Constant Linguistic Effects in the Diffusion of Be Like

Abstract

This article examines change in social and linguistic effects on be like usage and acceptability. Results from two studies are presented. The first set of data comes from a trend study with samples of U.K. university undergraduates collected in 1996 and 2006. While the effect of subject person, morphological tense, and quote content is constant in the two samples, the effect of speaker sex decreases. The second study is a judgment experiment with 121 native speakers of U.S. English, examining acceptability of be like in environments biasing direct speech and reported thought readings. The analysis reveals no interaction between age and the reported thought/direct speech contrast, suggesting no support for change in this effect on be like acceptability in apparent time. The two studies therefore converge in suggesting no evidence of change in linguistic constraints on be like as it has diffused into U.K. and U.S. Englishes.

Keywords

quotatives be like British English real time sociolinguistic language variation and change

The spread of be like as a quote introducer has received much attention over the past fifteen years and has provided an unprecedented scope for linguists to examine extremely rapid change across generations (Buchstaller 2006a, 2006b; Buchstaller & D’Arcy 2009; Cukor-Avila 2002; Dailey-O’Cain 2000; Ferrara & Bell 1995; Macaulay 2001; Romaine & Lange 1991; Tagliamonte & D’Arcy 2004, 2007; Tagliamonte & Hudson 1999). As much recent literature has noted, the process of the spread of be like now has sufficient time depth to allow for detailed comparisons across age groups (Buchstaller 2006a; Cukor-Avila 2002; Ferrara & Bell 1995; Tagliamonte & D’Arcy 2004, 2007). Much of the recent literature on be like has therefore focused on whether some well-known constraints on English quote introducers have changed through the course of diffusion of be like in various dialects. The most careful and detailed studies of this sort have been Tagliamonte and D’Arcy’s (2004, 2007) studies of change in Toronto English, which indicate ongoing change in social and linguistic constraints on be like as the form enters the local grammar. Most notably, among later age cohorts, Tagliamonte and D’Arcy find an increase in the speaker sex effect with women favoring be like and a weakening of the effect of quote content, where internal dialogue (reported thought) contexts favor be like over direct speech contexts. Tagliamonte and Hudson (1999:167-169) and Tagliamonte and D’Arcy (2004:511, 2007:202) suggest that these patterns represent general pathways of grammaticalization of be like; that is, it’s a change that, ceteris paribus, all dialects into which be like is diffusing will undergo.

The goal of this article is to assess evidence for similar changes in constraints on be like in varieties of English where quotatives have been less studied. In particular, this article considers two sets of data—one with U.K. English speakers and a second with U.S. English speakers—and tests whether the interaction between age group and other well-known social and linguistic effects on be like is more general in nature. The first set comes from a trend study comparing quotative usage in a 2006 corpus of speech from University of York undergraduate students aged eighteen to twenty-two with those from a similarly constructed sample from 1996. This comparison allows us to assess possible cross-generational change in constraints on be like use in York. The second set of data comes from a judgment experiment, conducted in 2009 with speakers of American English, comparing acceptability of be like and say in contexts biasing direct speech and reported thought readings.

Our results suggest two main findings. First, the corpus study shows a significant interaction between sex and age group with a smaller sex effect in the later data set. These results are in keeping with previous studies suggesting mutability of social factors across age groups and dialects as be like expands (Buchstaller & D’Arcy 2009; Ferrara & Bell 1995; Tagliamonte & D’Arcy 2007). Second, the corpus data and the experimental results both reveal no interaction between age and the classic linguistic constraints on be like discussed in much previous literature. The contrast between our findings and previous corpus findings from other locales therefore supports skepticism with regard to universal tendencies in constraint shifts in grammaticalization of be like. Instead, our results support findings from the historical syntax literature of a constant rate effect on syntactic change—that is, for any single abstract process of grammatical change, a constancy in the effect of different contexts across generations of speakers advancing the change (Kroch 1989, 1994, 2001).

Our discussion is organized as follows. We begin by discussing previous literature examining change on constraints on be like usage. We then present our two original studies focused on change in linguistic effects on be like. The final section summarizes some implications of the data presented.

The Expansion of Be Like

In contemporary English speech there is considerable variation in verbs which can be used to introduce direct speech as in (1)–(3).

(1) I was like “Easy tiger.”

(2) She said “Let’s go.”

(3) He went “Calm it love.”

Over the past two decades an extensive body of literature has focused on the expansion of the be like variant in (1). This literature has documented the spread of be like in several English-speaking societies globally starting in the United States (Cukor-Avila 2002; Ferrara & Bell 1995; Singler 2001), and later in Canada (Tagliamonte & D’Arcy 2004, 2007; Tagliamonte & Hudson 1999), the United Kingdom (Buchstaller 2006a; Macaulay 2001; Tagliamonte & Hudson 1999), and New Zealand (Buchstaller & D’Arcy 2009).

Following work by Ferrara and Bell (1995) and Tagliamonte and Hudson (1999), much of this literature has focused on continued grammaticalization of be like during this process of expansion. In an effort to track these changes, several recent studies have examined changes in the way be like use is constrained internally and externally in corpus data. Below we introduce four constraints frequently discussed in this literature.

Speaker Sex

Several studies have reported a speaker sex effect on be like usage. The most typical finding in the literature is that women tend toward innovative be like more than men (Blyth, Recktenwald, & Wang 1990; Ferrara & Bell 1995; Macaulay 2001; Tagliamonte & D’Arcy 2004; Tagliamonte & Hudson 1999). This pattern is in keeping with evidence from perceptual data. In particular, Dailey-O’Cain’s (2000) U.S. study and Buchstaller’s (2006b) U.K. study both suggest that quotative be like use is associated with young women.

Much of the literature, moreover, indicates that sex effects on quotative use are mutable diachronically and across communities. In a three-year trend study in Texas in the early 1990s, Ferrara and Bell (1995) found evidence of neutralization of the sex effect: in 1990 women used be like twice as frequently as men (15 vs. 29 percent); however, in later, similarly constructed samples in 1992 and 1994, men and women used be like at roughly equal rates. During this period, the overall rate of be like use increased steadily. These data might be taken to indicate that as be like diffuses, the effect of speaker sex may weaken. However, Tagliamonte and D’Arcy (2004) describe the opposite pattern in Canada: with increasing use of be like, a sex difference emerged between 1995 and 2002–2003. Buchstaller and D’Arcy (2009) also find conflicting results for gender across American, British, and New Zealand corpora. Based in part on these results, Tagliamonte and Hudson (1999) and Tagliamonte and D’Arcy (2004) hypothesize the opposite relationship between diffusion of be like and sex differentiation: the further be like diffuses, “the more likely it is to differentiate male and female speech” (Tagliamonte & Hudson 1999:167).

Quote Content

A second constraint on be like usage discussed in recent literature concerns the interpretation of the quoted material. The availability of be like in contexts such as (1) appears to have emerged through a reanalysis of sequences of be + discourse marker like. That is, prior to its emergence as an introducer of direct speech, be like could be used to describe states of individuals in contexts with predicate adjectives (4), with nonlexicalized sounds as in (5), and in “internal dialogue” as in (6).

(4) I was like devastated.

(5) She was like “Ugh.”

(6) I was like “Never again.”

In studying the increase of quotative be like, several authors have examined its distribution across age cohorts in some of the above contexts. Early studies report that be like is disfavored in contexts introducing direct speech and favored before nonlexicalized sounds and internal dialogue (Ferrara & Bell 1995; Macaulay 2001; Tagliamonte & Hudson 1999). More recently, however, Tagliamonte and D’Arcy (2004) show that for one of the three age groups in their 2002–2003 sample (seventeen- to nineteen-year-olds), speakers used be like to a greater extent in direct speech than internal dialogue. Similarly, Tagliamonte and D’Arcy (2007) demonstrate that young speakers, aged seventeen to twenty-nine, showed the weakest effect of quote content in their sample. The authors plausibly interpret these results as evidence of continued grammaticalization of be like as an introducer of direct speech.

Subject Person

Results reported in the literature concerning the effect of subject person on the variation have been more consistent. As it emerged as a dialogue introducer, be like appears to have been originally favored with first-person respondents, a fact plausibly related to its role as an introducer of internal dialogue. Ferrara and Bell (1995) report that this effect weakens during the period covered by their study, and they interpret this shift as evidence of expansion of function of be like. Most other studies, however, have reported much greater constancy in the effect of subject person, with first-person respondents favoring be like usage and second- and third-person respondents disfavoring it (Buchstaller & D’Arcy 2009; Cukor-Avila 2002; Tagliamonte & D’Arcy 2004, 2007; Tagliamonte & Hudson 1999).

Tense and Aspect

Finally, several authors have focused on the effect of tense on be like usage. In particular, Blyth, Recktenwald, and Wang (1990), Romaine and Lange (1991), and Singler (2001) all report that be like is favored in present tense contexts and disfavored in the past tense. More recently, Tagliamonte and D’Arcy (2007) use a three-way coding for tense, distinguishing past, present, and historical present. Their analysis of variation in a corpus of speakers ranging in age from nine to thirty-nine in Toronto indicates that, across several age groups, historical present contexts most consistently favored be like use, followed by present tense contexts and finally past tense contexts. Buchstaller and D’Arcy (2009) corroborate this pattern for the corpora of American and New Zealand English as well; for their U.K. corpus past contexts were most favored. Our study uses a two-way distinction focusing solely on the tense of the verb; the token numbers of present are too few in one of our data sets (twelve tokens, or 2.4 percent) to examine it separately.

To our knowledge, none of the literature on grammaticalization of be like has touched on variation and change in the aspectual behavior of be like. In particular, the claim that be like is undergoing a process of reanalysis from stative be + discourse like to a quotative verb akin to accomplishment verbs such as say and go suggests the possibility that be like will come to behave like the latter in progressive environments. That is, stative verbs like have, know, and be are unlike activity and accomplishment predicates in that they are canonically poor in progressives, as illustrated in (7)–(11).

(7) Tina is eating the sandwich.

(8) Terry was smoking.

(9) *I’m having money.

(10) *I’m knowing French.

(11) *Terry is being ill.

If be like is indeed coming to behave syntactically like go and say, then we might expect it to begin to appear in progressive contexts, as in (12)–(14) from the 2006 data set.¹ We examine this possibility below.

(12) The cat is going “Meow.”

(13) She was going “Don’t touch.”

(14) He was saying “Uhm oh we’ve got to move her.”

Tables 1 and 2 summarize the evolution of the above four constraints on be like as described in two influential sets of real-time studies of quotatives in North America: Ferrara and Bell’s (1995) data from Texas and Tagliamonte and D’Arcy’s (2004, 2007) data from Toronto.

Table 1.

Ferrara and Bell’s (1995) Constraint Evolutions for Be Like

Constraint	Early stages	Later stages
Sex	Women > men	Neutralization
Subject person	First > third	Expansion into third person
Content	Internal dialogue > direct speech	Expansion into direct speech

Table 2.

Tagliamonte and D’Arcy’s (2007) Constraint Evolutions for Be Like

Constraint	Early stages	Later stages
Sex	Women > men	Increased differentiation
Subject person	First > third	Constancy of effect
Content	Internal dialogue > direct speech	Expansion into direct speech
Tense	Present tense favored	Historical present favored

Our goals in view of the foregoing literature are twofold. First, we aim to examine to what extent the patterns of change in be like described in these North American varieties are also observed in England. Second, we aim to assess evidence for change in constraints on be like in real time using evidence from a trend study of usage and a controlled judgment study.

A Real-Time Study of Quotative Change in the United Kingdom

Data and Method

To test Tagliamonte and D’Arcy’s predictions about grammaticalization of be like, we compare patterns of variation in quotative usage in two data sets, which we describe in turn below.

We gratefully acknowledge the assistance of Sali Tagliamonte and Rachel Hudson in providing us with data from the York storytelling corpus, first reported on in Tagliamonte and Hudson (1999). This data set is a corpus of one-on-one sociolinguistic interviews with forty-four University of York undergraduates collected in the summer of 1996. The interviewers were fellow University of York undergraduates working under the direction of Sali Tagliamonte. Tagliamonte and Hudson (1999) do not specify the ages of the participants in this corpus; however, in view of University of York undergraduate enrollment during this period, it is likely that all or nearly all of the respondents were between the ages of eighteen and twenty-two. All of the forty-four respondents, who were evenly divided by sex, were native speakers of U.K. English. These data produced 397 tokens.²

For the purpose of inferring generational change in quotative use, we compare Tagliamonte and Hudson’s 1996 data with a similarly constructed data set gathered in the spring of 2006. This data set consists of one-on-one interviews of thirty-one University of York undergraduates—fourteen women and seventeen men—collected by the latter four authors. The interviews were conducted using a standard battery of questions intended to elicit maximally unselfconscious narratives, comparable to the data in Tagliamonte and Hudson’s 1996 corpus. All the respondents are native speakers of U.K. English. From transcripts of these interviews, all instances of quoted speech were extracted and coded following Tagliamonte and Hudson’s (1999) procedure. These data yielded 955 tokens of quotatives.

To be able to fully compare the two data sets and to address the above issues to do with grammaticalization, we defined our token set slightly differently from some previous studies. First, tokens of quotative form it’s like, as in (15), where the subject is impersonal, were excluded from the analysis.

(15) As soon as he came back in the room, it’s like “Oh no!” (2006 data set)

Although these tokens are found in both data sets and at roughly equivalent rates (twenty-six tokens, i.e., 5 percent of the total in 1996; and forty-one, i.e., 4 percent of the total in 2006), following Tagliamonte and D’Arcy’s (2004:504) procedure we excluded it’s like tokens as they have “an exceptional status.” The occurrence of the pronoun it, for example, differentiates be like from other quotative verbs, which can appear only with subjects that are in some way animate. The construction is thus not fully variable.

Second, to facilitate comparison with Tagliamonte and D’Arcy’s (2004, 2007) results, we excluded tokens of nonlexicalized sounds. This removed thirty-nine tokens in the 1996 data set and eighty in the 2006 data.

Third, because we are studying the effect of morphological tense on variation in quote introducers, we excluded non-tense-bearing forms including zero quotatives, infinitives, and participles. This culling removed 90 tokens from the 1996 data set and 127 tokens from the 2006 data set. This protocol departs from previous studies but was required for the present data set to fit a coherent model (Tagliamonte & D’Arcy 2004, 2007; Tagliamonte & Hudson 1999). Removing these tokens allowed us to focus more closely the factors affecting be like.

Finally, our initial analysis of the use of quote introducers with progressives and participial adjuncts, as in (16) and (17), showed that this context was not fully variable.

(16) The cat is going “Meow.” (2006 data set)

(17) And he was standing there going “Ooh not again girls.” (2006 data set)

Recall that the use of be like in these contexts could be interpreted as a sign of further grammaticalization of the quotative system. In these environments, in which state predicates are canonically poor, be like was found to be strongly disfavored, although there were some signs of increase of use with be like. The 1996 data set contains 94 such examples (about 23 percent of the overall number of tokens), none of which occur with be like. In the 2006 data set, there are 105 such tokens (accounting for 12 percent of the overall number), two of which contain be like.³ For these reasons, we treated this variation as categorical and excluded all tokens in these environments from the analysis.

The near categorical absence of be like in these environments suggests an important limit on the extent to which be like has come to be reanalyzed as a quotative on a par with accomplishment verbs like say and go. The presence in our data of these two tokens of be like progressives demonstrates that for some speakers, at least, be like may indeed behave as a true event (nonstate) predicate.⁴ Nevertheless, the relative scarcity of such tokens indicates that be like’s conservative competitors remain preferred in usage in unambiguously eventive contexts in this sample.⁵ We return to these semantic considerations later in this article.

For our analysis to be as comparable as possible to previous multivariate results, we follow an approach used first by Tagliamonte and Hudson (1999) and subsequently elsewhere (Tagliamonte & D’Arcy 2004, 2007; Buchstaller & D’Arcy 2009), whereby all of be like’s main competitors, including go, say, and think, are included in the envelope of variation. Think and say, are not, of course, semantically equivalent since the former tends to introduce reported thought quotes and the latter direct speech. We note, however, that the adoption of this approach has no consequence for our main measures of interest, namely, the interaction between age and the internal and external constraints introduced above.

Results

A comparison of these two data sets suggests a sharp increase in be like usage over the ten years between 1996 and 2006. Among York undergraduates, be like has become the most frequently used quotative, increasing from 19 percent in 1996 to 68 percent in 2006. This increase comes at the expense of be like’s main competitors—say, think, and, to some extent, go—all of which show lower rates of use in the 2006 sample than in the 1996 data. Use of other quotatives—shout, and others—is marginal in both samples. Figure 1 provides an overall distribution of five of the most frequent quotatives in our two data sets.

Figure 1.

Overall distribution of variants for two data sets

A similarly vertiginous increase in be like use is also reported in a study of Canadian youth by Tagliamonte and D’Arcy (2004). This study compares patterns of quotative use in two corpora of sociolinguistic interviews—one consisting of data from University of Ottawa students in 1995 and a second from Toronto youth aged ten to nineteen, collected in 2002 and 2003. Like the present York data, Tagliamonte and D’Arcy’s Canadian data show that be like has become the most frequent quotative used in narratives among young people, rising from 13 percent in the 1995 Ottawa data to 63 percent in the Toronto data for seventeen- to nineteen-year-olds (the age group closest to the Ottawa University students). We reproduce Tagliamonte and D’Arcy’s results in Figure 2. The present data, then, suggest that be like is diffusing into the speech of British youth with a pace similar to that described in the literature on North American Englishes.

Figure 2.

Tagliamonte and D’Arcy’s real-time results from Canada

A further goal of this article is to test change in real time in the effect of the well-studied social and linguistic constraints on be like use described earlier. To this end, we fit a logistic regression model using both data sets and tested for significant interactions between our sample factor (with levels 1996 sample and 2006 sample) and our four social and linguistic factors: speaker sex, subject person, quote content, and morphological tense. The model is summarized in Table 3, which shows the contribution of different factors to be like usage. The first column lists the factors in the model. The second column gives log odds ratios for the treatment level given in brackets,⁶ and the next three columns provide the standard error, z score, and p value for these coefficients.

Table 3.

Contribution of Factors to Be Like Usage

Factor	Coeff.	SE	Wald z	p
Intercept	−0.5547	0.1773	−3.13	.0018
Sample (2006)	1.7704	0.1905	9.29	.0000
Sex (male)	−1.5562	0.2879	−5.40	.0000
Person (third)	−0.8101	0.1293	−6.26	.0000
Tense (present)	0.8448	0.1718	4.92	.0000
Content	—	—	—	ns
Sample:sex^a	1.1479	0.3224	3.56	.0004
Sample:person	—	—	—	ns
Sample:content	—	—	—	ns
Sample:tense	—	—	—	ns

With treatment levels 2006 sample and male.

We illustrate these effects in Figure 3, which shows proportional use of be like versus other quotatives by condition for the two samples.

Figure 3.

Proportional use of be like versus other quotative verbs by speaker sex and context

The analysis summarized in Table 3 reveals a significant main effect for sample. As also illustrated in Figures 1 and 3, be like is much more frequent in the 2006 corpus than the 1996 corpus. In addition, Table 3 shows a significant main effect for sex, with women favoring be like, and a significant interaction between sex and sample. Figure 3 demonstrates that the effect of speaker sex in the 1996 sample is much greater than in the 2006, suggesting an attenuation of the sex effect over time.

These data are surprising from the perspective of Tagliamonte and D’Arcy’s (2004, 2007) studies of quotative use in Toronto, Canada, where the youngest speakers, with the highest rates of be like use, showed the strongest sex effects. Older speakers in the sample, whom Tagliamonte and D’Arcy suggest were among the be like innovators in the community a generation earlier, used be like relatively little and showed a much weaker sex differentiation. Based on these data, Tagliamonte and D’Arcy propose that sex differentiation in be like use has emerged in Toronto as it continues to diffuse into local speech.

The present results, in contrast, suggest a mild form of neutralization of the speaker sex effect (Ferrara & Bell 1995). These findings, alongside Tagliamonte and D’Arcy’s (2004, 2007) results, imply that the diachronic development of the sex effect in be like usage is variable across communities. One possible explanation of this U.K.–Canadian difference in sex effects is that it represents different stages of change for be like. Much of the diachronic literature has suggested that sex effects vary across the trajectory of change, with strong sex associations appearing at early stages of change and a weakening effect appearing at later stages (Labov 2001:308; Tagliamonte & D’Arcy 2009). From this perspective, one possibility is that the diffusion of be like is slowing in the United Kingdom but not Canada, which would explain why the effect of speaker sex is becoming neutralized.⁷ We are aware of no independent evidence in support of this supposition, however.

The divergent sex effects in the United Kingdom and Canada are nevertheless in keeping with some matched guise evidence suggesting a much stronger gender association for be like in North America than in Britain. In particular, in Dailey-O’Cain’s (2000) U.S. study, 80 percent (24/30) of respondents identified be like with women. In a similarly constructed study in the United Kingdom, Buchstaller (2006b) found that only 34 percent (65/191) of respondents associated be like with women. Our data lend further support to Buchstaller’s suggestion of cross-societal differences in the social meaning of using be like. As be like continues to diffuse globally, the social meaning associated with its use does not necessarily diffuse along with the surface form. Rather, individual communities adapt the innovation in the context of local social and economic conditions and local symbolism (Eckert 2000).

The quote content data in Table 3 are also unexpected from the perspective of much previous literature in that the analysis returned no significant main effect or interactions for quote content. Figure 3 shows that, in both of the above samples, be like is used more frequently in reported thought contexts than with direct speech, but this difference does not contribute significantly to the model. This result, again, contrasts with Tagliamonte and D’Arcy’s (2004, 2007) Canadian findings, which propose an expansion of be like into direct speech contexts over time. In particular, Tagliamonte and D’Arcy report that the youngest and most advanced set of be like users—the seventeen- to nineteen-year-olds—favor be like in direct speech contexts and disfavor it in internal dialogue. The present data provide no evidence of a change in the effect of the reported thought/direct speech contrast. The fact that the analysis summarized in Table 3 and Figure 3 returned no main effect for quote content suggests the possibility that the effect of quote content weakened in England at an earlier stage relative to overall usage of be like than in Canada. That is, in Canada, this effect weakens substantially only among cohorts for which be like is the majority variant in spoken interview-style data. In the U.K. sample, this effect is fairly weak even among the earlier users for whom be like is a minority variant (see Figures 1 and 2). We return to these issues later.

The effect of subject person in Table 3 and Figure 3 is negligible. The analysis returned a significant main effect for subject person, with first-person respondents favoring be like. No significant interaction with sample was returned, indicating no evidence of inconstancy in the effect of subject person between the two samples, as illustrated in Figure 3. This result is expected from the perspective of most previous studies, which suggest a remarkable consistency of the subject person effect across contexts (Buchstaller & D’Arcy 2009; Tagliamonte & D’Arcy 2004, 2007; Tagliamonte & Hudson 1999).

The tense results in Table 3 are also in keeping with results in previous literature. Table 3 and Figure 3 show that morphological present tense favors be like while past tense favors other quotatives (Singler 2001; Tagliamonte & D’Arcy 2007). No significant interaction with sample year was returned, suggesting no inconstancy in the effect of tense over time.

From the perspective of the literature reviewed above, the most important outcome in these data is the difference between speaker sex and the three linguistic factors—tense, subject person, and quote content—in terms of their interaction with sample group. Again, the data show a weakening of the sex effect from 1996 to 2006, but a constancy of the three linguistic factors. The mutability of the speaker sex effect in itself is unremarkable in light of much previous literature on be like and other phenomena showing that the way linguistic variation indexes social meaning often changes over time in a community as the variable undergoes social reanalysis and constant symbolic reappropriation (Dyer 2002; Eckert 2000, 2008; Tagliamonte & D’Arcy 2004). The fact that the sex effect in our trend study has changed in the direction opposite to that indicated in Tagliamonte and D’Arcy’s studies is in keeping with results from Buchstaller and D’Arcy’s (2009) multicommunity comparison. This suggests that, as be like has diffused geographically, social constraints on quotative variation often change from community to community, as patterns of linguistic variation map onto local social and stylistic differences in community-specific ways.

The constancy of the linguistic factors contrasts with findings from much of the literature implying that there is a change in the effect of some of these constraints (Ferrara & Bell 1995; Tagliamonte & D’Arcy 2004, 2007). The gap between these findings and previous corpus results from other locales therefore supports skepticism with regard to universals of grammaticalization of be like that dictate change in linguistic effects in be like, particularly quote content. We consider these issues further in light of judgment data presented in the following discussion.

Direct Speech and Nonspeech Interpretations of Be Like Quotatives in Judgment Data

The second set of data for this study comes from a judgment experiment focusing on age effects on direct speech and reported thought interpretations of be like and say quotatives. As discussed above, early work on be like described it not as an introducer of direct speech but rather exclusively as an introducer of reported thought. This variant of be like may have emerged as a reanalysis of descriptions of states of individuals in sequences of be + focuser/discourse marker like + predicate adjectives or nonlexicalized sounds (Butters 1982; Tannen 1986) (see examples in (4) and (5)).

Much subsequent corpus-based work on be like however has reported that quotes introduced by be like could be used to describe not just states of individuals as in (18a), but also saying eventualities as in (18b).

(18) Aaron was like “Ok, fine.”

a. Aaron thought/felt like saying “Ok, fine.”

b. Aaron said “Ok, fine.”

A disadvantage of usage corpora for analyzing semantic variation of this kind is that the intended reading can be difficult to discern from the context. To complement the corpus-based findings discussed above, we report on a controlled judgment experiment intended to examine cross-speaker differences in the availability of be like quotatives in contexts biasing these different readings. If younger speakers are coming to reanalyze be like quotative predicates as descriptions of speech events rather than reported thought, then we expect age to interact with the reported thought/direct speech difference in judgments of quotative sentences. The following discussion describes an experiment designed to test this claim. Acceptability judgments are a different kind of linguistic performance from speech in a sociolinguistic interview context, which raises the question of the comparability of judgment data and speech corpus data for the present purposes. Some recent research in fact has suggested that judgments of acceptability are sensitive to and in fact closely mirror the relative probabilities of semantically equivalent competing forms in production (Bresnan 2007; Bader & Häussler 2010a, 2010b; Melnick, Jaeger, & Wasow 2011). We assume, based on these studies, that acceptability judgments are indeed a suitable kind of data for the purpose of inferring cross-speaker differences in linguistic effects on quotative variation.

Data and Method

The participants were 121 self-described native speakers of American English aged eighteen to seventy-three (M = 31.3, SD = 11.6)—71 women and 50 men.⁸ All had at least some university education, and in this respect our sample is similar to those reported on in several previous studies on be like change (Ferrara & Bell 1995; Dailey-O’Cain 2000; Tagliamonte & Hudson 1999). Previous literature on be like has indicated no strong regional effects on be like usage within the United States, and no effort was made to stratify the sample by region. No regional effects were observed in the analysis. Participants were recruited online through the contacts of the researchers and were not paid for their participation.

The experiment compares scores for be like and say sentences in six environments. A first, baseline context was created with no stativity/eventivity bias, as in (18). Four additional contexts—progressives, imperatives, force . . . to complements, and pseudoclefts with do—were used as ways of biasing eventive readings; all of these are contexts in which eventive predicates are fine, but true states are poor (Dowty 1979). We illustrate this contrast in (19)–(22), which compare stative have $100 with eventive spend $100 in each environment.

(19) She was *having $100/spending $100. (progressives)

(20) Just *have $100/spend $100. (imperatives)

(21) Tim forced him to *have $100/spend $100. (force . . . to)

(22) What she needs to do is *have $100/spend $100. (do pseudoclefts)

In the following experiment, we use these environments to compare acceptability of eventive, direct-speech readings of be like and say, as illustrated in (23)–(26).

(23) She was being like/saying, “They’re coming tomorrow at 11:00” (progressives)

(24) Just be like/say, “They won’t ever do it.” (imperatives)

(25) Tim forced him to say/be like, “Fine, I’ll do it next week.” (force . . . to)

(26) What she needs to do is say/be like, “John already quit.” (do pseudoclefts)

The final environment biased nonspeech be like readings using for adverbials. As illustrated in (27), temporal for phrases are fine with atelic predicates in simple tenses but poor with eventives (Dowty 1979).

(27) For an hour, Mark had $100/*spent $100. (for adverbials)

We use such contexts to diagnose the availability of stative, nonspeech interpretations of be like and say quotative predicates, as in (28).

(28) For an hour, Mark was like/said, “Let’s go to McDonald’s.” (for adverbials)

Two lexicalizations were created for each environment, each assigned either to a be like or say condition yielding two test sets. Each participant therefore saw each condition once. Respondents were randomly assigned to test sets, and a unique random order of the twelve test sentences and eighteen fillers was created for each respondent. A list of the experimental sentences is provided in the appendix.

The data were gathered through a self-paced online magnitude estimation procedure using WebExp2 software (Mayo, Corley, & Keller 2008) in the summer of 2009. In syntactic magnitude estimation experiments, respondents judge stimulus sentences not on an abstract n-point scale but rather in relation to a positive numerical score arbitrarily assigned to a benchmark (“modulus”) sentence (Bard, Robertson, & Sorace 1996; Keller & Sorace 2003). If the stimulus sentence is judged to be twice as acceptable as the benchmark sentence, the participant gives it twice the benchmark score; if it is half as acceptable, half the benchmark score; and so on. In the present experiment, the benchmark sentence used was that in (29), which native speakers of English typically find to be of intermediate well-formedness.

(29) I wouldn’t give to the boy the difficult puzzle.

After they had given consent to participate, respondents were asked to provide some background information, including age, sex, highest level of education completed, and hometown. Respondents were then introduced to the magnitude estimation procedure and then given two sets of slides providing practice in applying this technique. In the first set, respondents used magnitude estimation to measure lengths of lines; the second set provided sample sentences to judge. The experimental phase followed, which respondents typically completed in five to ten minutes.

Following Bard, Robertson, and Sorace’s (1996) procedure, raw scores were normalized by dividing them by the benchmark score. The base-ten logarithms of these scores were then taken to make data normally distributed and suitable for parametric tests. In the following discussion, we report these normalized, log-transformed values.

Results

To examine the effect of speaker age on acceptability scores, we fit mixed-effect linear models for each condition using the lme4 package for R (Bates & Maechler 2010; R Development Core Team 2008). The dependent variable was the log-transformed values for each condition, with age and verb as fixed effects and subject and item as random effects. Following Baayen’s (2008) procedure, p values were simulated by Markov chain Monte Carlo sampling (ten thousand samples) using the LanguageR package for R (Baayen 2008, 2010). To examine cross-generational difference in acceptance of be like in each of these environments, we focus not on the effect of age but rather on the age–verb (be like vs. say) interaction. We chose this measure to account for a possible age effect in preferences toward direct speech versus reported speech. The results are summarized in Figure 4, which plots say and be like scores by respondent age for each condition and reports a p value for the verb:age interaction variable. Zero on the y-axis corresponds to the respondent’s value for the benchmark sentence in (28). Values above zero, therefore, reflect a judgment for the stimulus sentence better than that for (29), and scores below zero reflect a judgment below that for (29).

Figure 4.

Say and be like scores by age for six conditions

The plots in Figure 4 show, unsurprisingly, that the say-be like acceptability gap is inversely correlated with age. Respondents under thirty years of age in the sample accept be like in each of these six conditions, roughly on a par with say, indicating that be like is natural for many young speakers in contexts biasing both eventive and stative interpretations of be like.

Nevertheless, the interaction between age and verb reaches significance at α = .05 only for three of these environments: the baseline context, pseudoclefts, and imperatives. The interaction for force . . . to complements is suggestive at p = .052. For for-adverbials there is no interaction between age and verb, and in fact no main effect for verb. These judgment data therefore align only partially with corpus data suggesting diffusion of be like in direct speech and nonspeech contexts. The absence of more consistent age effects in these data may be partially attributable to the fact that our sample is relatively youthful, with a mean age of 31.1.

More directly relevant to the issues considered here is the fact that the regression lines for the be like conditions in Figure 4 are all roughly parallel. To examine possible interaction between age and the speech/nonspeech contrast, we fit four separate mixed effect linear models with scores for for-adverbials and each of the four speech event-biased conditions as levels in a fixed factor; age was an additional fixed effect with random effects speaker and item. The analyses returned no significant interaction at α = .05 for any of the four comparisons, suggesting no evidence of an age difference in the effect of the direct speech/reported thought contrast (for for adverbials vs. force . . . to as the eventive/stative comparison, p = .542; for adverbials vs. pseudoclefts p = .316; for adverbials vs. progressives p = .930; and for adverbials vs. imperatives, p = .317). These judgment results from U.S. English speakers align with the corpus results presented earlier in that they suggest no change in the effect of the reported thought/direct speech contrast over time, which diverges from the findings reported in Ferrara and Bell (1995) and Tagliamonte and D’Arcy (2004, 2007). The contrast between our findings and previous corpus findings from other locales again supports skepticism toward strong universal pathways of grammaticalization of be like.

Conclusion

Results from two studies—one a usage study and the other a controlled judgment study—suggest no interaction between age and the typical linguistic effects described in foundational work by Ferrara and Bell (1995) and Tagliamonte and Hudson (1999). Most importantly, results from our studies fail to support an age difference in the effect of the direct speech/reported thought contrast as reported in Tagliamonte and D’Arcy (2004, 2007). From the perspective of Tagliamonte and D’Arcy’s proposals and like-minded work suggesting rigid grammaticalization pathways in syntactic change (Jespersen 1917; Roberts and Roussou 1999; Saxena 1995), one possible interpretation of the above results, alluded to earlier, is that the Toronto dialect sits at a different point in the trajectory of grammaticalization from the varieties represented by our U.S. and U.K. samples. Again, the multivariate analysis summarized in the section examining the longitudinal data turned no significant main effect or interactions for quote content. This result suggests the possibility that the leveling of the quote content effect hypothesized by Tagliamonte and D’Arcy came about at an earlier stage than in Canada relative to overall usage of be like. This would mean that the reanalysis responsible for the change in contextual effects on be like is independent of overall frequency of usage of be like, a plausible assumption.

While the constancy in linguistic effects in the two data sets reported on here is surprising from the perspective of some previous corpus findings, it is expected from the perspective of findings in the quantitative diachronic syntax literature. Much previous quantitative work on historical corpora has shown that for any single abstract process of syntactic change, contextual effects are typically constant over time—a phenomenon known as the constant rate effect (Fruehwald, Gress-Wright, & Wallenberg, in press; Kroch 1989, 1994, 2001; Pintzuk 1991; Santorini 1992).

In particular, Kroch (1989, 2001) attributes this constancy to individuals’ language-independent faculty for tracking frequencies of experienced events. As learners acquire and increment new forms, they learn from input sources the relative propensities of use of variants in different contexts, with the consequence that contextual effects are propagated across generations of speakers, all other things being equal. Occasionally, linguistic factors can come to interact with social factors in new ways, which may have the effect of changing the effects across time, but this is the exception rather than the rule, to judge from the published literature (Kroch 1989, 2001). Kroch and colleagues’ model and the present results lead us to anticipate that the typical pattern of evolution for be like will not involve changing linguistic constraints as in Tagliamonte and Darcy’s (2004, 2007) findings but rather will be characterized by consistency in linguistic effects on be like. Future work might usefully address this possibility.

Footnotes

Appendix

Acknowledgements

We are grateful to the respondents of this study and to Sali Tagliamonte and Rachel Hudson for sharing their 1996 data set with us. Thanks also to Helen Lawrence and colleagues at the University of York for help in contacting potential informants. Thanks to Alex D’Arcy, Tony Kroch, Sali Tagliamonte, and audiences at NWAV, DiGS, LingEvid 2010, and University of Glasgow for comments on some of the material presented here. We are responsible for all errors.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research and/or authorship of this article: This research was supported by ESRC Grant 061-25-0033.

Notes

Bios

Mercedes Durham is a lecturer at the University of Aberdeen. Her work focuses on language variation and change in English, particularly in terms of the acquisition of variation by children and nonnative speakers.

Bill Haddican is an assistant professor at CUNY Queens College. His work focuses on language variation and change and syntax in dialects of Basque and English.

Eytan Zweig is a lecturer at the University of York. His work focuses on the semantics and pragmatics interface, from both theoretical and psycholinguistic perspectives.

Daniel Ezra Johnson received his PhD in sociolinguistics from the University of Pennsylvania. His work focuses on dialectology and language change as well as quantitative modeling of language data.

Zipporah Baker, David Cockeram, Esther Danks, and Louise Tyler are former undergraduates in linguistics in the Department of Language and Linguistic Science at the University of York.

References

Baayen

R. Harald

. 2008. Analyzing linguistic data: A practical introduction to statistics. Cambridge, UK: Cambridge University Press.

Baayen

R. Harald

. 2010. LanguageR library for R. http://cran.r-project.org/ (10 September 2011).

Bader

Markus

Häussler

Jana

. 2010a. The primacy of grammaticality. Konstanz, Germany: University of Konstanz manuscript.

Bader

Markus

Häussler

Jana

. 2010b. Toward a model of grammaticality judgments. Journal of Linguistics 46. 273-330.

Bard

Ellen

Robertson

Dan

Sorace

Antonella

. 1996. Magnitude estimation of linguistic acceptability. Language 72(1). 32-68.

Bates

Doug

Maechler

Martin

. 2010. Lme4 package for R. http://cran.r-project.org/ (10 September 2011).

Blyth

Carl

Recktenwald

Sigrid

Wang

Jenny

. 1990. I’m like, “say what?!”: A new quotative in American oral narrative. American Speech 65. 215-227.

Bresnan

Joan

. 2007. Is syntactic knowledge probabilistic? Experiments with the English dative alternation. In Featherston

Sam

Sternefeld

Wolfgang

(eds.), Roots: Linguistics in search of its evidential base, 77-96. Berlin: de Gruyter.

Buchstaller

Isabelle

. 2006a. Diagnostics of age-graded linguistic behaviour: The case of the quotative system. Journal of Sociolinguistics 10(1). 3-30.

10.

Buchstaller

Isabelle

. 2006b. Social stereotypes, personality traits and regional perception displaced: Attitudes towards the “new” quotatives in the UK. Journal of Sociolinguistics 10. 362-381.

11.

Buchstaller

Isabelle

Alex

D’Arcy

. 2009. Localized globalization: A multi-local investigation of quotative like. Journal of Sociolinguistics 13. 291-331.

12.

Butters

Ronald

. 1982. Editor’s note [on “be like”]. American Speech 57. 149.

13.

Cukor-Avila

Patricia

. 2002. She say, she go, she be like: Verbs of quotation over time in African American Vernacular English. American Speech 77.3-31.

14.

Dailey-O’Cain

Jennifer

. 2000. The sociolinguistic distribution of and attitudes toward focuser like and quotative like. Journal of Sociolinguistics 4.60-80.

15.

Dowty

David R.

1979. Word meaning and Montague grammar. Dordrecht, Netherlands: D. Reidel.

16.

Dyer

Judy

. 2002. “We all speak the same round here”: Dialect levelling in a Scottish-English community. Journal of Sociolinguistics 6. 99-116.

17.

Eckert

Penelope

. 2000. Linguistic variation as social practice: The linguistic construction of social meaning in Belten High. Oxford, UK: Blackwell.

18.

Eckert

Penelope

. 2008. Variation and the indexical field. Journal of Sociolinguistics 12.453-476.

19.

Ferrara

Kathleen

Bell

Barbara

. 1995. Sociolinguistic variation and discourse function of constructed dialogue introducers: The case of be like. American Speech 70. 265-289.

20.

Fruehwald

Josef

Gress-Wright

Jonathan

Wallenberg

Joel

. In press. Phonological rule change: The constant rate effect. In Kan

Seda

Moore-Cantwell

Claire

Staubs

Robert

(eds.), NELS 40: Proceedings of the 40th meeting of North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association.

21.

Jespersen

Otto

. 1917. Negation in English and other languages (Historiskfilologiske Meddelelser I.5). Copenhagen: A.F. Høst.

22.

Keller

Frank

Antonella

Sorace

. 2003. Gradient auxiliary selection and impersonal passivization in German: An experimental investigation. Journal of Linguistics 39. 57-108.

23.

Kroch

Anthony

. 1989. Reflexes of grammar in patterns of language change. Language Variation and Change 1. 199-244.

24.

Kroch

Anthony

. 1994. Morphosyntactic variation. In Beals

Katharine

(ed.), Proceedings of the 30th annual meeting of the Chicago Linguistic Society, 180-201. Chicago: Chicago Linguistic Society.

25.

Kroch

Anthony

. 2001. Syntactic change. In Baltin

Mark

Collins

Chris

(eds.), The handbook of contemporary syntactic theory, 699-730. Oxford, UK: Blackwell.

26.

Labov

William

. 2001. Principles of linguistic change: Social factors. Oxford, UK: Blackwell.

27.

Macaulay

Ronald

. 2001. You’re like “why not?”: The quotative expression of Glasgow adolescents. Journal of Sociolinguistics 5. 3-21.

28.

Mayo

Neil

Martin

Corley

Frank

Keller

. 2008. WebExp2. www.webexp.info/ (5 December 2011).

29.

Melnick

Robin

Florian Jaeger

Wasow

Thomas

. 2011. Speakers employ fine-grained probabilistic knowledge. Pittsburgh: Linguistic Society of America paper.

30.

Pintzuk

Susan

. 1991. Phrase structures in competition: Variation and change in Old English word order. Philadelphia: University of Pennsylvania dissertation.

31.

R Development Core Team. 2008. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.

32.

Roberts

Ian

Roussou

Anna

. 1999. A formal approach to “grammaticalization.” Linguistics 37. 1011-1041.

33.

Romaine

Suzanne

Lange

Deborah

. 1991. The use of like as a marker of reported speech and thought: A case of grammaticalization in progress. American Speech 66. 227-279.

34.

Santorini

Beatrice

. 1992. Variation and change in Yiddish subordinate clause word order. Natural Language & Linguistic Theory 10. 595-640.

35.

Saxena

Anju

. 1995. Unidirectional grammaticalization: diachronic and cross-linguistic evidence. Sprachtypologie und Universalienforschung 48. 350-372.

36.

Singler

John

. 2001. Why you can’t do a Varbrul study of quotatives and what such a study can show us. University of Pennsylvania Working Papers in Linguistics 7. 257-278.

37.

Tagliamonte

Sali A.

Alex

D’Arcy

. 2004. “He’s like, she’s like”: The quotative system in Canadian youth. Journal of Sociolinguistics 8. 493-514.

38.

Tagliamonte

Sali A.

Alex

D’Arcy

. 2007. Frequency and variation in the community grammar: Tracking a new change through the generations. Language Variation and Change 19(2). 199-217.

39.

Tagliamonte

Sali A.

Alex

D’Arcy

. 2009. Peaks beyond phonology: Adolescence, incrementation, and language change. Language 85(1). 58-108.

40.

Tagliamonte

Sali A.

Rachel

Hudson

. 1999. Be like et al. beyond America: The quotative system in British and Canadian youth. Journal of Sociolinguistics 3. 147-172.

41.

Tannen

Deborah

. 1986. Introducing constructed dialogue in Greek and American conversational and literary narrative. In Coulmas

Florian

(ed.), Direct and indirect speech, 311-332. Amsterdam: de Gruyter.