2.1. Be going to and gonna in traditional grammars

A crucial question regarding be going to and gonna is that of their status. How can we categorize them? Depending on the level of analysis, two perspectives may explain the grammatical status of such forms.

First, these forms are described as belonging to the ‘modal’ domain: Huddleston & Pullum et al., in their grammar, speak of ‘modal idioms’ (Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 1227) to refer to be going to, but also to other forms such as have to, got to and want to. In other linguistic works, they are referred to as ‘quasi-modals’ (Collins Reference Collins2009), ‘emerging modals’ (Krug Reference Krug2001), ‘semi-modals’ (Leech Reference Leech, Facchinetti, Palmer and Krug2012) or even ‘periphrastic modals’ (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021: 484). All these terms highlight the fact that the forms belong to the modal realm, while not exactly corresponding to typical modal auxiliaries such as will, must, can, etc. This may be due to their periphrastic structure, as well as to the fact that be going is followed by to-infinitives, whereas modals are typically followed by bare infinitives. Semantically speaking, be going to refers to an event occurring in the future: ‘future as outcome of present circumstances’ (Palmer Reference Palmer1988: 146). More specifically, the form expresses prediction, either based on the speaker’s intention (or ‘future fulfilment of present intention’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 214)) or inference (or ‘future fulfilment of present cause’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 214)), respectively represented in examples (3) and (4):

As be going to and gonna are modal-like forms that express futurity, they are often said to compete with the modal will. It is therefore no surprise that they should belong to the modal realm, since modality is defined by the expression of possibility or necessity and the expression of the ‘attitude of the speaker’ (Lyons Reference Lyons1968: 306).

Secondly, from a syntactic viewpoint, be going to and gonna are described as resembling auxiliaries. In their grammar, Quirk et al. call be going to a ‘semi-auxiliary’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 137), in an intermediate position in their auxiliary verb – main verb scale, in-between ‘modal idioms’ (such as have better, have got to) and catenative verbs (appear to, seem to). This analysis seems to apply to gonna as well, since gonna is referred to as a ‘non-standard spelling’ of be going to in Quirk et al.’s grammar (Reference Quirk, Greenbaum, Leech and Svartvik1985: 898). Hopper & Traugott, for their part, go as far as categorizing be going to as an ‘auxiliary’ (Hopper & Traugott Reference Hopper and Traugott2003: 69).

These two approaches complete one another in helping us understand these forms: Quirk et al. define the class of semi-auxiliaries as ‘verb idioms which express modal or aspectual meaning [and which are introduced by one of the primary verbs have and be]’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 143, emphasis added). Both auxiliary status and modality, as scalar concepts, seem to help us categorize the verbal forms under analysis. Moreover, the two are often linked – especially when referring to the future. As Quirk et al. write: ‘Futurity, modality, and aspect are closely interrelated, and this is reflected in the fact that future time is rendered by means of modal auxiliaries, by semi-auxiliaries, or by the simple present or present progressive forms’ (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 213).

Although terminology may vary, two key dimensions help refine our understanding of be going to and gonna: their evolving syntactic role as auxiliaries and their association with modal meaning. Thus, it may be said, as a basis for my analysis, that be going to, and therefore gonna, lean towards being emergent modal auxiliaries.

In traditional grammars, be going to receives only limited attention, and gonna is typically mentioned only in passing, often as a ‘non-standard spelling’ (Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 1616; Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 898). The form is never analyzed independently – unsurprisingly given that be going to itself is only briefly examined. Over the past two decades, however, linguists have started to analyze the potential independence gonna may be taking from be going to, especially in the theories of grammaticalization and, more recently, Construction Grammar.

2.2. Be going to and gonna in previous research: grammaticalization and Construction Grammar

The verbal periphrasis be going to is not a recent development. Assigning a precise date of origin is difficult, but the construction can already be found in Shakespeare’s works, although its use with a future-oriented meaning remained relatively rare at the time. Recent studies (e.g. Petré & Van de Velde Reference Petré and Van de Velde2018) have shed light on the early stages of be going to’s grammatical development, suggesting that the initial phase of its grammaticalization most likely took place and stabilized between 1620 and 1640. According to Petré & Van de Velde (Reference Petré and Van de Velde2018), earlier attestations are considered non-conventional usages, rather than clear evidence of an established construction.

If the form is not new, its use has increased over time, particularly in recent decades. Figure 1 shows the frequencies of be going to and gonna in the Corpus of Historical American English (COHA, Davies Reference Davies2010), and we can note that COHA’s first occurrence of gonna dates back to 1917, in Thomas Augustus’ play The Copperhead.

Figure 1.

Occurrences of going to V and gonna in COHA, from 1890 to 2010Footnote ²

As shown in figure 1, the use of be going to increased steadily in the twentieth century, notably from the 1930s onwards. As for gonna, determining its first appearance is even more difficult, most likely because the form is considered non-standard and therefore avoided in transcriptions and writings, as Col & Duchet (Reference Col and Duchet2000) suggest: ‘L’apparition de gonna est difficile à dater, du fait de l’évitement graphique dont cette forme semble être victime’ (‘gonna’s first appearance is hard to date, due to the graphic avoidance of which this form seems to be a victim’; my translation). Petré & Van de Velde (Reference Petré and Van de Velde2018) cite 1806 as the earliest attestation of gonna, referencing the Oxford English Dictionary (OED). However, the form remained infrequent throughout the nineteenth century and only began to rise in frequency during the twentieth century. Figure 1 displays the increase of both forms in American English, with a marked rise from the 1930s for be going to, and from the 1970s for gonna, at which point the number of gonna occurrences per decade begins to rival that of be going to. This trend aligns with the more widespread use of semi-modals in American English in general, which tend to replace their modal auxiliary counterparts, as evidenced by Daugs (Reference Daugs2017).

This rise in frequency is crucial: an increase in usage frequency can lead to morphological reduction, a phenomenon Lorenz identifies as a ‘symptom of advanced grammaticalization’ (Lorenz Reference Lorenz, Hasselgård, Ebeling and Ebeling2013: 133). Krug considers phonological variants such as wanna and gonna to represent ‘different stages in the evolution of new auxiliaries’ (Krug Reference Krug2001: 314). Accordingly, the more reduced gonna becomes, the more it may be perceived as auxiliary-like, as part of its process of auxiliation. This hypothesis will be tested against my corpora in section 4, in order to assess which variant exhibits more auxiliary-like behavior. Furthermore, as several authors point out, grammaticalization is unidirectional (Haspelmath Reference Haspelmath1999, Reference Haspelmath, Fischer, Norde and Perridon2004; Hopper & Traugott Reference Hopper and Traugott2003: 99): it always progresses towards more grammaticalization. However, gonna itself is only minimally addressed in such works. As Col & Duchet (Reference Col and Duchet2000) note, when gonna is mentioned, it is typically regarded merely as a contracted or reduced form of be going to (‘elle est considérée, dans le meilleur des cas, comme la forme réduite de (be) going to’, ‘in the best case scenario, it is considered to be a reduced form of (be) going to’; my translation). It is treated as a ‘non-standard spelling’ (Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 1616; Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 898) or an ‘informal variant’ (Leech Reference Leech2004: 58).

Nonetheless, some authors point out that, with the weakening of infinitival to and its ‘incorporation into the preceding word’ (Collins Reference Collins2009: 18) in gonna, the motion meaning that could be found with be going to (as in I’m going to London) has entirely disappeared: *I’m gonna London is ungrammatical. As Krug (Reference Krug, Heine and Narrog2011) notes, be going to is thus semantically closer to the lexical verb go, while gonna will tend to be more abstract. This implies that the to-contraction is not merely phonological but also morphosyntactic in nature (Pullum Reference Pullum1997). Furthermore, research in the 2000s began to suggest that gonna might be evolving into a morphosyntactically and semantically independent form, distinct from be going to (Col & Duchet Reference Col and Duchet2000, Reference Col, Duchet, Col and Roulland2001; Krug Reference Krug2000). This idea – the potential autonomy of gonna – is a prominent issue examined in the present study and will be tested in section 4.

More recently, other authors have advocated for a potential independence of gonna, including Machová (Reference Machová2015), Lorenz (Reference Lorenz2012a, Reference Lorenz and Botinis2012b, Reference Lorenz, Hasselgård, Ebeling and Ebeling2013, Reference Lorenz, Sommerer and Smirnova2020), Lorenz & Tizón-Couto (Reference Lorenz, Tizón-Couto and Botinis2016, Reference Lorenz, Tizón-Couto, Hohaus and Schulze2020, Reference Lorenz and Tizón-Couto2024) and Daugs (Reference Daugs2017, Reference Daugs, Hilpert, Cappelle and Depraetere2021). Among other findings, Machová notably observes that the auxiliary be can be dropped with gonna, especially with first- and second-person plurals in American English. She hypothesizes that the form may acquire additional operator properties in the future. Lorenz, Tizón-Couto and Daugs, in particular, approach emergent periphrastic modals through the lens of Construction Grammar. Within this framework, constructions are defined as ‘form-meaning pairing[s] organized in a network’ (Traugott & Trousdale Reference Traugott and Trousdale2013: 1), cognitively stored in the speaker’s mind. This model assumes a ‘lexicon-syntax continuum,’ rejecting a strict boundary between lexical items and syntactic rules (Hoffmann & Trousdale Reference Hoffmann, Trousdale, Hoffmann and Trousdale2013: 1). In this framework, Lorenz (Reference Lorenz2012a: 192) identifies multiple variants for different emergent modals. Lorenz & Tizón-Couto (Reference Lorenz and Tizón-Couto2024) argue that once a ‘pronunciation’ variant reaches a high frequency compared to other variants, it is being stored cognitively as an independent lexical item. This implies that speakers do not derive gonna from be going to on the spot but rather access gonna directly from their mental lexicon as an independent form. The example of such contractions offers valuable insights into how the human mind ‘processes and stores linguistic experiences’ (Lorenz & Tizón-Couto Reference Lorenz and Tizón-Couto2024: 2). This view is further developed by Daugs (Reference Daugs2017, Reference Daugs, Hilpert, Cappelle and Depraetere2021), who also adopts a Construction Grammar perspective to argue that contractions should be distinguished from phonological reductions. In the case of gonna, the form does not merely represent a pronunciation variant of the same variable but rather reflects a true lexical choice on the part of the speaker, as the contractions are institutionalized. Daugs advocates for variants – contracted forms in both his analysis and mine – to be perceived as constructions in their own right and not mere equivalents of the mother form. Although Construction Grammar involves a form–meaning pairing, the modal contractions are distinct from their full counterparts in terms of morphology, distribution and collocational patterns. In this respect, Construction Grammar offers a productive framework for examining the transition from – and choice between – be going to to gonna, as well as the relationships established between the two forms, but also with other forms that exist in-between and beyond this network. While ‘meaning’ is a core dimension of constructions, this article focuses exclusively on the formal dimension of the pairings. This focus represents the first step of the analysis, and further research will shed light on the semantic-pragmatic component. The relationship between form and meaning – preliminarily explored elsewhere in this special issue by Mikkelsen & Morin (Reference Mikkelsen and Morin2025) – falls outside the scope of the present article.

3.1. The corpora

To better understand the heterogeneous network at work, all forms resembling be going to and gonna were extracted from two different corpora. The selection of forms was based on semantic and orthographic or phonological resemblance, depending on the corpus. The analyses are based on two different mediums: a written corpus and a spoken corpus. Since gonna is said to be the result of a phonological reduction first, it seemed interesting to investigate potential new variations used in place of gonna in spoken corpora. Moreover, previous studies indicate that be going to is especially frequent in conversation (Daugs Reference Daugs2017; Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021), which motivated the choice of a spoken corpus as one of the datasets. This also apply to gonna, but the form has rarely been studied by itself. The literature states that the proportion of be going to (and gonna) is far higher in spoken corpora than in written ones. One possible explanation for it is what Leech refers to as the ‘prestige barrier’ (Reference Leech, Marín-Arrese, Carretero, Hita and van der Auwera2013: 110) – that is, the fact that the use of semi-modals might be slowed down by ‘the taboo surrounding the use of highly colloquial forms in written (especially printed) texts.’ For this reason, a web corpus was selected as the second dataset, rather than a traditional written corpus. Online writing tends to be more spontaneous, allowing users to break free from the norms of traditional grammar (McCulloch Reference McCulloch2019). As Tagliamonte puts it: ‘variation in language is most readily observed in the vernacular of everyday life’ (Reference Tagliamonte2011: 2). In this light, comparing a spoken corpus with a web corpus offers a fruitful approach to better understand the spectrum of variations with be going to and gonna.

The written corpus is a thematic Twitter corpus consisting of geotagged American tweets related to Climate Change (Climate Change Tweets Ids, Littman & Wrubel Reference Littman and Wrubel2019; hereafter referred to as CCTweets). It was collected between September 2017 and May 2019 by a research team at the George Washington University. After filtering out tweets not geotagged in the US, the corpus comprises approximately 1,300,000 words across more than 55,000 tweets. Despite its thematic focus, the corpus retains many of the typical features of computer-mediated communication, and Web 2.0 in particular (Herring Reference Herring and Chapelle2012), such as unconventional typography, emojis, hyperlinks and other genre-specific markers. In this context, the communicative genre appears to take precedence over the thematic constraints of the corpus. This is clearly illustrated by example (5), which combines an emoji with unconventional typography – the use of capital letters for emphasis – within a single sentence. Such features reflect the user’s attempt to convey emotion, highlight key points and engage the audience in a manner typical of Web 2.0 discourse:

This type of corpus was selected because it reflects language as it is used online, in everyday contexts, by a wide and diverse range of speakers, offering valuable insight into ongoing linguistic innovation. However, we acknowledge certain limitations inherent in this dataset: the thematic focus and the limited number of characters may constitute biases in my study. These biases will be taken into account in my discussion of the results.

The spoken corpus is the Santa Barbara Corpus of Spoken American English (SBC, Du Bois et al. Reference Du Bois, Chafe, Meyer, Thompson, Englebretson and Martey2000–5). It consists of recordings of naturally occurring conversations between two or more people all over the United States in the 1990s and contains 249,000 words. Both corpora are openly accessible for academic research under Creative Commons licenses (BY-ND 3.0 US and 1.0 Universal).

A first analysis (table 1) presents raw and normalized frequencies of going to and gonna in the two corpora in comparison with the total number of words. For comparison, the raw and normalized frequencies in the Corpus of Contemporary American English (COCA, Davies Reference Davies2008–) were also added. Notably, normalized frequencies in COCA – which has included a web section since 2020, covering data from 2012 to 2019 – and CCTweets are similar, whereas the results differ markedly for the spoken corpus. This suggests that CCTweets, despite being a web-based corpus, may exhibit more traditional usage patterns than initially expected. Nonetheless, a marked decrease in the frequency of going to is observed in CCTweets compared to COCA. In contrast, within the SBC, be going to occurs much less frequently, while gonna is prominently represented in conversation.

Table 1.

going to V and gonna in CCTweets, the SBC and COCA

These two forms – going to and gonna, the more institutionalized ones – are not the only ones that can be found in my corpora, as the next subsection will demonstrate.

3.2. Presentation of the data

The transcription of the SBC only gives two potential transcriptions: going to and gonna. However, additional variants do exist, as confirmed by my web corpus. Table 2 illustrates the presence of other variants such as gunna, gon and imma, among others.

Table 2.

Typology of variants in the be going to / gonna paradigm in the SBC and CCTweets (normalized frequencies per million words)

Two important observations should be made regarding this typology. First, identifying such variants in the Twitter corpus is particularly challenging, as it requires prior knowledge of their possible orthographic realizations in order to retrieve them automatically from such a wide dataset. Second, categorizing forms in the spontaneous spoken corpus presents its own difficulties: the SBC transcriptions only include the forms going to and gonna,Footnote ³ which means the analysis initially had to rely on auditory inspection. Every occurrence of going to or gonna in the transcriptions of the sixty recordings was carefully listened to, categorized in the typology, and this process was double-checked when needed,Footnote ⁴ but I acknowledge that some degree of subjectivity may remain in the categorization. Moreover, establishing a typology that would encompass both mediums proves to be complex, as each corpus presents its own set of issues – orthographic variation in the written corpus and phonological variation in the spoken corpus. For this reason, table 2 provides two distinct typologies: phonological variants for the SBC and orthographic variants for CCTweets.

Table 2 presents an attempt at a typology of the different variants identified in the two corpora, along with an example for each variant. While the specific forms differ slightly across datasets, eight distinct variants were observed in each corpus. The traditional semantic values typically associated with going to and gonna – such as intention and inference – remain observable in the examples. Although the two corpora differ markedly in nature, several forms are attested in both: goin(g) to / [ˈɡoʊɪŋ/n tʊ/ə]; gon(’) / [ɡɔn] / [ɡən] / [gəm] and Ima, though the phonological realizations and orthographic representations may vary. Other forms, however, are exclusive to one of the two mediums. In the SBC, we find variants such as [ˈɡoʊjnə]; [gə]; [nə] / [na] and [ˈa(ɪ)məna] / [a(ɪ)mənə]. Conversely, Twitter allows for a variety of orthographic – and morphological – representations of ‘Ima’: I’ma; Imma; I’mma; Ima. It also features interesting orthographic variants of gonna: gunna and gone.

Taken together, both typologies appear to map out a continuum, with be going to at one end of the spectrum and, at the opposite extreme, the most reduced form found in both corpora: Ima. This contraction is particularly striking: the only morpheme/phoneme left of gonna in this word is <a> / [a]; [ə]. Besides, the variant exhibits what I have termed a hypercontraction, involving the first-person singular pronoun I, conjugated be (am or its cliticized form’m), and gonna, all compressed into a single unit: I(’m)ma. Although my data is synchronic, the SBC provides insight into the possible trajectory of change of these variants with an interesting continuum of forms from the phrase I’m gonna: [aɪm ˈɡɑnə/a] → [am ɡənə] → [amənə] → [amə/a]. I believe this may illustrate how synchronic variation can serve as a window onto ongoing language change. It is important to note that while such reduction may partly result from increased speech rate – especially in elements with lower semantic content such as grammatical constructions – speech rate alone cannot fully account for these forms. Indeed, the extreme phonetic reduction found in spoken data also surfaces orthographically in the written corpus (‘Ima’), suggesting a more entrenched, cognitively stored form rather than a fleeting pronunciation effect.

3.3. Defining criteria

Section 3.2. established the broad range of forms belonging to the be going to / gonna paradigm, which appears to constitute what Lorenz & Tizón-Couto (Reference Lorenz and Tizón-Couto2024: 29) describe as a ‘network of variants’. The SBC presents different realization variants, which could appear to be only phonetic variants, due to some specific phonetic context. Yet many of these same forms, with slight orthographic adaptations, also surface in the CCTweets corpus. This raises a key question: if some of these phonetic variants recur in written form, could it indicate that there are some identified forms becoming conventionalized in some way?

For the purposes of this study, I decided to investigate the immediate syntactic environment surrounding the variants (N-2; N+2). Three morphosyntactic criteria and one syntactico-semantic criterion were selected for analysis. While specifically semantic or pragmatic criteria may provide valuable insights into the potential differences between the variants – particularly regarding meanings such as intention or prediction (see Mikkelsen & Morin Reference Mikkelsen and Morin2025) – it appears that the semantic differences are too subtle to be discerned solely through an analysis that would distinguish between intention and prediction or inference, and are prone to interpretative results. For this reason, I chose to begin with a distributional analysis, which yields less interpretative results than semantic-pragmatic analyses. Nonetheless, some of the selected criteria, once analyzed, may allow us to make assumptions about meaning.

Drawing on previous works by Gesuato & Facchinetti’s (Reference Gesuato and Facchinetti2011), Col & Duchet’s (Reference Col and Duchet2000) and Berglund’s (Reference Berglund, Lewandowska-Tomaszczyk and Melia2000a, Reference Berglund2005), four criteria were selected:

• – The type of subject

• – The presence of be

• – The presence of negation

• – The verb following the variant

The final criterion, although more semantically related, was retained because it falls within the immediate syntactic context of the variants. It also offers preliminary semantic insight into potential differences between the various forms. These four criteria have previously been used to distinguish among grammatical structures and explore whether distributional patterns reflect the growing autonomy of certain forms within a paradigm.

Other possible criteria – such as the presence of adverbs or the type of clause – were not included, as my priority was to analyze the closest syntactic environment of the variants, as well as elements likely to be consistently present across all instances. Additionally, while factors such as phonetic context or speech rate are undoubtedly relevant and insightful, they apply exclusively to the spoken corpus. Given my aim of identifying common criteria that could reliably be examined across both corpora, the analysis was deliberately limited to elements that allowed for systematic comparison. This approach also helps support the hypothesis that some variants may have become conventionalized to some degree. For the analysis of the SBC, all transcriptions were manually corrected so that the variant under investigation corresponded to the acoustically identified form rather than simply be going to or gonna. Analyses for both corpora were first conducted using the concordance software AntConc (Anthony Reference Anthony2021) and subsequently manually verified to ensure that no relevant occurrences (e.g. presence of subject, presence of be) were overlooked due to the N-2/N+2 window limitation. For instance, in (6), both the subject and the auxiliary be occur outside the immediate window surrounding gonna and were therefore retrieved through manual inspection:

5.1. The status of gonna

Results from my corpora show that gonna – both in its written form and its phonetic realizations – remains one of the most frequent forms within the paradigm, alongside going to. It is the only variant able to ‘compete’ with going to in terms of language change.

My results have several implications. Firstly, the proportion of contracted be with gonna and more contracted variants compared to be going to can be interpreted in two different ways. First, one can surmise that formality is at play here: gonna is traditionally associated with more informal settings and contexts (Leech Reference Leech2004) and the contracted form of be is therefore expected as it contributes to an overall ‘informal harmony’ within the structure. However, it is worth noting that be going to was initially also associated with very informal settings (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021) and the fact that full forms of be are found with it may consequently be harder to explain. I hypothesize that, rather than mainly being a matter of formality, the contracted form of be associated with gonna is indicative of processes of grammaticalization and auxiliation. This also explains why, the closer we get to more contracted variants of be going to / gonna, the more likely it is to find be elided. These variations in the contractions, contracting gonna even more in forms such as gon(’), are a sign of the auxiliation process at play here. If gonna or other contracted forms are to be considered modal auxiliaries, the presence of the auxiliary be becomes redundant and its systematic elision can be expected.

Secondly, the absence of a verb following gonna (three times more frequent with gonna than with going to) may highlight a pragmatic difference between going to and gonna, particularly in spoken conversation. Gonna may have acquired a metadiscursive role, being used as a way to keep the conversation flowing or regain the floor (Azorin Reference Azorin, Patrukhina and Bosbach2025).

The evidence from gonna calls into question its status as an emergent modal auxiliary. Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 137) classify it as a ‘semi-auxiliary’, but the scale may need to be revisited. Gonna, in comparison to going to, has acquired syntactic characteristics that make it more akin to modal auxiliaries such as must or can. With the elision of be and the removal of to, gonna may become an auxiliary in its own right. Consistent with this, and similarly to true modal auxiliaries, gonna shows no subject agreement when be is absent: forms such as *gonnas do not appear in my data. Its univerbation may thus steer it away from the category of ‘periphrastic modals’ (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021: 484) to make it a full-fledged modal auxiliary. From a morphosyntactic perspective, my findings strongly support this trajectory for gonna.

Finally, my results – in line with recent works – indicate that gonna truly seems to distance itself from be going to, not only morphologically but also syntactically, resulting in new variants developing out of gonna, not going to. One such example is Im(m)a, a contraction of I’m gonna, and not I’m going to. Although the form is not fully stabilized yet, as evidenced by variations at both the orthographic and phonetic levels, it is noteworthy that this form is created out of gonna, which is independent enough to generate variants of its own. The emergence of such forms confirms that gonna has begun to function as an independent construction. This shift is not merely formal; it also signals a syntactic reorganization that could, over time, warrant the recognition of gonna as a construction in its own right within the framework of Construction Grammar. Whether a semantic distinction also exists between going to and gonna Footnote ⁷ remains an open question. A sociolinguistic and pragmatic approach – though beyond the scope of this article – would also offer valuable insights into potential differences between going to and gonna. If we consider the network of forms associated with gonna within Construction Grammar, new nodes appear to have emerged, connecting it to its variants – gon/ima/imma/[a(ɪ)mənə] in my study. While gonna still maintains a link to going to, its mother form, the link may no longer be strictly hierarchical (i.e. vertical link), but rather horizontal, suggesting that gonna may now be seen as another possibility (‘lexical choice’) for going to, instead of a simplified spelling and production of going to.

5.2. Other variants

The morphosyntactic analyses revealed differences in distribution not only for be going to and gonna, but also for more marginal forms. The raw frequencies for these more marginal forms do not always allow for significant statistical results but it should be noted that these forms nevertheless exist. They are attested in a web corpus and/or a spoken corpus, whereas they do not occur, or very rarely, in more traditional corpora such as COCA. In the SBC, these more marginal forms account for over 30 percent of all forms within the paradigm (32.2 percent of all forms). While these forms may be partly genre-related, they are likely not exclusively so. As Hundt (Reference Hundt2004) notes, patterns which are viewed as genre-specific can spread and have an ‘impact on the systemic possibilities and the overall spread of a grammaticalizing pattern’. Moreover, the contracted forms seem to come from spoken language first, and then spread into written production, maybe starting with more loosely normed productions (such as Twitter here). This explains why many variants found orthographically are often called ‘phonetic’: phonetic respellings. However, the emergence of innovative spellings in digital platforms, such as Twitter or online forums, challenges the adequacy of this label. The comparison between these spoken and written modes of communication offers unique insight into how new linguistic forms emerge, stabilize and sometimes become conventionalized. As argued by Budts & Petré (Reference Budts and Petré2016) and Bybee & Hopper (Reference Bybee and Hopper2001), these marginalized forms may, in time, become conventionalized. Even though Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan2021) argue that such reductions are governed by conventions up to a certain point, and that one should therefore rarely come across gona or gunna for example, my corpora do present such occurrences. The very existence of these forms indicates that speakers are indeed using them for various purposes, whether it be style or register, or for sociolinguistic reasons (age, social class, ethnicity, etc.).

With respect to the morphosyntactic criteria, the analysis of subjects highlights one morphosyntactically constrained variant or variants in the forms of ‘Ima’ and [amənə]. It raises the question as to why such variation appears specifically with the pronoun I, rather than with others. At present, frequency appears to be the most plausible explanation. Also, the absence of an explicit subject is noteworthy in light of the grammaticalization process affecting these forms. However, this observation must be contextualized with respect to the corpora investigated: a spontaneous spoken corpus and a web corpus – Twitter in particular, which obeys strict rules such as character limits (280 characters at the time of data collection). Further research into subject distribution would be worthwhile, especially in relation to semantic parameters. Given that I is the most frequently used pronoun in the SBC, it would be useful to explore whether it is systematically associated with intention. Conversely, is there a pronoun which favors the inference meaning? A deeper semantic analysis of pronoun distribution across the different variants is therefore warranted. As for the absence of be, the co-occurrence of this phenomenon with very contracted variants derived from gonna is particularly interesting in relation to the ongoing process of auxiliation.

One theoretical issue that emerges from the paradigm of forms concerns the categorization of these more marginal variants, or their definition. Lorenz distinguishes ‘variation’, of which gonna would be an instance, from ‘reduction’ such as [amənə]. According to Lorenz, reductions are shaped by how someone speaks, considering situational parameters such as speech rate or preceding phonetic element, while variations are determined by who speaks, taking into consideration social factors (Lorenz Reference Lorenz, Hasselgård, Ebeling and Ebeling2013: 146). This distinction is conceptually useful, but in actual speech, these parameters often interact in complex ways, making it difficult to categorically separate variants into purely ‘variations’ or ‘reductions’. The typology presented in this article does not allow for such separation, as the more marginal forms also seem to exhibit specific distributional patterns.

A final remark concerns the notion of dichotomy, as introduced in the title of this article. While the aim of this study was to challenge and move beyond the traditional be going to / gonna dichotomy, the findings suggest that this binary opposition remains highly relevant. On the one hand, the presence of a variety of marginal and innovative variants demonstrates that the paradigm is more complex and dynamic than a mere two-form system. On the other hand, the fact that these less frequent variants typically derive from either be going to or gonna reinforces the idea that these two forms continue to operate as central anchors within the paradigm. This is further evidenced by the frequency distributions observed: the analysis of verb types was restricted to be going to and gonna alone, not only for methodological consistency, but also because of insufficient frequencies for the other variants. Thus, rather than fully breaking free from the dichotomy, the analysis points to a structured system where binary oppositions coexist with, and may even give rise to, more gradual forms of variation. This gradient variation manifests in a continuum of forms that seem to differ in terms of phonetic reduction, morphological structure and degree of grammaticalization.