The most stable it's ever been: the preterit/present perfect alternation in spoken Ontario English

Abstract

English tense/aspect-marking is an area where variation abounds and where many theories have been formulated. Diachronic studies of the preterit/present perfect alternation indicate that the present perfect (e.g. I have eaten already) has been losing ground to the preterit (e.g. I ate already) (e.g. Elsness 1997, but see Hundt & Smith 2009, Werner 2014). However, few studies have examined this alternation in vernacular speech. This article fills this lacuna by analyzing spoken data from Ontario, Canada, from an apparent-time perspective. Using a large archive of multiple communities and people of different generations, we focus on linguistic contexts known to be variable, viz. with adverbs of indefinite time. Results indicate that, in contrast with previous studies, the alternation is mostly stable. We find evidence of change only with the adverb ever. Where there is evidence of change, this change is different from the predictions in the literature, with the preterit increasing in frequency. We suggest that a minor constructionalization process operates in tandem with ongoing specialization of the preterit/present perfect contrast. Taken together, these results provide another example of the importance of including speech in research on language variation and change and of the unique contribution certain constructions make to more general systems of grammar.

1 Introduction

This article focuses on variation and change between the present perfect and the preterit in spoken informal Ontario English. In (standard) English, the present perfect consists of an auxiliary have and a past participle, e.g. I have called, I have laughed. For most verbs, the preterit is formed as the base form of the verb + -ed, e.g. I called, I laughed, although irregular formations exist as well, e.g. I went, I had. In many contexts, both the preterit and the present perfect are possible. In (1a, b) and (2a, b), for example, the contexts are very similar and we find both present perfects and preterits in naturalistic speech.

1. (1)

   1. (a) I find that unbelievable and like I haven't told her yet that … I have the library job. (C. Felipe, F, 20, Toronto)²

   2. (b) Oh no, I didn't tell her yet. I am going to tell her now though. (K. Wilson, F, 18, Temiskaming Shores)

2. (2)

   1. (a) I’ve known trees since I was a kid, like individual trees. (B. Dailey, M, 61, Haliburton)

   2. (b) Kids really don't know what they want to do but I always knew since I was in grade eight or nine. (B. Barnes, M, 67, Kirkland Lake)

Variation and change in the past-referring morphemes of English is a topic that has been researched from various perspectives and in great detail, as evidenced by several book-length studies (e.g. Elsness 1997; Davydova 2011; Werner et al. 2016). Research on synchronic variation often revolves around the question of which functions can be expressed by the present perfect (cf. Werner 2014: 59–72 for a detailed overview). Variation between the present perfect and the preterit is then generally explained by the relationship between the point of reference and the time of speaking (e.g. Jespersen 1954: 60–1; Quirk et al. 1985: §4.18; Rastall 1999; Biber et al. 1999: 467; Huddleston & Pullum et al. 2002: 139–41): the preterit is used when the point of reference precedes speaking time, whereas the present perfect relates the point of reference to speaking time, due to its meaning of ‘current relevance’ (e.g. Quirk et al. 1985: 190), or ‘continuing present relevance’ (Comrie 1976: 52). Many scholars go on to distinguish several readings of the present perfect that exemplify prototypical ways in which the relationship between present time and past situation are linked (e.g. Comrie 1976: 52–65). Werner (2014: 67–72) provides an overview of different models, with, depending on the model, up to seven semantic categories being distinguished (e.g. the perfect in existential contexts, the hot-news perfect, the perfect with resultative meaning, etc.).

The notion of current relevance on which these theoretical accounts rely is not well defined (cf. Werner 2014: 62–5). According to Quirk et al. (1985: 193), for example, ‘the choice between the present perfective and the preterit is often determined by whether the speaker has in mind an implicit time zone which has not yet finished’. They explain the difference between (3a) and (3b) below by arguing that in the former, John's stay in Paris has finished, whereas in the latter example, John lives in Paris at present as well (and may stay there in the future).

1. (3)

   1. (a) John lived in Paris for ten years.

   2. (b) John has lived in Paris for ten years. (Quirk et al. 1985: 189–90)

The methodology employed in these studies has been criticized by Van Herk (2010). His major problem is that they rely on subjective interpretations to explain the use of the present perfect (for an overview of other problems with defining the variable context, see Van Herk 2010: 49–52). For example, many studies have relied on the assumed intended meaning of the speaker to explain their choice of the present perfect. With regard to (4a) and (4b) below, for instance, Quirk et al. (1985: 193) argue that the only difference is that ‘[t]he first … implies that the Exhibition is still open; the second that the Exhibition has finished’ (also see (3) above). If nothing other than the actual form of the verb indicates what the assumed intent of an individual may be, it is impossible to determine this objectively in naturalistic data, and doing so would be circular. In contrast, if (4a) and (4b) had contained temporal adverbs like yet or when it was here, these explicit references to time would have provided the researcher with objective means to probe the intended meaning of the speaker.

1. (4)

   1. (a) Have you seen the Javanese Art Exhibition? [yet]

   2. (b) Did you see the Javanese Art Exhibition? [when it was here] (Quirk et al. 1985: 193)

A similar argument is laid out in Werner (2014: 59–79). Werner even argues that ‘it has repeatedly been observed that many of these accounts are incompatible with each other’ (Werner 2014: 59–60). To solve this problem of ‘identifying categories by introspective reasoning’ (Werner 2014: 59), Van Herk and Werner argue that research should rely on formal properties available in corpus evidence to analyze the variants.

In addition, studies that focus on the interpretation of the perfect often do not take into account other variants that may express the same function (Van Herk 2008: 55). Thus, a variationist perspective which examines the factors that (dis)favour the use of perfect vis-à-vis the preterit or other variants is necessary to adhere to the principle of accountability. This principle entails ‘that any variable form … should be reported with the proportion of cases in which the form did occur in the relevant environment, compared to the total number of cases in which it might have occurred’ (Labov 1969: fn. 20). In this way, variationist studies on the present perfect construction contrast it with other variants that may be used to express the same meaning (usually the preterit, although the picture is more complex in some varieties, e.g. Tagliamonte & Poplack 1988).

An extensive overview of the distribution of the preterit vis-à-vis the present perfect in the history of the English language is found in Elsness (1997: 237–48). This study contains both synchronic and diachronic analyses of the relationship between the two forms. Elsness’ general conclusion with regard to the diachronic patterns is that, after a long history of gaining new functions, the present perfect has been ‘losing ground’ (Elsness 1997: 341) to the preterit since the Modern English period and that this evolution is more pronounced in American English than in British English. For instance, in Elsness’ 1750–1800 British English corpus, an example such as (5) still occurs. In this example, a present perfect is used with the definite time adverb yesterday, a context that would only allow for the use of a preterit in present-day standard English (see below).

1. (5) I am told he has had another execution in the house yesterday. (Elsness 1997: 342)

In present-day English as well, the ‘encroachment’ (Roy 2014: 36) of the preterit on the present perfect is the standard view in the literature (e.g. Vanneck 1958; Yao & Collins 2012). Other diachronic patterns have been noticed as well: Hundt & Smith (2009) and Werner (2014: 21) argue that only minor changes in the distribution of the preterit/present perfect alternation are found nowadays and that the variation is generally stable.

From these variationist studies, a number of factors have been shown to influence variation between the present perfect and the preterit. Van Herk (2010: 52–3), in a diachronic study on different stages of the evolution of the present perfect, mentions over 60 different hypotheses put forward in previous literature, though he argues that many studies on this alternation are not multivariate in nature. For this reason, he analyzes a more limited set of factors (temporal distance, adverb, clause type, object type, verb semantics, discourse type) and we largely follow his approach in this article.

Next to language-internal factors, another line of research has paid attention to language-external factors that may affect variation between the present perfect and other variants. Studies in this vein have mostly focused on regional differences, either comparing standard British English and American English (e.g. Elsness 1997, 2009; Wynne 2000; Hundt & Smith 2009) or varieties of English across the world (e.g. Tagliamonte & Poplack 1988; Tagliamonte 1997; Van Herk 2008, 2010; Davydova 2011; Yao & Collins 2012; Werner 2013). They have also shown that register effects impact the proportion of present perfects vis-à-vis preterits (e.g. Hundt & Smith 2009). In addition, Davydova (2011), a study on variation in present perfect contexts in L2 and foreign-speaker varieties, has examined the effect of sociolinguistic variables such as speaker sex.

The present article aims to contribute to the knowledge about preterit/present perfect alternation by focusing on spoken informal Ontario English and by taking an accountable variationist perspective, that is by including only contexts where the variants are ‘alternative ways of saying the same thing’ (Labov 1969: fn. 20; also see Van Herk 2010 for discussion). As will be explained in section 2, the variable context is defined by taking into account only adverbial contexts that have been shown in previous literature to allow both the perfect and the preterit. Section 3 describes the methodology of our study. Section 4 presents the results of our distributional analysis and of our statistical modeling. We discuss the relevance of these results with regard to what is known about the preterit/present perfect alternation in section 5 and end with a conclusion in section 6.

2 Defining the variable context

Most variationist studies on the present perfect vis-à-vis the preterit have been devoted to a comparison of varieties of English around the world, aiming to answer the question of whether the ratio between the present perfect and the preterit³ is stable across these varieties. Studies in this vein either focus on a single variety (e.g. Tagliamonte & Poplack 1988; Tagliamonte 1997; Van Herk 2008; Werner & Fuchs 2016; Suárez-Gómez 2019), or take a cross-varietal perspective (e.g. Yao & Collins 2012; Werner 2013, 2014; Fuchs 2016).

These studies have shown that one of the most important language-internal properties available in corpus data that governs the variation is the presence of adverbs that refer to either definite or indefinite time (e.g. Elsness 1997; Van Herk 2010; Davydova 2011; Yao & Collins 2012; Werner 2013; Werner & Fuchs 2016). Definite time adverbs, like yesterday, three days ago, last Monday etc., refer to a (more or less) clearly defined time in the past. Because they explicitly specify that the proposition takes place at a specific time point that precedes, and is distinct from, speaking time, they are said to always occur with the preterit (e.g. Quirk et al. 1985: 183–4, 194–5; Hundt & Smith 2009: 55–7). Another group of adverbs carry a meaning of indefinite time: they do not anchor the proposition wholly in the past. Werner (2013: 206) argues that clauses with the forms he investigates, already, yet, always, ever, never, recently, just and since, are the contexts where the present perfect and the preterit are ‘truly variable’ (also see Elsness 1997: 353–4). Further, he shows that the use of the preterit/present perfect with the group of indefinite time adverbs is not homogeneous.

He conducted an analysis of the use of the preterit and present perfect with indefinite time adverbs (as well as with definite time adverbs) in twelve varieties of English, using the ICE corpora (e.g. Greenbaum 1996). His results show that the frequency of the two constructions differs per adverb. For the indefinite time adverbs, he proposes a hierarchy of decreasing perfect-friendliness (figure 1), with some adverbs very often taking either the present perfect (since, already and yet) or the preterit (never and recently), and others (always, ever and just) showing more variation (Werner 2013: 232). This hierarchy is proposed to be global, i.e. it holds for all the varieties under investigation, although minor variability within a variety may occur.

Figure 1. Werner's (2013: 211) hierarchy of decreasing perfect-friendliness across twelve varieties of the ICE corpora

Building on these findings, the present study aims to contribute new understanding of the preterit/perfect alternation in three ways. First, following the reservations of Van Herk (2010) and Werner (2014) with regard to defining the variable context on the basis of subjective interpretations, we investigate variation and change between the present perfect and the preterit by using a formal approach to query the data. More specifically, we rely on the occurrence of indefinite time adverbials to define the variable context, the context where the variants are ‘truly interchangeable’ (Werner 2013: 206; a similar approach is used in Werner & Fuchs 2016). We use this method to investigate whether the preterit is ‘encroaching’ on the present perfect in variable contexts. It is crucial to take into account that two definitions of the ‘encroachment’ of the preterit on present perfects contexts have been used in the literature (Roy 2014: 36). On the one hand, some scholars have paid attention to the fact that the preterit is increasing in frequency in contexts which are traditionally reserved for the present perfect (e.g. Vanneck 1958), such as contexts with resultative or continuative meaning. As explained by Van Herk (2010), it is impossible to distinguish these contexts without relying on subjective interpretations of the speaker's intent. For this reason, we follow the other interpretation that Roy proposes: he argues that in several studies, an overall increase in frequency of the preterit vis-à-vis the present perfect has been interpreted as the former form appearing in contexts that were formerly reserved for the present perfect. Thus, if we find an overall shift in the frequency distribution between the present perfect and the preterit in our data, we take this as evidence that the present perfect is losing ground to an incoming variant, the preterit.

Second, we add to Werner's (2013) study on the ICE corpora by focusing on vernacular spoken data. Whereas the ICE corpora contain some face-to-face conversations and phone calls (about 20 per cent per variety, or 100 out of 500 texts of 2,000 words), the majority of the included corpus texts consist of written materials and (mostly formal) monologues.⁴ By basing the analysis on spontaneous spoken data from informal conversations, we can determine the extent to which the proposed hierarchy of perfect-friendliness holds in vernacular speech.

Third, we employ the apparent-time construct (Bailey et al. 1991) to examine whether the preterit is invading contexts of the present perfect, as is often argued for present-day English (see above). Related to this, it is remarkable that only a few studies have investigated the importance of social factors (one exception is Davydova 2011, who showed that speaker sex plays a role in the use of the variants in learner varieties of English). If studies include external factors, they are generally limited to differences between regional varieties of English (as in Werner's 2013 study), or between different registers (e.g. Hundt & Smith 2009). Yet social differences are a well-known explanation for the generational transmission of a linguistic change (e.g. Weinreich et al. 1968; Labov 2001). We fill these gaps by including in our analysis spoken vernacular data and broad social factors that have been shown to be relevant for linguistic change in general.

3 Methodology

3.1 Corpus and data extraction

To investigate variation and change in the preterit/present perfect alternation, we use data from the Ontario Dialects Project (ODP), a long-term ongoing project that has been collecting spoken Ontario English since 2002 in the main urban centre of the region, Toronto, and other communities throughout Ontario.⁵ These materials comprise sociolinguistic interviews with individuals ranging in age from 9 to 98 and also socially stratified by sex, occupation and education (for further discussion see Jankowski & Tagliamonte 2017). The communities contrast by founding populations, urban versus rural locations, population size, economic base, and distance from Toronto. This allows for insights into variation across a number of geographic and social factors in a single variety of English circumscribed by political territory (Ontario) and majority language (English) in the Canadian context. The individuals in the corpora were born between 1879 and 2001, which provides a proxy of usage across the entire course of the twentieth century. This century encompasses the time frame for the development of Canadian English more generally (see e.g. Chambers 2006). At the time of writing, 19 communities throughout Ontario were available for analysis. The materials were collected between 2002 and 2018, although for some communities, archival materials collected at earlier points in time are also available. To extract the data, we queried the corpus with AntConc (Anthony 2018) for all occurrences of five indefinite time adverbs: since, already, yet, ever and recently, as well as the 100 characters preceding and following the adverb. As explained above, clauses with these adverbs represent contexts where the preterit/present perfect alternation is considered to be ‘truly variable’ (Werner 2013: 206). The specific adverbs included in the study are situated at different points along the hierarchy of perfect-friendliness from Werner (2013: 211), including the most extreme points.⁶ The total number of tokens extracted using this procedure is 10,157.

After extracting the data for these adverbs, we manually verified that each token analyzed represents a truly variable example. More specifically, we removed contexts without perfective meaning (6)–(10), tokens where the adverb does not represent indefinite time semantics (11)–(13) and tokens with a modal verb, where a preterit is not possible (e.g. Hundt & Smith 2009: 50; Werner & Fuchs 2016: 134) (14). In addition, preliminary distributional analyses showed that tokens with progressive aspect are extremely infrequent in our data (N = 60) and that they mostly occur with the present perfect (N = 52, or 87%). Because of the low frequency of progressive tokens, which does not warrant them to be analyzed separately in a multivariate model, they were excluded as well (15). Finally, we excluded false starts and ambiguous tokens, resulting in a dataset that comprises 2,764 observations.

1. (6) No main verb

   We saw “My Friend Flicka” … my first video, or first film ever. (B. Verhoeven, M, 77, Tay Valley)

2. (7) Main verb present tense

   I'm sixty-nine, I'm not over seventy yet. (G. Washington, F, 69, Lakefield)

3. (8) Main verb future tense

   I thought, “Oh my God, now what do I do? She's going to hit me yet!” (A. Drazkowski, F, 71, Wilno (Barry's Bay))

4. (9) Main verb past perfect

   Back in those days, probably hardly any of us had ever been to Niagara Falls. (L. Hubbert, F, 72, Beaverton)

5. (10) Past-tense narratives

   Interviewer: And what brought- I assume you came with your family?

   P. Forsanto: We came as immigrants to … my grandfather who was here already and my Uncle-Dorian who was here already. They called us over because the war ended. … And um, we were running out of money (laughs). And there was no jobs. So we had to come to Canada to renew our lives.

   Interviewer: Wow. So when did your grandfather come here?

   P. Forsanto: I believe it was ah, nineteen-nineteen, roughly. (P. Forsanto, M, 74, Kirkland Lake)

6. (11) Since with definite time meaning (conjunction)

   I’ve made some very good friends moving here, since the kids have started school. (N. Jalna, F, 39, Toronto)

7. (12) Yet with other meaning than indefinite time

   My mother's parents were from Russia, my father's parents were from Austria, yet both of my parents grew up on Grey Street. (E. Jelenic, F, 49, Toronto)

8. (13) Ever with other meaning than indefinite time

   Was I ever glad to get into high-school. … Oh was I ever! (D. Best, F, 85, Almonte)

9. (14) Modal verb

   Oh Nicky-Nine-Doored once. … We wouldn't’ve got caught ever. (T. Bowers, M, 19, Belleville)

10. (15) Progressive aspect

    And most recently you know hydro rates were going up and up and up and up. (J. Ward, M, 69, Seguin Township)

3.2 Independent variables

As outlined above, much of the previous research on the preterit/present perfect alternation employs subjective semantic interpretations to explain the variation. Our alternative strategy considers only tokens that occur with the five indefinite time adverbs listed above (since, already, yet, ever and recently). As a next step, we code each token for the relevant language-internal variables that have been shown to affect the alternation in previous work: adverb, sentence type, grammatical person of the subject and object type. Although Van Herk (2008) and Davydova (2011), for instance, also analyze the effect of other factors, such as temporal location, semantic context, Aktionsart/verb semantics and transitivity, these factors are not included in our analysis. On the one hand, some of these frequently mentioned factors are irrelevant in our study. Temporal location, for example, is indefinite for every token in the dataset as they all occur in a context with an indefinite time adverb. Aktionsart/verb semantics and transitivity have only been found significant in research on learner varieties of English (e.g. Collins 2002 and contributions in Fuchs & Werner 2020). More specifically, Davydova (2011: 155) argues that these factors are mostly relevant for varieties where the present perfect is not fully grammaticalized. On the other hand, other factors rely on subjective interpretations of the tokens – exactly what we try to avoid in this study. Semantic context, for example, requires an interpretation of the speaker's intent by the researcher (Van Herk 2008: 58).

First, sentence type has been mentioned as an influential factor in Elsness (1997: 321), among others. He finds that negated clauses favour the present perfect (also see Van Herk 2010, Davydova 2011, but see Wynne 2000: 119 and Werner 2013: 224). Our distributional analysis (section 4.1) will show that sentence type plays only a minor role in the variation.

Second, Elsness (1997: 197–201, 355) also shows that the ratio of present perfect versus preterit is higher for plural subjects than for singular ones and higher for first-/second-person subjects than for third person, with noun phrases taking up an intermediate position (contrasting the findings of Wynne (2000: 118) for the present perfect alone, who found no distributional patterns). Elsness explains these patterns by referring to the different functions that these grammatical categories and the present perfect and the preterit may have with regard to contextual integration. More specifically, according to Elsness, first-person pronouns tend to have a higher degree of situational integration than other subject types, linking the scene that is being described to the moment of speaking or introducing new information into the discourse. In his model, the present perfect is more likely to be used in this context because this category typically expresses ‘extra-textual situational reference’ (in contrast with reference to events described in previous discourse) and for ‘an interactive use of language’ (Elsness 1997: 198). Elsness calls this ‘new information’. In contrast, third-person pronouns, as well as NPs and other types of pronouns (like demonstratives), often have an anaphoric function, linking the proposition to the previous context. In Elsness’ model, this closely aligns with the function of the preterit, which is used for ‘given time’, expressing a ‘high degree of textual integration’ (Elsness 1997: 355). Elsness’ results are largely confirmed in our analyses in section 4.2.

A third factor relates to object type. Van Herk (2008, 2010) analyzes the presence of the object in a clause, distinguishing between (non-anaphoric) NPs, pronouns and intransitive or other verbs (also see Davydova 2011: 117 on the effect of transitivity across varieties). According to Van Herk, the type of object in a clause is related to information structure, with non-anaphoric NPs introducing new information and pronouns referring to given information. In line with Elsness’ (1997; see section 4.1.2) model, he argues that clauses with new information (and, thus, with non-anaphoric NPs) are expected to favour the present perfect: ‘New (recent, relevant) information must be specified (with an NP), while given information can be described with an anaphoric pronoun’ (Van Herk 2008: 63). As section 4.2 will show, object type does not reach significance in our analysis.

In addition, we code for the following sociolinguistic factors: individuals’ perceived gender, education level, occupation level and year of birth. These factors have all been shown to influence changes in apparent time (cf. Labov 2001). Table 1 provides an overview of the data sample by these characteristics, showing that the tokens come from a dataset that is relatively balanced with regard to perceived gender and education level. With regard to occupation level, there is less data for the students.

Table 1. Sample characteristics by broad social factors

Figure 2 shows the distribution of the data by year of birth by means of a histogram and boxplot. It shows that most of the data comes from individuals born between 1920 and 1970 and from people born between 1980 and 1995.⁷ Furthermore, analyzing correlations between these sociolinguistic factors (including the individual's year of birth) indicates that there are some correlations in the data: individuals with post-secondary education are also more likely to have a white-collar job (67% of more highly educated individuals have a white-collar job, versus 24% for people of lower education). More problematically, students are on average much younger than people with a different occupation level (mean_blue-collar = 1947, sd_blue-collar = 28 years; mean_white-collar = 1949, sd_white-collar = 23; mean_student = 1990, sd_student = 6). For this reason, interactions between the sociolinguistic variables will not be considered. Instead, we will test each broad sociolinguistic factor as a fixed effect in the statistical modeling.

Figure 2. Distribution of year of birth in the dataset

4 Results

We first use distribution analyses to examine whether the hypotheses from the literature hold in our data. As a next step, we conduct a mixed-effects logistic regression analysis. In the regression model, we also include meta-data information on the individuals to examine the effect of broad social predictors. The data and analyses are available at the following OSF repository: https://osf.io/vfdyh.

4.1 Distributional analysis

Table 2 provides an overview of the number of tokens per adverb, as well as of the number of preterits and present perfects per adverb. The table confirms the findings of Werner (2013: 232) that the frequency of present perfect versus preterit differs between the indefinite time adverbs.

Table 2. Counts and frequency of preterits and present perfects per adverb

In the remainder of this section, we explore the data using distributional analysis and cross-tabulations. These procedures expose important categorical contexts and correlations in the data that have not been reported in the literature previously, which enables us to circumscribe the variable context of the preterit/present perfect alternation in new ways.

4.1.1 Sentence type

Elsness (1997: 321) found that that negated clauses favour the present perfect vis-à-vis the preterit. Our dataset may be biased for sentence type because some of the adverbs under scrutiny are Polarity Items (PI), linguistic constructions that are limited to particular contexts and sentence types. Two types of PIs exist: Positive Polarity Items (PPI), which are linguistic constructions that are not felicitous in negated clauses (Giannakidou 2017), and Negative Polarity Items (NPI), linguistic items that occur only in negated clauses. In the present study, both types of PIs occur: already is a PPI, and ever and yet are two NPIs.

In previous work on NPIs, the contexts that are NPI-licensing have been shown to be broader than simply the presence of negation (e.g. Haspelmath 1997: 33–7; Israel 2011; Giannakidou 2011, 2017). According to Haspelmath (1997: 34), NPIs are also licensed in ‘non-negative contexts that only carry a negative implication of some kind’. Thus, to examine the distribution of the adverbs over different sentence types, we coded each token in the data for a broad set of NPI-licensing contexts described in the literature for ever (Israel 1998; Hoeksema 2010; Giannakidou 2017) that occur in the dataset.⁸ These polarity contexts are reproduced in table 3, together with their frequency.

Table 3. Polarity contexts in the dataset (all examples recorded in Toronto)

Table 4 shows the frequency of each adverb in the dataset by polarity context as detailed in table 3. It confirms that both ever and yet are very restricted to the contexts that are argued to be NPI-licensing and that they hardly occur in affirmative declaratives (N = 5 for yet and N = 10 for ever). Instead, ever occurs in the negated context (N = 485), as well as in all other NPI licensing contexts. The use of yet is more restricted; it predominantly occurs in negated sentences (N = 256). Furthermore, the affirmative declarative contexts where ever and yet occur, as in (16) and (17), are highly colloquial and may even sound ill-formed for native speakers of English (see the Appendix for the full list of tokens with ever and yet in affirmative declaratives). These contexts are excluded from further analysis, resulting in a dataset of 2,749 tokens.

1. (16) But ah, I would say that ranks as a pretty good second vacation I've ever taken. (P. Gilmore, M, 54, Lakefield)

2. (17) I sold some, and I’ve got good start for this year in the barn yet. (C. Winter, M, 87, Temiskaming Shores)

Table 4. Frequency of polarity contexts per adverb

Furthermore, the adverbs that are not NPIs (since and recently) are very frequent in affirmative declaratives. The PPI already occurs only six times in a negated context. In all these negated contexts, already is either in the protasis of a conditional clause as in (18) (N = 5), or is used in a relative clause with no-negation in the main clause as in (19) (N = 1), which may be the reason why the PPI is licensed in these tokens.

1. (18) You should tell anyone you know, if they haven't heard already, they're trying to … break the Guinness Book of World Records. (R. Gruensten, M, 26, Toronto)

2. (19) Like there are no bassists I know around, that haven't already been snatched up. (M. Symanski, M, 19, South Porcupine)

Thus, there is a correlation between the polarity contexts in which each of the adverbs occurs and the sentence type of that context. As a result, we suggest that sentence type as an explanatory variable is impossible to disentangle from adverb as an explanatory variable.

To solve this problem, table 5 shows the proportion of present perfects and preterits per sentence type (affirmative declarative, negation or interrogative), but only for the two adverbs that can occur in affirmative, negative and interrogative contexts, i.e. since and recently. Interrogative clauses are very infrequent in the data (N = 6). Although the vast majority of tokens occur in affirmative contexts, the table reveals a trend: present perfects are much more frequent in negative contexts (85.42%) than in affirmative declarative contexts (57.36%), as was found in Elsness’ study. Because the other adverbs are highly correlated with sentence type we do not include this variable as a predictor in the multifactorial analysis below. We will come back to this point in the discussion.

Table 5. Preterit and present perfect per sentence type for since and recently

4.1.2 Subject realization

Elsness (1997) found that the subject of the clause (plural versus singular and first-/second-person versus third) favours the use of the present perfect instead of the preterit. We coded the grammatical subject of each token in the database into 16 categories (see table 6). Most tokens occur with first-person singulars. This is not surprising given the nature of our data, consisting of conversational interviews in which interviewees often recount narratives of personal experience (cf. Labov 2013: 2). Further, N = 53 tokens in the dataset do not have an explicit subject, due to ellipsis. These tokens are excluded from the remainder of the analysis, resulting in a dataset of 2,696 tokens.

Table 6. Overview of grammatical person (examples from Toronto unless otherwise specified)

Figure 3 shows the proportion of present perfects by grammatical person. Infrequent categories (N < 40) are collapsed as shown on the x-axis. The figure reveals that some of the patterns predicted by Elsness are found in our data as well, while other hypotheses do not hold up. First, there is a difference in the number of perfects for first-/second-person subjects and third-person subjects, with the present perfect used more with (singular and plural) first-/second-person subjects, as predicted. Elsness’ explanation for this finding is that clauses with first-/second-person subjects typically have a higher degree of extra-textual, situational integration (new time), whereas third-person clauses tend to be used for textual integration (given time). In Elsness’ model, the perfect is more suitable for new time, whereas the preterit is more suitable for reference to given time.

Figure 3. Percentage of present perfects (y-axis) per grammatical person (x-axis). The white text in each bar shows the number of tokens per category.

Second, this interpretation may also explain the results for other types of pronouns (indefinites, demonstratives/relatives/interrogatives), which are typically used for anaphoric reference. As a result, the perfect is less frequent with these pronouns than with personal pronouns or NPs. Third, in existential constructions, the present perfect seems to occur relatively frequently. Since these are also constructions that typically introduce new information into the discourse, this is not surprising. Nonetheless, they are sparse in the data (N = 40), making these results less reliable.

One factor that does not seem to play a role in these data is grammatical number. While Elsness found that plural subjects prefer the perfect more than singular subjects, figure 3 does not reveal distinct differences in number.

In the remainder of this article, we use a reduced version of the grammatical person variable. We distinguish six categories: first/second singular (including second-person indefinite, N = 1,593), third singular (N = 290), first/second plural (N = 220), third plural (N = 134), NP (singular or plural, N = 185), other pronoun (N = 274).

4.1.3 Object type

Van Herk (2010), in his diachronic study on the preterit/present perfect alternation in British, African American and Canadian dialects, found that the present perfect is used in increasingly more transitive (but not intransitive) contexts over time. More specifically, he found that the present perfect is favoured in transitive constructions with an NP or pronoun as the object, whereas it is not in intransitive constructions. He explains these findings by referring to the process of grammaticalization, whereby a new construction spreads from more concrete to more abstract contexts. Davydova (2011) argues that transitivity may play only a minor role in varieties where the perfect has fully grammaticalized, although she does not investigate whether there are differences between object types.

In this article, we took a similar approach to Van Herk (2010) and also coded the tokens for object type, initially distinguishing between intransitive verbs (N = 310) and transitive verbs (N = 2,386). For transitive verbs, we coded the object type: NPs as an object (N = 1,089), (personal or demonstrative) pronouns as an object (N = 565), other objects, like subordinate clauses, verbal or adverbial phrases (N = 604), and ellipsis (N = 128).

VPs with a transitive verb are by far the most frequent in the data. Figure 4 shows the proportion of present perfect per object type. The first four bars on the left show the distribution of this variant for the constellations distinguished for transitive verbs. In contrast with Van Herk's findings, the figure shows that present perfects are not more frequent with NPs than with pronouns. In comparison, other objects have heightened use of present perfects. In addition, for most categories, the proportion of present perfects is around 50 per cent, indicating that the object type may not have a large effect on the variation in Ontario data and with the indefinite temporal adverbs. One exception to this pattern is the clauses with ellipsis, where only 40 per cent of the tokens occur with the present perfect.

Figure 4. Present perfect (percentage) with respect to object type

4.2 Statistical modelling

As a next step, we use mixed-effects logistic regression, with random intercepts for the individuals and for the main verbs. All analyses were carried out with R (R Core Team 2021), with the lme4 library (Bates et al. 2015). Verbs that occur less than five times in the data are recategorized as ‘other’. Given the predictions in the literature, that the preterit is encroaching on the present perfect, the model is set up to show the use of the preterit, the form that is expected to be increasing in frequency. As we will see, our results reveal a different trend. Positive significant estimates record a higher probability of the preterit, whereas negative significant estimates record a higher probability of the present perfect. In the maximal model, the independent variables included were: adverb, grammatical person of the subject, object type and their interaction with the individual's year of birth, as well as gender, education level and occupation level. Year of birth is centred around the mean to reduce multicollinearity with higher-order effects (cf. Iacobucci et al. 2016). Using Wald's chi-square tests, we remove non-significant interaction effects and main effects from the model. Analyzing the goodness-of-fit of the model (with regard to influential observations, multicollinearity and heteroskedasticity) does not reveal any problems.

Table 7 shows the results of the final model. This model correctly predicts 78 per cent of the observations, compared to a baseline of 51 per cent, for a model that always selects the most frequent variant. The C index of 0.86 confirms that the model has predictive power. The conditional R-squared value is 0.389 and the marginal R-squared values is 0.275, confirming that about 28 per cent of the variance in the data is explained by the fixed effects alone, while about 11 per cent is explained by the random effects. The following predictors did not reach significance in the multivariate environment: object type, gender and occupation level. Furthermore, none of the predictors except for adverb shows a significant interaction with time. We interpret this to mean that the constraints on the variation between the present perfect and the preterit have remained consistent throughout the twentieth century.

Table 7. Output of the mixed-effects logistic regression model

The effects of year of birth, adverb and the interaction between them are visualized in figure 5. Recall that the decreasing hierarchy of perfect-friendliness from Werner (2013) would predict the preterit occurring more to less frequently in this order: since – already – yet – ever – recently. Even though Werner's hierarchy is based on a different type of dataset, which may also influence the results, figure 5 shows that the regression model only partly confirms the hierarchy. While it shows that there are differences between the adverbs, the preterit is more frequent only with already, ever and recently and not with since. Yet occurs with the preterit the least, and significantly less often than since. Interestingly, Werner's (2013: 238) data for Canadian English also shows that yet is surprisingly infrequent with the preterit (in the ICE corpora). This adverb occurs with the present perfect more in Canadian English than in any other variety; however, this result is based on only 19 tokens for yet in the Canadian ICE-data. Werner (2013: 220–1) explains the strong correlation between yet and the present perfect by appealing to the fact that in most of his tokens, yet occurs in a negative sentence. A similar observation is made by Hundt & Smith (2009: 54). As explained above, the occurrence of negation in a sentence and the adverb that is used are highly correlated. Due to the fact that yet is one of the NPIs that are restricted to very specific contexts (including negation), the claim that yet occurs more frequently with the preterit because of negation is problematic. In fact, for yet, it is impossible (and arguably not necessary since yet occurs only in one sentence type) to disentangle the effect of sentence type from the impact of the adverb itself.

Figure 5. Predicted proportion of preterit by year of birth (centred) and adverb. The error bars show the standard errors.

The pattern for the individuals’ year of birth shows that the distributions of the preterit and present perfect are mostly stable over time. The predicted pattern for already (orange) shows an increase in the use of the preterit but this pattern is not significant.⁹ In fact, the model finds a significant effect only for year of birth with the adverb ever (pink). Importantly, this effect is in the opposite direction to what is predicted in the literature: the proportion of preterits decreases over time.

Next, the regression model confirms the patterns found above for the effect of grammatical person: in comparison with first- and second-person singular personal pronouns, the preterit occurs significantly more with third-person singular. In addition, it also occurs more with pronouns other than personal pronouns. These findings are in line with Elsness’ model where the preterit is predicted to be used for reference to extra-textual, given information. Elsness also predicted that plural first- and second-person pronouns would be combined with the present perfect more frequently, but we find the opposite effect.

Finally, we address the effect of education. In the extensive history of research on the preterit/present perfect alternation in English, there is almost no information on the patterning of variants by social factors. Most studies focus on varietal differences, regional contrasts (e.g. British vs North American) and intra-variety register differences (e.g. spoken vs written). So the significant effect of education level returned by the regression model requires explanation. The model shows that people with post-secondary education use the preterit significantly less often than people with less education; in other words, they use the present perfect significantly more. We hypothesize that these findings are due to more highly educated people having more exposure to the standard variety. While our dataset does not consist of materials from more than one register, there are two reasons why we believe this hypothesis may hold, though further research is necessary.

On the one hand, the use of the present perfect is unchanging in apparent time – at least in the context of most indefinite time adverbs and in spoken Ontario English. We argue that, consistent with Hundt & Smith's (2009: 51, 58) position, the distribution of this variable is the result of stable layering, or stable variation, where multiple linguistic forms may be used to express the same function. As is well known, in cases of stable variation, a contrast between levels of formality, style, register or education is predicted. ‘Stable sociolinguistic variables are essentially monotonic functions of position in the socio-economic hierarchy’, which includes various indicators including education (Labov 2001: 182).

On the other hand, Biber et al. (1999: 462) have reported that the present perfect is more frequent in news reporting than in conversation, suggesting that it is used more in more formal (written) contexts. Thus, the negative coefficients for the use of the present perfect by the more educated individuals in Ontario English may be the result of register adjustments due to increased exposure to the standard language. Further research on a dataset stratified along various registers and with information about speakers’ education levels is needed to confirm this interpretation.

5 Discussion

This article has shown that variation between the preterit and present perfect is largely stable in spoken vernacular Ontario English from speakers born in the twentieth and twenty-first centuries. We examined the effect of language-internal and -external predictors, which allowed us to uncover key generalizations in the ongoing evolution of these variants. After circumscribing the variable context to tokens where the variants are interchangeable, the multivariate patterns reveal that in the context of adverbs of indefinite time, variation between the present perfect and the preterit is – for the most part – a relatively stable system.

Nonetheless, there are some situations where the preterit is preferred over the present perfect, although the regression model indicates that the impact of most of these factors has remained the same over time. First, we find that the preterit is used more frequently in clauses with a third-person singular personal pronoun or with non-personal pronouns than in tokens that contain a first- or second-person singular personal pronoun as the subject. We explained this finding by referring to Elsness’ (1997) model, which argues that the former subjects are more likely anaphoric in nature, thereby referring to given time for which the preterit is well-suited as it integrates sequential pieces of discourse. The latter subjects, in contrast, are more typically used in contexts of higher situational integration, for which the present perfect is also more suitable.

Second, the regression model also indicated that more highly educated speakers use the present perfect more. We explained this finding in terms of stable sociolinguistic variation, whereby synonymous forms may be distributed across the language according to language-external features. Particularly in this case, we hypothesized that more highly educated language users may have more exposure to the present perfect because this variant is, according to Biber et al. (1999), more frequent in more formal registers (news reporting versus conversation).

Third, with regard to the effect of adverb, the distribution of the preterit and present perfect reflects the cline proposed by Werner (2013) but there is one outlier, yet.

In addition to these variables which have a stable effect over time, there is one context in which the distribution between the present perfect and the preterit has changed, namely in the context of superlative-like clauses with ever. Furthermore, it goes in a different direction than expected: ever is becoming more frequent with the present perfect, instead of with the preterit, e.g. Of all the places I ever lived → Of all the places I’ve ever lived. We propose three possible explanations for this finding.

One explanation is that the meaning of ever is particularly well suited for the prototypical uses of the present perfect. It has the meaning of ‘at any time; in any case, in any degree’ (OED, s.v. ever, adv. and adj.), resulting in a natural semantic connection between this adverb and the present perfect. However, this semantic alignment does not explain why the use of ever with the perfect changes over time: our data show that the present perfect with ever becomes more frequent among the younger generation. According to the OED, the meaning of the adverb ever is not a recent semantic evolution: the oldest examples are attested as far back as Old English. Thus, the semantic connection between the meaning of the adverb and the verb form is not sufficient to explain why we find a change over time in the twentieth century.

A second explanation relates to the use of ever in a specific context, namely with a superlative as in (20), (21) and (22).

1. (20) Eighteen pound lake trout, was the biggest one that I’ve ever caught. (Q. Reczynskie, M, 85, Wilno)

2. (21) That was the best gelato I’ve ever had. (E. Gadek, F, 73, South Porcupine)

3. (22) That church … was like the coolest thing I've ever seen. It was massive. Biggest church I've ever seen in my life. (L. Dean-Reynolds, M, 12, Beaverton)

In these examples, the construction with ever arguably emphasizes and boosts the meaning of the clause without necessarily referring to its superlative nature. This use of the construction to express a high degree of some value is very frequent in our data and has even been borrowed into another language, Dutch (Zenner et al. 2018). Furthermore, in some examples, like (22) above, the meaning of the sentence is no longer literal: the church is unlikely to have been the absolute coolest thing the individual has seen in his life, but he describes it as such because it is an impressively big church as he explains in the next sentence. Thus, we have evidence that this use of ever has properties of a construction whose ‘form, meaning or use is not strictly predictable from other aspects of the grammar’ (Goldberg 1995: 13). Therefore, an alternative reason for why ever is becoming more frequent over time with the present perfect may be due to constructionalization, ‘the development of form_new-meaning_new combinations that can be contentful (lexical) as well as procedural (grammatical)’ (Traugott & Trousdale 2013: 22). In the case of ever combining with the present perfect, we suggest that the meaning of the construction has bleached as it became more fixed. This new construction has the meaning ‘intense X’ and has a set grammatical structure with some flexibility with regard to the lexical elements (a similar argument was made in Zenner et al. 2018 for Dutch NPs with borrowed ever), as in (23).

This explanation would hold if the construction is becoming more fixed over time as a distinct form–meaning pair, i.e. (i) the superlative with ever is becoming more frequent over time and (ii) within this construction, ever is becoming more fixed with the present perfect.

Figure 6 shows the frequency of the constructions, i.e. the polarity contexts of ever, in the dataset per decade of birth of the individuals. The figure reveals that, minor fluctuations aside, the superlative context (the solid line) is the only polarity context that is becoming more frequent. The other contexts with ever are more stable, which suggests that the superlative with ever is gaining in popularity.

Figure 6. Frequency of the polarity contexts with ever per individuals’ decade of birth

Next, we need to establish whether the construction with ever is becoming more fixed with the present perfect over time as well. Figure 7 shows the frequency of the present perfect in all the polarity contexts over time. The figure does not completely confirm our second hypothesis, that ever is becoming more frequently used with the present perfect in superlative contexts only. Instead, the increasing use of the present perfect with ever is changing over time in every context. However, this overarching systemic change does not preclude the simultaneous development of a constructionalized entity because it is still possible that the constructionalization of the ever-superlative is further influenced by the fact that ever is becoming more frequent with the present perfect in general.

Figure 7. Proportion of present perfect per polarity context of ever by individuals’ decade of birth

The third explanation bridges across constructionalization and the unique status of NPIs. First, all the NPIs that we have studied either take the present perfect frequently or are changing towards more present perfects. This is true in both Werner's (2013) study and in our own. Second, both since and recently are not NPIs; they take the present perfect more frequently in NPI-licensing contexts (negatives and interrogatives, see section 4.1.1). Both these trends lead us to wonder whether what we are observing in these patterns of change is the grammar's response to functional overlap. According to other scholars, the grammatical system is unlikely to tolerate ‘doublets’, that is, two forms that are not functionally differentiated (Kroch 1994). Where ‘competing variants’ (Kroch 1994: 17) exist, either one form takes over the grammatical space of the other or specialization results in the two variants carving up the functional space in distinct ways. Given that this study started by circumscribing the variable context to ‘truly variable’ contexts of the preterit/present perfect alternation, the trajectories of change in the data nicely mesh with this possibility.

6 Conclusion

Much of the previous literature on the preterit/present perfect alternation – both over a long time span (Elsness 1997) and in more a limited diachronic scope (Vanneck 1958, Yao & Collins 2012) – has argued that the preterit is taking over functions that were previously reserved for the present perfect alone. However, this is not what we have found: our results mirror the findings of Hundt & Smith (2009) and Werner (2014), who have argued that the preterit/present perfect alternation has been stable over the last century. The apparent-time perspective from a corpus of spoken vernacular Ontario English provides further evidence for this conclusion: for speakers born between 1879 and 2001, variation between the preterit and the present perfect can be considered stable. In a few contexts (e.g. depending on the adverb or grammatical person), one or the other variant is preferred, yet sociolinguistic factors play a minor role. Only one context shows a significant diachronic development, namely an increase in present perfect constructions with the adverb ever, as in the coolest thing I've ever seen.

The findings of our study add to the existing literature in three ways. First, it is the first to analyze spoken vernacular data from Ontario, Canada, thereby adding information about a register and variety of English that has not been investigated in detail in the extant literature. Second, we have considered contexts where both variants are possible, namely in the context of indefinite time adverbs, obviating reliance on subjective interpretations. Finally, use of the apparent-time construct has allowed us to uncover a specific change in progress, but not of the preterit encroaching on the present perfect. Instead, present perfect forms are increasingly used in constructions with ever. Outside this minor area, the present perfect is robust and steadfast in vernacular Ontario English over the course of the twentieth century, continuing the longitudinal specialization of the preterit/present perfect contrast. Taken together, these results provide another example of the importance of including vernacular speech in research on language variation and change and of the unique contribution certain constructions make to more general systems of grammar.