‘When intuitions (don't) fail’: combining syntax and sociolinguistics in the analysis of Scots

Abstract

A perennial problem for sociolinguists interested in morphosyntactic variation is that such forms are often low frequency, making quantitative analysis difficult or impossible. However, sociolinguists have been generally reluctant to adopt methodologies from syntax, such as acceptability data gleaned from speaker intuition, due to the belief that these judgments are not necessarily reliable. In this article we present data from the Scots Syntax Atlas, which employs sociolinguistic methodologies in spoken data alongside the results of acceptability judgments. We target three morphosyntactic variables and compare and contrast these across the two data types in order to assess the reliability of the judgment data at community level. The results show that reliability is variable-dependent. For some variables, there is clear correlation; with others, it appears that, as Labov (1996) phrased it, ‘intuitions fail’. We discuss how factors such as salience, social stigma and local identity combine to govern the reliability of judgment data.

1 Introduction

At the core of the variationist paradigm is the quantitative analysis of spontaneous speech, where variable use of forms is the focus in uncovering structured heterogeneity (Weinreich, Labov & Herzog 1968). While there is a wealth of research on phonetic variation (see e.g. Drager & Ketig 2021 for an overview), work on morphosyntactic variation is comparatively uncommon in variationist sociolinguistics (for discussion, see e.g. Rickford et al. 1995: 106; Cheshire 1999: 59; Cornips & Corrigan 2005: 2) despite wide-ranging variation, such as (1a–g).

One reason for the lack of studies is frequency: many of these forms would not occur frequently enough in spontaneous speech for the type of quantitative analysis employed in the variationist paradigm. A potential solution is for sociolinguists to employ methodologies used by syntacticians, specifically ‘elicited judgments, the intuitions of the native speakers’ (Labov 1996: 77), in which speakers are asked to introspect on what they believe themselves to say. However, sociolinguists have remained sceptical of introspective data (Labov 1972; Carden & Dieterich 1981; Bender 2001; Green 2010), even going so far as to suggest it may ‘fail’ (cf. Labov's (1996) ‘When intuitions fail’) to capture the linguistic situation, in that speakers may ‘agree that a certain form is completely unacceptable, yet use it themselves freely in every-day speech’ (Labov 1996: 78). This impasse has meant many morphosyntactic variables remain understudied in sociolinguistic research, leaving a significant gap in our knowledge of variation across all levels of the grammar. This leads Rickford (2019) to suggest that the utility of introspective data should be revisited, and specifically:

What is needed is a concentrated effort to determine what kind of intuitive judgments are more robust than others, what factors influence their variability, and what methods we might use for calibrating them against observational and other evidence. (Rickford 2019: 102)

In this article we address these issues using data from the Scots Syntax Atlas, herein SCOSYA (scotssyntaxatlas.ac.uk) (Smith et al. 2019). SCOSYA includes two types of data:

1. 1. Likert-scale responses to a questionnaire of 200 dialect syntax forms given by over 500 speakers (‘acceptability judgments’)

2. 2. A total of 275 hours of conversation gathered through sociolinguistic interviews with the same speakers (‘spontaneous spoken data’)

With access to both these data types from the same speakers, we can assess the reliability of the judgment data against spoken data. In doing so, we address the questions of where and why speakers’ acceptability judgements depart from their use of variables in everyday speech, and more broadly, how judgment tasks, and other introspective methods, can be used in combination with variationist analysis in future sociosyntactic research.

2 Contrasting methods in the analysis of morphosyntactic variation

2.1 Spontaneous speech

Vernacular speech is considered the gold standard in variationist sociolinguistics, primarily collected through the sociolinguistic interview (Labov 1972), which aims to mitigate the Observer's Paradox: how to tap the most unmonitored style of speech when a participant is being monitored. The Principle of Accountability (Labov 1966: 49) is also key to variationist study, and in particular, the linguistic variable, where a particular form is studied in relation to the form(s) it is in variation with. Moreover, the linguistic variable must be ‘high in frequency’ (Labov 1984: 49) for systematic patterns of variation to emerge. While debates on equivalence are no longer at the forefront of discussion (Lavandera 1978; Buchstaller 2009), the issue of frequency with morphosyntactic variables remains, as they ‘often involve special semantic and pragmatic circumstances which may occur rarely or unpredictably in interview’ (Rickford et al. 1995: 106). For example, Henry's (2005: 1609) work on the after-perfect in Belfast English found only three tokens in 720,000+ words, and Coats (2023: 706) finds 2.4 double modals per one million words in a corpus of Scots, despite both forms being robustly attested as part of the respective varieties. Studies on morphosyntactic variation thus often converge on a limited set of variables which appear frequently enough for analysis, including agreement (e.g. Cheshire & Fox 2009; Rupp & Britain 2019), relative pronouns (e.g. Guy & Bayley 1995; Tagliamonte, Smith & Lawrence 2005) and negative concord (e.g. Cheshire 1982; Smith 2001).

2.2 Acceptability judgments

Acceptability judgments have been the go-to methodology for theoretical syntacticians since the 1960s (Chomsky 1966),² and their reliability for investigating at least standard English has been experimentally demonstrated (e.g. Sprouse & Almeida 2017; Langsford et al. 2018; Goodall 2021). The methodology has been extensively employed in the study of non-standard morphosyntactic variation, from early introspective work (Rizzi 1982; Poletto 1993) to recent large-scale crowdsourced projects (e.g. Zanuttini et al. 2018). The judgment methodology has been refined for non-standard varieties, highlighting factors such as participant selection, conducting the task in the local dialect, and oral presentation of stimuli (Cornips & Corrigan 2005; Cornips & Poletto 2005; Barbiers & Bennis 2007; Buchstaller & Corrigan 2011), and some studies have shown that judgment data can usefully complement spoken data. For example, Henry (2005) compares the judgments of nine Belfast English speakers on a set of examples with non-standard agreement patterns, to the agreement patterns found in a 130-hour corpus of Belfast English speech. She finds that the corpus gives useful information about frequency, but the judgments present an opportunity to find out what is ungrammatical as opposed to simply infrequent. However, concerns about whether reported judgments of acceptability truly reflect community speech patterns remain (Henry 2005; Green 2010; Eide & Åfarli 2020).

What factors might influence the accuracy of judgments in non-standard varieties? In the next section, we summarise Labov (1996), who proposes reasons as to why speaker intuitions, as expressed through judgment tasks, might ‘fail’ to match up with speech patterns.

2.3 Why might intuitions fail?

Labov (1996: 85) provides an overview of some linguistic variables where comparison between such intuitions and spontaneous speech is available.³ One such case is positive anymore (2), ‘roughly equivalent to nowadays’:

1. (2) Cars are sure expensive anymore.

Despite widespread use in conversation data, speaker intuitions about the form were ‘very erratic indeed’ (Labov 1996: 84), with some speakers denying knowledge of the form despite using it in the same interview.

Another variable, ain't, has a complex set of sociolinguistic rules which provided a good testing ground for judgment data. With ain't for isn't (3), there is a direct correlation between intuitions and production: speakers in communities across the US produce it (Labov et al. 1968; Feagin 1979; Hazen 1996), and recognise it as grammatical (Labov 1996: 90).

1. (3) He ain't too smart.

However, in African American English (AAE), ain't can also be used as a variant of didn't (4). In Labov's data, AAE speakers recognise (4) as grammatical to the same extent they do unattested forms, such as (5) in which ain't is used for don't. ⁴ Both receive 40–50 per cent acceptance.

1. (4) He ain't see her yesterday.

2. (5) I ain't really wanna do that.

Two issues arise for the use of introspective judgments here: rejection of forms attested in speech (positive anymore, ain't for didn't), and acceptance of forms not found in speech data (ain't for don't). From these observations, Labov (1996: 98) proposes reasons why speakers’ intuitions might ‘fail’:⁵

1. I. A ‘socially superordinate norm’ may take precedence because:

   1. a. The availability of a non-regionally specific variant may override judgment of a regional form;

   2. b. There may be social stigma attached to the form.

2. II. Infrequency: ‘judgments that a form is ungrammatical may actually be motivated by the fact that it is rare’ (Labov 1996: 99).

Regarding (Ia), as we saw above, intuitions about positive anymore were ‘erratic’ among speakers who used it. Labov states that ‘the social bias [against anymore] is not at all obvious’ (1996: 98), but nevertheless the non-regionally specific nowadays overrides speakers’ ability to introspect about anymore. Regarding (Ib), social stigma is complex, involving factors such as class (e.g. Niedzielski & Preston 2003), race (e.g. Alim, Rickford & Ball 2016), region (e.g. Preston 1989), age (e.g. Cheshire 2005). What is considered prestigious varies across communities (Coupland & Bishop 2007), although the ‘standard’ variety tends to hold sway across the board. Furthermore, stigma is closely related to social salience (Labov 1994, 2001; Labov et al. 2011), and to the divisions between indicators (socially stratified but not identifiable by speakers), markers (controlled depending on speech context) and stereotypes (‘the overt topics of social comment’) (Labov 1994: 78). While markers and especially stereotypes may be stigmatised, indicators are, by hypothesis, below the level of consciousness.⁶ The salience of a feature, and its possibility of being stigmatised, is therefore relevant to understanding why intuitions might fail. The influence of stigma on the results of judgment tasks can be seen in Blanchette's (2017: 2) study of negative concord. Although her participants report negative concord as unacceptable due to ‘heavy social stigma’, they ‘have grammatical knowledge’ of it, demonstrated in subtly different Likert-scale acceptability judgments for different types of sentences.

In terms of (II), absolute frequency may not be particularly important for acceptability. For example, it-clefts in English appear in ‘less than one tenth of a percent of all sentences’ (Roland, Dick & Elman 2007: 353) but, given appropriate pragmatic contexts, are judged more acceptable than unclefted sentences (Destruel, Beaver & Coppock 2019). This seems to apply also to non-standard features: for example, Wood et al. (2020: 4) find that although the datives they investigate (1e) are ‘not frequently attested in written forms or even spoken corpora’, they are ‘robustly accepted’ in judgment tasks.

However, evidence exists for correlation between acceptability and the relative frequency of variants – although the picture gets more complicated at lower frequencies (Featherston 2005; Kempen & Harbusch 2005; Arppe & Järvikivi 2007; Bresnan 2007; Divjak 2017; Bader & Häussler 2010; Bermel & Knittl 2012). What is considered ‘high frequency’ varies, from over 50 per cent (Bermel & Knittl 2012) to just 3 per cent (Bader & Häussler 2010), but what is consistent is that relatively frequent variants receive high acceptability ratings. On the other hand, grammatical but rarely produced variants are neither categorically accepted nor rejected. Acceptability rates ranged from 20 to 60 per cent for low-frequency grammatical options in Bader & Häussler's (2010) study of German word order, compared to over 80 per cent  acceptability for more frequent grammatical word orders, and < 10 per cent acceptability for ungrammatical word orders. This gives a good starting point for considering how relative frequency might impact the results of judgment tasks, though there may be additional factors at play when considering a non-standard variant alongside standard ones.

2.4 Summary

A range of factors may affect speakers’ abilities to introspect about the acceptability of morphosyntactic features, including frequency and social stigma. However, it remains unclear when these factors come into play and how they combine to influence the results of acceptability judgment tasks. Furthermore, there have been considerable adaptations made to judgment tasks to collect data from non-standard varieties. It is unclear to what extent the factors discussed in the context of earlier research continue to influence these adapted judgment tasks.

In the remainder of the article, we address these questions by comparing judgment data with spontaneous spoken data from the same speakers. In section 3, we introduce the social context of Scots and Scottish English in Scotland, SCOSYA, and its data collection methods, before comparing three different phenomena in section 4.

3 The Scots Syntax Atlas

3.1 Scots and Scottish English in Scotland

Historically, Scots was a distinct Germanic language within the British Isles, spoken in Lowland Scotland. However, over the centuries following the political union of Scotland and England in the 1600s, the role and prestige of English within Scotland grew. In the present day, Scots – which is distinct from English lexically, morphosyntactically and phonetically – exists on a continuum with (Standard Scottish) English. The existence of the continuum at an interspeaker level intersects with various sociodemographic issues – e.g. broader Scots is often associated with working-class speakers – but on an intraspeaker level, many speakers of Scots style shift along the continuum depending on their interlocutors and the linguistic task at hand. Use of broad Scots in a professional context may be liable for disapproval, but so may use of Standard Scottish English when speaking to members of a speaker's own community. For more information on the continuum and the context of Scots in Scotland, see e.g. Aitken (1984), Johnston (2007), Smith (2012).

In Aitken (1984), linguistic features of Scots are divided into covert, overt and vulgarisms, generally mapping to Labov's indicators, markers and stereotypes. Many morphosyntactic features of Scots are covert – used by speakers across class boundaries ‘unself-consciously’. Speakers are often unaware of alternative (standard English) ways to express the same meaning (Aitken 1984: 18). For example, Scots covertly employs modal verbs can and will in place of may or shall respectively. Overt Scots features are used freely at the broad Scots end of the continuum, but can also be employed stylistically in more Standard Scottish English ‘on occasions when it seems desirable to claim membership of the in-group of Scots’ (Aitken 1984: 22); for example, in proceedings of the Scottish Parliament, using the Scots-specific form kent in place of the (Standard Scottish) English knew in ‘Burns kent better’ (quoted in Corbett & Stuart-Smith 2012: 86). Finally, vulgarisms are features of Scots that are often considered ‘Bad Scots’ and condemned across social contexts (Aitken 1984: 24). These tend to be variables that are most associated with urban Scots varieties and/or younger speakers, despite the fact that they may be used across social and regional groups – such as the pronoun youse (see also Corbett & Stuart-Smith 2012 for how these different categories are realised in both spoken and written data).

It is important to remember when considering Aitken's distinctions that there is considerable variation within Scots itself, with distinct regional varieties found across the country. In different regions, there are different features that will be overt and covert, and that may have different levels of social stigma attached to them (see e.g. Smith 2001).

Recalling Labov's (1996) proposed reasons as to why ‘intuitions may fail’, detailed in section 2.3 above, the social context of Scots and English in Scotland therefore provides us with an ideal testing ground to tease apart the most important factors. Features of Scots’ morphosyntax of course vary in frequency, but there are also distinctions between highly local features and region-wide non-standard features, as well as standard (English) ways of ‘saying the same thing’, allowing us to explore the role of regionality. There is also detailed existing research into the salience and perception of many grammatical variables across Scots varieties (Aitken 1984), allowing us to take account of how social stigma may affect judgments. Attempting to combine these factors leads us to the choice of features investigated in this study, detailed in section 4. Before this, we present SCOSYA and the data that will be used in this article.

3.2 Overview of SCOSYA

The Scots Syntax Atlas (SCOSYA) builds on the growing enterprise of dialect syntax atlases, e.g. SAND (Barbiers & Bennis 2007), ASIS (Benincà & Poletto 2007), the Nordic Syntax Database (Lindstad et al. 2009) and YGDP (Zanuttini et al. 2018). While these atlases were built primarily from judgment data, SCOSYA also incorporates the practice of a number of dialect syntax corpora, e.g. AAPCAppE (Tortora et al. 2017), gathering both judgment data and spoken data from the same speakers.⁷ In total, SCOSYA combines 275 hours of sociolinguistic interview data with over 100,000 acceptability judgments across 200 morphosyntactic phenomena found in varieties of Scots. The data were collected between December 2015 and July 2018 from 530 speakers in 146 locations across Scotland (figure 1).

Figure 1. All locations in the SCOSYA data

We divide the data into 15 broad geographic regions, based on traditional dialectological areas of Scotland (Grant 1931; Aitken 1984; Johnston 1997; Miller 2007). See scotssyntaxatlas.ac.uk/linguists-atlas for more information.

The data were collected by community insider fieldworkers, as insider status is key to gaining access to the relevant non-standard forms (Labov 1972). The fieldworkers were responsible for recruiting participants using a standard set of sociolinguistic criteria (Labov 1984): participants were born and brought up in the area; had not spent any significant time away; had parents who were from the area; and had not gone on to higher education.⁸

Dialect syntax atlases have tended to focus on older speakers to capture the ‘traditional’ dialects of a particular language or region. When younger speakers have been included, such as in the Nordic Dialect Corpus, there is no direct comparison with older speakers in the same community. In SCOSYA, pairs of participants were recruited in two age groups: 65+ and 18–25 (thus four participants in each location), in order to systematically investigate change in apparent time. Each pair generally consisted of friends or family members, and were usually also friends or family members of the fieldworker. Data collection was conducted in participants’ homes to encourage relaxed, naturalistic data in a comfortable setting (Labov 1984). By following these criteria in the data collection process, the SCOSYA data hones in on participants’ most vernacular speech, rather than a more formal style that might elicit the Standard Scottish English end of their speech continuum. We now detail the two data collection methods used.

3.3 The judgment task

The first part of the data collection process was an acceptability judgment task using a questionnaire developed by the SCOSYA team. The questionnaire targeted key non-standard morphosyntactic forms. All were features of Scots dialects; we use the term ‘non-standard’ to indicate the range of variation within the forms, from hyperlocal dialect features to more general English ‘vernacular universals’, that were tested. For example, some features were noted in the literature to occur throughout Scotland, such as possessives or determiners where standard English would have a bare noun (6). Other examples were attested for particular dialect areas, such as gonnae imperatives (7), attested in Glasgow. Some pandialectal forms found in varieties of English both across the UK and beyond were also included, such as non-standard preterit verb forms (8).

1. (6) I'm going to my bed.

2. (7) Gonnae you leave us alone!

3. (8) I seen that last week.

A total of 182 features were tested with all participants. Additional items of specific interest were judged in certain locations only, so each questionnaire contained approximately 200 examples. In total, over 100,000 judgments were given.

The questionnaire was delivered in an adapted version of the interview method (Barbiers & Bennis 2007). Each target example was presented following a short context which included relevant referents and contextual information (9):

1. (9) You're telling me you saw me and a friend earlier. You say:

   I saw youse earlier on.

The fieldworker read out the contexts and examples to the first participant in each pair. The participant verbally rated each example on a five-point Likert scale (figure 2), indicating their intuitions about whether/how the sentence would be used in their community. Each point on the scale was labelled, and participants were given a copy of the scale to refer to throughout the task. Scores were recorded by the fieldworker. The first participant subsequently adopted the role of ‘fieldworker’ and delivered the questionnaire to the second participant.

Figure 2. SCOSYA five-point Likert scale

3.4 The interview

The second part of the data collection process was a standard one-hour sociolinguistic interview, conducted between the pairs of speakers in each age group, with the fieldworker present. Due to the insider status of the fieldworker, the Observer's Paradox was reduced as far as possible, and conversation was generally relaxed and open (see further in Labov 1984).

Example (10) demonstrates the kind of talk that arose in these interviews, with bold indicating some of the non-standard forms:

The recordings were text-to-speech transcribed in Transcriber (Barras et al. 2001). The full corpus of over 3 million words is available at scotssyntaxatlas.ac.uk/about/accessing-the-spoken-corpus

4 The data

4.1 Selecting the variables

We focus on three variables (11a–c) which differ across geographic and social dimensions in Scots. Unless indicated, all examples are taken from the SCOSYA corpus.

Choosing to investigate these features allows us to differentiate the roles of factors suggested by Labov (1996) as leading to the ‘failure’ of intuitions. Regarding frequency, while we do not have prior information on the relative frequency of variants, we do know about their likely absolute frequency. Variable contexts where need + PAST can possibly occur are known to be ‘low frequency’ (Strelluf 2022) but possible contexts for periphrastic div may be much higher. Negative concord (NC) may be situated somewhere in between.

The features also vary with regard to regionality. Div is highly local within communities, while need + PAST is supralocal across Scots varieties and NC is supralocal across Englishes. Crucially, all also have a ‘standard’ variant, which speakers may also have access to.

Finally, the features vary with regard to their salience and social stigma. Need + PAST is recognised as a covert feature of Scots (e.g. Aitken 1984:21). It is therefore unlikely to be salient, nor stigmatised. We know div is a socially salient feature of Tyneside English (Rowe 2007: 362; Pichler 2009: 290); however, there is little evidence regarding the salience of div in Scots. The features defining div as salient for Tyneside varieties of English – grammaticalisation to set phrases, strong usage by young working-class men (Rowe 2007; Pichler 2009) – may not be relevant for div in Scots. Our third feature, NC, is one of the most socially salient variants in any variety of English (Cheshire 1982; Labov 2001; Smith 2001; Anderwald 2005; Blanchette 2017). Aitken (1984:25) categorises NC as a Scots vulgarism, subject to ‘explicit condemnation’, while Smith & Holmes-Elliott (2022) find speakers in Buckie change their rates of NC if speaking to a community insider as opposed to an outsider. This controlled usage highlights its salience, and the associated stigma.

We now provide a more detailed description of what is already known about each of these features, before presenting the results of the SCOSYA corpus search for each and contrasting it with the judgments.⁹

4.2 need + PAST

4.2.1 Background

In English, a passive construction can be embedded under a matrix verb. Usually, the infinitival form to be combines with the past participle (12a). It is, however, also possible to have a present participle follow the matrix verb (12b).

There is a third variant available in some varieties in which certain matrix verbs (need, want and like (Murray & Simon 2002)) combine directly with a past participle (12c–e).¹⁰

We adopt the term need + PAST (Strelluf 2020) for this construction, as we focus purely on cases with need.

need + PAST constructions have been attested in some US Englishes (Stabley 1959; Murray, Frazer & Simon 1996; Maher & Wood 2011; Duncan 2021), and Irish English (Montgomery 1997). In Scots, need + PAST is understood as a feature of Scots generally (Brown & Millar 1980; Montgomery 1997); we might thus expect it to be attested and judged highly across Scots varieties.

4.2.2 need + PAST: spoken data

All three variants are used in the SCOSYA spoken corpus:

Table 1 shows how these are distributed across the data. Table 1 reveals that need + PAST (13a), the non-standard form, makes up over 57 per cent of examples. In this case, the non-standard variant is produced at the highest rate in the corpus.

Table 1. Proportion of need constructions in the SCOSYA corpus

Figure 3 plots where need + PAST occurs. The faded circular dots indicate communities where conversation was recorded, but no attestations of need + PAST were found. The black dots indicate communities where need + PAST was attested. Figure 3 shows that attestations of need + PAST are spread across Scotland. We note that absence of a form does not necessarily mean no use, but instead may arise from the general infrequency of any of the three constructions with need.

Figure 3. Locations with attestations of need + PAST in SCOSYA

4.2.3 need + PAST: judgment data

Participants judged one example of a need + PAST construction.

1. (14) The postman pulls up in his van and it's filthy. You say:

   His van needs washed.

Recall that participants judged items on a 1–5 scale. ‘5’ indicated the participant would ‘definitely say’ it and it was ‘very natural’, while ‘1’ indicated they ‘wouldn't say’ it, and it was ‘very unnatural’. We treat locations where two or more participants gave an example 4 or 5 as ‘acceptable’, following e.g. Zanuttini et al. (2018) and Thoms et al. (2019). We do not treat the data as continuous, and report median and mode statistics throughout.

Participants generally gave the example in (14) high ratings. The median and mode scores were 5, while the individual-level acceptance rate (percentage of speakers who rated the example 4 or 5) was 80 per cent. Figure 4 shows the widespread acceptance, with dark dots indicating locations where at least two participants rated the example 4 or 5.

Figure 4. His van needs washed, accepted by at least two participants

Comparing figures 3 and 4 demonstrates that speakers’ intuitions align with respect to need + PAST. It is both accepted and used throughout Scotland, despite the low absolute frequency of the construction in the corpus.

4.3 Div for do

4.3.1 Background

In some varieties of Scots, div varies with periphrastic do in the present tense in all subject types except third-person singular (15a–e).

Div can be used in negatives (16a), questions (16b) and tags (16c) but not imperatives (16d).

In terms of geographic spread, div is attested in the North East (Smith 2000; Dictionary of the Scots Language 2004) and the Borders.

4.3.2 Div: spoken data

In the corpus, div is found in emphatic positives (17a), questions (17b) and tags (17c).

There are over 15,000 possible contexts in which div could alternate with do. Table 2 shows the relative frequency of the two forms.

Table 2. Proportion of periphrastic do/div in the SCOSYA corpus

Despite thousands of contexts where div could be used, it is very rarely used. Figure 5 plots the attestations. The faded circular dots indicate communities where conversation was recorded, but div was not attested. The black dots indicate communities where div was attested. Figure 5 shows the majority of attestations are in the North East. In the Borders, we also find attestations. The distribution of div in the SCOSYA corpus aligns with what we would expect based on previous reports of the form.

Figure 5. Locations with attestations of div in SCOSYA

4.3.3 Div: judgment data

Participants judged one example of div in an emphatic polarity context.

1. (18) You're sitting down with a cup of tea and a scone. You say:

   I div like a scone!

Figure 6 shows the distribution of judgments. In figure 6, the dark dots indicating acceptance are geographically clustered. While div was not rated highly across the whole data set (median and mode both 1), in the North East, there is a median and mode of 5, while in the Borders the mode is 5 and the median is 3.5.

Figure 6. I div like a scone, accepted by at least two participants

Just as with need + PAST, the spoken data in figure 5 and the judgment data in figure 6 align with respect to div, suggesting reliable intuitions regarding this feature.

4.4 Negative concord

4.4.1 Background

Standard English includes two ways of marking negation with indefinites. With sentential negation (19a), the negative marker not or -n't appears after the verb, and scopes over an indefinite any- form. Alternatively, the negative is incorporated into the indefinite, realised as a no- form (19b). A further alternative exists in non-standard dialects of English: negative concord (NC).¹¹ In (19c), negation is marked on the negative marker -n't and the indefinite, nothing.

Despite denigration of forms such as (19c) over centuries, NC is widespread and ‘recurs ubiquitously all over the world’ (Chambers 1995: 242), with numerous contemporary reports (e.g. in Britain, Hughes & Trudgill 1979; Cheshire 1982; Coupland 1988; Beal 1993; Edwards 1993; Anderwald 2005). Examples like (19c) have been reported across Scots varieties (Macaulay 1991; Cheshire et al. 1993; Smith 2001; Anderwald 2005; Macafee 2011), although at varying rates (e.g. 49 per cent in the North East (Smith & Holmes-Elliott 2022), but 8 per cent in Glasgow (Childs 2017)). Based on previous research, we may expect NC would be produced and judged acceptable in Scots varieties, but perhaps at variable rates across communities.

Note that there are two forms of contracted negative marker in Scots, the broad Scots form -nae and the (Standard Scottish) English form -n't. Speakers who have -nae also have -n't (Smith, Durham & Richards 2013):

For the purposes of NC, -n't and -nae have the same syntactic properties (see Thoms et al. 2023), and we include examples with both contracted negative markers.

4.4.2 Negative concord: spoken data

In the judgment task (section 4.4.3 below), we discuss judgments for NC with two postverbal indefinites: nothing (21a) and nowhere (21b). We also focus on these two forms in the spoken data, though note the corpus also contains examples of NC with other indefinites, such as nobody, none and no NP.

Table 3 shows NC by indefinite type.¹²

Table 3. Proportion of negative concord by type in the SCOSYA corpus

Turning first to NC with nothing (21a), figure 7 shows the distribution of the 148 tokens. The faded circular dots indicate communities where conversation was recorded, with no attestations of NC with nothing. The black dots indicate communities where NC with nothing was attested. Figure 7 shows widespread use throughout Scotland. As with need + PAST, we cannot rule out NC with nothing in the communities with no attestations due to low absolute frequency.

Figure 7. Locations with attestations of NC with nothing in SCOSYA

There was only one attestation of NC with nowhere, from the North East (21b), and the total percentage of NC in the nowhere/anywhere case is only 1 per cent. From the corpus data, then, we see variation in NC depending on the indefinite.

4.4.3 Negative concord in the SCOSYA judgment data

All participants judged examples of NC where sentential negation was combined with nothing (22a) and nowhere (22b).¹³

We will firstly discuss the example with nothing in (22a).

Cannae see nothing

There is a great deal of variability in the ratings for (22a). It was not rated highly across the regions (mode=1, median=3). Nevertheless, it had a 43 per cent acceptance rate (ratings of 4 or 5). Figure 8 presents an acceptability map for this example, with black dots indicating communities in which 2 or more participants gave the example 4 or 5.

Figure 8. I cannae see nothing, accepted by at least two participants

It is clear from figure 8 that the moderate acceptability rate cannot be explained by variation across place; although there are some areas with attestations in almost every community (e.g. Tayside & Angus), there is not the strong geographic clustering of judgments we saw with div (see section 4.3.3).

Didnae see it nowhere

The example in (22b) was not rated highly across the regions (mode=1, median=2), with a 21 per cent acceptance rate. Figure 9 presents an acceptability map for this example, with black dots indicating communities in which 2 or more participants gave the example 4 or 5.

Figure 9. I didnae see it nowhere, accepted by at least two participants

The map in figure 9 shows some geographic clustering. Although there are communities with acceptance dotted around e.g. Fife and Stirling, there is a particular concentration of black dots in the North East. There, the median is 4 and mode is 5, with an acceptance rate of 56 per cent. This geographic concentration of acceptance is also in the region where we find the singular attestation of NC with nowhere in the corpus.

4.4.4 Negative concord: combining spoken data and judgments

In contrast to need + PAST and div, figures 7 and 8 show little alignment between spoken attestations and judgments summed at community level with respect to NC with nothing. While need + PAST was infrequent in the corpus, the wide geographical distribution mapped to widespread acceptability, suggesting it was down to (in)frequency that it was not attested in our data from more communities. On the other hand, while attestations of NC with nothing in the spoken corpus are spread across the communities sampled, judgment data is patchier, with no clear distribution of (un)acceptability.

It is worth exploring whether these are individual community-based distinctions. Comparing the maps, we can see that some communities attest NC with nothing and accept it. Some communities do not attest NC with nothing, but accept it in the judgment data. This is what we see for need + PAST and div, where we assume the phenomenon was not attested in those communities for frequency reasons. However, other communities do attest NC with nothing, but reject it in the judgment task – for example, this is a clear pattern for communities in Caithness and the Highlands. With NC with nothing, then, judgments do not clearly map to the spoken data.

As there was only one attestation of NC with nowhere in the corpus, it is difficult to draw conclusions about its distribution; however, the single attestation came from a location in the North East, where there is also a cluster of acceptability in the judgments. The judgment data for NC with nowhere maps more clearly to the spoken data than NC with nothing, though the picture is not as sharp as for need + PAST and div.

4.5 Summary

We have examined three variables from SCOSYA: need + PAST, div for periphrastic do and NC. Each of the variables has a different pattern when comparing spoken data to judgments. Need + PAST is produced and accepted across regions. Div for do is geographically circumscribed to the North East and Borders in the corpus, and this is matched in the judgments. For NC with nothing, results are mixed, with attestations across regions but inconsistent acceptability in the judgment task. Here, it seems the results of judgment tasks diverge from the usage picture. However, NC with nowhere exhibits a clustering of acceptability around the single attestation in the North East. In section 5 we will revisit the social context of Scots and Scottish English in Scotland presented in section 3.1 to propose reasons for the varying levels of match/mismatch in the data.

5 Discussion

In section 2.3, we outlined Labov's (1996) reasoning as to why speakers’ intuitions might ‘fail’ to match speakers’ usage in everyday speech, and section 4.1 we laid out how the particular features investigated in this study allow us to try to tease apart the importance of these factors. Here, we consider each of the factors individually – frequency, regionality and social stigma – and summarise how they do, or do not, explain the data. We also draw on evidence from other features investigated in the SCOSYA data collection where relevant patterns seem to emerge.

5.1 Frequency

The first possible reason for the disparity between the outcome of introspective judgment tasks and speakers’ behaviour in conversational settings is (relative) frequency. Looking firstly at the data for need + PAST, despite low absolute frequency, need + PAST accounts for 58 per cent of possible contexts (that is, its relative frequency is high). Need + PAST is the most frequent variant in production, and – in alignment with this – judgments are accurate. However, when we look at the difference in accuracy between judgments on div and those on NC, we find variation that cannot be explained in terms of relative frequency alone. With div, across Scots varieties, relative frequency in our data is low – less than 1 per cent. Nevertheless, speakers are accurate in their judgments. However: at what level should we be measuring frequency? Frequency in the corpus overall does not necessarily indicate frequency at community level, given the wide regional distribution of participants. Indeed, the frequency of div varies by area: within the North East and Borders, the relative frequency of div increases to 4 per cent (73/1,676 possible contexts), while it falls to 0 per cent in the rest of the regions. Div's rejection in the rest of the regions, combined with its lack of attestations there, is a fairly reliable indication it is unacceptable in those communities.

However even in the North East and Borders, div remains a relatively low-frequency variant. As discussed in section 2.3, what is considered ‘low frequency’ varies across studies, from 3 to 50 per cent. The frequency of div is at the lower end of this range even in the North East and Borders, and may provide evidence that the cut-off point for being ‘rare’ enough to affect ratings is really quite low.

While relative frequency is aligned with the accurate judgment patterns for need + PAST and div, at least at a local level, the same cannot be said of the judgments given to NC. NC with nothing occurs at a rate of 10 per cent across the corpus. While this is not a high rate, it is certainly above the 4 per cent that saw div rated as acceptable in the communities in which it was produced. And yet, as discussed above, judgments on it are inconsistent.

Given the variable production rates for NC across Scots varieties in the literature (see section 4.3.1), we should consider whether that 10 per cent production frequency is a flat rate across the regions. Perhaps unsurprisingly, it is not. In the North East, NC with nothing occurs at a rate of 34 per cent. In a further three regions, NC with nothing occurs at a rate over 10 per cent (Ayrshire, Fife and Tayside & Angus). At the other end of the scale, three regions exhibit NC with nothing at a rate between 0 and 3 per cent (Orkney, Stirling & Falkirk and the Western Isles). In the remaining eight regions, the rate is 4–8 per cent.

However, the differences in rates do not necessarily parallel differences in judgments. In the North East and Tayside & Angus, which have high relative production rates, there are also high acceptance rates (53 per cent and 61 percent 4–5 ratings respectively). The highest acceptance rate (63 per cent) is, however, found in Shetland, where NC with nothing was attested in only 6 per cent of possible contexts. Participants in the Western Isles had a relative frequency of 1 per cent in production, arising from a single attestation, but an acceptance rate of 36 per cent (median=2, mode=1). In Fife, where the relative frequency is 13 per cent, the acceptance rate is 31 per cent (median=2, mode=1).

We cannot rule out the possibility of low relative frequency of a variant in a community leading to the failure of speakers’ intuitions, as evidenced by the judgment tasks. However, frequency is not able to explain the judgments on NC with nothing. We therefore move on to the second factor, regionality.

5.2 Regionality

Labov (1996: 22) proposes that ‘any grammatical pattern that is perceived as regional may be suppressed in introspection’. The features we have investigated are different in terms of their regional distribution: div for do is highly localised; need + PAST is supralocal across Scots varieties, while NC is supralocal across Englishes. Does this variation affect judgments?

As we saw in section 4.2, judgments for div mapped accurately to attestations. This success is replicated with other highly localised features. For example, double modals are attested as a feature of Southern/Borders Scots (Brown 1991), and there are two examples of double modals in the corpus, from southern varieties (23a–b).

In the judgment data, double modals are rated highly across these areas, and low elsewhere. It seems when a morphosyntactic feature has a local distribution, speakers accurately judge its presence, or lack thereof, in their community.

The acceptability of NC with different indefinites also supports this. While NC with nothing is found across Scots varieties, NC with nowhere appears limited to the North East. This reflects other work showing that production of NC holds at different rates for different lexical items in different varieties (Cheshire 1982; Smith 2001; Anderwald 2005; Robinson & Thoms 2021). For localised NC with nowhere, participants’ judgments accurately reflect local spoken use; on the other hand, for supralocal NC with nothing, judgments diverge from local usage patterns. This might lead us to conclude that it is with supralocal features that judgments diverge from spoken usage patterns, but of course, this does not hold true for need + PAST.

Unlike need + PAST, NC with nothing is a vernacular universal (Chambers 1995) across Englishes. It is possible, then, that being a non-standard variant that is not specifically a Scots feature could affect judgments. However, another vernacular universal tested in the corpus is non-temporal never, where never acts as regular sentential negation (24a–b).

Although not a feature of standard English, this form is widespread across varieties of English (Cheshire et al. 1993; Kortmann & Szmrecsanyi 2004) – including Scots (Macafee 2011). Judgments for non-temporal never in SCOSYA are very clear. It is the most highly rated example in the data set, with a median and a mode of 5 across all participants, and an acceptance rate of 88 per cent.

In summary, judgments for highly localised features are very reliable. When intuitions do seem to diverge from spoken usage, it is when judging more widespread features (table 4), but supralocality in itself does not predict ‘failure’. We use Labov's shorthand of ‘failure’ and ‘success’ of intuitions to mean that they fail to track or succeed in tracking the statistical patterns in corpus data. We will explore this further in section 5.4.

Table 4. Intuition ‘success’ and ‘failure’ in the SCOSYA judgment tasks

5.3 Social stigma

The salience of and social stigma surrounding the variants in this study was detailed in section 4.1. Firstly, need + PAST is known to be a covert feature of Scots (Aitken 1984), and it is therefore unsurprising that social stigma does not interfere in speakers’ intuitions here.

There was little previous evidence about the salience of div in Scots. From the SCOSYA data, we can see that it appears to be overt to speakers outside the North East and Borders. For example, there were four examples of div in our corpus that were instances of metalinguistic discussion (these examples were removed from our count of tokens of this variable) (25).

The speakers in (25) are from Caithness. Evidently the feature is salient to them, although they ascribe it to speakers in a different part of Caithness, who do not use the feature.¹⁵ Within the North East and Borders, we don't see this same kind of discussion – indicating div may not be salient for speakers who actually use it. So, although we cannot rule out div as salient for speakers in the North East or Borders, there is no evidence to suggest it is. It falls out from this that div is unlikely to be stigmatised, and thus stigma is not relevant to judgments of this feature.

Our third feature, NC, is highly salient and highly stigmatised, and it seems that even with a judgment task designed specifically to reduce the Observer's Paradox as far as possible, an overtly stigmatised variant like NC is still at risk of diverging from vernacular patterns. However, as we saw in section 4.4, participants are relatively successful at judging NC with nowhere: this variant is only attested in the North East, where it was rated more highly in the judgment task as well. There is no evidence that NC with nowhere is subject to different social evaluation as NC with nothing. In Smith & Holmes-Elliott's (2022) study, for example, speakers in Buckie ‘control’ their usage of NC with outsiders regardless of the indefinite used (e.g. nothing, nowhere). It seems, therefore, that social stigma alone cannot account for speaker intuitions diverging from the speech patterns in corpora. Instead, in the final section, we will consider the interplay between social stigma, regionality and the construction of a local identity in judgment tasks.

5.4 Stigma x local identity

As with the features in our study, Aitken's (1984: 25) list of ‘vulgarisms’ contains features which vary in their regionality. Some are highly regional, such as so it is tags (26), attested in the south west of Scotland, and sentence-final but (27), which means something like ‘though’ and is attested in Glasgow/Ayrshire.

However, some are vernacular universals – namely, NC, and ‘the well-known syncretism of past tense and past participle forms’ (Aitken 1984: 25) (28a–b).

As we saw, speakers across Scots varieties often did not match the attested speech patterns when judging NC with nothing. The same issue arises at least for older speakers when it comes to irregular past tense forms. Older speakers rate examples like (28a–b) low (seen – 35 per cent acceptance rate; median=2, mode=1; done – 47 per cent acceptance rate; median=3, mode=1), despite attestations across Scotland. Younger speakers, on the other hand, accept the construction at a higher rate (seen – 56 per cent acceptance rate; median=4, mode=5; done – 70 per cent acceptance rate; median=5, mode=5). We suggest this is due to a change in levels of stigma rather than a usage change, as irregular past tense forms have been part of vernacular Scots for at least one hundred years (Grant & Dixon 1921).

The results for these two features contrast with what we saw for NC with nowhere, despite it also being a vernacular universal. We thus propose that, given a well-controlled judgment task, a salient variant which is stigmatised can lead intuitions to fail if the variant is not able to contribute to the construction of a local (cultural) identity. We define local identity following Hazen (2002: 241): ‘how speakers conceive themselves in relation to their local and larger regional communities’.

Judgment tasks are ultimately metalinguistic, allowing speakers to project their identity by expressing their intuitions – and identity formation is locally oriented both in production (e.g. Labov 1963; Eckert 2000; Hazen 2002; Stuart-Smith, Timmins & Tweedie 2007) and judgments (e.g. Jamieson 2020). Reflecting on the relationship between Scots and (Scottish) English (see section 3.1), for the participants in the SCOSYA corpus, we would expect a stronger association with the broad Scots end of the continuum. The participants came from families who had been in the community for generations, and had themselves remained in the communities and networks they grew up in. Their cultural identity is likely to be oriented to that community, and so when given the opportunity to construct a self through a metalinguistic task – particularly in a task which is administered by someone else from the community, and specifically designed to encourage them to access the broad Scots end of their linguistic continuum – it is this alignment we would expect to emerge.

Both covert and overt highly regional forms allow successful judgments that map to their attested usage patterns – even if they are stigmatised, like NC with nowhere. In relation to identity construction, this makes sense – if the form is covert, it is acceptable with little hesitation; if it is overt, participants may be aware of stigma, but can reframe this as pride in a local variant. However, it appears that at least in the Scots context, English vernacular universals which are used across varieties do not contribute to local identity, and so any desire to push past prescriptive stigma and index regional associations is lessened. If the variant is covert, this may not affect judgments – as we see with non-temporal never. This is simply perceived as a feature of the grammar. However, if the variant is overt, intuitions may fail – as with NC with nothing. Here, the feature is stigmatised in such a way that it is known as an ‘incorrect’ feature of English. Participants who wish to align themselves with the broad Scots end of the Scots language continuum also wish to distance themselves from the non-local, ‘English’ end of the continuum (e.g. Le Page & Tabouret-Keller 1985).

It may be the case that (overt) English vernacular universals cannot regularly contribute to identity construction more broadly across dialects; this would require further study. However, certainly in the case of the Scots language continuum, the pull to construct a local identity appears to lead judgments to fail when a feature is both stigmatised and used by speakers with whom the participants do not wish to align themselves.

6 Conclusions

In this article we investigated how speakers’ intuitions map to production data using SCOSYA, a large-scale data set which allows us to compare acceptability judgments from 530 speakers with over 3 million words of spoken data from those same speakers. Speakers’ intuitions, as expressed through the judgment task, broadly matched the corpus patterns for need + PAST and div, but failed to do so for NC with nothing. However, introspective judgment tasks are more successful in tracking the patterning of NC with nowhere, a more localised variant, in the corpus. We discussed the results in terms of frequency, regionality and salience/stigma, arguing that speakers are generally successful in judging covert features regardless of whether they are stigmatised, while stigma attached to salient features can affect speakers’ judgments. However, if a stigmatised variant can be ascribed as a marker of local identity, speakers are more likely to accept it, while overt vernacular universals are more likely to be rejected.

As noted in the introduction, Rickford (2019) calls for a:

concentrated effort to determine what kind of intuitive judgments are more robust than others, what factors influence their variability, and what methods we might use for calibrating then against observational and other evidence. (Rickford 2019: 102)

The analysis presented here provides the foundations for future research across a much wider range of morphosyntactic variables in uncovering the complexities of when intuitions (don't) fail.