Forms of contrastive negation in the data

The previous section explored the frequency of contrastive negation on a general level. In this section, I zoom in on the constructional forms. First I discuss the replacives, then the additives, and finally the restrictives, before summing up with some general tendencies concerning the semantic types and the constructional forms that they take.

4.2.1 Replacives

The data on replacives has been divided into constructions based on what grammatical rank the contrasting elements have (i.e., phrases and words for simple forms, matrix clauses for expanded ones) and how the affirmative and negative parts are connected (i.e., by a conjunction, by another word or with no overt marking). Examples (22)–(27) illustrate the negative-first constructions:

The not X but Y construction is marked by the presence of contrastive but. [14] Its asyndetic variant is not X, Y. This construction is always simple; if a construction includes two complete matrix clauses, it is classified as an expanded negative-first construction. There is also a middle form of these two, which I have labelled the prefaced negative-first: it has a phrasal negative as a preface constituent, followed by a clausal affirmative. I have also included the not so much X as/but Y construction in the list since it sometimes comes with a contrastive but, suggesting a high degree semantic similarity with not X but Y. Finally there are certain negative-first forms that were difficult to categorise.

Examples (28)–(31) illustrate the negative-second constructions:

As to the negative-second forms, the canonical one seems to be the asyndetic X not Y. It has a close relationship with the overtly coordinated X and not Y. The X but not Y construction is formally related to the two but its meaning is more aptly described as restrictive rather than replacive. The data also includes expanded negative-second forms as well as forms that are not easy to place in any of the other categories.

My data also includes a few examples that have a tripartite structure, possibly due to lack of planning or attention on the part of the speakers.

Figure 1 displays the orderings of replacive constructions in the two data sets. [15] Table 6 shows a more detailed breakdown of the constructional forms of replacive contrastive negation in speech and writing.

 

Table 3 showed that speech has more contrastive negation than writing but this difference was seen to be virtually the product of replacives alone. Table 6 demonstrates that the difference is actually created by one construction type: expanded negative-first constructions. Interestingly, this construction is much more frequent than not X but Y, which is generally taken to be canonical. As the variant without an adverbial is more prevalent, it seems that spoken language can manage to create a negative-contrastive construction with rather little syntactic work. An example is seen in extract (33):

Perhaps surprisingly in light of previous research, the expanded negative-first construction is by far the most common way to create contrastive negation even in writing. The not X but Y formula, previously regarded as canonical, is much more frequent in the written data than speech, but its frequency is no match to that of the expanded negative-first construction.

In Tottie’s data, negative-second was more frequent in speech, while writing was dominated by negative-first. In my data, only the latter generalisation holds true since both speech and writing are heavily biased towards negative-first forms. In fact, due to the extreme position of the expanded negative-first construction, my speech data is more likely to have negative-first forms than writing.

4.2.2 Additives

Additives are much less frequent than replacives, and their classification is simpler. Negative-first forms of additives include not only X but (also) Y, which has a corrective but; its asyndetic variant not only X, Y, which does not; and an expanded as well as a prefaced form. Not only X but (also) Y is much more relaxed about the syntactic rank of its fillers than the replacive not X but Y, as example (34) shows:

Negative-second forms of additives are also akin to those of replacives. The simplest one is X not only Y, which is similar to an overtly coordinated variant, X and not only Y. In addition to these simple forms, there is also the possibility of forming an expanded negative-second additive. My data does not include tripartite forms of additives.

Figure 2 and Table 7 show the breakdown of additives in my data:

 

Additives pattern very differently from replacives. In speech, additives are on the whole very rare, and the only form that has any kind of sustained presence in my data is the expanded negative-first. In writing the construction that includes a corrective but is by far the most common, although the expanded negative-first form also appears quite regularly. Negative-second forms are very rare in both speech and writing.

4.2.3 Restrictives

Restrictives are much rarer than replacives but more frequent than additives. The category of restrictives also mirrors the replacives in terms of form. The negative-first forms include a form with a corrective but (not X but just Y), an asyndetic sub-clausal coordination without but (not X, just Y) and an expanded as well as a prefaced construction without but. There are also other, more marginal forms:

The negative-second forms of restrictives include the X but not Y construction as well as forms that mirror those of the other semantic classes: the asyndetic just X not Y, its close cousin just X and not Y:

There are also tripartite forms of restrictives:

Figure 3 and Table 8 show the distribution of restrictives in the data:

 

Again, speech and writing differ markedly from one another. In writing, the X but not Y construction is by far the most prevalent: indeed, restrictives in writing are the only category where negative-second constructions outnumber negative-first constructions. By contrast, in speech, the expanded negative-first construction dominates the scene: the vast majority of restrictives in my spoken data have this form.

4.2.4 Summary

We have seen above that there are many ways of expressing the three kinds of contrastive negation in English. In this section, I draw attention to the parallel forms and their functional differentiation. I focus on five forms whose replacive correlates are: (i) not X but Y; (ii) not X, Y; (iii) expanded negative-first; (iv) X not Y; and (v) expanded negative-second. The way in which these forms are distributed among the three semantic classes in the written corpus is shown in Figure 4:

The figure shows that the forms are partially specialised for different semantic classes: not X but Y shows an equal proportion of replacives and additives while not X, Y is strongly biased towards restrictives. X not Y has restrictive semantics almost exclusively although this is partially the result of not counting X but not Y among its forms. The results are statistically very highly significant (χ² ≈ 827.6, d.f. = 8, p < .001) though they should be approached with caution since two of the cells contained fewer than five tokens.

Figure 5 shows the corresponding data for speech:

Because of the high number of cells with values below five, statistical testing was not conducted. The numbers do indicate that restrictives tend to appear in the extended negative-first form. However, this form is even more frequent with replacives. This suggests that different principles are at play when a negative-contrastive construction is chosen in the different modes of communicating.

5. Discussion

In this article, I set out to investigate whether and to what extent contrastive negation is a register-sensitive grammatical phenomenon. My strategy was two-pronged. On the one hand, I have surveyed the variation within one register, broadsheet newspaper writing. This part of my study uncovered substantial differences in the prevalence of contrastive negation in closely related sub-registers, which suggests that the constructions studied here vary according to communicative dimensions that are probably not dissimilar to those uncovered by Biber (1989; 2009). On the other hand, I compared newspaper writing to a rather distinct register in terms of both production circumstances and the purpose of communication, namely casual conversation between people who usually know each other very well. This revealed that the strategies used to create contrastive negation are very different indeed in the two registers. The findings of this part of my investigation support Biber’s (2012: 25) contention that register is “arguably the most important predictor of grammatical differences”.

In combining apples and oranges, the road from data to evidence required employing two methodologies – corpus linguistics and interactional linguistics. It also included a large dose of sifting through raw data. This inevitably meant that I had to make compromises as to the amount of data that I have used. The question of what is big data in grammar depends on the aspect of structure being researched. For a phenomenon that is not readily found using simple corpus queries, even hundreds of thousands of words may be big.

I hope that the findings of this study will also be of use in more theoretically oriented studies of negation. One of the most important functions of negation is to manipulate information that is taken to be in the common ground (cf. Givón 1978). Contrastive negation does this explicitly by not only identifying the part of the common ground that is in need of repairing but also by supplying something that can go in its stead. From this point of view, it is interesting that the negative prototypically precedes the affirmative in negative-contrastive constructions. This property would follow from the mental spaces account of negation according to which negative sentences create two mental spaces while the corresponding affirmative only creates one (Verhagen 2005; Sweetser 2006; Dancygier 2012). In order for a negative sentence to be used cooperatively, it needs to be predicable at least in theory – there needs to be a corresponding state of affairs in the affirmative for the sentence to be relevant. [16] Therefore, negation is a projective construction type, i.e., one that creates room for upcoming talk or text (Ford 2002; Auer 2005). By contrast, the affirmative sentence is unmarked for the contrast and for the possible projection, which means that when the subsequent negative arrives in affirmative+negative clause combinations, it forces a reanalysis of the affirmative. [17] This causes an additional processing cost to the hearer, thus violating the Gricean maxim of manner, specifically the sub-maxim “be orderly” (Grice 1975: 46; see also Hawkins 2003: 144–147). For this reason, a negative sentence followed by an affirmative one is cooperative, but an affirmative sentence followed by a negative one is less so.

Another explanation, which is not necessarily incompatible with the previous one, is negative-first, the general propensity for negation to come early in a sentence (Jespersen 1917). Failure to understand that a piece of linguistic content is negated is catastrophic to successful communication, and there is a strong functional motivation to make the negation of something that is assumed to be in the common ground available to the hearer as early as possible. A negative-first construction achieves just that, while also showing the way forward.

Acknowledgements

This work was funded by HELSLANG, the Doctoral Programme for Language Studies at the University of Helsinki, and Langnet, the Finnish Graduate School for Language Studies, which I gratefully acknowledge. The ideas in this paper were significantly improved by discussions with the audience at the d2e conference. I also wish to thank Matti Miestamo, Minna Palander-Collin, the referees and the editors, whose comments on earlier drafts of this paper developed both my thinking on these issues and its expression. None of these people should, of course, be blamed for any of its shortcomings.

Notes

[1] Tottie uses the terms neg-first and neg-second. This is somewhat unfortunate since neg-first already has an established meaning in linguistics, referring to the tendency of negative elements to be placed before the element that they negate and, more generally, earlier rather than later in a sentence. The idea, though not the name, goes back to Jespersen (1917). McCawley (1991) employs the notions “basic form” and “reverse form” to refer to my negative-first and negative-second; I have rejected these, too, since I do not endorse the transformational relationship between not X but Y and X not Y that he postulates. [Go back up]

[2] All BNC examples in this paper have a source references that contains the code of the text (in this case A2J), the number of the s-unit (i.e., sentence unit, in this case 184) and the genre code as per Lee’s (2001) classification (in this case W:newsp:brdsht_nat:editorial, which means that the s-unit is in the written component of the corpus, more specifically in the Editorial sub-corpus of the national broadsheet newspaper category). In the conversation extracts in this paper, the interlocutors are referred to by their initials to protect their anonymity. [Go back up]

[3] In this paper, I call the two parts of the constructions “affirmative” and “negative”, for the sake of convenience. However, it is possible, though rare, for negation to be contrastive with another negative element. For instance, the first negation in both (i) and (ii) is analysed here as contrastive:

1. Not only will it not be like its parents, it probably won’t want to be. (A7Y, 26, W:newsp:brdsht_nat:social)
2. Yet he did not oppose all toxic waste movement (after all, the same union represents workers at Rechem itself); he simply didn’t like the idea that big industrial nations like Canada could dump their dirty problems on this country and his members. (A1U, 151, W:newsp:brdsht_nat:misc)

Why is only the first negation contrastive? In (i), contrastive negation is outside the two clauses that are juxtaposed, both of which happen to be negative in this case. In (ii), the judgement is based on the adverb simply, which functions as a persuasive element that presents the writer’s point of view as truth (Aarts 1996). This makes it compatible with “affirmative” rather than the “negative” element in more prototypical examples of contrastive negation. As indicated by its position before the verb, the adverb is not under the scope of the negation, which is why this example is classified as restrictive, not additive (see sections 2 and 4.2). [Go back up]

[4] Tottie’s spoken material is a 170,000-word subset of the London-Lund Corpus of Spoken English (LLC), while her written material comes from the Lancaster-Oslo/Bergen Corpus of British English (LOB); the total number of words of her written corpus is not specified but it consists of “texts from categories A, B, C, (press: reportage, editorial, and reviews, respectively), F (popular lore, e.g., magazines), G (belles lettres, biography, essays), and J (learned and scientific writings)” (Tottie 1991: 106). Since this is only a small part of the LOB, which itself is only one million words, the total number of words in Tottie’s two corpora is unlikely to be much over 500,000. [Go back up]

[5] In English linguistics, words such as unhappy and disallow are known as affixal negation (e.g., Horn 1989; Tottie 1991; Zimmer 1964). This term is unfortunate, however, since in a number of languages the standard negative construction is an inflectional affix added to the verb (Dryer 2013). [Go back up]

[6] The distinction between replacives and restrictives is connected to the type of elements being contrasted. Elements that are mutually exclusive alternatives require a replacive construction (Huddleston & Pullum 2002: 1313), as in (iii):

3. a. They died in 1984, not 1983.
   b. *They died in 1984 but not 1983.

A central question relates to sentences such as M doesn’t have (just) three kids – she has four, in which the default (descriptive) interpretation of the negation has to give way to a secondary (metalinguistic) one. The main reference is Horn (1985; 1989), whose work on metalinguistic negation sparked a long discussion that is still on-going. [Go back up]

[7] See http://corpora.lancs.ac.uk/BNCweb/ (accessed on 29 June 2015). The main sources of documentation for BNCweb are Berglund et al. (2002) and Hoffmann et al. (2008). [Go back up]

[8] Like Tottie, I exclude inherent negatives (such words as absent, fail and lack), functional negatives (You must be joking) and incomplete/fuzzy negatives (such as few and seldom). Unlike Tottie, I exclude affixal negatives, i.e., negative prefixes (such as un- and dis-) and suffixes (such as -less). It is unclear if the excluded forms are even capable of contrastive readings, at least productively. A possible counter-example is Renault’s slogan Stop watching, start living. [Go back up]

[9] Nae is originally a feature of Scottish English, which has spread to Irish English (Anderwald 2002: 47, 53–56). It has properties of both not and no. Nowt and naught are similar to nothing. I thank one of the anonymous reviewers for pointing out that nowt is a feature of the Yorkshire dialect (Petyt 1985: 239); Huddleston & Pullum (2002: 812) claim that the form is “archaic”. In my data, nowt mainly appears in the file KCX, which records the speech of mainly middle-aged women who were living in Yorkshire in the early 1990s. [Go back up]

[10] A downside of this data set is the fact that approximately one third of the material is classified as “miscellaneous”, findings related to which are very difficult if not impossible to explain functionally. [Go back up]

[11] The files used here are given in http://www.phon.ox.ac.uk/SpokenBNCcontents (accessed on 19 January 2016). [Go back up]

[12] In what follows, the significance of the quantitative results reported here has been tested using the χ² test. Though widely used, it is not optimal for testing the differences in the frequencies of linguistic items due to the assumptions underlying it (Kilgarriff 2005; Lijffijt et al. 2016; Säily 2014: 45–46). For this reason, the results of the significance testing should be seen as only indicative. [Go back up]

[13] The query also yielded some false positives. Notably, no was used as a shorthand for number in some cases. These are not included in the number of hits reported above. [Go back up]

[14] Not as the most frequent negator stands for all kinds of negation here. The same goes for the words just and only when restrictives are part of the construction in additives and restrictives. [Go back up]

[15] Note that figures 1–4 are scaled differently. [Go back up]

[16] This is in line with the fact that the lexical fillers of negative-contrastive constructions need to be coherent. Thus the negated information is not suppressed, as is often assumed, but remains at least somewhat activated. See Giora et al. (2005) for discussion and experimental support. [Go back up]

[17] That negation is here seen as a marker of projection should not be understood as deterministic; in fact, Auer (2005: 17) notes that negation only exhibits weak projectivity in the denial + account sequences identified by Ford (e.g., 2002), of which negative-contrastive constructions of the [negative clause + affirmative clause] type are a subset. [Go back up]

References

Aarts, Bas. 1996. “The rhetorical adverb simply in present-day British English”. Synchronic Corpus Linguistics: Papers from the Sixteenth International Conference on English Language Research on Computerized Corpora (ICAME 16), ed. by Carol Percy, Charles F. Meyer & Ian Lancashire, 59–68. Amsterdam & Atlanta: Rodopi.

Anderwald, Lieselotte. 2002. Negation in Non-Standard British English: Gaps, Regularizations and Asymmetries. London & New York: Routledge.

Audacity Team. 2014. Audacity®: Free Audio Editor and Recorder. Version 2.0.6. http://audacity.sourceforge.net/

Auer, Peter. 2005. “Projection in interaction and projection in grammar”. Text - Interdisciplinary Journal for the Study of Discourse 25(1): 7–36. doi:10.1515/text.2005.25.1.7

Berglund, Ylva, Sebastian Hoffmann, David Lee & Nicholas Smith. 2002. BNCweb Manual. http://rdues.bcu.ac.uk/bncweb/manual/bncwebman-home.htm

Biber, Douglas. 1989. “A Typology of English Texts”. Linguistics 27(1): 3–43. doi:10.1515/ling.1989.27.1.3

Biber, Douglas. 1993. “Representativeness in corpus design”. Literary and Linguistic Computing 8(4): 243–257. doi:10.1093/llc/8.4.243

Biber, Douglas. 2009. “Multi-dimensional approaches”. Corpus Linguistics: An International Handbook, ed. by Anke Lüdeling and Merja Kytö, 822–855. Berlin: Walter de Gruyter. doi:10.1515/9783110213881.2.822

Biber, Douglas. 2012. “Register as a predictor of linguistic variation”. Corpus Linguistics and Linguistic Theory 8(1): 9–37. doi:10.1515/cllt-2012-0002

Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman Grammar of Spoken and Written English. Harlow: Longman.

Bolinger, Dwight. 1977. Meaning and Form. London & New York: Longman.

Coleman, John, Ladan Baghai-Ravary, John Pybus & Sergio Grau. 2012. Audio BNC: The Audio Edition of the Spoken British National Corpus. Phonetics Laboratory, University of Oxford. http://www.phon.ox.ac.uk/AudioBNC

Couper-Kuhlen, Elizabeth & Margret Selting. 2001. “Introducing interactional linguistics”. Studies in Interactional Linguistics, ed. by Margret Selting and Elizabeth Couper-Kuhlen, 1–22. Amsterdam & Philadelphia: John Benjamins. doi:10.1075/sidag.10.02cou

Dancygier, Barbara. 2012. “Negation, stance verbs, and intersubjectivity”. Viewpoint in Language: A Multimodal Perspective, ed. by Barbara Dancygier and Eve Sweetser, 69–96. Cambridge: Cambridge University Press. doi:10.1017/CBO9781139084727.006

Dik, Simon, Maria E. Hoffmann, Jan R. de Jong, Sie Ing Djiang, Harry Stroomer & Lourens de Vries. 1981. “On the typology of focus phenomena”. Perspectives on Functional Grammar, ed. by Teun Hoekstra, Harry van der Hulst & Michael Moortgat, 41–74. Dordrecht: Foris Publications.

Dorgeloh, Heidrun & Anja Wanner. 2010. “Introduction”. Syntactic Variation and Genre, ed. by Heidrun Dorgeloh & Anja Wanner, 1–26. Berlin & New York: Mouton de Gruyter. doi:10.1515/9783110226485.0.1

Dryer, Matthew S. 2013. “Negative morphemes”. The World Atlas of Language Structures Online, ed. by Matthew S. Dryer and Martin Haspelmath. Leipzig: Max Planck Institute for Evolutionary Anthropology. http://wals.info/chapter/112

Ford, Cecilia E. 2002. “Denial and the construction of conversational turns”. Complex Sentences in Grammar and Discourse: Essays in Honour of Sandra A. Thompson, ed. by Joan L. Bybee and Michael Noonan, 61–78. Amsterdam: John Benjamins. doi:10.1075/z.110.05for

Gates Jr., Dave L. & Orin Dale Seright. 1967. “Negative-contrastive constructions in standard modern English”. American Speech 42(2): 136–141. doi:10.2307/453965

Giora, Rachel, Noga Balaban, Ofer Fein & Inbar Alkabets. 2005. “Explicit negation as positivity in disguise”. Figurative Language Comprehension: Social and Cultural Influences, ed. by Herbert L. Colston and Albert N. Katz, 233–258. Hillsdale, NJ: Erlbaum.

Givón, Talmy. 1978. “Negation in language: Pragmatics, function, ontology”. Syntax and Semantics 9: Pragmatics, ed. by Peter Cole, 69–112. New York: Academic Press.

Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press.

Goldberg, Adele E. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press.

Grice, H.P. 1975. “Logic and conversation”. Syntax and Semantics 3: Speech Acts, ed. by Peter Cole and Jerry L. Morgan, 41–58. New York: Academic Press.

Grieve, Jack, Andrea Nini, Diasheng Guo & Alice Kasakoff. 2015. “Big data for the analysis of language variation and change”. Paper presented at the From Data to Evidence: Big Data, Rich Data, Uncharted Data (d2e) conference, University of Helsinki, 19–22 October 2015.

Günthner, Susanne. 2010. “Grammatical constructions and communicative genres”. Syntactic Variation and Genre, ed. by Heidrun Dorgeloh & Anja Wanner, 195–217. Berlin & New York: Mouton de Gruyter. doi:10.1515/9783110226485.2.195

Hawkins, John A. 2003. “Efficiency and complexity in grammars: Three general principles”. The Nature of Explanation in Linguistic Theory, ed. by John Moore and Maria Polinsky, 121–152. Stanford: CSLI Publications.

Hiltunen, Turo. 2010. Grammar and disciplinary culture: A corpus-based study. PhD dissertation, University of Helsinki. http://urn.fi/URN:ISBN:978-952-10-6464-7

Hoffmann, Sebastian, Stefan Evert, Nicholas Smith, David Lee & Ylva Berglund Prytz. 2008. Corpus Linguistics with BNCweb – a Practical Guide. Frankfurt: Peter Lang.

Hopper, Paul J. 2011. “Emergent grammar and temporality in interactional linguistics”. Constructions: Emergent and Emerging, ed. by Peter Auer and Stefan Pfänder, 22–44. Berlin & Boston: de Gruyter. doi:10.1515/9783110229080.22

Horn, Laurence R. 1985. “Metalinguistic negation and pragmatic ambiguity”. Language 61(1): 121–174. doi:10.2307/413423

Horn, Laurence R. 1989. A Natural History of Negation. Chicago: University of Chicago Press.

Huddleston, Rodney & Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press.

Jespersen, Otto. 1917. Negation in English and Other Languages. Historisk-Filologiske Meddeleser. Copenhagen: Høst. Reprinted in The Selected Writings of Otto Jespersen, 3–151. London: George Allen & Unwin.

Kerz, Elma & Daniel Wiechmann. 2015. “Register-contingent entrenchment of constructional patterns: Causal and concessive adverbial clauses in academic and newspaper writing”. Journal of English Linguistics 43(1): 61–85. doi:10.1177/0075424214564364

Kilgarriff, Adam. 2005. “Language is never, ever, ever, random”. Corpus Linguistics and Linguistic Theory 1(2): 263–276. doi:10.1515/cllt.2005.1.2.263

Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus, and the Mental Representations of Discourse Referents. Cambridge: Cambridge University Press.

Laury, Ritva & Tsuyoshi Ono. 2014. “The limits of grammar: Clause combining in Finnish and Japanese conversation”. Pragmatics 24(3): 561–592. doi:10.1075/prag.24.3.06lau, https://benjamins.com/#catalog/journals/prag.24.3.06lau/fulltext

Lee, David Y.W. 2001. “Genres, registers, text types, domains and styles: Clarifying the concepts and navigating a path through the BNC jungle”. Language Learning & Technology 5(3): 37–72. http://llt.msu.edu/vol5num3/lee/

Lijffijt, Jefrey, Terttu Nevalainen, Tanja Säily, Panagiotis Papapetrou, Kai Puolamäki & Heikki Mannila. 2016. “Significance testing of word frequencies in corpora”. Digital Scholarship in the Humanities 31(2): 374–397. doi:10.1093/llc/fqu064

Mair, Christian. 2006. “Tracking ongoing grammatical change and recent diversification in present-day standard English: the complementary role of small and large corpora”. The Changing Face of Corpus Linguistics, ed. by Antoinette Renouf and Andrew Kehoe, 355–376. Amsterdam & New York: Rodopi.

Manyika, James, Michael Chui, Brad Brown, Jacques Bughin, Richard Dobbs, Charles Roxburgh & Angela Hung Byers. 2011. Big Data: The Next Frontier for Innovation, Competition, and Productivity. McKinsey Global Institute. http://www.mckinsey.com/business-functions/business-technology/our-insights/big-data-the-next-frontier-for-innovation

McCawley, James D. 1991. “Contrastive negation and metalinguistic negation”. Proceedings of the 27th Annual Meeting of the Chicago Linguistics Society (CLS) 27(2): 189–206.

Petyt, K.M. 1985. ‘Dialect’ and ‘Accent’ in Industrial West Yorkshire. Amsterdam & Philadelphia: John Benjamins.

Pitts, Alyson. 2011. “On salience and enrichment in expressions of negation”. Salience and Defaults in Utterance Processing, ed. by Kasia M. Jaszczolt and Keith Allan, 103–150. Berlin: Mouton de Gruyter. doi:10.1515/9783110270679.103

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik. 1985. A Comprehensive Grammar of the English Language. Harlow: Longman.

Säily, Tanja. 2014. Sociolinguistic Variation in English Derivational Productivity: Studies and Methods in Diachronic Corpus Linguistics (Mémoires de la Société Néophilologique de Helsinki XCIV). Helsinki: Société Néophilologique. http://urn.fi/URN:ISBN:978-951-9040-50-9

Sweetser, Eve. 2006. “Negative spaces: Levels of negation and kinds of spaces”. Proceedings of the conference “Negation: Form, Figure of Speech, Conceptualization”, ed. by Stéphanie Bonnefille and Sébastien Salbayre, 313–332. Tours: Publications universitaires François Rabelais.

Tottie, Gunnel. 1991. Negation in English Speech and Writing: A Study in Variation. San Diego: Academic Press.

Wang, Tricia. 2013. “Big data needs thick data”. Ethnography Matters. http://ethnographymatters.net/blog/2013/05/13/big-data-needs-thick-data/

Verhagen, Arie. 2005. Constructions of Intersubjectivity: Discourse, Syntax, and Cognition. Oxford: Oxford University Press. doi:10.1093/acprof:oso/9780199226702.001.0001

Zimmer, Karl. 1964. Affixal Negation in English and Other Languages. Word 20(sup1), Monograph No. 5. http://www.tandfonline.com/toc/rwrd20/20/sup1

Corpora

BNC = British National Corpus. 1991–1994. BNC Consortium. http://www.natcorp.ox.ac.uk

ICE-GB = International Corpus of English, British Component. 1990–1998. Survey of English Usage. http://www.ucl.ac.uk/english-usage/projects/ice-gb/

LLC = London-Lund Corpus of Spoken English. 1959–1990. Jan Svartvik, Lund University. http://www.helsinki.fi/varieng/CoRD/corpora/LLC/

LOB = Lancaster-Oslo/Bergen Corpus of British English. Geoffrey Leech (project leader), Stig Johansson (project leader), Knut Hofland (head of computing), Roger Garside (head of computing, POS-tagged version). http://www.helsinki.fi/varieng/CoRD/corpora/LOB/