Future time reference and viewpoint aspect: Evidence from Gitksan

In many languages, future time reference can be conveyed in more than one grammaticized way. An example is English, which uses will and be going to . These two forms make different semantic and pragmatic contributions, and the source of the contrast is a matter of debate. For example, Copley (2009) argues that both will and be going to have a modal component, but be going to also contains progressive aspect. Klecha et al. (2008) and Klecha (2011) also posit modality for both forms, but argue that will introduces obligatory modal subordination; crucially for them, be going to does not contain the progressive. In this paper, we address the following three questions: (a) Do any other languages show a contrast between will - like and be going to -like futures? (b) Is there cross-linguistic support for the proposal that some futures contain progressive aspect? (c) Can cross-linguistic data shed light on the debate about English? Our answer to all three questions is ‘yes’. We show that (a) Gitksan (Tsimshianic) displays a contrast between will - like and be going to - like futures; (b) their distribution provides support for progressive aspect in the latter type of futures; and (c) Gitksan contributes cross-linguistic evidence to the debate about the nature of futures in English. We provide an analysis that combines elements of both Copley’s (2009) and Klecha’s (2011) accounts. More generally, we argue that different future constructions across languages are derived by combining at least the following three building blocks: prospective aspect, a modal and the progressive.


Overview
Many languages have more than one grammaticized way to convey future time reference; an example is English, which has will and be going to.In these languages, the different forms typically convey different semantic and/or pragmatic information.For example, will is felicitous in an offer on an advertising billboard, while be going to is not (Copley 2009). (1) [A sign seen (a) and one not seen (b) on the highway.]a. We'll change your oil in Madera.b. #We're going to change your oil in Madera.(Copley 2009: 77) There is debate about how to analyze these contrasting future forms; in particular, there is controversy about whether be going to contains progressive aspect: Copley (2009) argues that it does, while Klecha et al. (2008) and Klecha (2011) argue that it does not.According to Klecha and colleagues, a core difference between the two future constructions is that only will introduces obligatory modal subordination.
In this paper we bring cross-linguistic evidence to bear on the interaction between future time reference and viewpoint aspect.We present new data from Gitksan (Tsimshianic), a language which also possesses more than one grammaticized way to refer to the future.We focus on the contrast between the plain future morpheme dim, and a complex form yukw dim, which contains the progressive marker yukw.We demonstrate that these two Gitksan futures are semantically strikingly similar to English will and be going to, respectively.We argue that the presence of the progressive marker yukw in constructions similar to English be going to provides support for Copley's (2009) proposal that be going to contains progressive aspect.In addition, we argue that the contrast between dim and yukw dim requires the adoption of modal subordination in the former but not in the latter, following proposals by Klecha et al. (2008) and Klecha (2011).Gitksan future constructions thus require features of both Copley's and Klecha et al.'s analyses.
In the remainder of the introduction we provide background about the Gitksan language and outline our data collection methodology.In section 2 we overview the two English futures and the debate about their analysis.In section 3, we show that Gitksan also has two grammaticized forms for future time reference: a plain dim, and a complex form yukw dim.We show, via a range of empirical tests, their semantic similarity to will and be going to, respectively.
Section 4 presents our analysis, according to which dim is a non-modal prospective aspect and yukw is a modal progressive.In plain dim constructions, modal semantics is introduced by a phonologically null modal.Section 5 situates our approach within the debate about the denotations of different future forms, and argues that future constructions cross-linguistically are derived from smaller semantic building blocks that can include prospective aspect, a modal, and the progressive. 1

Language background
'Gitksan' (ISO code git) is the English name for a continuum of Interior Tsimshianic dialects spoken in the northwest Interior of British Columbia, Canada.Gitksan had approximately 523 fluent speakers at the time of writing of Dunlop et al. (2018).There are active community efforts towards language revitalization and retention.
Gitksan is a predicate-initial language, but the predicate may be preceded by a nominal that has undergone fronting, as in (2)b; the movement of smax 'bear' is signaled by the intransitive subject extraction morpheme -it.

Methodology
Un-cited data in the paper come from our fieldwork.The data come from seven speakers, from several dialects.Primary consultants are Vincent Gogag (from Git-anyaaw (Kitwancool), Hector Hill (from Gijigyukwhla'a (Gitsegukla)), Jeannie Harris (from Ansbayaxw (Kispiox)), and Barbara Sennott (from Ansbayaxw).Additional consultants are Ray Jones (from Prince Rupert and Gijigyukwhla'a), Herb Russell (from Gijigyukwhla'a), and Louise Wilson (from Ansbayaxw, and seasonally Prince Rupert).
The following fieldwork methodologies were used: translation tasks (in both directions), elicited production tasks, acceptability judgment tasks, and forced choice tasks.Translations from English to Gitksan were elicited within specific discourse contexts (conveyed verbally).
The consultant was first given the context, and then asked to produce a Gitksan version of an English sentence within this context.Elicited production tasks differ only in the lack of the intermediate presence of English; consultants were given a discourse context and asked to produce an appropriate Gitksan sentence in that context.
Acceptability judgment tasks involved the consultant evaluating a Gitksan utterance in a particular discourse context.The consultants were not given a strict response scale, but simply responded verbally about whether the sentence sounded acceptable in the given context.Where relevant, we provide spontaneously volunteered comments by the consultant.
Forced-choice tasks involved the consultant being given a discourse context followed by two alternative Gitksan utterances, which had previously been either volunteered or judged as acceptable in some discourse context (therefore were known to be grammatical).Consultants were asked to choose which utterance was better in the context.In these cases, there is no explicit '#' judgment for the forms that were not chosen, because the task for the consultant was simply to pick which form they preferred.

Two futures in English
Several distributional differences between English will and be going to have been discussed in the descriptive and formal literature.Here we focus on three contrasts brought to the forefront by Copley (2009), Klecha et al. (2008) and Klecha (2011).
The first contrast concerns offer contexts.As shown in (1) and further illustrated in (6), will is felicitous inside offers, but be going to is dispreferred.It should be noted that (6)b is not totally The second contrast involves warnings, as in ( 7)-(8).Will is preferred when the explosion is contingent on some action by the addressee, as in ( 7), and be going to is preferred when the explosion will happen no matter what, as in ( 8).We will refer to these two types of contexts as conditional warnings and inevitable warnings, respectively.
(7) [There is a bomb which explodes when somebody opens the door.does not seem to rely on evidence that is only present in the utterance situation.We believe that the important feature of the contexts in ( 9)-( 11) is instead that they are discourse-initial; we will therefore refer to this set of examples as discourse-initial. 5Binnick (1971) and Klecha (2011) also refer to the relevance of discourse-initialness in their discussion of the contrast between will and be going to. 4 There is some disagreement in the literature about the status of utterances like (7)b.Klecha (2011) claims that (7)b is acceptable in contexts similar to that given here, although he notes that Binnick (1971) had claimed that be going to is not compatible with this type of context.In our judgment, will is better than be going to in this type of context, and in section 3, we will show that the going to-type future in Gitksan is clearly dispreferred here.See footnote 9 and section 5.3 for further discussion. 5The reader may notice that the notion of inevitability also comes into play here, as it did with the warning data.
In (11), for example, the speaker's upcoming marriage is conceived of as inevitable, rather than contingent on some other event or mental attitude.This is not an accident and we will show in section 4 that one unified analysis accounts for all these facts.As we have already hinted, there are two main contrasting approaches to these differences between will and be going to.Both Copley (2009) and Klecha et al. (2008); Klecha (2011) propose that will and be going to convey modal semantics and futurity.However, Copley (2009) argues that be going to also contains a semantic progressive aspect, while Klecha et al. (2008) and Klecha (2011) argue that it does not.For Klecha and colleagues, will and be going to (which they call gonna) are truth-conditionally equivalent.They differ in that only will contains a presupposition that enforces modal subordination.In what follows, we outline the core conceptual ideas behind each of these analyses, while staying away from formalism.

Copley's account
Copley's analysis is designed primarily to account for the offering data. 6She argues that the reason will allows offers, and be going to does not, is because of the composition of the two forms and the pragmatics of offering.Will conveys a necessity modal and combines either with no viewpoint aspect or with a generic aspect (we set aside generic will here).Roughly speaking, the analysis of will is paraphrased in ( 12).
(12) An utterance of will(p)(w)(t): presupposes that there is a 'director' (an agent who determines what will happen in the future) of p in w at t; asserts that in all worlds maximally consistent with the commitments of the director in w at t, the proposition denoted by p is realized (adapted from Copley 2009: 69). 7 For be going to, Copley proposes that it contains the same necessity modal as will, and additionally includes progressive aspect.One core effect of the progressive is to introduce a super-interval of time surrounding the reference time; the relevant event takes place at that larger interval in some possible worlds (see section 4.3 for more details).The relation between the reference time t and the super-interval introduced by the progressive, t', is schematized in (13). (13) 6 Our summary of Copley's analysis draws in part on the presentation in Louie (2014: 104ff).Any errors are our own.
7 (12) does not force p to be realized at a time after t.At some points, Copley leaves futurity out of the formulas and suggests that it follows from the nature of the director relation (e.g., pp.43, 67), and at other points, she hardwires the futurity (e.g., pp.37, 69, 83).We set these details aside, but what is important is that t is a time at which the director's commitments hold.It is not necessarily equivalent to the utterance time.
Copley claims that in be going to, the progressive takes a proposition q which already contains the universal bouletic modal in (12); in other words, the progressive's prejacent q has itself the form will(p). Due to the semantics of the progressive, the following happens: instead of p being realized in all worlds maximally consistent with the director's commitments at the reference time t, it is realized in all worlds maximally consistent with the director's commitments throughout an interval t' surrounding t.Crucially, t' begins before t.This means that in a be going to assertion, the director is already committed to the future event before t.
Copley's idea is that the semantics of will is compatible with the pragmatics of offering, but the semantics for be going to is not.She captures the restrictions on felicitous offering as in ( 14).
This condition considers both the role of the speaker (director) and the desires of the hearer.If the hearer wants what is offered, it will happen.If she doesn't want it, it won't.
(14) An offer q by a director d to a hearer h: presupposes that d directs q (has the ability and commitment to make q true); asserts that if h wants q, q; asserts that if h doesn't want q, ¬q.

(adapted from Copley 2009: 79)
There is also crucially a temporal component to offering: the time at which the hearer wants what is offered must be the same time that the speaker is prepared to carry it out (Copley 2009: 78).For example, if the hearer wants what is denoted by q at the utterance time but won't want it at the future time when q will happen, q shouldn't happen, as in (15).Similarly, if the hearer wanted q in the past but no longer wants it at the utterance time, proposing q at the utterance time does not make a good offer, as in (16).Neither (15) nor ( 16) are successful offers.
(15) [A is getting married this afternoon and the florist hasn't shown up!A asks B if B can provide any flowers for the wedding.B replies:] #I will bring some flowers tomorrow.(16) [Yesterday, A got married and the florist never showed up!A couple of hours before the wedding, A texted B asking if they could bring flowers, but B didn't receive the text until today.Today, B replies:] #I will bring some flowers right now.
Putting all this together, the reason that offers don't work with be going to is because the speaker is committed to q at a super-interval t' of t, where t is the time at which the hearer's desires are checked.This means that the speaker is committed to q throughout an interval that begins before the hearer's desire for q is confirmed.In other words, the speaker is committed throughout t' to q happening, even though it is possible that at some point during t', the hearer does not want q to happen.Since the time of the hearer's desires and the speaker's commitment do not necessarily align, be going to doesn't make an acceptable offer.Conversely, will does not involve any superinterval t'.Thus, the speaker is committed to q at t, which is the point in time when the hearer's desires are checked.Given that the speaker's commitment and the hearer's desires are checked at the same time, the speaker can act based on the hearer's desires, making will felicitous in an offer context.

Klecha's account
The analysis of Klecha et al. (2008) and Klecha (2011) is designed primarily to account for the warning data in ( 7)-( 8).Unlike in Copley's analysis, Klecha and colleagues argue that will and gonna are truth-conditionally equivalent, and gonna does not include progressive.The difference between them is that will has, and gonna lacks, a presupposition that obligatorily enforces modal subordination.
Modal subordination (Roberts 1989) is domain restriction on the set of worlds quantified over by a modal, and it involves anaphora.An example with will is given in ( 17).

(17)
If Edna forgets to fill the birdfeeder, she will feel very bad.The birds will get hungry.(Roberts 1989: 683) The second sentence in (17) means that in all the worlds in which Edna forgets to fill the birdfeeder, the birds will get hungry.The if-clause in the first sentence restricts the set of worlds that will in the second sentence takes into account; the domain restriction of will is anaphoric to the if-clause.
According to Klecha, will obligatorily requires a discourse antecedent that provides the domain restriction, and gonna does not need one.He thus draws a parallel between will and definites, which similarly introduce familiarity presuppositions (Heim 1982).In (18), a dog restricts the domain of assignment functions for it; it is felicitous only if there is a discourse antecedent, in this case a dog.
(18) A dog walked in.It sat down.(Klecha 2011: 370) Given that Klecha incorporates an obligatory antecedent in his analysis of will, the prediction is that will always needs an explicit or implicit if-clause (such as If you touch the door), which provides domain restriction on the set of worlds quantified over. 8 This account explains why will works well for conditional warnings, but is inappropriate for inevitable warnings.Will quantifies over every world in which, for example, the hearer goes near the bomb.In a context where the bomb will explode no matter what, there is no discourse 8 There is a similarity between this idea and an intuition of Copley's (although Klecha and Copley implement the idea in different ways and apply it to different subsets of the data).According to Copley, offers of the form will(q) are interpreted as if they had a silent If you want q clause (Copley 2009: 79-80).
antecedent necessary to provide the modal restriction for will.Conversely, be going to is not restricted to quantifying over worlds where the hearer goes near the bomb. 9 Unlike Copley's (2009) account, Klecha's (2011) analysis does not involve decomposition of the different future forms into smaller parts.Instead, Klecha argues that the contrast between them is simply a matter of 'lexical idiosyncracy ' (2011: 367).He proposes that there are in fact three separate lexical items: will, gonna, and offering-will.Klecha treats offering-will as a separate lexicalized item for two reasons: (a) in some languages, offers are not expressed using future constructions (so there is a lexically distinct form for offers), and (b) when will is used in an offer, it has a performative flavor, unlike when it is used in non-offer scenarios.This is shown in the contrast between ( 19) and ( 20).
( We therefore do not believe that a separate lexical item is needed for offering-will.Instead, we will argue that the offer and warning data (and, indeed, the discourse-initial data) can all be captured under a unified analysis that combines features of both Copley's and Klecha's accounts. 9Klecha's analysis, as it stands, seems to predict that gonna is available either with or without domain restriction; however, one also might expect there to be a pragmatic preference for will over gonna when will's presupposition is satisfied.As noted above, some speakers do in fact disprefer gonna in conditional warning contexts, while others are fine with it.

The two accounts contrasted
The contrasts between Copley's and Klecha's analyses are summarized in ( 23). (23) Copley: bare will = necessity modal be going to = prog + necessity modal

Klecha:
will bomb = necessity modal with modal subordination presupposition gonna = necessity modal without modal subordination presupposition will offer = separate lexical item Above, we introduced three core sets of data from English, involving offers, warnings, and discourse-initial contexts.We have shown that Copley's analysis appears promising for offers, while Klecha's analysis offers important insights about warnings with will.However, Klecha's analysis alone cannot account for the judgments of speakers for whom be going to is infelicitous in conditional warnings.
What about the discourse-initial cases, as in ( 9)-( 11)?Copley aims to account for these contrasts through the Subinterval Property (SIP), which bare will lacks and progressive be going to has (2009: 70ff). 10Roughly speaking, these contexts are situations in which the director is asserted to be committed to p at the moment of the utterance time. 11Since bare will lacks the SIP, it cannot be applied to the single moment of the utterance time.Conversely, be going to is fine because it does have the SIP, due to the progressive: the situation of the director being committed to p holds throughout the entire larger interval.While this analysis captures the contrast, it is not quite clear to us why bare will lacks the SIP; Copley writes only that 'Bare futures we might expect to be -SIP ' (2009: 70).
For Klecha, the distribution of will and be going to in discourse-initial contexts follows from the modal subordination account and requires no additional stipulations.In particular, will is infelicitous in discourse-initial cases because there is no antecedent to provide the domain restriction will requires.On the other hand, be going to does not require a domain restrictor and is thus felicitous in cases with no antecedent.
Given the equivocal nature of the evidence coming from English, it is worthwhile turning to cross-linguistic data to attempt to shed light on the debate.So far, cross-linguistic work has not got very far in this respect.While similar semantic contrasts to the offer data in (1) and ( 6) have been 10 A predicate p of times has the Subinterval Property if and only if for all times t, for all subintervals t' of t, the truth of p(t) entails the truth of p(t') (Bennett and Partee 1978; this formulation taken from Copley 2009: 5). 11The director may include 'the world', in the case of rain.
found in Indonesian and Turkish (Copley 2009), St'át'imcets (Salish;Glougie 2007), and Siamou (Niger-Congo;Toews 2015), the forms which are dispreferred in offer contexts do not normally overtly contain progressive morphology.Blackfoot (Algonquian; Reis Silva 2009) is, as far as we know, the only non-English language in which a correlation has been established between nonoffer contexts and progressive (or more precisely for Blackfoot, imperfective) morphology.
In the remainder of this paper, we address the following questions: (a) Do any other languages show a contrast between will-like and be going to-like futures?(b) Is there cross-linguistic support for the proposal that some futures contain progressive aspect?(c) Can cross-linguistic data shed light on the debate about English?
In the next section we begin answering these questions by turning to Gitksan.We show that (a) Gitksan displays a contrast between will-like and be going to-like futures; (b) their distribution provides cross-linguistic support for modal subordination in the former type of futures, and for progressive aspect in the latter type, and (c) Gitksan contributes to the debate about the nature of futures in English.We argue that Copley is right about be going to-type futures -they do contain progressive -and that Klecha is right about will-type futures: we can cover all three core facts about will if we assume modal subordination.Gitksan future constructions can thus be captured with a blend of parts of Copley's and Klecha's analyses.We also argue that the Gitksan data do not support Klecha's proposal that offering-will and gonna are lexically idiosyncratic, because Gitksan, like English, uses the same lexical item for ordinary futures as for offers, and it transparently composes non-offering (gonna-type) futures by using the progressive.

Introduction to dim and yukw
Future time reference in Gitksan is expressed by the pre-predicative marker dim (Rigsby 1986: 279).Dim is both necessary and sufficient for a future interpretation.Its obligatoriness is illustrated in (24).The temporal adverbial t'aahlakw 'tomorrow' conveys future time reference, but nevertheless dim is obligatory.
*(prosp) sing=pn / sick=pn James tomorrow ʻJames will sing / be sick tomorrow.'(Matthewson 2013: 357) Dim's sufficiency for a future interpretation is illustrated in (25), where dim is incompatible with eventualities that are ongoing at the utterance time, and (26), where dim is incompatible with past adverbials.See also Jóhannsdóttir and Matthewson (2007), Matthewson (2013), and Matthewson and Todorović (2018)  The English translations in (31) reflect the usual tendency for how speakers translate the two forms.In the following subsections, we present diagnostics to show that plain dim corresponds to the behaviour established for will, while yukw dim has a distribution and meaning which closely parallel be going to.Note that in all the contexts discussed in sections 3.2-3.4,there is always a clear preference for the form that we predict will fit the given context.However, it is not always the case that the other form is outright rejected in the context.Sometimes, it is a matter of preference for one form over the other, which is established via a forced choice task.

Offer contexts
In offer contexts, plain dim is preferred, and the progressive future yukw dim is dispreferred ((32)-( 34)).In some cases, we give two slightly different versions of the sentences, to show the responses from consultants who speak different dialects.These data show that just like in English, the progressive future in Gitksan is dispreferred in offer contexts, while the plain future form is preferred.

Conditional vs. inevitable warnings
The Gitksan plain vs. progressive futures also pattern like English will vs.be going to in the warning cases.( 35)-( 36) show that dim is acceptable when a bomb's explosion is dependent on the addressee's actions (a conditional warning) and yukw dim is acceptable when the bomb is going to explode no matter what the addressee does (an inevitable warning).(35) [There is a bomb which explodes when somebody opens the door.A third consultant accepts both dim and yukw dim in this inevitable warning context, but in a forced-choice task chooses the yukw dim version; this consultant also consistently volunteers the yukw dim version in this context (as do all the consultants).
In summary, we have seen that plain dim is preferred for conditional warnings, and progressive yukw dim is preferred for inevitable warnings.This matches our judgments for English will vs.be going to.As noted above, Klecha (2011) assumes that be going to is good in both types of warning in English.His analysis works well for Gitksan plain futures with dim, but does not fully account for the facts with yukw dim; applying Klecha's analysis of gonna to yukw dim would incorrectly predict that it should always be acceptable in conditional warnings.

Discourse-initial contexts
Recall that in English, will is infelicitous in certain discourse-initial contexts and be going to is fine.The same is true for the Gitksan plain vs. progressive futures, as shown in (37).For some speakers, the plain dim form is not outright rejected in these contexts; nevertheless, on forced-choice tasks, yukw dim versions are consistently preferred over dim versions.In ( 38)-( 39), the order of the forms differs, reflecting how they were presented to the consultant, to avoid an ordering bias. (

Empirical summary
The data presented in this section have shown that there is a striking parallel between plain dim and will, and between progressive yukw dim and be going to.The parallels hold for all three datasets: offers, warnings, and discourse-initial contexts.

Analysis
In this section we present our analysis of Gitksan, which is extendable also to English, as will be argued in section 5.2.We adopt a neo-Reichenbachian analysis of tense and aspect, following Klein (1994).In this framework, tenses are elements that restrict the temporal relation between an evaluation time (in matrix clauses, usually the utterance time) and a reference (a.k.a.topic) time (the time about which the clause makes a claim).Aspects are elements that restrict relations between reference times and event (a.k.a.situation) times; the relations may either be of inclusion (e.g., perfective or imperfective / progressive), or of precedence (e.g., perfect or prospective).
We begin with our analysis of non-future temporal reference in Gitksan.

Non-future tense
We showed in section 3.1 that dim is necessary and sufficient for future time reference in Gitksan.
However, the language does not grammaticize other tense distinctions: there is no overt past or present tense morphology.For example, the sentence in ( 41) is compatible both with contexts in which Yoko's running was happening before the utterance time or is happening right now.
(41) Bax̱ =t Yoko.run=pn Yoko 'Yoko ran' / 'Yoko is running.' (Jóhannsdóttir and Matthewson 2007) Following Jóhannsdóttir and Matthewson (2007), we assume that Gitksan possesses a covert non-future tense, that restricts the reference time to being at or before the evaluation time t 0 .In matrix contexts, t 0 is the utterance time. 13This is modeled in (42) using a pronominal analysis of tense; the tense denotes a time interval, which precedes or coincides with t 0 .
The value of the reference time interval is a contextually salient time, given by the assignment function g.

Dim
We now turn to future time reference.We analyse dim as a prospective aspect rather than as a future tense, and we adapt ideas of Bohnemeyer's (2014) for how to model prospectives. 14 13 See Aonuki (2021) and Todorović (2021a) for arguments that the Gitksan non-future tense is a relative tense; it is not always evaluated with respect to the utterance time.In section 4.2.we show examples in which t 0 is not equal to the utterance time. 14We will continue to refer to dim and yukw dim informally as plain future and progressive future respectively; this does not mean that they contain literal future tenses.
As shown in (43), dim applies to a predicate of events P and outputs a predicate of times and events.For any event e and reference time t, e holds at t and e causes a P-event eʹ that happens at a time tʹ following t.The pre-event e can be thought of as some event (or state) that is a precursor of the later P-event.
For example, in a case of predicting rain, e might be the gathering of clouds while eʹ is the raining event itself.
Dim co-occurs with the non-future tense, which provides the reference time argument t.
This is parallel to a common analysis of English will/would, as containing an abstract futurity marker woll which combines with present/past tense (Abusch 1985).In English, woll + present (pronounced as will) places the event time after the utterance time, while woll + past (pronounced as would) places the event time after some past reference time.As Gitksan does not distinguish present from past, and the non-future tense is phonologically null, the combination of dim and tense is always pronounced as dim.
This analysis, involving a prospective aspect dim combining with the non-future tense, correctly predicts that dim can be used either for ordinary future time reference, or for 'past futures', as in (44); see Jóhannsdóttir and Matthewson (2007) iija-m anaax-t.tell-t-tr=pn Aidan obl-1sg.iiprog prosp fry-attr bread-3.ii'Aidan told me that he was going to cook fried bread.'So far, we have not assigned any modal semantics to future time reference in Gitksan.Like the majority of the literature, we assume that reference to future events involves modality (see Cariani and Santorio 2018 and references therein for recent contributions to this debate 16 ).
However, rather than assigning modal semantics to dim itself, we propose that dim combines with a separate modal, which can be phonologically null.
The main motivation for proposing that dim can co-occur with a covert modal is that dim freely co-occurs with overt modals.They encode various modal flavours and strengths, as in ( 46)-( 47).( 46 2013) argues that when dim co-occurs with these modals, dim merely provides temporal ordering (i.e., prospective aspect); dim provides the temporal orientation of the modal, in the sense of Condoravdi (2002).Thus, in ( 46), the modal da'akhlxw conveys circumstantial possibility, and dim gives future orientation.Henry being able to cook means it is possible for him to cook at some time after the reference time (which is either the utterance time or some other salient time, depending on the context).In (47), sgi is a circumstantial necessity modal and again, dim provides future orientation.Lisa's return home happens after the time of the obligation, either the utterance time or some salient previous time.(For the idea that circumstantial modals are future-oriented, see Abusch 2012;Thomas 2014;Klecha 2016;Chen et al. 2017, i.a.). Finally, (48) shows that dim can co-occur with epistemic modals, providing future temporal orientation.Crucially, when dim is not there, as in ( 49), temporal orientation is present or past.
Given that the modal flavor and strength stay the same in ( 45)-( 46), but the temporal orientation changes, these examples provide evidence that dim is responsible for the temporal orientation of the modal and does not itself contribute modality.For the idea that futurity can stem from a prospective aspect below a modal rather than the modal itself, see Kratzer (2011), Matthewson (2012), Rullmann and Matthewson (2018), among others. 16Cariani and Santorio (2018) propose a re-analysis of future modals to deal with well-known problems when these modals combine with negation.According to them, future modals do not involve quantification over possible worlds.
These issues are outside our concerns.For parallelism with these cases, we propose that in plain future dim constructions, there is a covert modal, for which dim provides the temporal orientation.That is, the prospective aspect dim requires some kind of modal licensor, and if there is no overt modal, there is a null one. 17We do not go into the details of the flavour or strength of this null modal, but, for concreteness, we assume it is some kind of circumstantial necessity modal. 18We also assume that this modal has a modal subordination requirement, as proposed by Klecha (2011) for English will; in other words, it is not the prospective dim that requires modal subordination, but the null modal. 19  The denotation of the null modal is given in (50).
(50) ⟦ mod ⟧ g,f,h is defined only if f is a circumstantial modal base that is familiar in the discourse and h is a stereotypical ordering source If defined, ⟦ mod ⟧ g,f,h = λQ <i,<l,st>> λt λe λw .∀wʹ [wʹ ∈ best h(w,t) (∩f(w,t)) → Q(t)(e)(wʹ)] (51) shows the syntactic structure that supports the composition of Gitksan futures.The vP provides the predicate of events P that is the first argument of dim; mod introduces modality, and a P-event is asserted to take place in all the best worlds given by the modal base and the ordering source, at a time after the time provided by the non-future tense.
17 For the idea that future-denoting elements need to be syntactically licensed, see Todorović (2021b).
18 Some research explicitly investigates the conversational backgrounds of future modals.For example, Tonhauser (2011) claims that the Paraguayan Guaraní future marker has a circumstantial or an epistemic modal base, and Giannakidou and Mari (2018) claim that Italian and Greek futures are epistemic. 19We are providing only an informal rendition of the modal subordination requirement (a presupposition that the modal base is familiar in the discourse).Klecha (2011) provides a more in-depth definition, but his analysis is couched within dynamic semantics, which we are not using. (51) The semantics of the sentence in ( 52), uttered in a world w 0 at time t 0 , are given in (53).We assume that the contextually salient reference time is t 0 , the utterance time.
(52) Dim wis.prosp rain 'It will rain.' According to (53), utterance of (52) in w 0 at t 0 asserts the following: there is an event e such that in all worlds wʹ that are compatible with the circumstantial modal base (which is familiar in the discourse), and that are ranked as best by the stereotypical ordering source, e takes place at t 0 and there is an event eʹ and a time tʹ such that tʹ follows t 0 , e causes eʹ, and eʹ is a raining event that takes place at tʹ.Put simply, ( 52) is true if and only if in all circumstantially accessible stereotypical worlds, it rains within a time interval which follows the utterance time.

Progressive futures
We adopt a modal analysis of the progressive yukw.Modal analyses of the progressive have a long line of support in the literature for English (Dowty 1979;Portner 1998; among many others).As mentioned above, Schwan (2019) also argues for a modal approach in Gitksan.In addition, there is extensive evidence that reference to events is necessary for an empirically adequate account of the progressive; see, for example, Landman (1992).The analysis of Portner (1998) combines modality with reference to events, and we adapt a version of it here.
Our denotation for the progressive is given in (54).We follow Portner (1998) in assuming that the progressive relies on a circumstantial modal base; this modal base picks out the set of propositions that are relevant to whether the event is completed as an event of the right type.
The progressive also relies on an ordering source, a set of propositions that ensure that the event is not interrupted.When applied to a predicate P in a world w 0 , the progressive asserts that there is an event going on in w 0 at t 0 which, if it is not interrupted, will become an event of type P (see Portner 1998: 774).
(54) ⟦ prog ⟧ g,f,h = λP <l,st> .λt λe λw.The application of progressive yukw to a non-future predication is illustrated in ( 55)-( 56).In a context where the salient reference time g(i) is the utterance time, the sentence asserts that there is an event e going on at UT and in all the best worlds in which e is not interrupted, e is part of a larger event e' which is an event of raining.
(55) Yukw=hl wis.prog=cn rain 'It is raining.' Now we turn to the final step: progressive futures.The progressive denotation given in (54) takes as its first argument a predicate of type <l,<s,t>> (a predicate of events, provided by the vP in cases like ( 55)).In yukw dim constructions, we need a type-shifted version of the progressive: dim has already introduced a time argument, so the first argument of yukw needs to be of type <i,<l,<s,t>>>.The type-shifted version of yukw is given in (57).
(As shown in ( 44)-( 45), 'past futures' also exist, and the extension to them is straightforward.) This says that (58) is true if and only if there is an event e in w 0 at t 0 such that in all the best worlds wʹ in which e is not interrupted, e is part of an event eʹ and there is a time tʹ that includes t 0 , and in all the best worlds that are accessible from wʹ and tʹ, eʹ takes place at tʹ and eʹ causes a further event eʹʹ that takes place at a time tʹʹ following tʹ, and eʹʹ is a raining event in wʹʹ.
This complex denotation makes reference to three separate events: there is an event e in w 0 at t 0 which, if it is not interrupted, will become an event eʹ, and eʹ precedes (and causes) an event eʹʹ of it raining.This complexity is a result of the combination of our prospective analysis of dim (which involves preparatory events that cause the event described by the predicate), and Portner's event-based analysis of the progressive.In fact, Portner suggests something along similar lines for futurates (constructions which convey future time reference without overt future marking); he refers to Dowty (1979) for the original idea. 20 (i) Max was running the next day (but he got sick and had to stay in bed).
These could be handled within the present system in terms of a preparatory or planning event.The idea is that Max's event e of planning/preparing to run may be sufficient to make it the case that he runs in all worlds in Best (Circ, NI, e). 21In particular, we need to allow for a notion of interruption that applies to preparatory events, and say that Max's preparations were interrupted by his getting sick.(Portner 1998: 776) The idea is similar in our analysis of progressive futures: it is the preparatory event that is progressivized.In other words, the precursor event of something happening has already started at the reference time in the actual world, and continues/completes in all the best accessible worlds.A similar example to Portner's (i) is felicitous with yukw dim in Gitksan (60); here, what is progressivized is the preparatory state before Hector talks: (60) [I asked Hector to tell me a story about his feast last month.He wanted to share it only with me, and he almost started, when Michael walked in and he never told me the story.]Yukw (hli) 22 dim mehl-i=s Hector=hl wila wil-t dis wil=hl kwʹitxw=s prog (hyp) prosp tell-T=pn Hector=cn manr lvb-3.iitime lvb=cn arrive=pn Michael.Michael 'Hector was going to tell a story when Michael showed up.' Importantly, our analysis preserves Copley's original intuition that one core difference between will and be going to is whether the modal backgrounds are calculated at a moment, or at a larger interval.A progressive future sentence asserts that there is an event that follows t 0 in all worlds 20 See also Cipria and Roberts (2000) and Arregui et al. (2014) for a similar idea for Romance and Slavic imperfectives.
21 'Circ' is the circumstantial modal base and 'NI' is the non-interruption ordering source.
22 For Tarpent (1987: 417), writing about the closely related Nisga'a, hli dim encodes 'an event that could happen or could have happened but did not.'Indeed, our consultant's comment for (60) with hli is that "there is a hint that it didn't happen." does not fit the context, where it is not yet determined at UT that the bomb will explode (i.e., it depends on whether someone opens the door). 23(62) [There is a bomb which explodes when somebody opens the door.and 9).This is because for Klecha, gonna has optional modal subordination, and is therefore acceptable in contexts involving modal subordination.In Gitksan, the progressive future is degraded in conditional warning contexts, and applying Klecha's analysis of gonna directly to yukw dim would incorrectly rule in (62).
The null modal that co-occurs with dim in (64)a enforces modal subordination, yielding the interpretation 'if X happens, it will rain'.However, since it is already pretty certain that the rain will happen, an if-clause does not make sense here and the modal subordination requirement of the null modal is not satisfied.The progressive future yukw dim is correctly predicted to be felicitous in this context, because the future raining is already determined: in all the best worlds compatible with the facts at a time beginning before the utterance time, it is going to rain.
The same reasoning can be extended to the jumping and marriage announcement cases, repeated in ( 65)-( 66).In ( 65), it looks as if the person on the bridge is necessarily going to jump.This clashes with the semantics of plain dim: due to modal subordination, dim would force the interpretation to be 'He will jump if X happens', but no X is provided by the context.
Progressive yukw dim again makes reference to a super-interval surrounding the utterance time, so the jumping is determined before the utterance time, which matches the context. (

Summary of proposals
Our first goal in this paper was to establish whether a contrast similar to that between English will and be going to can be found in languages other than English.We investigated Gitksan plain vs.
progressive future constructions, and we provided evidence that these two forms of future time reference show the same meaning contrasts as English will vs.be going to.In both languages, the plain future (will or dim) is compatible with offers and conditional warnings, and incompatible with inevitable warnings and discourse-initial contexts.And in both languages, the progressive future (be going to or yukw dim) is dispreferred in offer and conditional warning contexts, and fully acceptable in inevitable warnings and discourse-initially.
We have analyzed Gitksan dim as a prospective aspect, which places the time at which a predicate holds after some precursor event that takes place at the reference time.Dim co-occurs with an overt or covert modal; this modal provides the quantification over possible worlds appropriate for assertions about the future.When dim co-occurs with a covert modal, we have plain future constructions; these involve obligatory modal subordination (lexically specified on the covert modal, following the spirit of Klecha 2011).Dim constructions also involve covert nonfuture tense; this gives rise to either ordinary future readings, or past-future readings, depending on whether the non-future tense picks out the utterance time or a past time.
Our second task was to determine if there is cross-linguistic support for positing progressive aspect in some futures.The answer is yes.We showed that Gitksan yukw dim transparently contains the progressive yukw.The effect of the progressive above dim is to extend the time interval of the preparatory event which precedes the prejacent's future event.Since that longer time interval crucially begins before the relevant reference point, this gives rise to the meaning that the future event is already determined (due to happen in all the best worlds compatible with the relevant facts).Such an analysis correctly predicts that inevitable warnings, which involve events determined before the utterance time, are felicitous with the progressive future yukw dim.The same holds for discourse-initial contexts.Yukw dim asserts that the future event of, for example, jumping in (65), is already determined prior to the reference time, so it is correctly ruled in.The analysis also correctly predicts that yukw dim is incompatible with offers or conditional warnings; these do not make assertions about already decided events, but are crucially dependent on the addressee's desires or actions.
On the other hand, plain future dim is felicitous in exactly those contexts (offers and conditional warnings).More generally, plain future dim is felicitous whenever there is sensitivity to the circumstances at the utterance time; this is due to the obligatory modal subordination on the null modal.This also explains why plain dim is infelicitous with inevitable warnings or discourse-initial contexts, because in those contexts, the utterance time circumstances do not affect the future outcome of the action.
In sum, we have advanced an analysis of Gitksan dim and yukw dim future constructions which draws on elements of both Copley's (2009) and Klecha's (2011) analyses.The analysis is compositional and uses only independently-required building blocks: the prospective aspect dim, the progressive yukw, and a covert modal element in plain dim constructions.
offers or conditional warnings, but allow discourse-initial contexts, are composed of progressive (-ing or yukw) plus prospective.If we tentatively assume that English uses the verb go as a prospective aspect, we have overt prospective in all cases. 24In

Future research
Much more research remains to be done.One issue that deserves further investigation is the status of English be going to in conditional warning contexts (Don't touch the door!The bomb is going to explode!).As pointed out in footnote 9, Klecha's analysis predicts that be going to is semantically acceptable here, but might be pragmatically dispreferred as will is a better choice because its presupposition is met.Given the disagreements about judgments in the literature, a first useful step would be to establish some firm empirical generalizations.In doing so, it would be good to avoid some complicating factors, such as the presence of a definite article in the bomb, which introduces potentially interfering presuppositions into the context.
In terms of the analysis of be going to, we showed that our inclusion of progressive semantics rules be going to out in conditional warnings, because the progressive asserts that the exploding will happen in all the best worlds compatible with the facts at a time beginning before the speaker touches the door.This works well for Gitksan, where the progressive future is degraded in conditional warnings, and for some English speakers, but the analysis would need tweaking for English speakers who allow be going to here.
Within Gitksan, there is one more building block that appears in future constructions: the pre-predicative marker hlaa, which we tentatively analyze as a marker of proximity, as does Jóhannsdóttir (2006).The effect of hlaa on plain dim futures is illustrated in (67).Futures with hlaa place the event time close to the reference time; a compositional analysis remains to be developed for these constructions.There is also more investigation to be done on whether other aspects, including yukw, can appear below dim.According to our analysis, progressive yukw scopes over prospective dim, and dim selects for a vP (see ( 51)).If yukw needs to be located in an AspP, we predict that yukw cannot appear under dim. 25  The facts in this regard are not clear-cut.On the one hand, sentences with yukw dim yukw are dispreferred, and often judged as unacceptable.In (68) the speaker volunteered a single yukw, and when asked about the possibility double yukw, commented that "It's okay in English but it doesn't sound good in Gitsenimx [Gitksan]."(68) [Your relatives are coming for a visit.They say they are going to arrive at 5pm.You say:] Nem=dii di~t'aa ji hlaa bekw-si'm, yukw dim (*yukw) ixw-i'y.neg+fut=foc dur~sit irr prox arrive-2pl.iiprog prosp (prog) fish-1sg.ii'I won't be home when you guys get here, I'm going to be fishing.'Further, yukw is rejected under plain dim, as shown in (68).(70) [You are arriving late to a ceremony and you're being told to be discreet when you enter because there will be a man singing and you shouldn't interrupt.]a. Yukw dim yukw limx=hl get wil ts'in-in.prog prosp prog sing=cn man comp enter-2sg.ii'The man is going to be singing when you go in.' b.Yukw dim yukw=hl limx=hl get, ji hlaa ts'in-in.prog prosp prog=cn sing=cn man irr prox enter-2sg.ii'People will be singing when you come in.' c.Yukw dim limx=hl get, ji hlaa ts'in-in.prog prosp sing=cn man irr prox enter-2sg.ii'People will sing when you come in.'One possible explanation for the acceptability of lower yukw for some speakers comes from an independently attested homophonous verb yukw, which means 'busy'.In other words, the examples in (70) could be paraphrased as 'people will be busy singing upon your entry'; the verbal yukw forms a resultative verbal compound with limx 'sing' meaning 'busy singing'. 26  However, because there is no consensus on the acceptability of yukw dim yukw cases, we have to leave this question for future research.
It is also worth asking how speakers who reject (yukw) dim yukw express future progressive meanings (as opposed to the progressive future meanings we have been discussing throughout the paper).Can they simply use yukw dim?In other words, can the construction that we argued means 'be going to V' (progressive future) also be used to convey 'will be V-ing' (future progressive)? 27 Further research is needed here, but we do not believe so.Even though, as shown in (68), speakers can use yukw dim in a context where the event will already be in progress at some later salient time interval, yukw dim is also perfect in contexts where the event is not already in progress at a future time.This includes, for example, the discourse-initial contexts such as 'It is going to rain' (≠ 'It will be raining') and 'He is going to jump' (≠ 'He will be jumping.'). 28  In general, much more remains to be revealed about viewpoint aspect in Gitksan.It is noteworthy that the syntax of the progressive yukw -sitting very high in the tree and taking a 26 Progressive yukw is likely a recent grammaticization from yukw 'busy'.The two yukws unambiguously appear in different syntactic contexts.The progressive yukw triggers dependent agreement, but for independent reasons, these changes would not show up in the structure embedded under the lower yukw in (70).We leave these issues for further research. 27We thank a reviewer for prompting us to clarify this point. 28There are additional problems for treating yukw dim as ambiguous between a progressive future ('be going to V') and a future progressive ('will be V-ing').For example, this would raise the question for English of why we have a fourway distinction between I will fish, I will be fishing, I am going to fish, and I am going to be fishing.Gitksan-internally, this idea would also have the prima facie problem that yukw is higher in the syntactic structure than dim, so compositionally it makes more sense to have yukw semantically applying after dim does, as in our analysis.subordinate clause -is quite different from that of the English progressive.There is also obviously much to be done in a wider range of languages on interactions between future markers and viewpoint aspect.We do not necessarily expect that our analysis will extend to all languages that have two contrasting grammaticalized means of expressing future time reference.Nevertheless, the close parallels between English and Gitksan, two unrelated languages, suggest that it is at least worth entertaining the hypothesis that our analysis has wider cross-linguistic validity.
We also believe that the unification of offers with conditional warnings, as both deriving from obligatory modal subordination, provides a step forward in our understanding of the English facts.

touch=cn door prog prosp explode=cn
bomb 'Don't touch the door, the bomb is gonna explode!' Consultant's comment: "When you say Yukw dim xhluxwhl bomb, there is a certainty at some point the bomb will explode whether you touch the door or not."There is a time-bomb on the door which is set to explode in two minutes from now.irr toward go-2sg.iiobl-3.iiprog prosp explode-3.ii=qudd'Don't go near it, it's going to explode!' The proposed analysis also explains the facts for inevitable warning contexts, such as in (63), repeated from (36).The time-bomb is set to explode in two minutes, i.e. the explosion is determined before the utterance time, so yukw dim is felicitous.Plain dim is infelicitous in this context, since the if-clause required for modal subordination does not fit here; the bomb will explode regardless of the actions of the hearer.

Table 1 ,
'offer' stands as shorthand for the collection of properties that plain futures have.

Table 1 :
Building blocks in plain vs. progressive futures in English and Gitksan.