2.1 High aspect

Descriptions of be going to often gloss over a specific view of where the construction fits into the syntax. However, claims that sentences with be going to are monoclausal (Krug 2000), with be going to being semantically a modal (Klecha 2014) and taking an auxiliary or auxiliary-like role (Quirk et al. 1985; Krug 2000), seem to offer an implicit argument for locating be going to in the inflectional domain above vP. Likewise, although the syntax of the construction itself is typically not explicitly addressed, suggestions that be and going to contribute separately to the construction’s meaning (Nicolle 1998) may indicate that its pieces occupy two or more functional heads.

In her work on the semantics of the future, Copley (2009) formalizes such an approach. She notes that future be going to appears to maintain some progressive aspect in its meaning, and argues that this would mean that the construction includes some sort of functional head (henceforth Asp, following van Gelderen 2003) that triggers progressive morphology on go. In Copley’s (2009) account, be takes this Asp position while going to introduces the future reading. This introduces a complication, as progressive be is quite low in the inflectional domain (Adger 2003) while the future would be introduced higher in Mod. An account placing be going to fully in the inflectional domain must therefore posit a novel functional head: either as part of an AspP above the modal, or as a modal appearing below ProgP. Copley (2001; 2009) opts for the former, proposing that Asp takes a high position above other auxiliaries.¹ Crucially, this AspP is distinct from the typical English ProgP; canonical progressives are still inserted in the position assumed under standard approaches to the English inflectional domain (cf. Adger 2003). In her formalization, Copley (2001: 95) proposes that going to appears in the WollP proposed by Abusch (1985):

Nie (2015) incorporates a modified version of this account into her approach to the Canadian French periphrastic future, placing the future in ModP. I follow Nie in taking Copley’s (2001; 2009) use of WollP to be equivalent to ModP. Under the approach to the inflectional domain that I assume, this would give the following proposed structure for be going to:²

This approach to be going to fits the broader generalizations about the construction noted above. Given that be is in Asp and going to/gonna is in ModP, the following verb is in fact the main verb of a monoclausal sentence. At the same time, placing going to in Mod makes it syntactically a modal while fitting the characterization of it as semantically modal (Klecha 2014). It is not clear from Copley’s proposal where to is located in going to. She notes that unlike in other contexts with to as a finite tense head, the two items cannot be separated. Where to lies is exactly is beyond the scope of this paper, but I will effectively treat going-to as a single lexical item here, as the internal structure will not be relevant to the experimental design or analysis.

Note that while I follow Copley’s assertion of a progressive reading in placing be in Asp, this is not necessarily crucial to the analysis. Be could of course occupy a functional head without being aspectual. At the same time, even if there was not a second progressive aspectual head this high in the inflectional domain, placing going to/gonna in Mod would most likely necessitate insertion of be as an auxiliary because going to/gonna has progressive morphology directly. This would seem to block it from being tensed, which would mean it could not raise to T as the highest modal/auxiliary present. This failure to be tensed would lead to default insertion of be above going to/gonna (Bjorkman 2011). The key part of the high aspect account, therefore, is the positing of going to/gonna being inserted in Mod, as there are multiple reasons why be could appear above Mod.

2.2 Little v

As seen, the high aspect account essentially accounts for be going to having modal qualities (Klecha 2014) by making it syntactically a modal. However, Klecha et al. (2008) note that in Copley’s (2009) account be going to is effectively the marked form of expressing the future because its aspect head above ModP is additional material not present in will. In this sense be going to is probably more syntactically marked as well, but Klecha et al. focus on semantics. They propose the opposing view: be going to is the unmarked future, and will is marked by having a context restriction operator. As evidence that be going to is the unmarked future they note that children acquiring English adopt use of be going to before use of will, and in more contexts than will early on. Because their focus is on the semantics of be going to, Klecha et al. (2008) unfortunately do not offer a syntactic account of what this unmarked approach to be going to would look like, but we can consider two possibilities.

One possibility involves be going to being high in the inflectional domain like in Copley (2009). To do so would mean locating the entirety of be going to as a single lexical item within ModP, so as to not be more syntactically marked than will. This seems unlikely, as be is clearly an auxiliary within the construction. It participates in inversion, is tensed, and precedes negation (6–8).

In order to act as a typical English auxiliary, it seems quite clear that the be in be going to is a head distinct from going to.

The alternative is to keep the syntax of be going to as reflective of linear surface order as possible. However it is implemented, such an approach would locate be in a head below ModP. Because the construction involves be introducing a verb with progressive morphology in linear surface order, a clear possibility is that be is in ProgP and going in v or V. This appears to be the approach roughly adopted by Cournane (2015: 103), who describes be going to as a little v modal in her study of L1 modal acquisition:

As seen, going to is lower in the clause. The status of be is less clear, as the phrase is left unlabeled in the original work. However, that it is unlabeled suggests to me that there is no novel head posited, as that would necessitate some discussion. Cournane (p.c.) confirms that this tree treats be as some aspectual head while remaining agnostic as to where it is placed. As with Copley’s (2001) high aspect proposal, it is not entirely clear where to lies in the syntax since it is compiled into VP with the remainder of the sentence, although it is most likely the typical finite tense head. Taking this head to be Prog gives the little v account of be going to:

Similar views take going to/gonna to be a Restructuring phenomenon (Pullum 1997; Roberts 1997; Goodall 2006) that selects for something smaller than a full CP. My use of VP as the complement of going to/gonna is consistent with these views, but the complement could in principle be larger if the construction is indeed a case of Restructuring. Because Restructuring phenomena have monoclausal-like properties (Wurmbrand 2003), this would mean Krug’s (2000) suggestion that be going to is analyzed as monoclausal is accounted for in this account.

The account in (10) captures a view of be going to present in work like that of Cournane (2015). However, even if such a view were not present in the literature, it would make sense to adopt it as an alternative hypothesis to the high aspect account because it is effectively a null hypothesis: rather than adopt a novel functional head, assume future be going to is minimally distinct from main verb be going to in the syntax.

2.3 Interaction with the inflectional domain

In sentences where be going to is present without other modals/auxiliaries, it is difficult to distinguish between the high aspect and little v accounts.³ However, because they insert be going to in such different positions within vP and the inflectional domain, this does influence what is predicted to be able to precede the construction. In obtaining such predictions, I continue to assume standard approaches to the English inflectional domain (Roberts 1985; Pollock 1989; Adger 2003; van Gelderen 2003). Importantly the highest merged auxiliary raises to T, with no additional movement of lower heads. In (11–12) respectively I repeat the high aspect and little v accounts. In these trees, I include more of the inflectional domain for illustrative purposes.

The trees in (11–12) point to a clear difference in what strings of auxiliaries are permitted in each account. In the little v account, the progressive be is merged below ModP and PerfP. As such, we would expect examples such as (13–14), in which another auxiliary is merged and tensed above be, to be licit:

In contrast, the high aspect account merges going to/gonna in ModP, and be above other potential auxiliaries. Because in this account be is the highest auxiliary, PerfP should not be able to be merged above be going to. At the same time, because going to/gonna is in ModP, other modals are blocked from being inserted. For these two reasons, the high aspect account predicts that future be going to cannot occur as in (13–14). This is a clear contrast between the two competing accounts.

Note that these accounts are not clearly contrasted with respect to what may follow be going to. It is clear that the high aspect account predicts that auxiliary modals may not follow it, as going to/gonna is in ModP blocking other modals (15). Likewise, the account predicts that semi-auxiliaries like have to and perfects may follow be going to, as these can be merged below ModP (16–17).

However, it is less clear what the predictions are for the little v account because it is unclear what going to/gonna selects as a complement. We might assume it to be VP, but if we take be going to as a Restructuring phenomenon (Pullum 1997; Roberts 1997; Goodall 2006), this would mean that going to/gonna simply selects for something smaller than a CP. As such, the little v account could permit (15–17), block (15–17), or allow a subset of these. For this reason we will focus on the clear contrast between accounts in what may precede be going to.

The prediction that future be going to is blocked when preceded by a modal/perfect in the high aspect account, but not the little v account, takes two forms depending on whether the sentence includes going to or gonna. Gonna is only allowed with a future reading, and cannot be used as a main verb to indicate motion.

If the high aspect account is correct, this means that since future be going to cannot be preceded by a modal/perfect, examples like (13–14) are ungrammatical with gonna, as the only context in which it is permitted is blocked.

In contrast, going to can act as a main verb denoting motion. In this context its progressive would presumably be derived in the ProgP typically assumed of English and adopted by the little v account. As such, sentences such as (13–14) would not be immediately ungrammatical. However, the structure for the future reading is still blocked in such examples. In this case, the prediction is that the sentence is forced to be read as motion when be going to is preceded by a modal/perfect.⁴

It is important to note that even though the predictions when going to is present involve the meaning of the sentence, this is solely a syntactic effect. Other periphrastic future constructions, as well as the progressive, are both licit when preceded by modal/perfect, as shown in (19–20):

These predictions make clear, testable contrasts. When gonna is present, the question is whether a sentence with preceding modal/perfect is grammatical. When going to is present, the question is whether a sentence with preceding modal/perfect must indicate the future. While the questions are clear, answering them is less clear. I certainly do not trust my own judgements; sentences with modal/perfect be gonna seem quite degraded, but I do utter such examples (including 13) from time to time. It is unclear to me, then, whether such utterances are grammatical or speech errors. At the same time, corpus studies of the English future which code data for syntactic context do not account for these potential environments (Szmrecsanyi 2003; Tagliamonte et al. 2014; Denis & Tagliamonte 2017). Two issues arise with this omission: it is unclear whether the omission is due to a lack of examples or a lack of looking for them, and even if few/no instances were found, it would be difficult to prove this is a systematic gap rather than an accidental gap. For these reasons, I turn to experimental methods to obtain a larger number of non-linguist judgements on the contrasts. Judgements obtained experimentally tend to correlate with more informally obtained judgements (Sprouse et al. 2013), and in this case the extra judgements may prove helpful in resolving the research question.

2.4 Be going to and grammaticalization theory

The reanalysis of be going to to indicate the future is held up as a classic example of grammaticalization (Hopper & Traugott 1993; Krug 2000; Eckardt 2006). In this context, any syntactic description of the construction is inherently a description of some stage of its grammaticalization. We should therefore consider what the competing accounts of be going to offered above imply about syntactic change during grammaticalization. Because its grammaticalization and adoption as a dominant form of future expression is quite advanced in many English varieties (Tagliamonte et al. 2014), it makes sense to think of the high aspect and little v accounts of be going to as representing the outcome of the grammaticalization process. Taken as endpoints of this process, the two approaches assert differing manners for grammaticalization to proceed.

Both accounts share the claim that to has moved somewhere in the main clausal spine, as it no longer heads a prepositional phrase. Beyond this, the conversion of be going to into a Restructuring phenomenon is the primary syntactic change to occur in the little v account. Relatively little changes in the hierarchical positioning of heads; because be remains in ProgP, going to/gonna can at highest be merged into v (Cournane 2015). In contrast, the high aspect account changes where be going to is located quite drastically by placing the entirety of the construction in the inflectional domain since going to/gonna is in ModP and be is in a novel functional head above this (Copley 2001; 2009).

While we cannot ascribe the changes yielding the modern synchronic form to movement without historical evidence per se, it is worth noting that the high aspect account is consistent with arguments that a lexical item grammaticalizes by moving into and up the clausal spine, whether due to principles of economy (“merge-over-move”, van Gelderen 2004) or parametric changes and markedness (Roberts & Roussou 2003). The high aspect account, in which to is now in the clausal spine and future be going to as a whole is higher up the clausal spine than main verb be going to, reflects the kind of outcome we would expect from this kind of grammaticalization process. The little v account involves some of the movement assumed in this view as well, as the entry of to into the clausal spine and movement of going to/gonna to v are instances of such movement. However, the view of grammaticalization put forward by Roberts & Roussou (2003) and van Gelderen (2004) is not sufficient to obtain this account. Rather, the little v account appears to posit reanalysis as a Restructuring construction as a diachronic grammaticalization process in and of itself.

Testing which account best captures the facts surrounding the place of be going to in the inflectional domain thus offers more information beyond simply informing us about the structure of a particular phenomenon. Indeed, comparing the high aspect and little v accounts of be going to informs our understanding of the process of grammaticalization. Evidence in favor of the high aspect account would constitute independent evidence in support of views of grammaticalization as primarily movement up the clausal spine (Roberts & Roussou 2003; van Gelderen 2004). However, evidence in favor of the little v account would suggest that the process they describe is not the only syntactic process involved in grammaticalization, and would offer a path forward for exploring potential other processes.

3.1 Methods

This experiment utilized a grammaticality judgement task. Participants were presented visually with novel sentences (n = 87), constructed by manipulating examples found online through a Google search. Sentences were presented in a randomized order. Of these, 36 items were key test stimuli (Supplementary file). These included 6 uses of be gonna without additional inflectional material (21), 6 uses with a preceding modal (22), 6 uses constructed as a perfect (23), and a contrast set using be about to in these same three environments (6 uses in 3 environments gives n = 18, 24–26).

The remaining items in the task were fillers: clearly grammatical and ungrammatical items (n = 6 each, 12 total), as well as non-test conditions involving be gonna in other syntactic environments (n = 39). Participants were asked to rate these sentences for acceptability, defined as how natural the sentence seemed to the participant, on a sliding scale from 1–100 in which a rating of 1 indicated a sentence was completely unacceptable and a rating of 100 indicated a sentence was completely acceptable. Because the visual presentation of examples included gonna, participants were explicitly told that their rating of a sentence’s acceptability should not reflect whether the sentence is acceptable in formal writing. Participants were asked to test the scale on two control items before proceeding to the task. Following the task, participants gave some basic demographic information and indicated whether they would be willing to participate in a follow-up study.

The grammaticality judgement task was produced and hosted on the PCIbex Farm (Zehr & Schwartz 2018). Participants were recruited via Prolific, a platform designed to enable the screening, recruitment, and fair payment of participants in academic studies (Peer et al. 2017; Palan & Schitten 2018). The task was estimated to take 20 minutes to complete, and participants were paid the USD equivalent of £2.50 for their time (as of November 2020, a £7.50 hourly rate). Payment was approved after confirming that the participant passed attention checks which indicated that they understood and executed the task. Attention checks were defined as the 12 clearly grammatical and ungrammatical items in the task: if visual inspection of violin plots of the distribution of ratings for these conditions showed that they clearly differed, the participant passed. Participants were instructed to take the study on a desktop computer. Fifty participants were targeted for inclusion in the study. To achieve this, 52 total people engaged with the task. One returned it without completing it (i.e., reserved a space, chose not to participate, and released the space back to the broader participant pool), and one was rejected for failing the attention checks.

Because some literature suggests that be going to is more established in American Englishes than other varieties (Szmrecsanyi 2003; Tagliamonte et al. 2014: 81), participants were prescreened to live in the United States. Upon review of the demographic information that participants provided, this was too broad; although passing the attention checks shows the participants to be fluent English speakers, a few participants indicated that English was not their first language and/or they were not originally born in the US. Rather than exclude these participants, the experiment was re-run to select for a more targeted population. Fifteen participants were targeted for the re-run; 19 in total engaged with the task. One participant returned the task, one exceeded Prolific’s time limits for completion, and two were rejected for failing attention checks. The prescreening criteria used for the re-run were that the participant lived in the US, was born in the US, spoke English as their first language, and did not take the first run of the study. Comparison of both trial groups shows that any differences in ratings are minimal. The sole significant difference is that sentences with bare gonna, as in (20), are rated as slightly less natural by the second trial group (Figure 1).

Figure 1

Grammaticality ratings for bare gonna by trial group.

As seen in Figure 1, both groups rate such sentences highly despite the minimal between-group difference. This suggests to me that any differences in ratings are based more on sociolinguistic evaluation than syntactic constraints. For this reason, I present the experimental results with the two groups combined, but will account for trial group where relevant in statistical analyses.

3.2 Predictions

The key test items introduce two independent variables: Verb (gonna vs. about to) and Context (bare vs. modal vs. perfect). Regardless of the syntax of the construction, the test items in the bare context are expected to be rated as grammatical. Likewise, the data in section 2 indicates that the test items involving about to are expected to be rated as grammatical regardless of context. The high aspect and little v accounts differ with respect to gonna + modal/perfect (Table 1).

The high aspect account predicts ungrammaticality in these contexts. In the modal context, this is because insertion of the modal gonna into Mod blocks insertion of other modals in this head. Meanwhile, in the perfect context be gonna is inserted above have. This means that a statistically significant interaction between Verb and Context in the expected direction constitutes evidence in favor of the high aspect account.

In contrast, the little v account predicts gonna + modal/perfect to be grammatical, because in both ModP and PerfP can be inserted above be in ProgP. The prediction is therefore a nonsignificant interaction. Note that since it effectively is the null hypothesis, this means this experiment cannot directly support the little v account. However, nonsignificance in conjunction with high grammaticality ratings would certainly favor the little v account in comparison to the high aspect account, although further research would be required to confirm this.

3.3 Results

The data to be analyzed includes the test items for 64 participants (n = 2304; age 18–75; self-identified gender 34 female/28 male/2 other; approximately 60% White; approximately 60% with BA or higher education level). This includes 49 of the participants from the first run and all 15 from the second. One participant was excluded after completing Experiment 2 (below) and failing the attention checks. Upon review of the Experiment 1 data, it was found that this participant had very nearly failed the Experiment 1 checks as well, as the distribution of clearly grammatical/ungrammatical ratings overlapped to a large degree with means differing in the correct direction. It was concluded that the participant likely should not have passed the checks for this experiment due to the overlap, even if the difference in means was on the whole correct. Another participant who failed the Experiment 2 attention checks did pass the Experiment 1 checks, and is included in the data reported here.

Data was normalized by z-scores to account for participants using differing ranges of the sliding scale. Low values indicating overall ungrammaticality correspond to negative z-scores, and high values indicating to overall grammaticality correspond to positive z-scores. Visual inspection of the results (Figure 2) shows that, as expected, bare about to and gonna are rated as grammatical. However, they differ in how ratings pattern in the modal/perfect contexts. About to ratings decrease but tend toward grammaticality when preceded by modal/perfect. In contrast, gonna ratings are quite low and indicate ungrammaticality when preceded by modal/perfect.

Figure 2

Grammaticality ratings of be gonna.

Linear mixed effects regression, using the lme4 package (Bates et al. 2015) in R (R Core Team 2020), was used to further evaluate the results. The model was first run using random effects only. Participant and Experiment Item were included as random intercepts, and Presentation Order, representing when in the task a given item was encountered, was included as a by-participant random slope. The step function was then used to determine by a stepdown process which, if any, effects should be kept in the final model. This process kept the random intercepts, but eliminated the random slope for Presentation Order. Next, the fixed effects were added into the model alongside the random effects. I begin maximally, including Verb, Context, and Trial Group as main effects and all combinations of two- and three-way interactions. The step function was then used again to determine which fixed effects should be kept in the final model. This approach yielded the final model in (27):

I use bare about to as the baseline condition for this model (Table 2).

As seen, although the main effect of Trial Group and its interactions with Verb and Context are kept in the model, these effects are the smallest of any in the model and only the interaction between Verb and Trial Group is statistically significant. This represents the difference illustrated earlier in Figure 1: sentences with gonna are rated slightly less grammatical in the second group than in the first. In contrast, we see stronger, significant main effects of Verb and Context. As noted above, the main effect of Verb may well be due to sociolinguistic evaluation as the estimate for the effect is quite a bit smaller than the intercept estimate (–.205 vs. 1.363). This indicates that while bare gonna is rated slightly worse than bare about to, both are seen by participants as grammatical. The main effect of Context may be a grammatical effect whereby some participants find about to grammatical when preceded by modal/perfect and others do not. It may also be a task effect whereby more complex sentences are rated worse than simpler sentences. This would make sense, given that the Context effects in conjunction with the Intercept estimate still yield positive ratings. However, the interaction between Verb and Context shows that gonna is rated significantly worse than about to when preceded by modal/perfect. The large estimate values, combined with the fact that we have already observed a main effect of Verb and Context, strongly suggests that the interaction reflects a grammatical constraint, as any task effects should have already been accounted for. Finding the significant interaction between these factors is uniquely predicted by the high aspect account of be going to.

4.2 Predictions

The key test items introduce two independent variables: Phrase (VP vs. PP) and Context (bare vs. modal vs. perfect). Regardless of the syntax of be going to, a main effect of Phrase is expected: going PP is expected to favor a motion reading over going VP. Table 3 shows the competing accounts predicted outcomes in the VP condition’s three possible contexts.

The high aspect account, in which be is high and going to is in Mod like gonna, predicts a statistically significant interaction between Phrase and Context. Because in this account future going to competes for insertion with other modals and is inserted above PerfP, the future reading is expected to be incompatible with preceding modal/perfect contexts. In contrast, the little v account predicts that the presence of modal/perfect will not influence whether the sentence takes a motion or future reading.

4.3 Results

The data includes the 36 test items, normalized using z-scores, from 61 total participants (n = 2196). Trained linguists who took a pilot version of this experiment gave feedback indicating that the task was quite difficult.⁵ This feedback suggested that the difficulty lay in making the judgements rather than understanding the instructions. Indeed, participants’ responses to the filler items show that the task itself was understood well. Figure 3 shows the distribution of ratings for the items that were designed to clearly yield a motion or future reading. Because the values are z-normalized, negative values correspond to a judgement that the sentence indicated motion (a low overall rating), while positive values correspond to a judgement that the sentence indicated a future event (a high overall rating). As seen, the categories are quite distinct and nearly separated, with only a handful of outlier responses for one category overlapping with responses for the other. This is, admittedly, a post hoc assessment; participants were excluded from the study if the distribution of ratings for the two categories did not show at least some separation. However, the degree of separation, and that responses to the fillers largely fell into the extremes of participants’ ratings, suggests that the task was understood and achievable.

Figure 3

Motion/future ratings for filler items.

That participants could easily categorize sentences as having a motion or future reading in the obvious cases suggests that the difficulty reported by my piloters had to do with the test items. Figure 4 illustrates one reason why this may be the case.

Figure 4

Results of motion/future rating experiment by Phrase and Context, faceted by Trial Group.

We had expected that bare going PP would yield a clear motion reading. This was not the case; ratings ranged widely for this condition between motion and future readings. Participants evidently interpret going as indicating the future to a degree regardless of context, which introduces a bit of a semantic confound to the experiment.⁶ Aside from the results indicating that going to includes a future component, we see that bare going VP is rated as more indicative of the future than bare going PP, and that this difference is largely neutralized when a modal or perfect precede the sentence.

Visual inspection of the results therefore is suggestive of the main effect of Phrase and interaction between Phrase and Context predicted by the high aspect account. The question remaining is whether these effects are significant, or whether the wide range of ratings and small differences in means make the overall results inconclusive. The z-normalized results were subjected to linear mixed effects regression, with the basic plan of testing for Phrase/Context main effects and interaction. The model was initially built with random effects alone. Participant and Experiment Item were included as random intercepts, and Presentation Order (scaled to z-scores) was included as a by-participant random slope. The step function in R was used to determine if any random effects should be dropped from the model; all were kept. The fixed effects were added to the model including these random effects. I included Phrase, Context, and Trial Group as main effects and all possible two- and three-way interactions between them. Step was again used to determine if any effects should be dropped from the model. Trial Group was dropped both as a main effect and from all interaction contexts. This process yielded the final model for analysis described in (37).

The results of this model are reported in Table 4. Two significant effects here reflect the overall tendency of participants to rate bare going PP as involving some degree of a future reading. The intercept, representing bare going PP as a baseline, has a positive estimate indicating this future reading. There is also a main effect of being in a perfect: sentences in this condition have ratings indicating a motion reading. The remaining effects involve the going VP context. A main effect shows that this context is rated as involving a future event significantly more than the PP context. The interactions with preceding context show that going VP is rated as involving the future less when not in the bare context.

Despite the confound of going to PP involving some semblance of a future reading, the results thus reflect the predictions of the high aspect account. The main effect of Phrase is expected by all accounts, but the high aspect account uniquely predicts the significant interactions between Phrase and Context favoring the motion reading. Note that these results do not reflect a clear motion reading for going VP when preceded by modal/perfect. Rather, the effects appear to undo the main effect for the bare context. When preceded by modal/perfect, going VP has the same degree of motion/future meaning as going PP. In this light, the results are wholly consistent with the high aspect account.

5.1 Alternative approach

The interpretation of the experimental results offered above assumes that standard approaches to the syntax of the English inflectional domain (Roberts 1985; Pollock 1989; Jaeggli & Hyams 1993; Adger 2003; van Gelderen 2003), in which linear surface order of modals/auxiliaries reflects hierarchical structure and only the highest merged modal/auxiliary raises to T, are correct. My interpretation also assumes that these standard approaches apply as normal to be going to in both the high aspect and little v accounts. Relaxing this latter assumption, however, could allow the little v account to generate the experimental data. Here I briefly consider how this and another adaptation of the little v account could result in future be going to being blocked from having a preceding modal/perfect.

The issue to address here is essentially that we find in Experiment 1 that “semi-auxiliaries” (Quirk et al. 1985) are not in fact a cohesive group. Be gonna and be about to are treated differently by experiment participants. The high aspect account is easily amenable to this; the two constructions are syntactically different. But, the constructions are much more alike syntactically according to the little v account because be is merged quite low, and going to and about to may in fact both be located in v (Cournane 2015). To be viable, the little v account must be able to explain why there is a difference between be gonna and be about to. I suggest that for the little v account to derive the experimental results syntactically, the assumption that the highest merged modal/auxiliary raises to T cannot hold. Since be is inflected for tense in future be going to, we know that it does raise to T when no other auxiliaries are present. One way for the little v account to block future modal/perfect be going to is to make this raising obligatory: the be in be going to must raise to T. Such a rule would be rather arbitrary and apparently lexically specific. We see from Experiment 1, after all, that be about to is relatively acceptable with a preceding modal/perfect, which would suggest that be is merged low and stays low in this case. Likewise, other progressives permit a preceding modal/perfect. For the little v account to generate the appropriate facts syntactically, it therefore must arbitrarily introduce a new rule to the inflectional domain which applies solely to be, if and only if it co-occurs with gonna/going to.

Another possibility is that the restriction is semantic rather than syntactic. Perhaps be going to is simply incompatible with modals and perfects. Such a restriction would seem to be compositional; however, the same facts that are inconvenient for the above approach are as inconvenient here. Be going to appears to have some future and progressive components. As such, if these were to be compositionally incompatible with modals or perfects, we would expect either or both of other future constructions and the progressive to also be incompatible in these environments. This is not the case. As such, it appears that for the little v account to assert an incompatibility with modals and perfects on semantic grounds, this must again be lexically specified as only applying to be going to. Such a claim is not readily falsifiable due to its specificity, but it is unclear whether there is a principled way to justify such an approach.

What these approaches have in common is that for the little v account to be viable, there must be some novel restriction that is lexically constrained in application to be going to in comparison to other constructions like be about to. For this reason, any evidence of a wider distribution of that restriction would be difficult to explain, as an account asserting a lexically constrained rule will have to explain why the restriction applies to one subset of contexts but not another. This wider distribution of the restriction may in fact exist: usage of be fixing to, commonly associated with Southern US Englishes (Myers 2014), is reported as being quite restricted in co-occurrence with the past perfect and will (Staub & Zentz 2017).⁷ In varieties of African American English, fixing to can additionally be contracted to finna. In such varieties, we would expect to see overt be/been preceding finna in the contexts discussed in this paper (Green 1998; 2002). Google searches show that co-occurrence of finna with modal/perfect are restricted in a similar way to gonna (Table 7).⁸ I searched for several constructions that can indicate the future in their bare form, and following be and been (reflecting co-occurrence with modals or have, respectively). As seen, while co-occurrences are a low overall proportion of search results for about to and planning to, these proportions are orders of magnitude lower for gonna and finna. The rarity of examples in these cases suggests a restriction on co-occurrence.

For the little v account to be viable, it will have to explain why the same restriction exists for gonna and finna, but not for planning to and about to. This cannot simply be a matter of referring to these forms being contracted, as Experiment 2 shows that the future reading of going to is relatively blocked in the uncontracted form. Rather, there would need to be a principled semantic distinction among these constructions or reason for some of these constructions to trigger obligatory raising of be but not others. In contrast, the high aspect account would take finna to be inserted into Mod, like gonna. Thus, the grammaticalization of be + X constructions in English may involve placing X in ModP, with be in a higher functional head, more generally than be going to alone.

5.2 Grammaticalization revisited

In section 2, I noted that the high aspect and little v accounts represent differing approaches to grammaticalization. While in both accounts to appears to have moved into the clausal spine because it is followed by an infinitive, they crucially differ in what other changes must have occurred diachronically. The high aspect account effectively suggests that future be going to has moved up the clausal spine over time, as the construction is higher in the clause than main verb be going to in the present day. In contrast, that the little v account keeps be going in the same position as the progressive and a quite close position to the main verb suggests that relatively little of such movement has occurred. For this account, reanalysis of be going to as a Restructuring construction is sufficient to obtain the modern future be going to.

Taken broadly, syntactic approaches to grammaticalization suggest that one common diachronic process involves a lexical item moving into and up the clausal spine (Roberts & Roussou 2003; van Gelderen 2004). The high aspect account of the structure of be going to as a synchronic phenomenon is thus quite consistent with being an outcome of this approach to grammaticalization. For this reason, the experimental evidence in favor of the high aspect account lends additional support to the approaches advocated by Roberts & Roussou and van Gelderen. It is worth noting that the synchronic evidence offered here is consistent with a fairly maximal movement up the clausal spine, as be going to has ended up at the upper edge of the English inflectional domain. However, an anonymous reviewer notes that items in IP regularly grammaticalize into CP, so further grammaticalization of be going to remains possible in the future. Some varieties of Scots may in fact be engaging in such further grammaticalization. Sailor & Thoms (2019) observe that for some Scots speakers, gonnae can be used to introduce an exhortative clause:

They suggest that in such sentences, gonnae appears in the CP domain.

While ending up with a high be going to is consistent with syntactic approaches to grammaticalization, this does not mean that future be going to must have developed this particular structure in one fell swoop. Although synchronic experimental data cannot shed light on this, it is quite possible that there were one or more intermediate steps in the development of future be going to. Semantic evidence would seem to support such a view, as be going to V first developed as a purpose clause (Danchev and Kytö 1994; Eckardt 2006). One possible intermediate step is in fact the structure proposed in the little v account: to has moved into the clausal spine, perhaps here developing the purpose clause reading. This would enable semantic reanalysis of be going to as a future marker (Eckardt 2006) as well as syntactic reanalysis of the construction as a Restructuring phenomenon. From there, be going to could be subsequently reanalyzed upwards out of vP and fully into the inflectional domain. This would mean sentences containing the construction progressed from biclausal with two TPs, to monoclausal-like, to truly monoclausal over time.

It is worth noting that describing the high aspect account as the endpoint of the grammaticalization process may not be the complete story. Krug (2000: 253–54) notes further reduction of gonna (40), and suggests it may represent further grammaticalization of the construction.

However, this is quite reminiscent of double contraction, as in (41). As seen, the perfect auxiliary have can be contracted to the already-contracted modal would.

If going to/gonna is indeed in ModP, it would not be surprising for it to be able to contract to I and be. The high aspect account of be going to can therefore account for the appearance of reduced forms like imma without additional grammaticalization taking place. However, further research would be necessary to determine if imma is simply an instance of contraction.