3.1 The ‘pattern preservation’ hypothesis

The pattern preservation scenario essentially hypothesizes a functional repair mechanism. It was first suggested by Crowell (Reference Crowell1959), and, as it were, implicitly adopted by Krug (Reference Krug2000: 64ff.) and Tagliamonte (Reference Tagliamonte, Rohdenburg and Mondorf2003: 533). Crowell's argument runs as follows. A general trend towards vowel reduction in unstressed syllables from the fifteenth century onwards affected the forms have and has (both as auxiliary and as main verb), whose vowels were reduced to schwa. In the sixteenth century, this schwa began to undergo further reduction to zero, leading to the contraction of have/has with the preceding item, i.e. the now common forms ’ve and ’s.Footnote ¹ These contracted forms appear in the written language in the sixteenth century, as in (2).

1. (2) Though by my thoughts I've plunged into my life's bondage, I yet may disburden a passion. (OED 1593: Sir Philip Sidney, Countess of Pembrokes Arcadia 90)

Crowell further assumes that the pattern subject + verb, as ‘one of the strongest habits or “patterns of structure” of English’ (Crowell Reference Crowell1959: 282), needs to be complete and recognizable. This pattern is easily identified when have is stressed as in (3) or when it serves as an auxiliary for another verb as in (4). A contracted have in the function of a main verb, however, obscures the pattern, and it is here that have got comes in to preserve it (5a–b). Thus, ‘have got was substituted for have to strengthen the subject + verb construction in places where, through reduction of stress on have, that construction was not clearly signaled’ (Crowell Reference Crowell1959: 280). As possessive have got emerges in the late sixteenth century (see above), Crowell argues that this is a direct consequence of the increase of have-contraction.

1. (3) John hasn't been there, but I have. (Crowell Reference Crowell1959: 283)

2. (4) I've been living with a genius. (ibid.)

3. (5)

   1. (a) I've two hands. (ibid.)

   2. (b) I've got two hands.

The pattern preservation hypothesis invokes the competition of the opposing motivations of economy and explicitness (see Haspelmath Reference Haspelmath1999): contraction is employed in the service of economy, while pattern preservation is a matter of explicitness. The overuse of a strategy to increase economy (have-contraction) led to a problem which had to be repaired by increasing explicitness (got-insertion). Thus, possessive have got emerged because it was needed to mend the system: ‘Where there is a “gap” or inconsistency in the system of a language, something usually happens’ (Crowell Reference Crowell1959: 284). This line of thought is found in many accounts of language change, such as, for example, the notion of ‘renewal’ in cyclical change (van Gelderen Reference Gelderen and van Gelderen2009).

Despite the impression of intentionality that Crowell's description sometimes makes, the claim of got-insertion is not that speakers consciously decided to insert got in order to patch up the subject–verb pattern. It is an unconscious, cumulative process; the only intention that needs to be assumed is that speakers try to communicate effectively. However, got-insertion is teleological in that it serves the goal of pattern preservation.

The pattern preservation hypothesis is in line with the historical record as far as first attestations (of have-contraction and possessive have got) go. However, as Schulz (Reference Schulz2012) points out, early examples of have-contraction are rare and restricted to auxiliary have rather than the possessive use. Yet the theory of got-insertion hinges on the frequent contraction of main verb, possessive have. Schulz (Reference Schulz2012: 135) rejects it on this basis, concluding that

Krug (Reference Krug2000: 64f.) argues that contraction was certainly more common in spoken interaction than in the written record, yet the problem remains that we are left without direct quantitative evidence from historical data. I will suggest a way of resorting to indirect evidence in my analysis below.

3.2 The ‘conventionalized implicature’ hypothesis

The conventionalized implicature hypothesis (see Lorenz Reference Lorenz2010; Schulz Reference Schulz2012) takes a slightly different approach to the origin of possessive have got – rather than asking how got came to be added to have, it focuses on how a present perfect of get came to denote stative possessive meaning. Its central assumption is that have got underwent a gradual semantic shift from ‘have obtained’ to ‘possess’. This view is already advocated (but not spelled out) by Visser (Reference Visser1973: 2202), who speaks of an ‘extremely slow and formally entirely indiscernible process’.

The idea of conversational implicatures leading to new meanings was tentatively explored in diachronic linguistics in the 1970s, e.g. by Bolinger (Reference Bolinger1971: 522): ‘[I]nferential meanings may in time become referential’. Traugott & König (Reference Traugott, König, Traugott and Heine1991) and Traugott & Dasher (Reference Traugott and Dasher2002: 34ff.) spelled out the idea in what they call the Invited Inferencing Theory of Semantic Change. The invited inference in the case of have got is that a past event of ‘obtaining’ or ‘acquiring’ with present relevance (as expressed by the present perfect) implies, or invites the inference of, ‘possession’. This holds irrespective of the specific verbs in use, so that statement (6b) follows from (6a) in exactly the same way as (7b) follows from (7a).Footnote ²

1. (6)

   1. (a) I have got(ten) a new camera.

   2. (b) I have a new camera.

2. (7)

   1. (a) I have bought a new camera.

   2. (b) I own a new camera.

With the form have got (and only this form), this implicature was gradually conventionalized, leading to a reanalysis of have got from dynamic to stative possessive. This is a semantic reanalysis as defined by Eckardt (Reference Eckardt2006: 236):

Thus, if (8a) is read in its initial sense, the acquisition is asserted, and the consequent possession implied; if read in the reanalysed sense, the sentence states that ‘some’ now own these estates. The completion of the reanalysis is evinced, for instance, in the use of have got for inalienable possession, as in (8b). A detailed formal account of this process of implicature conventionalization is provided in Schulz (Reference Schulz2012: 137ff).

1. (8)

   1. (a) and some that were hardly able to shew their heads, have now got great Estates. (1703: CED D4HOMEMB)

   2. (b) By all that's good, he has got the sorry cunning of our Sex. (1696: CED D4CMANLE)

The gradualness of the process rests not only in its spread across the population – there is clearly also a gradualness in the ambiguity of have got, i.e. in how easily it is analysed as stative possessive rather than perfective. In (9), it is possible that the first speaker meant to ask what food the addressees had procured, but the answer shows that What have you got was understood as stative possessive.

1. (9) What have you got to eat? We have a Piece of Beef on the Spit. (1731: CED D5HGBEIL)

Based on this gradual reanalysis, Schulz (Reference Schulz2012: 138ff.) describes three stages of invited inference. In the first stage, ‘possession’ may occur ad hoc as a conversational implicature of perfective have got. This is akin to what Eckardt (Reference Eckardt2006: 239) calls the PRE-stage of reanalysis. The activation of this implicature then leads to the second stage, in which ‘possession’ is a ‘generalized invited inference’ (see Traugott & Dasher Reference Traugott and Dasher2002: 35), rendering have got pragmatically polysemous between the perfective and the possessive reading. Its use and interpretation at this stage oscillates between those two poles, and is negotiated through context. In the third stage, the invited inference is semanticized, which is marked by uses of have got that allow only a possessive interpretation (such as (8b) above). This corresponds to Eckardt's (Reference Eckardt2006: 239) POST-stage. In my data analysis below, I will present an operationalization of these three stages.

The hypotheses of ‘pattern preservation’ and ‘conventionalized implicature’ are alternative explanations of the emergence of possessive have got, though they are not strictly mutually exclusive. Semantic change and structural pressure could work together in forwarding the new variant. The separate tests in the next sections allow for this possibility. Yet, the hypotheses differ crucially in the primary motivation of the change. Is it a structural repair mechanism or an implicature becoming conventional?

3.3 Data and method

The dataset used in this study is combined from the Parsed Corpus of Early English Correspondence (CEEC; Taylor et al. Reference Taylor, Nurmi, Warner, Pintzuk and Nevalainen2006) and the Corpus of English Dialogues 1560–1760 (CED; Kytö & Walker Reference Kytö and Walker2006). The CEEC is a collection of personal letters from late Middle English and Early Modern English; for the time span considered here, 1460 to 1700, it comprises 2.2 million words. The CED consists of various speech-related texts (drama, trial proceedings, didactic and fictional prose), with 1.2 million words in total and a time span from 1560 to 1760. Both corpora thus represent ‘speech-like’ data in the sense of Culpeper & Kytö (Reference Culpeper and Kytö2010: 17f.).

To examine the emergence of possessive have got(ten), all instances of this bigram were extracted from the combined CEEC/CED data, excluding only those in which get has a meaning clearly distinct from ‘obtain’, as in (10).

1. (10) 500 of the Mounster rebells have gotten into Beere Haven [. . .] (1602: CEEC CHAMBER,I,139.005.244)

This yields a database of 229 tokens. A possessive interpretation is possible with 165 of these. Despite this low overall number, a consistent increase in frequency is observed over time, most notably around the year 1600 (see figure 1). In figure 1 and the following analyses, five time periods of 60 years each are stipulated, following the organization of the CED.

Figure 1.

Frequency development of have got(ten) 1460–1760

3.4 Testing the hypotheses

We now turn to testing the hypotheses concerning the emergence of possessive have got, beginning with the ‘functional’ idea of got-insertion to preserve the subject–verb pattern.

As explicated above, the pattern preservation hypothesis assumes that early possessive have got was contingent on have-contraction, which is largely absent from the written record of Early Modern English but may have been common in the spoken vernacular. In the speech-like written data provided by the corpora, we can therefore only look for indirect evidence. If have-contraction is the trigger for possessive have got, it can be assumed that Early Modern English writers would preferably write have got in those contexts in which they would articulate ’ve got in speaking. These contexts can be determined. Labov (Reference Labov1969) identifies preceding vowels and pronominal subjects as favourable to the contraction of a copula or auxiliary (i.e. forms of be, have, do), and this is confirmed on a large database by MacKenzie (Reference MacKenzie2013: 23ff). Similarly, Krug (Reference Krug1998: 289) states that ‘pronouns favor and full nouns disfavor encliticization of adjacent reduced forms’, due to the high string frequency of pronoun + auxiliary. From this, a testable prediction of the pattern preservation hypothesis can be derived: early instances of possessive have got should be found in positions immediately following a vowel-final pronoun. Hence, the rise of have got from the fifteenth to the sixteenth century should be driven by exactly this type.

To test this prediction, those 165 instances of have got in the CEEC/CED data which allow for a possessive reading are grouped into two categories (see table 1): ‘Contraction environments’ are those with a preceding vowel-final pronoun, i.e. the strings I/thou/he/she/we/you/they have got, as in (11). All other tokens are labelled ‘non-contraction environments’, for instance with a full NP subject (12) or subject–auxiliary inversion (13); pronouns ending in a consonant (e.g. it, that) also belong in this category.

1. (11) I am glad to hear [. . .] of the little rise, which at the last yow have gott in your profession, (1615: CEEC HOLLES,I,66.015.336)

2. (12) I doubt your brother Williams hath got the grudging of an ague. (1608: CEEC CHAMBER,I,254.016.673)

3. (13) And what news have you got for me Sir? (1647: CED D3CTB)

Table 1.

Environments of have got(ten) in CEEC/CED

If got-insertion is triggered by the contraction of have, we should be able to observe an increased frequency of have got in ‘contraction environments’ in the crucial time, i.e. around 1600; the frequency in ‘non-contraction environments’ should remain stable, as these do not provide a trigger for got-insertion. Thus, the share of ‘contraction environments’ among the instances of have got is expected to increase. By the same token, this share should decrease with main verb have, if it is the contraction contexts that are taken over by have got.

Figure 2 presents the share of contraction environments across time periods for possessive have got and main verb have.Footnote ³ None of the above expectations is borne out. With have got the share of contraction environments is slightly increased in the latter half of the seventeenth century, but the differences between the time periods are not statistically significant (p(χ²) = 0.621.Footnote ⁴ Moreover, the same slight increase in contraction contexts is observed with main verb have, and throughout the data, main verb have even has a stronger tendency to occur after vowel-final pronouns (i.e. contraction environments). The χ² test yields a significant result (p < 0.001) for the changes in main verb have due to the larger number of tokens of this form. However, using χ² for a heterogeneity test (see Gries Reference Gries2013: 190ff.) to compare the results for main verb have and have got (disregarding the fourth time period) shows that the developments are not significantly different (p = 0.644).

Figure 2.

have got and main verb have in contraction environments

The data show no visible effect of contraction environments on the development of possessive have got. Two caveats should, however, be noted. Firstly, the data contain also all the ambiguous instances of have got, for which a possessive reading is possible but may not have been intended by the authors. Secondly, the evidence considered here is indirect, since first-hand data of have-contraction in the spoken vernacular of Early Modern English is not available. Therefore, what is presented here is not counter-evidence to the ‘pattern preservation’ hypothesis. What can be asserted is that no evidence was found to support it.Footnote ⁵

The hypothesis of conventionalized implicature makes very different predictions. The determinant here is not a specific phonological environment but rather the pragmatic meaning of the item. The pragmatic meaning is dependent on context. To operationalize the stages of invited inference (see Schulz Reference Schulz2012: 138), a model developed for grammaticalization by Heine (Reference Heine, Wischer and Diewald2002) is adapted to the case of have got. This is presented in table 2; the columns ‘Stage’, ‘Context’ and ‘Resulting meaning’ are directly adopted from Heine (Reference Heine, Wischer and Diewald2002: 86). The fourth column describes the model's application to have got.

Table 2.

Context model, adapted from Heine (Reference Heine, Wischer and Diewald2002: 86)

In the ‘initial stage’, have got is used with a focus on the event of ‘getting’, which is situated in the past. Lee (Reference Lee, Minkova and Stockwell2002: 374) defines the ‘existential perfect’ as denoting ‘a situation that occurred once or more in the past within a time span construed by the speaker as continuing up to the present’. For have got this is exactly the type which rules out a possessive reading; indicators in the context are adverbs of manner or time, and prepositional phrases that specify the ‘get’-event. The ‘bridging’ stage of have got is defined as giving rise to possessive meaning. This is the case in the ‘resultative perfect’, defined by Lee (Reference Lee, Minkova and Stockwell2002: 375) as ‘past action with present results’. The result of ‘get’ is almost necessarily ‘possess’. Thus, a ‘bridging context’ is present whenever the inference of ‘possession’ is possible, i.e. when there is no contextual cue to interpret it as an existential perfect. A ‘switch context’, then, is a context which is incompatible with the perfective meaning, that is, ‘possession’ is the only possible reading. This is identified by possession that is inalienable, and by adverbials that signal stativity. The last stage, ‘conventionalization’, is not clearly indicated by context; a possible candidate would be metaphoric uses of ‘possession’, such as (14). As this type is very rare in the data, it is here conflated with the ‘switch’ context. Table 3 summarizes the resulting three stages of have got and their operationalization in the CEEC/CED data.

1. (14) for he had got on a new Fustian Frock and Breeches, (1740: CED D5TGREEN)

Table 3.

The context model applied to have got

The conventionalized implicature hypothesis assumes a gradual progress through these stages. Its empirical prediction therefore is that bridging contexts increase over time, as it is this context that brings about the change. Switch contexts should appear only in the seventeenth century.

Figure 3 presents the relative frequencies of the three context types over time, as well as the results of a linear model taking context type as predictor for the year of occurrence (where ‘bridging’ as the ‘middle’ stage serves as the reference level). The developments clearly match the predictions of the conventionalized implicature hypothesis.

Figure 3.

The development of have got(ten) by context types

Note that ‘bridging contexts’ are those that contain no specific cue to one or the other interpretation. This type is therefore expected to be frequent at all times, both before and after the reanalysis. Nevertheless, ‘bridging’ increases its share throughout the sixteenth and seventeenth centuries, clearly at the expense of ‘initial’ contexts. In the last period, when the ‘bridging contexts’ decrease slightly, it is the ‘switch contexts’ that increase their share. Thus, the data suggest a scenario by which contexts that invite the inference of stative possession (‘bridging’) slowly but steadily increased in frequency, thus strengthening the inference, to a point where it became the default interpretation, and speakers/writers begin to use the construction in contexts that rely on a possessive interpretation (‘switch’). The ‘initial’ use which coerces a resultative reading falls behind, but remains in use – its declining relative frequency is balanced by the overall rise of the have got construction (see figure 1).

It is noteworthy that the variant have gotten – which in PDE denotes a perfective reading – is completely absent from ‘switch’ contexts. It appears that from the beginning, the shorter have got is the form of choice for expressing ‘possession’. However, have gotten gradually disappears from ‘initial’ and ‘bridging’ contexts alike (see figure 4). Thus, while have gotten is associated with perfective readings, the form have got does not serve to distinguish the emerging possessive reading, nor does this distinction arise over time. Note that in the last time period, 1701–60, no instance of have gotten is found, although ‘initial’ still occurs (with have got).

Figure 4.

The decline of have gotten versus have got

4.1 Data and method

Possessive have got can be said to have become established as a variant in the seventeenth century. For the analysis of early variation, I therefore chose data from 1720–50, extracted from the Old Bailey Corpus of trial proceedings (OBC), transcripts of spoken English in a court of law in London (Huber Reference Huber2007). I consider this the most useful kind of data for the present purpose, due to its relative faithfulness to the actual spoken language (see Kytö & Walker Reference Kytö and Walker2003 for an appraisal of trial proceedings in this respect). The time frame predates the earliest period in Noble's (Reference Noble1985) diachronic study of have and have got. This thirty-year subset of the OBC comprises approximately 2.5 million words. All instances of finite, stative possessive have and have got in the present tense are considered, thus excluding perfective have got (the ‘initial stage’ above) and dynamic uses of have (e.g. have breakfast), as well as past tense, imperative and infinitive have, where the alternative have got is either ruled out or so rare that these uses cannot be regarded as variational contexts. This selection yields a total of 1,833 tokens, 1,549 of which are bare have and 284 have got.

Following from the above discussion, the questions to be answered are: is have got favoured in questions and with negation? And is there already a differentiation between have and have got for different aspects of possession? In addition, the speaker's role in court will be considered as a social factor.

Regarding the first question, the data has been coded for three sentence types: ‘affirmative’, ‘negative’ and ‘question’; ‘negative’ is defined by the presence of the negation marker not/n't (19a–b)Footnote ⁷ , and ‘question’ by the inversion of subject and verb or auxiliary (20a–b).

1. (19)

   1. (a) I han't got so much Silver about me; (OBC_1_17250827)

   2. (b) I am a poor silly Soul, and have not a Farthing in the World. (OBC_1_17250224)

2. (20)

   1. (a) The Prisoner clapt a Pistol to my Breast, and said, Damn your Blood, what Money have you got? (OBC_1_17340630)

   2. (b) Damn your Eyes, says he, what Money have you about you? (OBC_1_17320906)

The semantic aspects are operationalized by the labels ‘basic’, ‘abstract’ and ‘inalienable’. ‘Basic’ here refers largely to temporal and concrete possessionFootnote ⁸ (21a–b), in particular Langacker's (Reference Langacker, Taylor and MacLaury1995: 56f.) types a, d–f, l and r (as listed above); ‘abstract’ applies to non-concrete states or qualities (types i, j, m; examples (22a–b)); ‘inalienable’ possessions are family members, body parts and physical features (types b, c, k; examples (23a–b)). Some abstract possessions are inherently inalienable (24); these are also coded as ‘abstract’ as their abstractness is seen as the primary characteristic. It should also be noted that due to the nature of the data, certain ‘possessions’ tend to recur, such as money and watches (‘basic’), wives and husbands (‘inalienable’), having something to say and having a certain character (‘abstract’). It is also in the nature of the data that the possessor is almost always a specific person and generic subjects are extremely rare; this factor is therefore left out of the present analysis.

1. (21)

   1. (a) G - D - ye, says I, You have got my Watch. (OBC_1_17350911)

   2. (b) she laying her Hand upon her, said, You have my Watch, (OBC_1_17221205)

2. (22)

   1. (a) and said to the Prisoner, Honesty, I was not alone; I have got Company. (OBC_1_17330628)

   2. (b) If you have either Honour or Honesty, come away directly; (OBC_1_17380412)

3. (23)

   1. (a) I know the Prisoner again, because he has got but one Hand; (OBC_1_17320223)

   2. (b) but I know she has another Husband, and his Name is Sheppard. (OBC_1_17370420)

4. (24) He is an Irishman, and has the Character of most of his Countrymen – a very bad Character in general. (OBC_1_17331010)

Finally, the categories for the speaker's role are simply ‘court’ and ‘other’, where ‘court’ is the judge and ‘other’ may be the prosecutor, the defendant, a witness or a legal counselor. (For detail on the setting and power dynamics in the Early Modern English courtroom, see Cecconi Reference Cecconi2012: 49ff.).