A focus on form

The literature is replete with discussions of innovative forms. In London, the periphrastic quotative this is + [speaker] has recently emerged (Cheshire, Kerswill, Fox, & Torgersen, Reference Cheshire, Kerswill, Fox and Torgersen2011). California has witnessed the advent and decline of be all (Buchstaller, Rickford, Traugott, Wasow, & Zwicky, Reference Buchstaller, Rickford, Traugott, Wasow and Zwicky2010; Rickford, Buchstaller, Wasow, & Zwicky, Reference Rickford, Buchstaller, Wasow and Zwicky2007). Suggestions that go is making inroads in contemporary varieties of English have also surfaced (e.g., Buchstaller, Reference Buchstaller2006a; Butters, Reference Butters1980). These forms are exemplified in (1). However, the development that has received the bulk of attention is be like.

1. (1)

   1. a. This is him, “Don't lie. If I search you and if I find one I'll kick your arse.” (London, England; Cheshire et al., Reference Cheshire, Kerswill, Fox and Torgersen2011)

   2. b. She's all, “What do you mean, gum?” (California, USA; Rickford et al., Reference Rickford, Buchstaller, Wasow and Zwicky2007)

   3. c. ...later on we just go, “Oh man, you shut your old prune face.” (Wilmington, North Carolina; Butters, Reference Butters1980)

By their very nature, innovations are interesting. This partly explains the tendency to focus on change rather than stability, as well as the tendency for the diachronic dimension of variation to be identified with change (Labov, Reference Labov1989). Innovations are also useful. Be like has proven a robust heuristic in the effort to develop an empirical theory of language change, providing key insights into such issues as age-grading, lifespan change, and the incrementation of change, as well as allowing detailed exploration of linguistic, social, and geographic facets in the diffusion of change. Replicability is a sign of good science, and it follows that research on be like should lead to more research on be like.

No individual form (or select subset of forms) can be the sole focus of variationist analysis. The Labovian framework is defined by what gets counted. All variants must be included in the quantitative model (Labov, Reference Labov1966, Reference Labov1972a). In the case of constructed dialogue, the variable context is defined not by form (e.g., be like) but by function (i.e., direct quotation). However, because individual verbs of quotation are defined as dependent variables, the quantitative procedure can foreground forms and background broader systemic considerations.

What we gain in viewing quotation from this perspective is a strong understanding of how a particular verb functions at a particular stage of the language. What we risk losing sight of is how the system itself operates. Though we may see strong probabilistic effects of contextual factors, we tend not to ask why those constraints operate as they do, let alone why they operate at all. What is their role within the quotative system more generally, and how is direct quotation (as a discursive mode) constrained? To move further in our understanding of quotation in spoken narrative, a broader interpretive context is required.

Synchronicity

Early work on be like was often restricted to younger speakers, as they were the only ones to use this form. More recent work has redressed this imbalance, targeting a broader age spectrum, but, as is well acknowledged, the primary shortcoming of apparent time is that it is insufficient for disentangling age-grading from generational change (Labov, 1984:84). The persistence of be like among adult speakers, its increasing rates of use, and evidence for lifespan change have all served to dispel the question of age-grading (Buchstaller, Reference Buchstaller2006a; Cukor-Avila, Reference Cukor-Avila2002; Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007). No longer an “innovative fad,” be like is well entrenched within the quotative repertoire and its apparent time trajectory has been corroborated by real-time analyses (Barbieri, Reference Barbieri2009; Buchstaller, Reference Buchstaller2006a; Buchstaller et al., Reference Buchstaller, Rickford, Traugott, Wasow and Zwicky2010; Cukor-Avila, Reference Cukor-Avila2002; D'Arcy, Reference D'Arcy2010; Durham, Haddican, Zweig, Johnson, Baker, Cockeram, Danks, & Tyler, Reference Durham, Haddican, Zweig, Johnson, Baker, Cockeram, Danks and Tyler2011; Ferrara & Bell, Reference Ferrara and Bell1995).

These real-time perspectives on be like have narrow time depths, ranging from just 2-year increments to 15-year increments. In targeting particular forms, it is both sufficient and logical to restrict the analytical window to one that provides insight to the phenomenon in question. Because be like is relatively new, short periods are more than sufficient for establishing ongoing shifts in use and constraint patterning; they also allow examination of the overall functioning of the major verbs of quotation immediately before and after be like. These comparisons reveal what appear to be recent, sudden, and catastrophic shifts within the overall configuration of the quotative system, yet they are constrained in their ability to address questions concerning the nature and causation of change.

Linguistic change is known to proceed somewhat chaotically in fits and starts (see Janda, Reference Janda1999:329; Joseph & Janda, Reference Joseph, Janda, Joseph and Janda2003:20; Lass, Reference Lass1997:304), but evolutionary “realignments” are rarely confined to a few frenetic years. Neither synchronic nor real-time analyses of direct quotation have illuminated what its substantive nature may have been at more distal stages in the recorded history of spoken English. We cannot assume that the system was stable immediately prior to the incursion of be like, or that the rise of new forms has shifted the distributional, functional, and/or pragmatic workloads of traditional verbs for direct quotation. Until synchronic aspects of constructed dialogue become informed by diachronic perspectives, our understanding runs the risk of being fundamentally teleological.

Buchstaller (Reference Buchstaller2011) is the only research published to date in which the longitudinal development of the English quotative system is scrutinized. Drawing on data from five decades of conversational recordings from the North-East of England, Buchstaller ascertained that change has not been abrupt; “continuous restructuring” has occurred over several generations, sparked, she argued, by the “intrusion” of be like and go.

The Mobile Unit

The Mobile Unit (MU) recordings were made between 1946 and 1948 by members of the Mobile Disc Recording Unit of Radio New Zealand. The primary aim was to collect personal reminiscences of life in rural towns at the beginning of British settlement. The speakers were the children of the first European colonizers. The oldest were in their 90s at the time of recording, but the median age is about 75 years (years of birth: 1851–1910).Footnote ²

The Intermediate Archive

The Intermediate Archive (IA) consists of oral history projects and, to provide continuity of data, interviews with some of the descendants of the original MU speakers. The various IA materials were recorded between 1990 and 1996 (years of birth: 1890–1935).

The Canterbury Corpus

The Canterbury Corpus (CC) is a sociolinguistic monitor corpus, growing annually since 1994 (speaker years of birth: 1926–1985). Because the sample does not include speakers over the age of 60 years, these data are supplemented by recordings of older speakers from the Darfield Corpus (DC), a regional sociolinguistic corpus compiled in 2006 (years of birth: 1918–1935).

The intersection of corpus and historical development

The ONZE collections were assembled for distinct purposes; their formats are not consistent. Methodological discontinuities of this nature pose an unavoidable challenge when constructing longitudinal collections, even when the framework is held constant (e.g., Buchstaller, Reference Buchstaller2011). Field methods, sample designs, and topics of conversation all affect comparability, and yet using composite corpora as windows on diachrony presupposes that the materials are in fact (sufficiently) analogous.

What unifies the ONZE collections is the type of talk they represent, which on the whole is dialogic. ONZE has proven an invaluable resource for the documentation of variation and change in New Zealand English (e.g., Gordon et al., Reference Gordon, Campbell, Hay, Maclagan, Sudbury and Trudgill2004; Hay & Schreier, Reference Hay and Schreier2004). It has also enabled key insights to a number of theoretical statements about language more generally (e.g., Hay & Sudbury, Reference Hay and Sudbury2005; Langstrof, Reference Langstrof2006; Trudgill, Reference Trudgill2004). In the case of direct quotation, the ONZE data are constrained by type of dialogue (Blyth et al., Reference Blyth, Recktenwald and Wang1990; Tannen, Reference Tannen and Coulmas1986). They are largely restricted to narrative complicating action (Schiffrin, Reference Schiffrin1981:58; see also Hymes, Reference Hymes1977; Wolfson, Reference Wolfson1978).Footnote ³ Because narratives are a naturally bound unit of discourse (Schiffrin, Reference Schiffrin1981:45), they present an ideal site for quantitative analysis. Their privileged status as the primary locus of direct quotation functions to further level methodological discrepancies between the ONZE collections.Footnote ⁴ Drier, more stilted recordings will feature little in the current analysis as they do not tend to contain direct quotation, creating comparability with respect to the overarching discourse genre from which the data were extracted.

Method

Following well-established traditions in variationist analysis of constructed dialogue (e.g., Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2004, Reference Tagliamonte and D'Arcy2007; Tagliamonte & Hudson, Reference Tagliamonte and Hudson1999), the variable context is defined functionally as direct quotation—the recreation of speech, thought, action, sound, or gesture. Delineating the envelope in this way resulted in over 4600 tokens. As detailed in Table 1, the analysis is based on data from 434 speakers with birthdates spanning 125 years.Footnote ⁵

Table 1. The ONZE quotative sample

Due to the nature of the historical materials, the only social factor that can be examined is speaker sex, though fewer women than men were interviewed by the MU. To maintain constancy across the collections, the analysis focuses on speaker age as an external correlate of variation and change in the quotative system.

A number of linguistic factors are implicated in synchronic variation among verbs of quotation. The two “classic” ones are grammatical person (of the matrix subject) and content of the quote (Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007:203). They relate in particular to be like (believed to have entered the repertoire in response to the developing niche for the reporting of first-person inner states through monologue; see Ferrara & Bell, Reference Ferrara and Bell1995:283; Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007:212), but have been examined for other quotatives as well. Say and go are associated with third-person speech; go is also predicted to occur with nonlexicalized sound (Blyth et al., Reference Blyth, Recktenwald and Wang1990; Buchstaller & D'Arcy, Reference Buchstaller and D'Arcy2009; Romaine & Lange, Reference Romaine and Lange1991; Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007).

Another much discussed linguistic effect concerns the intersection of tense and temporal reference (Blyth et al., Reference Blyth, Recktenwald and Wang1990; Buchstaller & D'Arcy, Reference Buchstaller and D'Arcy2009; D'Arcy, Reference D'Arcy2010; Romaine & Lange, Reference Romaine and Lange1991; Singler, Reference Singler2001; Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007). Be like tends to encode the Historical Present (HP), present tense morphology with past temporal reference. The HP is a well-known feature of English narrative more generally (Schiffrin, Reference Schiffrin1981; Wolfson, Reference Wolfson1981, Reference Wolfson1982) and has been reported to coincide with verbs of direct quotation in particular (Schiffrin, Reference Schiffrin1981).

Dialogue can be constructed using a regular speaking voice, but quotation can also report sounds and gestural content, as well as lexical material that is rendered expressively through changes in pitch, intonation, rhythm, and accent (e.g., Buchstaller, Reference Buchstaller2008; Klewitz & Couper-Kuhlen, Reference Klewitz and Couper-Kuhlen1999). Examples are given in (2). Be like has been linked with mimetic effects since the outset (Butters, Reference Butters1982; Romaine & Lange, Reference Romaine and Lange1991; Tannen, Reference Tannen and Coulmas1986); the role of mimesis within the broader repertoire is less studied.

1. (2)

   1. a. And it's like “[grunts].” (CC, fyn99-11a, b. 1968)Footnote ⁶

   2. b. ...they said, “[raises pitch] Oh my hands, my hands.” (DC, fe1, b. 1936)

   3. c. The old lady said, “[lowers pitch, slows rate] I'm not a robber.” (MU, Annie Hamilton, b. 1877)

Grammatical person, content of the quote, tense/temporal reference, and mimesis are the primary system-internal constraints on direct quotation. Their operation across time provides the foundation of the analysis presented here. Two additional quotative resources are also considered: specification of an addressee and verb placement.

For many verbs of quotation, valency is categorically invariant. Ask and tell are generally verbs of indirect speech (She asked me whether she), but they sometimes introduce speech directly. When used in the active voice, they require an indirect object (ask John, tell her). In contrast, the null form cannot be specified for addressee, and early literature discussed the inability of be like and go to specify an object (Butters, Reference Butters1980; Schourup, Reference Schourup1982). A single instance of each occurs in the ONZE materials, shown in (3), but as these constructions do not (yet) appear productive in New Zealand, they will not be discussed further (though see Buchstaller [2008] for British English). However, there is a handful of verbs for which valency is variably marked (say, think, call/cry/yell [out], remark, suggest).

1. (3)

   1. a. I was like to people “Just stop taking photos of me.” (CC, fyn99-22a, b. 1979)

   2. b. I went to my mum, “Mum, he did it to me.” (CC, fyn97-13a, b. 1968)

Finally, the canonical English position for quotative verbs is immediately before the quoted material (Romaine & Lange, Reference Romaine and Lange1991; Schiffrin, Reference Schiffrin1981). The main clause may be postposed, however, such that it follows the quoted material, as exemplified in (4). In speech, this strategy is restricted to say.

1. (4)

   1. a. “Oh I'll take thirty shillings for it,” the blacksmith said. “Right-oh,” said the engineer. (MU, Vivian Young, b. 1885)

   2. b. “Oh,” she says. “You're not drunk,” she says. (IA, Harvey Summers, b. 1905)

   3. c. I had a three-pence up my nose. “Well, pull it out then,” she said. (CC, mop97-20, b. 1942)

The questions this research seeks to address are straightforward. English speakers have a number of resources on which they can draw when constructing dialogue. Is the system in which these resources operate consistently constrained across time, or has it undergone change? Is synchronic change restricted to the number of lexical options available to speakers, or have other changes affected the construction of dialogue?

The nineteenth century, 1860–1894

Contemporary perspectives of direct quotation reveal a system that is robustly variable, both lexically and internally, via the operation of a number of constraints that function in tandem as a “choice mechanism” (Poplack, Reference Poplack, Narrog and Heine2011:213). The MU data provide no such perspective. The system can best be described as “contained”: What variability exists is highly circumscribed. The passages in (5) are exemplary of direct quotation in the MU.

1. (5)

   1. a. I said, “Oh I think I'd better get home.”

      And so Mrs. Hawkins said, “Have all the soldiers gone to bed?”

      And I said, “Oh I don't know.” (Mary Ann Turnbull, b. 1875)

   2. b. He said, “You'd shoot a man? You've got a gun?”

      So Peter says, “Come on then now.”

      He says, “I've got no gun now.” (Robert Templeton, b. 1887)

The overall distribution of forms is provided in Table 2. A single form, say, accounts for nearly 90% of direct quotation. The second most frequent strategy, the null form, is marginal, accounting for less than 8% of the data overall. Quotative think—widely considered a “conservative” variant—is virtually nonexistent. It occurs just six times. The remainder, comprising a mere 3% of the data, consists of a handful of forms (eight to be exact), none of which occurs more than three times. In short, the repertoire is minimal, and it is dominated by say.

Table 2. Overall distribution of forms in the MU

Note: Here and elsewhere, total percentage may not equal 100 due to the effects of rounding.

This lexical restraint is striking. Putative functions of direct quotation include increasing the dramatic element of a narrative and adding authenticity to evaluative claims (e.g. Schiffrin, Reference Schiffrin1981:60). Speakers thus “emphasize various aspects of the report” (Romaine & Lange, Reference Romaine and Lange1991:234) through the use of a range of speech reporting verbs. The historical materials contain more than one verb of saying, but such occurrences are exceptional.Footnote ⁷

Moreover, the pragmatic content of direct quotation is virtually invariant. The instances of think reported in Table 2 are the only cases of quoted inner dialogue (i.e., thought/attitude); writing is quoted just two times. In other words, the quotidian function of constructed dialogue in the MU is speech encoding (99%; n = 544). Other types of content are exceedingly rare.

Despite the lexical and pragmatic uniformity of quotation in the MU, narrators had other strategies available to them. As the most robustly variable of the internal constraints, the primary mechanism is mimesis. A full third of direct speech is rendered with voice effects (34%). Another strategy used by these speakers is tense variation, yet the available choices are limited. Past-tense encoding is the most frequent overall (69%), followed by the HP (25%); this latter configuration is categorically restricted to say. Use of the simple present is an absolute minority option (5%), and the past progressive occurs once. No other aspectual or modal distinctions are attested.

Grammatical person also fails to have a discernible effect. The system operates along a binary contrast skewed in favor of third persons. Quotation is largely a tool for the speech of others (80%). Speakers are less likely to quote themselves (20%), and second persons are virtually unattested (n = 2).Footnote ⁸

Another little used strategy is the specification of an addressee, which is variably employed just 35 times (7% overall). The majority of these instances occur with say. Other verbs with prepositional objects in the MU are suggest and remark, but each occurs just once.

Finally, speakers in the MU rarely place the main clause after the quoted material (7%), and when they do, only clauses with say are postposed.

Table 3 presents the results of a multivariate analysis. In this case, the analysis is concerned not with constructed dialogue as broadly construed in the methodological discussion but specifically with the phenomenon of constructed speech, as this is the function of quotation in the 19th-century ONZE materials. The application value is say (overall distribution in speech, 90%; n = 544). Only three factors can be modeled: speaker sex, speaker decade of birth, and mimesis. The remaining groups cannot be included because there is simply too little variation. Most if not all factors in each group have (near-) categorical distributions (see Guy, Reference Guy, Ferrara, Brown, Walters and Baugh1988).Footnote ⁹

Table 3. Multivariate analysis of factors conditioning the use of say in the MU (speech only)

Input .900

Note: Factors not selected as significant appear in square brackets.

Speaker age (reflected through decade of birth) has a significant main effect. Speakers born in the final two decades of the 19th century disfavor say. This result is largely attributable to increasing use of the null form, rising from just 4% overall among speakers born in the 1860s (n = 193) to 12% overall among those born in the 1880s and early 1890s (n = 165) (χ²p = 0.032).

In contrast, there is little evidence for grammatical conditioning. Historically, direct quotation in this variety appears fundamentally to have been a means to an end: the reporting of speech, with little elaboration of the quotative context (through voicing effects, tense, and aspectual nuances, verb movement, etc.). However, this aggregated model masks a difference in the operation of the variable grammar in apparent time. Though not significant, mimetic quotes are assigned a lower factor weight than are nonmimetic ones (.40 versus .55). As shown in Figure 1, the rate of say in nonmimetic contexts is fairly constant across the MU. In mimetic contexts, however, use of say declines slowly yet steadily.Footnote ¹⁰ Among speakers born in the final decades of the 19th century, a weakly significant contextual effect is manifest (χ²p = 0.0467). In other words, the effect of mimesis is emergent in the MU.

Figure 1. Quotative say and mimesis across apparent time in the MU.

To summarize, the MU data exhibit a constricted quotative ecology, with variation confined to small sectors within the system. Nonetheless, subtle shifts were taking place. Mimesis was developing a more active role as a systemic constraint and the lexical restriction to say was weakening.

Turn of the century, 1891–1935

The overall perspective from the IA both resembles and differs from that of the MU in critical respects. Shared characteristics include a restricted lexical repertoire and a restriction to speech encoding; say again preponderates. The overall distribution of forms is given in Table 4. The lexical containment of the historical data is replicated here, and the “other” category remains distributionally marginal and compositionally restricted, consisting of just seven forms. Notably, however, the repertoire includes quotative go, which occurs four times in total. All instances are provided in (6).

1. (6)

   1. a. You would hear him going “clank clink clink” down the hill. (Dorothy Hagitt, b. 1896)

   2. b. There would be all old ladies with their knitting needles going “click click click.” (Will Oliver, b. 1907)

   3. c. He said to me, “You know how much they cost?” And I went, “Yeah, about so and so and so and so.” (IA, Ivy Thomas, b. 1912)

   4. d. The kids would all go, “Ha ha, water works is on again.” (John Johnson, b. 1922)

Table 4. Overall distribution of forms in the IA

These examples are important in two respects. Go is long attested for introducing lexicalized sounds and gestures (first Oxford English Dictionary attestation: 1791). This is precisely how go functions in (6a, 6b). In (6c, 6d), however, it unambiguously introduces spoken content. These tokens thus capture the extension of go to a new pragmatic context, marking the transition point to the contemporary system in which go is used to introduce a wide range of content types. Moreover, go is often considered an innovative form. This is true only in so far as it has yet to be attested for speech encoding before the 20th century. These data clearly illustrate, however, that it is not a contemporary of be like with respect to its temporal genesis in speech (see also Buchstaller, Reference Buchstaller2006a).Footnote ¹¹

Lexical choices remain restricted in the IA, but use of think has increased markedly (1% in the MU versus 6% in the IA). The IA speakers thus appear to use direct quotation for reporting internal states more frequently than did those of the MU. I will return to this point. The critical observation at this point concerns pragmatic encoding in these early 20th-century data.

Summarized in Table 5, a broader range of content types are attested in the IA than what is found in the MU. Moreover, these content types are less restricted in terms of the verbs by which they are introduced. Two content types not attested in the MU are lexicalized sound, as in (6a, 6b), and hypothetical speech (she's going to say “. . .”; we couldn't say “. . .”). Quoted writing occurs more frequently overall (0.3%, n = 2 versus 3%, n = 21). Say, strictly a speech verb in the MU, also introduces inner monologue in the IA (I said to myself “. . .”). And the null form, likewise restricted to speech in the MU, introduces most content types attested in the IA. Thus, whereas direct quotation remains a fundamentally speech-oriented activity (90%), a functional expansion is evident in the increase in content types and by the generalization of most verbs across multiple content types.

Table 5. Content and verb collocations in the IA

The continued emphasis on speech reporting, however, again necessitates that exploration of the variable grammar be confined to this content type. Table 6 reports the multivariate results.

Table 6. Multivariate analysis of factors conditioning the use of say in the IA (speech only)

Input .899

Note: Factors not selected as significant appear in square brackets.

As with the MU, the only factor groups that can be included are speaker sex, speaker decade of birth, and mimesis. With the remaining groups, individual cells again display nearly categorical distributions and, in some cases, extremely low token numbers (second-person subjects, simple present tense, postposed clauses). Nonetheless, some important shifts have occurred. Where in the MU the majority of quotation was third-person speech, in the IA there is a marked increase in first-person reporting (MU 19% versus IA 34%; p < .0001), in part a reflex of the increased ability to quote inner monologue (primarily, though not exclusively, a first-person mode). The use of tense and temporal combinations beyond simple past, simple present, and HP is also noteworthy. A number of modal, aspectual, participial, and infinitival constructions, including a range of habitual collocations (would say, used to say, used to go around saying; n = 92), are used.Footnote ¹² This qualitative difference suggests that the scope of tense and aspect distinctions is broadening.

Overt addressees and main clause postposing remain minority options overall. Both phenomena continue to be (largely) restricted to say, but postposing, already infrequent in the MU, has decreased markedly and appears obsolescent in the IA (7% versus 2%; χ²p < .0001).

Corroborating the trajectory in Figure 1, mimesis exerts a significant and strong main effect in the IA, disfavored with say (FW: .33, range: 20). Additionally, speaker sex is selected as significant. The internal factor explains more of the variation, but the variable grammar now includes a nontrivial social condition. Women favor say; men disfavor it. The explanation lies in the rise of the null form. Still a minority contender in the system, its overall frequency has increased, albeit marginally (cf. Tables 2 and 4), and it is men who are primarily responsible (men: 12%, FW: .584; women: 8%, FW: .457).Footnote ¹³ In fact, given that say and zero together account for 99% of quoted speech in the IA, it follows that they should pattern in complementarity, say for nonmimetic speech among women, and zero for mimetic speech among men.

In sum, the IA results reveal the emergence of a systemic variable grammar for the construction of dialogue in discourse. Social and linguistic constraints have come to the fore (mimesis, speaker sex); other internal factors may be preparing for activation as well (e.g., tense/temporal reference, grammatical person).

The 20th century, 1918–1987

The CC (including the DC) captures New Zealand English across the bulk of the 20th century. Table 7 reports the overall distribution of forms among speakers aged 60 years and older. They are contemporaries of the youngest cohorts in the IA, enabling a comparison between the two. The similarities are striking: The lexical choices available to the oldest CC speakers are few (say, think, zero); the overall frequency of say is little different from that in the IA (83% versus 80%)

Table 7. Overall distribution of forms: Older speakers (b. 1918–1936)

At the same time, the IA system is fundamentally predicated on speech reporting. Although the main function of direct quotation among the oldest CC speakers is the construction of spoken dialogue, the rate does not approach categoricity (83%). Moreover, the overall frequency of think is twice that found in the IA overall. Given that the IA results in Table 4 collapse data from four and a half decades and that those in Table 7 overlap with just the final one and a half of those, it is possible that Table 4 obfuscates a more general trend. Figure 2 explores this possibility, tracking the frequency of think across apparent time in the IA and comparing these results with those from the oldest speakers in the CC. The trend is consistent with monotonic change: The propensity to report thoughts and attitudes through direct quotation increases incrementally over time. Notably, although roughly 13 years separate the collection of the IA recordings and the CC materials from speakers aged 60 years and older, the results indicate much stability with regard to the lexical and pragmatic organization of direct quotation.

Figure 2. Overall distribution of think: IA (1890–1935) and CC (1918–1936).

The MU and the IA recordings were collected over fairly circumscribed periods (most of the IA materials used here were recorded from 1993 to 1996). The recordings in the CC represent 13 years of continuous, real-time data. Despite the evidence for systemic maintenance across the lifespan for speakers over the age of 60 years, it is well-established that in contemporary use, the repertoire for speakers under the age of 30 differs significantly from that of older ones (Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007). From this point, the analysis focuses on quotation among speakers aged 19 to 59 years, and the recordings are segmented into three periods: interviews collected from 1994 to 1997, 1998 to 2001, and 2002 to 2006.

The overall distribution of forms across the CC is provided in Table 8. The number of forms accounting for more than 1% of the data has increased from three to five (say, think, zero + be like and go), and patent shifts take place from one period to the next. Say remains the most frequent form, but the familiar and established upward trajectory of be like is clear.

Table 8. Overall distribution of forms across time in the CC (b. 1936–1985)

Crucially, it is not simply that the repertoire has changed (see also Figure 4, which displays lexical items across time, according to speaker decade of birth). Concomitant with shifting lexical frequencies, the overarching organization of the quotative system has continued its pathway of restructuring, already evident in the MU and IA datasets. Most analytically significant of the changes is the function of direct quotation itself. Quotation has generalized across content types. Moreover, the trend suggested by Figure 2 (an increase in inner thought reporting via the steady rise of think) is corroborated by the CC. No longer exceptional, the construction of inner dialogue is a fully productive aspect of narrative reporting.

Across the CC, thought constitutes roughly one-quarter of all instances of direct quotation. Furthermore, thought reporting is no longer coterminous with think. Inner monologue is also introduced by say (to self), go, zero, be like, and a range of other, less frequent forms (e.g., be, feel, figure out, decide, realize). It is precisely for this reason that analyses of contemporary data do not partition constructed dialogue into pragmatic fields. The system is robustly variable both in terms of quoted content types and in terms of the verbs introducing them. All quotative contexts must be included to allow for developmental pathways to be discerned.

Three identical multivariate analyses are presented in Table 9, one for each CC period. To provide a comparative perspective with the results for the MU (Table 3) and the IA (Table 6), the application value is say. The declining rate of say is reflected by the input values, which decrease steadily.

Table 9. Multivariate analyses of factors conditioning the use of say in the CC

In Table 9, the complexity of the model sets these results apart from those of the MU and the IA. In the older materials, only speaker sex, age (date of birth), and mimesis could be modeled. Here, not only are content of the quote, grammatical person, and tense/temporal reference also included, they are significant in every period. The variable grammar is thus more finely articulated in contemporary spoken dialogue than it was historically, where the only evidence of grammatical conditioning was the (emergent) mimetic effect.

In the CC, mimesis remains a significant effect on say in all periods. It is consistently outranked, however, by two newly evidenced language-internal effects: content of the quote and tense. Indeed, the variable grammar operating on say is quite stable across the CC. Say is favored for nonmimetic, third-person, past-tense speech, and content of the quote is consistently the strongest factor. This last observation is unsurprising, given that say is a verbum dicendi. However, the inclusion of the content group in the model highlights the fundamentally altered state of the quotative system vis-à-vis the results from the MU (Table 3). Once pragmatically monolithic, content of the quote is now a robust feature of the variable grammar.

Despite the consistent hierarchies and rankings across the CC, subtle and important shifts are discernible. Most obvious is the effect of speaker sex, which remains significant in the first period (cf. Table 6, IA) but then levels out, failing to be selected in subsequent periods. The person constraint—stable with respect to direction, ranking, and significance—weakens relative to the other factors across time. By the second and third periods, it accounts for much less variation in the use of say than does its next closest competitor, mimesis. In contrast, age gains in strength. In the first and second periods, its effect is just slightly stronger than is that of mimesis. By the third period, it rivals tense; speakers under the age of 45 years strongly disfavor this traditional form.

These small shifts are independent of the overall frequency of forms, providing a view of the operation of the system itself. This is particularly striking in the case of pragmatic content. Content is consistently the strongest constraint affecting the probability of say, even though its use in speech contexts decreases over time. Its overall trajectory of decline is not a facet of its use in nonspeech contexts (compare the first and last periods). This difference between pragmatic environments is reflected in the relative weakening of the content constraint in the final period, where the margin between its strength and that of the next strongest constraint, tense, has narrowed. Content still “matters more,” but it accounts for less of the variation than it has in prior periods.

The effect of tense/temporal reference is less clear-cut, arising from fluctuation due to small cells when tense cross cuts other groups. Nonetheless, say is consistently most frequent in, and most favored for, simple past-tense contexts, whereas it is always strongly disfavored with the HP. This latter result is striking when considered in broader historical context. In both the MU and the IA (Tables 3 and 6, respectively), the HP occurs only with say. In Table 9, this categorical correlation is obliterated, a trajectory that is visible corpus-internally, in apparent time. It is this result that accounts for the discrepancy between factor weights and frequencies. The oldest speakers maintain the association of say with HP encoding, whereas among the youngest speakers, the reverse obtains, and the HP is eschewed with say. This is because other verbs have become probabilistically associated with this feature of English narrative complicating action clauses.

Discussion of quotative tense/temporal reference has not often taken in the wealth of configurations evidenced in contemporary discourse. This is in part the consequence of form-focused analyses, but it also derives from a tendency to consider contexts that have previously received attention in the literature (and for which claims exist): simple past, simple present, HP (Blyth et al., Reference Blyth, Recktenwald and Wang1990; Romaine & Lange, Reference Romaine and Lange1991; Schiffrin, Reference Schiffrin1981; Tagliamonte & D'Arcy, Reference Tagliamonte and D'Arcy2007). However, cursory examination of the “other” category in Table 9 reveals the robust presence of configurations beyond those usually discussed. In addition to the habitual constructions first attested in the IA, the CC includes a number of inceptives, modals, and aspectuals (kept on saying, remember saying, could say, should have said, having been told, etc.). It is also replete with progressive structures (past, present, and HP), as in (7).

1. (7)

   1. a. I was thinking, “What? That's gross.” (fyn04-8, b. 1982)

   2. b. ...whereas now I'm saying, “Why did I bother?” (fyn98-1, b. 1978)

   3. c. I'm saying, “Look, Nathan's not here.” (myn98-16b, b. 1977)

In contemporary narrative, the use of the progressive is well-attested and unremarkable (see Labov, Reference Labov1972b; Schiffrin, Reference Schiffrin1981). The ONZE materials suggest that its use with verbs of direct quotation is, however, recent. Unattested in the MU and the IA, past and simple present progressives (7a, 7b) are first used by speakers born in the 1940s. The HP progressive (7c) does not appear in these materials until later, with speakers born in the 1970s.

To summarize, the results from the CC suggest that during the 20th century, a complex system of internal constraints coalesced and became fully operative. Crucially, this is not simply a consequence of the genesis of be like, which did not appear in the repertoire of New Zealand English until relatively late. In the aggregate data for 1994 to 1997, be like accounts for (not quite) 4% of direct quotation (N = 1349). Crucially, an intricate system is already fully established, having built on trends in both the MU and the IA datasets. The ONZE data therefore demonstrate that be like is not responsible for “disruption” of the status quo; rather, it entered an already volatile system.