Grammaticalisation of habitual aspect in World Englishes: assessing trajectories, areal patterns and rates of change with synchronic corpora

2.1 Grammaticalisation theory

Grammaticalisation is a pervasive historical process with rich implications also for synchronic language structure. It can be defined as “the gradual unidirectional change that turns lexical items into grammatical items and loose structures into tight structures, subjecting frequent linguistic units to more and more grammatical restrictions and reducing their autonomy” (Haspelmath 1998: 344). In short, grammaticalisation is characterised by functional expansion and formal reduction. That is, meaning and form tend to co-evolve (see Bybee et al. 1994: 20). Accordingly, the natural units of this process are form–meaning pairings, which is why the theoretical framework of construction grammar (e.g. Goldberg 2006) has been particularly fruitful in the study of grammaticalisation phenomena. In line with this, we will adopt a usage-based constructionist approach when operationalising degrees of grammaticalisation for our case study (see Section 4.1).

The diachronic paths that grammaticalising constructions take can be captured by cross-linguistically valid clines. For each type of grammatical marker in the world’s languages, there are only very few possible lexical sources as well as a recurrent order of intermediate stages, which at a cognitive conceptual level tend to be linked through metonymy and/or metaphor (e.g. Kuteva et al. 2019). For example, the future-tense markers of numerous languages have been demonstrated to derive from the source concepts of either volition (e.g. will), obligation (e.g. shall) or movement (e.g. gonna) via a likely intermediate stage of ‘intention’. Analogy is widely recognised as another key mechanism, directing change in newly grammaticalising constructions based on perceived similarities to established grammatical constructions (e.g. Fischer 2008). Existing constructions acting as analogical models may either be part of the same language, or they may stem from another language in cases of contact-induced grammaticalisation (Heine and Kuteva 2020). Crucially, grammaticalisation is a strongly directional change; its irreversibility is a very robust statistical language universal (Hopper and Traugott 2003: Ch. 5).

For synchronic linguistic systems, the nature of grammaticalisation entails, among other things, that grammatical categories are fuzzy rather than discrete, and that grammatical markers are often polysemous, spanning contiguous functions on an established cline. Diachronic grammaticalisation paths hence translate into implicational scales of (poly)functional coverage in synchrony (see the concept of semantic maps, as discussed in the World Englishes context by Siemund and Davydova 2017).

2.2 Grammatical variation and change in World Englishes

Since its inception, the study of World Englishes has been dominated by investigations of linguistic variation. In addition to asking how varieties of English differ from each other and their historical input varieties on all linguistic levels, studies have also put forward a range of explanations helping to answer why they differ. What has emerged from this research is that grammatical variation in World Englishes is a highly complex phenomenon that cannot be explained monofactorially, since factors at various levels of granularity may influence language choices. Key aspects to consider include processes of language acquisition (e.g. Buschfeld 2020; Williams 1987) and language contact (e.g. Sharma 2009). In particular, patterns of variation may be explained by approaches involving regional distance or proximity, such as colonial lag and the epicentre hypothesis. In the following paragraphs, we introduce these key aspects as promising candidates in the explanation of the grammaticalisation of habitual [used to V].

World Englishes are contact varieties of English by definition, and novel features that end up as part of nativised and later endonormatively stabilised varieties are frequently the result of contact. A key aspect in any analysis of language contact scenarios involves the morphological typology of the involved languages, in particular the analytic/synthetic contrast. An example of the effects of language contact repeatedly found in English varieties in contact with Sinitic languages, which are considered isolating or at least heavily analytic, is that affixation tends to be reduced in favour of analytic structures, such as the of-phrase instead of the ’s-genitive (see Heller et al. 2017). Employing L1 structures in this way constitutes what Matras and Sakel (2007) call pattern replication, i.e. the modelling of L2 structures based on a pivot structure (see (replica) contact-induced grammaticalisation in Heine and Kuteva’s 2020 framework). The acquisition of L2 features is highly complex, with various strategies at play at different acquisitional stages. However, certain processes have repeatedly been shown to be of particular importance in World Englishes contexts. In their programmatic article on bridging second language acquisition (SLA) and World Englishes, for instance, Sridhar and Sridhar (1986) stress overgeneralisation as a critical strategy: when learners acquire an L2, certain features may be expanded to grammatical contexts beyond those typical of the language in L1 contexts. As we show later, [use(d) to V] is used more flexibly in some varieties featured in ICE. A potential outcome of overgeneralisation and other SLA processes is the nativisation or fossilisation of features in a variety. The core idea behind either term is that recurring patterns of use may eventually manifest as stable patterns in a variety. The factors of contact and L1/L2 status also feature prominently in Kachru’s (1985) model of Englishes, and in this regard Kachru’s typology is relevant to change, including grammaticalisation, despite being a static model: notions such as Inner Circle (aka L1) versus Outer Circle (aka L2) Englishes provide a macro-level classification of which varieties are predicted to behave alike in terms of rates and types of change relative to another group of varieties united by patterns of contact and language acquisition as outlined above. In their large-scale survey on 72 grammaticalisation-relevant features in 52 non-standard Englishes, Kortmann and Schneider (2011) gained results that are partly compatible with Kachru’s coarse-grained typology: they found that varieties that fall under “contact Englishes”, including the more fine-grained variety types of high-contact L1, L2, as well as pidgins and creoles, exhibit most of the innovative grammaticalisation features. As a reasonable generalisation then, high-contact varieties of English may be assumed to exhibit higher rates of innovation than, and hence more substantial differences to, low-contact L1 Englishes or the group of Inner Circle Englishes as a whole.^[1]

This trend is in diametrical opposition to one of the older narratives in World Englishes research, the hypothesis of colonial lag. According to this hypothesis, which was first proposed in the context of AmE versus BrE (esp. Marckwardt 1958), postcolonial Englishes should change at a lower rate than the norm-providing variety in the original homeland. Some corpus-linguistic findings are in line with this idea, including World Englishes studies that are highly relevant to the present investigation as they deal with differential grammaticalisation: for example, Collins’ (2009) and Loureiro-Porto’s (2019) results on modal auxiliaries can be read as cases of Outer Circle Englishes following, but lagging behind, the probabilistic usage shifts observed in BrE and AmE. However, since its inception in the mid-20th century, the narrative of colonial lag has lost traction overall, as researchers using new empirical resources and methods have been accumulating what seem to be just as many counter-examples, i.e. instances of “home lag” or colonial innovativeness (see Hundt 2009 for a survey).

In addition to spreading supraregionally, a linguistic feature or bundle of features may spread beyond country borders, a process which is at the heart of the epicentre hypothesis. In this hypothesis, “[t]he term epicentre is a metaphor rather than a model for the connectivity between varieties of English in a given region” (Peters and Bernaisch 2022: 321; italics in original). At their core, “linguistic epicentres can represent regionalised transnational varietal developments […], illuminat[ing] relationships between regional varieties, and the spread of areal features from a local lead variety to others” (Peters and Bernaisch 2022: 327). A major candidate and most intensively researched area for the epicentre hypothesis is South Asia, with Indian English (IndE) as the epicentre and neighbouring varieties, such as Bangladeshi English and Sri Lankan English (SLE), changing as a result of their proximity to the much more influential IndE (see Schneider 2022 for an overview). While the concept of linguistic epicentres has gained prominence through World Englishes research only since the mid-2010s, links can be established to a long tradition of grammaticalisation research, where influential typological work has always stressed the impact of areal patterns (see, e.g., Bisang and Malchukov 2020).

4.1 Data and methods

For our study, we analysed 13 national components featured in the International Corpus of English (ICE; Greenbaum and Nelson 1996). Eight of these represent Englishes that are widely considered Outer Circle varieties: East Africa (ICE-EA), Hong Kong (ICE-HK), India (ICE-IND), Jamaica (ICE-JA), Nigeria (ICE-NIG), Philippines (ICE-PHI), Singapore (ICE-SIN) and Sri Lanka (ICE-SL). The other five national components sample Inner Circle varieties: Canada (ICE-CAN), Great Britain (ICE-GB), Ireland (ICE-IRE), New Zealand (ICE-NZ) and USA (ICE-USA).^[2] Since there is no spoken part complementing the written section of ICE-USA in ICE itself, we included, as is common practice in the World Englishes context, the Santa Barbara Corpus of Spoken American English (Du Bois et al. 2000–2005).

Due to their easy accessibility, fairly high degree of balancing across spoken and written texts, and the range of included text types, the ICE corpora have been called “the Swiss army knife of world Englishes scholars” (Leuckert and Rüdiger 2021: 485). They are, to date, the only set of corpora that allow for a large-scale comparison of structural features in spoken and written World Englishes; many other World Englishes corpora, such as the Corpus of Global Web-Based English (GloWbE) (Davies and Fuchs 2015), tend to be limited to written contexts and often to specific registers. However, despite their usefulness, there are various limitations and caveats to the ICE family of corpora. Perhaps most importantly, the ICE corpora have been compiled at different points in time (sometimes decades apart), which introduces a diachronic dimension to their comparison and means that they are not fully representative of the varieties in their current state. In addition, as pointed out by Deshors and Götz (2020: 8), there are “differences and potential inconsistencies in corpus compilation procedures of the different ICE-components.”

We used AntConc (Anthony 2022) to extract all relevant forms of use(d) to in the ICE corpora and manually removed any false hits from the corpus output (e.g. passive be used ‘employed’ to do sth.).^[3] We also excluded the habitual, but adjectival, construction be used to sth. Moreover, as some texts are “inconsistent” in that they feature non-variety speakers, such as non-Hong Kongers in ICE-HK, we removed tokens produced by such speakers from the extracted concordances based on existing metadata. After this clean-up process, the resulting 2,338 instances of the auxiliary construction were coded for the nine variables listed in Table 1.^[4] These variables cover semantic, syntactic, morphological and text-linguistic criteria. In order to ensure inter-rater agreement, we took a random sample of 100 concordances which were then coded for all relevant categorical and nominal variables by all coders. We then calculated Cohen’s kappa, which was above 0.9 for all variables; in fact, for most variables, all coders were in perfect agreement.

The nine variables coded for reflect rich behavioural profiles (Gries 2010) of use(d) to, which were fed into a cluster analysis as described in Levshina (2015: Ch. 15). More specifically, we excluded the two text-linguistic variables of mode and token frequency from the analysis because of confounding details of corpus sampling and size and subjected the remaining seven variables to a hierarchical agglomerative cluster analysis (see Divjak and Gries 2009). The cluster analysis was conducted using the hclust function in R (R Core Team 2021). In the process, attested ratios of the variable levels were converted into a distance matrix for pairwise variety comparisons. The final cluster solution uses Euclidean distance and the merge algorithm “complete”, which we arrived at by comparing the agglomerative coefficients of competing solutions (incl. one with the merge algorithm “ward.D2”). The clusters were validated via multiscale bootstrap resampling, and the optimal number of “meaningful” clusters was determined via average silhouette width (Levshina 2015: 311–317). In short, the method makes varieties and variety bundles cluster together based on the degree of similarity of their use(d) to profiles.^[5]

A subsequent more fine-grained analysis of the resulting clusters employs so-called snake plots (again following recommendations by Levshina 2015: 313–315; see Divjak and Gries 2009). This visualisation method allowed us to zoom into and compare specific variety clusters. Essentially, the snake plots reveal how strongly specific variable levels contribute to the cluster patterns, thus highlighting distinctive structural features of the use(d) to construction in individual varieties or variety bundles.

When it comes to measuring degrees of grammaticalisation, two of the variables we coded for lend themselves to objectively represent a scale of grammaticalisation of the use(d) to construction. These are the two semantic variables of subject animacy and verb aspect/aktionsart. Recall from Section 3 that over the past six centuries the use(d) to construction has been grammaticalising along a functional cline: from habits in the literal sense, to grammatical habitual aspect, and most recently to an even more abstract ‘discontinuance’ meaning as its primary contribution. Especially from a construction grammar perspective (e.g. Coussé et al. 2018), it should become apparent how such semantic shifts manifest themselves through changes/differences in collocational patterns. Grammaticalising constructions essentially undergo a process of context expansion (Himmelmann 2004). The Middle English source of today’s use(d) to construction was confined to animate (typically human) subjects, which can actively follow a particular habit, and to dynamic verbs mostly, which qualify as habits to be performed (Neels 2015). The diachronic expansion to inanimate fillers in the subject slot is symptomatic of the semantic bleaching of use(d) to into a marker of habitual aspect; and so is the expansion to stative verbs, but stative verbs are even more characteristic of the discontinuance function of used to. Habituality is often no longer the primary meaning contribution of used to when combined with stative verbs, considering that states tend to have a long indefinite duration anyway. What typically motivates these combinations is the resulting meaning that the states no longer obtain at present. Since grammaticalisation is known to restructure linguistic categories in a highly gradual manner, novel fillers in the open slots of a construction are informative not only in terms of their first occurrences but also in terms of their increasing rates among other slot fillers. This is therefore how we measure semantic degrees of grammaticalisation of use(d) to, i.e. via the ratios of stative verbs and of non-animate subjects, including expletive it and there.

In essence, while only some of the structural variables enable a truly objective assessment in terms of “more” or “less” grammaticalised, all variables together can be used to measure overall degrees of similarities and differences regarding the probabilistic grammatical properties of the use(d) to construction across the 13 national Englishes.

4.2 Research questions and working assumptions for the corpus study

In the corpus study, we wish to investigate (i) whether synchronically World Englishes differ notably in the degree of grammaticalisation and overall usage profile of habitual [use(d) to V], and (ii) to what extent such differences pattern in ways that support any of the three general but partly conflicting trends regarding the development of World Englishes, i.e. colonial lag, contact-induced innovation and epicentres. The present section outlines how these three supposed trends translate into working assumptions for the concrete case of [use(d) to V] and how the corpus results should differ depending on which, if any, of the three trends or “narratives” outlined in Section 2.2 turn(s) out to manifest and dominate empirically.

The narratives of colonial lag versus contact-induced innovation share that they predict differential usage profiles of use(d) to based on lineage or variety type. Most generally, the five Inner Circle varieties, on the one hand, and the eight Outer Circle varieties, on the other hand, should cluster together, reflecting also the basic distinction between L1 Englishes as opposed to L2 Englishes with more potential for contact-induced transfer. Within this hypothesised pattern, lineage may still have a certain impact at a more fine-grained level, such that PhilE, for example, clusters sooner with AmE than a British-dominated L2 like IndE. More importantly, the competing narratives of colonial lag versus contact-induced innovation create opposite expectations too, however for other parts of the statistical analysis. Empirically teasing the two narratives apart in the case study will be possible as follows. The direction of the variety differences in terms of conservatism or innovativeness can be differentiated primarily thanks to the more specific analysis of the constructional semantics of use(d) to via subject and verb types, i.e. animacy and aktionsart. In addition, an inspection of grammatical structures that are categorically (aka qualitatively) unique or idiosyncratic (e.g. future-time will use to) will highlight innovativeness as well. Such structures can and will be pointed out in the analyses of snake plots. If idiosyncratic use(d) to variants unknown to the British (and American) norm are found primarily in Outer Circle (L2) Englishes rather than in other members of the Inner Circle (L1), this lends support to the narrative of contact-induced innovation.

The epicentre hypothesis is, in principle, compatible both with signs of innovativeness and with signs of conservatism. What distinguishes this hypothesised trend from the other two in the present case study is that if epicentres turn out to affect the behaviour of use(d) to the overall clustering results should (at least partly) reflect geographical proximity or the influence of a dominant variety in region, such as IndE in South Asia (see Schneider 2022 for a summary of studies on India as a potential South Asian epicentre). In our test case specifically, epicentres should thus lead to a greater similarity of the usage data from ICE-IND and ICE-SL or from ICE-NIG and ICE-EA, for instance.

4.3 Results

The results from the procedure described above are presented in this section. After an overview of absolute token frequencies (4.3.1), we present findings on the semantic degree of grammaticalisation (4.3.2), a cluster analysis (4.3.3) and a more detailed look at variety clusters and individual varieties using snake plots (4.3.4).

4.3.2 Semantic degree of grammaticalisation

The shifting semantics of the auxiliary use(d) to is reflected in differences in collocational patterns. In particular, as introduced in Sections 3 and 4.1, meaning shifts from literal habits to grammatical habitual aspect and to the even more abstract meaning of discontinuance (“anti-present-perfect”) are reflected in increasing rates of non-animate subjects as well as stative verbs. Figure 1 provides a synchronic snapshot of these rates in each of the 13 English varieties under scrutiny. The plot can be read to depict a semantic space where the upper right corner represents semantically innovative, i.e. highly grammaticalised, uses and the lower left corner stands for conservative uses of the use(d) to construction. In this semantic analysis, just like in the above results on the absolute token frequencies, BrE and AmE turn out to constitute “mean” values around which the other varieties pattern. What is somewhat surprising, however, is that the ways in which the postcolonial Englishes pattern based on the present semantic grammaticalisation index are at times quite opposite to the results of the more simplistic grammaticalisation index of token frequency. Most notably, IndE and JamE are now found to be by far the most conservative varieties, whereas HKE exhibits uses of the auxiliary that are semantically most advanced, in particular regarding the verb slot: as plotted in Figure 1 and exemplified in (4), about 45 % of all verbs combining with use(d) to in HKE are stative verbs, which are associated with the foregrounded discontinuance meaning of the construction.

Figure 1:

Probabilistic semantic context expansion of [SUBJ use(d) to V] in 13 national varieties of English.

(4)

Yeah I used to like him in Top Gun (ICE-HK: S1A-071#806:1:A)

Other varieties in which use(d) to is also highly grammaticalised (semantically speaking) include PhilE and EAE, although in the latter variety the two variables of aktionsart and subject animacy seem not to be correlated in the same fashion they are in the other Englishes.

These reported differences in grammatical patterns concern probabilistic usage preference rather than discrete categorical differences. Still, as advocated in frameworks such as probabilistic grammar (Bresnan 2007; Szmrecsanyi et al. 2016), distributional differences in grammatical usage can be assumed to be mentally reflected in a probabilistic component of linguistic knowledge. And with variation across varieties as broad as, for instance, 17 % versus 45 % of stative verbs in IndE versus HKE, it is not unreasonable to assume that structural nativisation has occurred regarding the functional properties of the use(d) to construction.

Overall, the clearest trend emerging from the semantic analysis is that L2 aka Outer Circle Englishes exhibit more extreme values and differences than the Inner Circle Englishes. Being already quite apparent from eyeballing Figure 1, this contrast between the two groups of varieties manifests itself in different mean Euclidean distance measures between the varieties in each group. While the mean Euclidean distance between the L1 varieties is 0.0503, the mean Euclidean distance between the L2 varieties is 0.1437. The difference in mean distances is statistically significant based on a Mann–Whitney U-test (U = 31, p < 0.001). These grammaticalisation data hence underscore the differentiation of varieties made by Kachru’s contact-related model. However, it is also apparent from Figure 1 that the Outer Circle Englishes are scattered in multiple directions relative to the more homogeneous group of Inner Circle Englishes. Consequently, the data do not consistently align with either of the two opposing narratives, i.e. colonial lag or contact-induced innovation.

4.3.3 Clustering all ICE varieties

To the semantic features presented above, we added morphological and syntactic variables (see Table 1 above) to create fuller profiles of the characteristics of use(d) to in each national variety. Clustering the 13 varieties based on these behavioural profiles gives the picture plotted in Figure 2. In the depicted tree, the further down two varieties or variety bundles connect, the more alike they are regarding the overall usage profile of the use(d) to construction.

Figure 2:

Cluster analysis of the behaviour profiles of habitual [use(d) to V] in 13 national varieties of English; rectangles signal the optimal number of clusters.

Figure 2 shows two main high-level splits, with the split on the left-hand side creating a cluster containing IndE and JamE. This pairing confirms the similarity of these varieties that was already noticeable in the semantic analysis in Figure 1, but it does so in consideration of their larger behaviour profiles. The second split creates two further clusters. The first of these clusters pairs HKE and PhilE, which, in a similar fashion to IndE and JamE, were already close to each other in Figure 1. Finally, the right-most cluster groups all remaining varieties. This cluster contains EAE, NigE, BrE, CanE, NZE, SLE, AmE, IrE and SinE. What becomes evident by looking at this selection is that all L1 aka Inner Circle varieties are part of this cluster. While there are well-known differences in how these varieties express habitual aspect (see, e.g., Hickey 2007 on the habitual in IrE), they appear similar enough in their use of use(d) to to form a cluster. The lower-level clusters, however, represent pairs of varieties whose proximity in the cluster proves challenging to explain. None of the varieties that form low-level immediate neighbours in the cluster analysis are geographically close. The higher-order adjacency of EAE and NigE does not constitute evidence for epicentre-like influences either. Overall, while some larger tendencies are observable (esp. all L1 varieties being part of the same large cluster), differences and similarities occur at finer levels of granularity, which we investigate in the form of snake plots in the next section.

4.3.4 Comparing and zooming into variety clusters and individual varieties

While the dendrogram examined in the previous section is, by itself, a “black box” when it comes to knowing what structural features from the behaviour profiles are most distinctive, snake plots are a convenient means for adding this information. For two clusters of interest, a snake plot visually contrasts relative differences in the attested ratios of variable levels. These ratio differences show up as transparent descriptive values on the x-axis of the plot.

Figures 3 and 4 zoom in on the two small main clusters established earlier, specifically on how each of them differs from the remaining 11 varieties taken together. Most prominently, both figures reaffirm the findings on semantic degrees of grammaticalisation presented in Section 4.3.2. That is, the ratio of stative verbs is 15 % higher in HKE and PhilE than in the other Englishes on average (Figure 3). For the cluster uniting IndE and JamE, the conservatively high ratios of dynamic verbs and animate subject are shown to be the most distinctive features, creating a 13 % (aktionsart) and a 9 % difference (animacy) to the other varieties in the analysis (Figure 4). Other differences per variable turn out to be more subtle. To some extent, this means that the behaviour of the use(d) to construction does not vary that strongly across national Englishes. However, this impression is also partly conditioned (i) by some of the inter-variety differentials being levelled out in the larger bundles of 9–11 varieties and (ii) by certain grammatical environments and non-standard features being infrequent in the data overall (e.g. negating use(d) to in general, and hence also negation via usen’t contraction in particular). Among the variable levels “exposed” in Figure 4, consider the new insight that IndE and JamE are characterised by a slightly higher rate of use(d) to with non-standard morphology, specifically absent agreement or tense suffixes (+2.77 % “morph_miss”), as illustrated in (5).

Figure 3:

Distinctive variable levels of main clusters: ratio differences of HKE and PhilE versus 11 other national Englishes.

Figure 4:

Distinctive variable levels of main clusters: ratio differences of IndE and JamE versus 11 other national Englishes.

(5)

Luna use to work for this man and his wife and Luna’s sister started having a relationship with him so in a way he use to be Luna’s boss. (ICE-JA: W1B-011#22:2)

Note, however, that the percentage points of the morphology features have to be taken with a grain of salt since the absence or presence of the past-tense suffix in particular should mostly be inaudible in the phonetic context of this auxiliary (see Section 3). Occurrences in the written mode, like Example (5), help attest such morphological idiosyncrasies with certainty, but the vast majority of tokens of the use(d) to construction come from the spoken mode, where transcription uncertainties apply. Still, what is at times more noteworthy about such non-standard features than the exact percentage points is their mere existence as deviations from Inner Circle/BrE/AmE norms. Consider tense-related properties of the use(d) to construction in HKE. Whereas the construction is restricted to past-tense contexts in standard varieties, HKE exhibits present-tense (6a) as well as will-combinations (6b).

(6)

a.

Yeah just like I I I type the Chinese word is even more faster than type the English / Because I use to type Chinese (ICE-HK: S1A-086#239:1:B)

b.

Will you use to be more authoritative than than Martin and I … [?] (ICE-HK: S1B-033#67:1:B)

These two unusual time-reference features make up only a small percentage of usage data from ICE-HK. Still, the existence of categorically different grammatical structures is meaningful from a qualitative perspective.

In fact, several, mostly Outer Circle varieties in the data exhibit such categorically distinct but rather infrequent structures. Below, we list some examples without additional visualisations. Besides HKE, present-tense versions of the use(d) to auxiliary are attested in the ICE data also for EAE, IndE, JamE, PhilE (hapax), SinE (hapax) and, above all, in the NigE data, where 15 tokens of present-tense use to make up 7.8 % of the sample. This is still a low relative frequency, but certainly enough to consider this non-standard structure an established feature of NigE rather than isolated idiolectal “slips”. Among the Inner Circle varieties, IrE stands out as the only variety in the entire sample that negates use(d) to exclusively as usen’t (7), hence preferring direct not-negation over the do-supported negation pattern otherwise prevalent across varieties.

(7)

I used to always get on well with my Mam but I usen’t get on well with my Dad (ICE-IRE: S1A-048$B)

At face value, the presence of the contracted form usen’t can be taken to indicate an above-average degree of grammaticalisation of IrE at the morphosyntactic and phonological dimensions since this negation pattern is a “mature” structure largely restricted to canonical English modals and primary auxiliaries.

Overall, however, translating the above-mentioned non-standard features into higher or lower degrees of grammaticalisation of the use(d) to construction is no trivial task on the basis of a synchronic snapshot. When it comes to the future and/or present-tense forms attested in EAE, HKE, IndE, JamE, NigE, PhilE and SinE, different grammaticalisation tendencies could come into play that impact one’s verdict in terms of “innovative” or “conservative” features. While the expansion to more tense forms seems to nicely match the grammaticalisation tendency of context expansion (Himmelmann 2004), the presence of these forms might also be read as lagging specialisation in the sense of Hopper’s (1991) grammaticalisation principle. Therefore, a full interpretation of the present-day grammatical differences between the national Englishes can only be reached when adding available historical evidence, as will be discussed further in Section 5.3.

5.1 Language contact and sociolinguistic contexts

Major (potential) explanatory factors in variation are language contact and the impact of sociolinguistic variables, such as age, gender and educational level. As briefly discussed earlier, the effects of language contact do not influence a feature in isolation but, instead, are intertwined with processes of SLA, target variety input and other factors. However, a closer look at the characteristics of the key contact languages of the varieties under consideration may potentially illuminate the reasons behind the variational patterns identified above.

Contact-induced transfer could be one of the reasons behind the HKE + PhilE cluster. According to Matthews and Yip (1994: 200), aspect markers are optional in Cantonese, the dominant language spoken in Hong Kong, but there is the aspect marker hōi that may express habituality. However, “[u]nlike the [standard] English used to construction, it is not restricted to past time, and in fact is typically used of the present” (Matthews and Yip 1994: 209). The presence of this temporally flexible habitual marker in Cantonese is a reasonable source of grammatical innovation in HKE through analogical thinking and pattern replication in the bilingual mind (see Sections 2.1 and 2.2). Tagalog as the main contact language of PhilE, in turn, marks aspect on the verb but, like Cantonese, requires additional lexical material to indicate tense. While this does not reveal much about the higher frequency of stative verbs occurring with use(d) to in the two varieties, it could be one of the reasons why present-tense use to is attested in PhilE and more frequent in HKE compared to other varieties in the dataset.

The varieties in the largest cluster are closely linked mainly by sharing BrE as their input variety (AmE, CanE, IrE, NZE), being (highly) advanced in their development in Schneider’s (2007) Dynamic Model of Postcolonial Englishes (AmE, CanE, IrE, NZE, SinE) and being in contact with inflectional languages with affixal habitual markers (EAE [contact language Swahili, Mpiranya 2015: 102], SLE [contact language Tamil, Schiffman 1999]).^[7]

While language contact may explain the data to some extent, there are obvious limitations. First, comprehensive grammatical descriptions of contact languages are not always easy to find, particularly when a specific feature is in focus. Second, contact does not act in a vacuum, and its impact is notoriously difficult to quantify. However, as the examples above have shown, some impact of contact appears likely in the usage patterns of use(d) to.

In sum, then, how do macro-level and micro-level analyses of varieties complement each other? A macro-level take-away from our corpus study was that constructional variants that categorically differ from the BrE and AmE norms were found almost exclusively in the L2 aka Outer Circle Englishes; and it is also members of this variety type that exhibit the greatest probabilistic differences in the semantic context expansion of use(d) to to non-animate subjects and stative verbs. Still, researchers should not be content with the coarse-grained potential explanation that language contact and SLA per se trigger extended, “overgeneralised” (see Section 2.2) uses of grammatical markers. The sociolinguistic specifics and intensity of contact certainly differ also within the group of Outer Circle Englishes. For example, take SLE as the odd one out in our dataset: why does SLE differ from the other seven Outer Circle varieties in not showing any categorically non-standard extensions of used to? A likely explanation lies in the sociolinguistic reality that meso- or acrolectal uses of English are less common as standard SLE is only used by a (typically highly educated) minority in the country (Ekanayaka 2020: 341; see also Kortmann et al. 2020) – a bias that the compilers of ICE-SL acknowledge to be reinforced by the over-representation of academics in their corpus (Bernaisch et al. 2019: 8). Therefore, at a local level of analysis, taking into account such sociolinguistic details as well as the grammatical systems of major contact languages in order to identify structures likely involved in cross-linguistic transfer appears to hold greater explanatory value than hypothesised macro-level trends.

5.2 Theoretical models and statistical modelling

In recent usage-based research on World Englishes, a gap has become apparent between major theoretical models, such as those by Kachru (1985) and Schneider (2007), and the increasingly sophisticated statistical models that seek to verify these theories but often fail to do so. Hundt (2021), in her survey on the issue, argues that this discrepancy partly derives from the fact that they target language structure at different levels of granularity. Indeed, only aggregate studies combining many structural features seem to replicate theoretical models, such as Kachru’s variety types, reasonably well (e.g. Szmrecsanyi and Kortmann 2009). For studies zooming in on single features, like habitual use(d) to here, more fine-grained “multifactorial” approaches have proved necessary that move past big-picture narratives and consider individual sociolinguistic constellations and contact languages more closely (e.g. Werner 2016, Leuckert 2019). For our own case study, this perspective has been sketched out briefly in the previous section. Moreover, we agree with Hundt in that:

[W]hat matters to speakers in terms of identity construction are not frequently used patterns and local patterns of co-variation with language-internal predictor variables but the use of salient patterns (often below the level of statistical significance and thus beyond the reach of probabilistic modelling) that are closely associated with language use in a particular region. (Hundt 2021: 15)

Categorically unique structures such as future-time will use to in HKE or contracted usen’t in IrE may be considered indicators of structural nativisation. These unique and non-standard features, however, occur with very low token frequencies, which make them prone to have little impact on statistical clusters, despite potentially being salient shibboleths for speech-community membership. In other words, such structural features may well be the linguistic reflexes of identity constructions shaping newer Englishes, but they can easily escape statistical modelling.

In a critical response to Hundt’s (2021) survey, Bernaisch et al. (2022) defend a quantitative approach for testing World Englishes models that focuses on constructions qualifying as “linguistic ants”. The metaphor of “linguistic ants” stands for largely schematic grammatical constructions with high usage frequencies and subtle, non-salient probabilistic variation. According to Bernaisch et al.’s argument, since these kinds of constructions bear the brunt of grammatical expression in a language, they should serve as the major test cases for verifying theories rather than focusing on “linguistic butterflies”. Butterfly collecting in linguistics pertains to non-canonical structures with low frequencies but typically high salience. Applied to the present case of habitual use(d) to, one may consider aforementioned structures like contracted usen’t and future-time will use to butterflies in this sense, whereas the common pattern of past-tense [used to V], including probabilistic variation regarding the main verb’s aktionsart, could be said to do the work of a linguistic ant. How to do justice to both the qualitative value of linguistic butterflies and the statistical impact of linguistic ants remains an open issue in World Englishes research as well as in usage-based linguistics more broadly.

To summarise, whether statistical analyses truly test general World Englishes theories seems to depend on whether they operate at a fitting level of granularity, including factors such as the number of varieties, the detail of classifying their sociolinguistic constellations, the number of structural features, and possibly also the “ant” or “butterfly” status of these features.

5.3 A matter of data resources: the limits of assessing change with synchronic data

The present case study can also serve to demonstrate that, although grammaticalisation processes tend to be fairly predictable in the sense of their strong inherent directionality of change, purely synchronic reconstructions of how conservative or innovative a grammatical marker of a given variety is may give rise to errors in interpretation. The concerns that Gries et al. (2018) raise about the uncritical application of the apparent-time method in contrastive World Englishes research, and which they substantiate for the case of the genitive alternation, therefore fully apply to investigating grammaticalisation processes.

Not all categorically deviant, non-standard variants of used to continue to appear as innovative once more detailed historical information is added, which is available for BrE and AmE at least thanks to rich diachronic records as reported in the OED Online, Visser (1969) and Neels (2015). For example, thanks to these sources, we know that present-tense use to has coexisted with past-tense used to for the longer part of the usage history of this auxiliary. Present-tense use to became obsolete in (standard) BrE and AmE only during the 19th century. Before, it was far from a being a minor variant, but diachronic corpus data discussed in Neels (2015) reveal that it was just as frequent as its past-tense counterpart during the Early Modern English period (ca. 1500–1700) before fading from usage during the subsequent centuries. For present-day L2 Englishes exhibiting present-tense use to, such as HKE and NigE among others, this triggers the open question if it is the outcome of innovation in the sense of a reinvention or if it was historically exported as part of the L1 founder dialects and survived until the present day. The latter scenario of transplanted, surviving dialect features (see Kortmann and Schneider 2011) would support the idea of colonial lag.

Similarly, in the ICE data, the negative contraction usen’t was identified as a formally highly grammaticalised variant exclusive to IrE. However, when consulting historical records, it turns out that it is far from clear, and perhaps rather unlikely, that IrE went this grammaticalisation step “on its own”. The OED Online lists several 19th-century and early 20th-century instances of usen’t from BrE prose (“use, v.” OED Online), suggesting that usen’t was rendered an “Irishism” from a synchronic present-day perspective only once this variant had disappeared from BrE usage.

To conclude, only through historical records do we learn that some variants that look like colonial innovations, and that possibly represent more advanced stages of grammaticalisation, have in fact a more complex history that would allow for a different interpretation, an opposite interpretation even when it comes to colonial lag. It is also conceivable that present-day grammatical differences between national varieties result not so much from different rates and kinds of change in the new Englishes but go back to differences in the original founder dialects (i.e. historical varieties that must not be equated with standard English). This underscores the methodological need for more diachronic/historical corpora for researching postcolonial Englishes. Creating such corpora, both as complements of synchronic ICE corpora (e.g. HiCE-Ghana, Brato 2019) and as independent corpora (e.g. The Diachronic Corpus of Hong Kong English; DC-HKE; Biewer et al. 2014), has become a recent trend in World Englishes research. For example, Evans (2014) has shown for HKE that data such as council proceedings, in his case prepared as the Corpus of Legislative Council Proceedings (1858–2012), often reach back quite far in time, allowing not only for historical but genuinely diachronic perspectives. Access to such corpora will provide critical access for studies of grammaticalisation and other processes of language change.