The semantics of weak imperatives revisited: Evidence from free-choice item licensing

: This paper provides a new analysis for the semantics and pragmatics of weak (permission/ acquiescence) imperatives. In a signiﬁcant modiﬁcation to the To-Do-List (or minimal semantics – strong pragmatic) theory of imperatives (Portner 2007; 2012; von Fintel & Iatridou 2017), I argue that in weak imperatives, the utterance of the imperative is directed not at the To-Do-List of the addressee, but at a separate list which contains the set of possible courses of action contemplated by the addressee (which I term the List of Actions Under Consideration by the addressee). I support this claim by a new detailed analysis of free-choice item licensing in imperatives (based on the dependent indeﬁnite analysis of FCIs put forth by Giannakidou 2001). I also show how my model correctly predicts that strong imperatives are felicitous out of the blue whereas weak imperatives need the prejacent to be already part of the context. It is also pointed out how this new approach helps us cut through the familiar controversy of whether weak imperatives can create obligations. Finally, I argue that the strong vs. weak imperative distinction is orthogonal to the degree of speaker endorsement ( pace von Fintel & Iatridou 2017): this claim is also supported by a new look at data from Rhaetoromance (Poletto & Zanuttini 2003), where the strong vs. weak imperative distinction is encoded overtly on the morphosyntactic level in a binary fashion.


Introduction
The goal of this paper is to provide a new analysis for the semantics and pragmatics of weak (permission/acquiescence) imperatives. Based on observations concerning the licensing of free-choice items in imperatives and other evidence, I will propose a significant modification to the To-Do-List (or minimal semantics -strong pragmatics) theory of imperatives (Portner (2007;; von Fintel & Iatridou 2017). I will argue that in weak imperatives, the utterance of the imperative is directed not at the To-Do-List of the addressee, but at a separate list which contains the set of possible courses of action contemplated by the addressee (which I will term the List of Actions Under Consideration by the addressee).

Tamás Halm
The first piece of evidence comes from the licensing of free-choice items (such as cualqier in Spanish, n'importe quel in French or opjosdhipote in Greek). Our starting point is the empirical observation that FCIs are acceptable in weak imperatives but not in strong imperatives. I will show that current theories of the semantics and pragmatics of imperatives cannot properly accommodate this fact: strong semantics approaches as well as TDL-style approaches appear to predict that FCIs should be licensed in strong and weak imperatives alike, contrary to fact.
I will argue, following the dependent indefinite analysis (Giannakidou 2001), that FCIs need a (non-singleton and non-empty) set of ⟨accessible possible world, individual⟩ pairs in order to be pragmatically felicitous. Furthermore, I will point out that weak imperatives in general (whether containing an FCI or not) are only felicitous if the fact that the action described in their prejacent is already being contemplated by the addressee is part of the common ground. Based on this, I will argue that in contrast to strong imperatives which affect the TDL of the addressee, weak imperatives affect a separate component of the common ground, the so-called List of Actions Under Consideration by the addressee, and their effect is the lifting of any prohibition that the addressee may have ascribed to the speaker with regard to the action described in the prejacent.
This List of Actions Under Consideration naturally contains alternatives, which satisfies the need of FCIs for a set of ⟨possible world, individual⟩ pairs. This explains why FCIs are felicitous in weak imperatives but not in strong ones. The observation that strong imperatives are felicitous out of the blue whereas weak imperatives need the prejacent to be already part of the context also follows from this model. Furthermore, this modification also helps us to avoid the unwelcome situation of weak imperatives creating obligations (something which has long been recognized as a very problematic prediction of the TDL approach): I will argue that whereas strong imperatives indeed add propertiesto-be-made-true to the TDL of the addressee and thus create obligations, weak imperatives do not directly affect the TDL and do not create any obligations.
I will also look at the issue of speaker endorsement. A well-known concept in declaratives and interrogatives, it has been introduced as a parameter of the semantics of imperatives by von Fintel & Iatridou (2017) in order to account for the weak vs strong imperative distinction. I will argue that the real difference between strong and weak imperatives lies in what kind of addressee-oriented list they manipulate: whether the To-Do-List or the List of Actions Under Consideration. The degree of speaker The semantics of weak imperatives revisited 447 endorsement is orthogonal to the strong vs. weak semantics distinction; it does, however, play a role in the subdivision of the TDL (into order, invitation, advice etc. imperatives) and of the LAUC (into permission, acquiescence, indifference etc. imperatives).
Finally, I will revisit Poletto and Zanuttini's (2003) analysis of imperative particles in Rhaetoromance. Their analysis was that these particles encode the "order" (given from the point of view of speaker) vs. "advice and permission" (given from the point of view of addressee) distinction. I will argue that based on the data presented in their paper (such as the fact that "advice and permission" particles are admissible in the first conjunct of imperative and declarative constructions whereas "order" particles are not), one can conclude that what these particles encode is in fact the strong vs. weak imperative distinction.
The fact that the strong vs. weak imperative distinction is obligatorily encoded as a clear binary distinction on the overt morphosyntactic level in some languages supports a "binary" model (such as the one proposed here, where strong imperatives affect the TDL whereas weak imperatives affect the LAUC) as opposed to a "graded" model (such as von Fintel and Iatridou's proposal where all imperatives affect the TDL and the strong vs. weak distinction is a function of speaker endorsement). A binary model predicts that an imperative is either strong or it is weak, with no shades in between (since it either affects the TDL or the LAUC). A graded model predicts that the strong vs. weak imperative distinction is to be conceived of as a spectrum: since speaker endorsement can freely range from the very weak to the very strong, one would expect that besides the prototypically strong and weak imperatives, there should be imperatives of "medium" strength as well (in the case of "medium" speaker endorsement). The evidence from Rhaetoromance thus supports a binary model over a graded model. The paper is structured as follows. First, I will present some data on FCI licensing to introduce the problem in more detail. This will be followed by a short overview of the main observations and theories of FCIs in general. Section 4 will be about previous approaches to the semantics and pragmatics of imperatives. After thus setting the stage, section 5 will contain a critical assessment of previous accounts for FCIs in imperatives. In section 6, I will present my analysis of the relevant data and the proposed model. In section 7, I will summarize the results and conclude the paper.

Imperatives and FCIs
The compatibility of free-choice items with imperatives is debated. In this section, I will present the main theoretical approaches and will also introduce some important data. I will argue that the main reason for the divergence of views in the literature is that most of the previous proposals have remained largely agnostic as to the details of the formal semantics and pragmatics of imperatives.
Standard treatments of FCIs (Quer 1999;Giannakidou 2001;Jayez & Tovena 2005) regard imperatives as an environment which licenses FCIs: (1) Pon cualqiuer excusa. put:IMP:2SG FCI excuse 'Give any excuse.' (Spanish) (2) Dhialekse opjodhipote forema. pick:IMP:2SG FCI dress 'Give any excuse.' (Greek) (3) Prends n'importe quelle carte. take:IMP:2SG FCI card 'Take any card.' (French) There is less consensus on whether both strong (command-type) imperatives and weak (permission) imperatives license FCIs. There are three main views in the literature: (A) some authors (cf. Aloni 2002;Kaufmann 2012) claim that imperatives containing FCIs lack a possibility reading (that is, FCIs are typically found in strong imperatives), (B) other authors (cf. Giannakidou 2001;Giannakidou & Quer 2013;Staraki 2018) argue that FCIs are grammatical in all imperatives, however, they are pragmatically infelicitous in (most) strong imperatives, (C) yet others (Strickland 1982;Haspelmath 1997) argue that in imperatives (as in contexts of necessity in general), FCIs are unacceptable, with the exception of weak imperatives, which are functionally more akin to contexts of possibility. 1 The semantics of weak imperatives revisited 449 This divergence of opinions is in large part due to the fact that until recently, most analyses of FCIs in imperatives remained agnostic as to the formal details of the semantics and pragmatics of imperatives, 2 and simply subsumed the scrutiny of FCIs in imperatives under the more general analysis of FCIs in necessity (and possibility) modal contexts; even though the difference between strong imperatives (expressing deontic necessity) and weak imperatives (expressing deontic possibility) has been a central concern of the formal semantics and pragmatics of imperatives for a long time (Lewis 1979;Hausser 1980;Portner 2007;von Fintel & Iatridou 2017). Below, I will put forward a new analysis of FCIs in imperatives that is firmly grounded within the formal semantics and pragmatics of imperatives. This approach will not only elucidate FCI licensing in imperatives, but will also help us to create a more precise formal model of the semantics and pragmatics of imperatives by adding a crucial modification to the model of von Fintel & Iatridou (2017).
In her comparative analysis of English any and French n'importe quel, Strickland (1982, 19-20) was the first to note that command imperatives and permissive imperatives crucially differ in terms of FCI-licensing. (4) is clearly unacceptable as an out-of-the blue command-type imperative: (4) # Bring me any chair. (out of the blue) However, in a context where the imperative is interpreted as a permisson, the FCI is licensed: In (6a), the main verb of the matrix clause shows that this is a command, and the sentence is accordingly infelicitious with the FCI. (6c) is an imperative where without context, both the command and the permission readings are accessible: hence the felicitousness of (6c) is uncertain. (6b) is a variant of (6c) where the additional most azonnal 'right now' indicates urgency from the speaker's side and makes the command reading more prominent and the sentence with the FCI is as a consequence infelicitous. (6d) contains the grammaticalized marker of permission/acquiescence and the sentence is felicitous with the FCI as expected. (6e) is an explicit permission where the FCI is felicitous.
In fact, (6) represents a scale of felicitousness. At both ends are utterances where the main verb of the matrix clause makes the command vs. permission force of the utterance unambiguous; and as a consequence, the felicitousness of the sentence with an FCI is straightforward. In the middle of the scale, we find imperatives without any clue added to favour either a command and a permission reading: accordingly, the felicitousness with FCIs is uncertain. Adding elements which either indicate urgency on the speaker's side (most azonnal 'right now') or overtly encode permission (nyugodtan 'permission/acquiescence marker') decisively tips the scale in either the command or the permission direction.
Further evidence is provided by looking at infinitival imperatives, which in Hungarian obligatorily have a strong (verging on the rude) command reading: a. out in in earlier accounts as well (Strickland 1982): FCIs are acceptable in weak (permission) imperatives but unacceptable in strong imperatives. 5

Free choice items cross-linguistically and in Hungarian
Intuitively, FCIs are elements that express free choice (Vendler 1967) and are further characterized by their (non-)availability in a number of specific environments (the Greek examples are taken from Giannakidou 1997 and2001). They are typically not licensed in affirmative contexts (10) and they are licensed in possibility modal contexts (11) and in generics (12) As far as negative episodic contexts are concerned, the licensing conditions are more nuanced. E.g., (13) is licensed only if the FCI is stressed and then the FCI has an absence of preference reading ('just any'): (13) Dhen idha opjondhipote not saw.PERF.1SG FC.person (i) # 'I did not see anybody.' 7 (ii) 'I did not see just anybody.' 5 An apparent counterexample to this claim from Greek will be discussed later in section 5, ex. (41). 6 For more details on licensing environments, see the detailed survey of Vlachou (2007;. 7 Note that English any appears to be an outlier among FCIs in that it is licensed under negation, which led many authors to assume that there are two anys: a polaritysensitive any and a free-choice any (see brief discussion below in main text). For a more nuanced view on the licensing of FCIs under negation, see Vlachou (2007;. One school of thought aimed to analyze FCIs as a class of polarity-sensitive items (Baker 1970), with Ladusaw (1979) distinguishing between two kinds of any: polarity-sensitive any (appearing in negative contexts) and freechoice any (appearing elsewhere). Kadmon and Landman (1993) proposed a uniform analysis of both kinds of any.
A key characteristic of the dependent indefinite approach is that the distribution of FCIs is derived from their lexical semantics. FCIs are represented as intensional indefinites, i.e., dependent indefinites which contain a possible world variable that must be bound by an appropriate intensional quantificational operator (i.e., it cannot be bound by text-level existential closure) in order to be licensed. The perceived universality of FCIs is derived from the presupposition of exhaustive variation: in each accessible possible world, a different value should be assigned to the FCI, and these assignments should exhaust all the available values from the domain of quantification. The licensing conditions of FCIs (they need to be under the scope of a nonv-veridical operator and the sentence they appear needs to be non-episodic) are derived from the lexical semantic properties of Concerning Hungarian, Halm (2013;2015;2016a;b) examined the behaviour of FCIs across a wide range of environments and constructions, including the standard tests of quantificational force (almost-modification, modification by exceptive phrase, donkey anaphora, predicative use, incorporation and split readings with modals) and various structural positions (identificational focus, contrastive topic). The results corroborate the analysis of FCIs in Hungarian as dependent indefinites (cf. Halm 2016a;b for a detailed account and a critical assessment of earlier proposals such as Hunyadi 1991;Abrusán 2007 andSzabó 2012). In terms of licensing environments, FCIs in Hungarian are not licensed in plain affirmative episodic sentences (14), they are licensed in possibility modal contexts (15), they are not licensed in generics (16)  Finally, it might be useful to consider how the landscape of FCIs in Hungarian fits into Vlachou's (2007; typology of full-set FCIs (which require that all alternatives without exception are considered), subset FCIs that express ignorance (which require that unknown alternatives are considered) and subset FCIs that express indifference (which require that non-preferred alternatives are considered). Hungarian, unlike some other languages, does not have separate lexical forms for these different kinds of FCIs: rather, the reading the FCI receives dependends on prosodic and syntactic fac-tors: that is, a lexical item such as bárki 'anyone' can be interpreted as a full-set FCI (anyone), a indifference-type subset FCI (just anyone) or an ignorance-type FCI (wh-ever) (for details, cf. Halm 2013; 2016a).

The semantics and pragmatics of imperatives
Any theory of the meaning of imperatives has the double task of accounting for both the denotational semantics and the illocutionary force of imperatives. Put simply, the imperative Go home! minimally has a denotation very similar to the declarative You go home., but crucially, the mere utterance of this imperative updates the common ground to the effect that the addressee now has the obligation to go home (In view of the speaker's wishes, you must go home.). Actual theories of imperatives, however, significantly differ in terms of the division of labour between denotational semantics and dynamic pragmatics in bringing about this change of context.
At one end of the spectrum, a rich denotational semantics model such as Kaufmann (2012) goes as far as to assume that imperatives, in fact, denote necessity modal propositions; that is, the denotational semantics of Go home! is essentially the same as that of You must go home and the difference between an imperative and the declarative is due to a presuppositional meaning component which accounts for the performative effects associated with imperatives.
On the opposite end of the spectrum, Portner (2007) argues for a very thin denotational semantics, where imperatives simply denote a property restricted to the addressee: (18) sit down! = λwλx : x is the addressee. x sits down in w As a consequence of this weak denotational semantics, much of the heavy lifting of eliciting imperative illocutionary force is done in the dynamic pragmatics component: the utterance of an imperative adds the property it denotes to the so-called To-Do-List of the addressee (much as a declarative denotes a proposition and adds it to the common ground).
Von Fintel and Iatridou (2017) reviewed the full spectrum of approaches and argued convincingly for a refined version of Portner's (2007) approach based on a close examination of weak (permission or acquiescence) imperatives and imperatives in conditional conjunctions. As I will argue below, facts from FCIs in imperatives support the main thrust of a thin denotational semantics approach, but also point to the necessity of some important modifications in the dynamic pragmatic component. (For

Weak imperatives
As we have seen, the difference between strong (command-type) and weak (permission or aqcuiescence) imperatives is crucial in terms of FCIlicensing. Therefore, we should look at the analysis of strong vs. weak imperatives in more detail.
Strong denotational semantics approaches assume an in-built necessity modality for imperatives, and as a consequence, accounting for weak (permission) imperatives is a major challenge for them. An assumption first raised by Wilson and Sperber (1988) and then formalized in Kaufmann (2012) is that weak imperatives are such that through contextual weakening, the modality is relativized to the hearer's desires (as opposed to command imperatives, where it is relativized to the speaker's desires) and this is how a permissive reading comes about: 8 (19) Go home. [command] ∼ 'in view of the speaker's wishes, the addressee must go home' (20) Go home.
[permission] ∼ 'in view of the addressee's wishes, the addressee must go home' I believe this approach has a number of shortcomings and cannot adequately explain the facts of FCIs in imperatives. Firstly, von Fintel and Iatridou (2017, 6-8) point out that if this contextual weakening mechanism did indeed exist, we would expect it to work not just with imperatives but also other related expressions such as directives, desideratives or deontic modals. This, however, is not the case. Secondly, Giannakidou (2001) shows that imperatives containing FCIs (whether we analyze them as commands or permissions) have an existential reading which is certainly problematic for any theory assuming an in-built necessity modality denotational semantics. 9 Also, von Fintel and Iatridou (2017) show that the existence of socalled imperative and declarative constructions (where imperatives appear in the first conjunct of conditional conjunctions) 10 also points to a weak (non-modal) denotational semantics of imperatives. The observation that imperatives are incompatible with certain sentence adverbials whereas the corresponding necessity modal sentences are not 11 is also an indication against closely modelling imperatives on necessity modals (Gärtner 2017).
Frameworks where an imperative has a weak denotational semantics with no in-built necessity modality are more promising for accomodating the facts concerning FCIs in imperatives and permission imperatives in general. The task of pinpointing the difference between command imperatives and permission imperatives then falls to the dynamic semantic component. (i) Study hard and you will pass the class. (ii) Ignore your homework and you will fail this class. (iii) Open the paper and you will find 5 mistakes on every page. 11 Cf. (Gärtner 2017, 123): You must (unfortunately/allegedly/presumably) stay here. vs.
Stay here ( *unfortunately/*allegedly/*presumably). Note that as far as unfortunately is concerned, this contrast is relevant even when the modal is used performatively. 12 Hans-Martin Gärtner (pc) has raised the question whether the modal approach (cf.
Kaufmann 2012) might have an advantage over the weak semantics approach (Portner 2007, von Fintel & Iatridou 2017 in analyzing so called descriptive/assertive uses of imperatives such as: (i) Péter azt mondta hogy János menjen haza Peter it:ACC say:PAST:3SG that John go:IMP:2SG home 'Peter said that John go (imperative) home.' (intended: 'Peter said that John should go home.') (ii) János szerint fogadd el az állást John according.to accept:IMP:2SG PRT the position:ACC '*According to John, take the job.' (intended: 'According to John, you should take the job') The relative order of the verb and the verbal particle is a clear proof that the sentences above contain a true imperative (as opposed to a subjunctive). In Kaufmann's (2012) model, (proto-)imperatives are, to begin with, underspecified in terms of descriptivity/performativity, and a presuppositional mechanism is used to derive descriptive vs. performative uses (cf. Gärtner & Gyuris 2012 for a Kauffmann-style analysis of the sentences above and Stegovec & Kaufmann 2015 for a discussion of Slovenian embedded imperatives). In the To-Do-List approach, Portner (2007) analyzes embedded imperatives such as (i) by resorting to "monsters" (operators capable of shifting the context of evaluation, cf. Kaplan 1989;Schlenker 2003): (iii) s say to h Φimp = w: C is a context representing what s says to h in w ∧ C + Φimp = C Evaluating the merits of these two approaches is beyond the scope of this paper: for To recapitulate, Portner (2007) assumes that imperatives simply denote a property restricted to the addressee: The imperative illocutionary force is elicited in the dynamic pragmatics component: the utterance of an imperative adds the property it denotes to the so-called To-Do-List of the addressee (much as a declarative denotes a proposition and adds it to the common ground).
It feels natural to try and differentiate strong imperatives from weak imperatives by assuming that they affect different chunks of the addressee's To-Do-List. Indeed, Portner (2007) argues that it is possible to subdivide the TDL according to the nature of the obligation (in parallel fashion to differentiating types of modals in term of the nature of necessity (or ordering source in the sense of Kratzer 1981)): 'Noah should sit down right now, given that he's been ordered to do so.' (deontic necessity) (23) Have a piece of fruit (invitation imperative) 'Noah should have a piece of fruit, given that it would make him happy.' (bouletic necessity) (24) Talk to your advisor more often (suggestion imperative) 'Noah should talk to his advisor more often, given that he wants to finish his degree.' (teleological necessity) The task of sitting down is added to the TDL based on the speaker's authority. The task of having a fruit is added to the TDL based on the addressee's desires. The task of talking to the advisor more often is added to the TDL based on the addressee's goals. Portner (2007) argues that the TDL is in fact divided into separate sections based on this, and that these our purposes, it suffices that both schools of thought have a way of accommodating embedded imperatives. For a syntactic analysis of embedded imperatives in Hungarian, see Varga (2014, 90-121). Turning to (ii), I propose that, pace Gärtner & Gyuris (2002), what we see there is the performative use of the imperative, the only "complication" being that the speaker and the source of the obligation are decoupled, but this can be readily accommodated in the TDL framework. That is, the only difference between János szerint fogadd el az állást and Fogadd el az állást is that in the former, the speaker and the source of obligation are different, whereas in the latter, they are identical. Otherwise, the mechanism is the same: a property-to-be-made-true is being added to the addressee's TDL by virtue of the imperative having been uttered.

Acta Linguistica Academica 66, 2019
Unauthenticated | Downloaded 09/14/23 04:09 AM UTC The semantics of weak imperatives revisited 459 categories have a grammatical reality, as evidenced by various phenomena such as the function of an overt subject signalling speaker authority in English imperatives (Potsdam 1996) or the use of particles differentiating between kinds of imperatives in Badiot (Poletto & Zanuttini 2003). 13 Portner (2007) does not discuss weak imperatives in detail, only suggesting that the difference between permissions and invitations may be "whether it is presupposed that the speaker has the authority to prohibit the act in question". Crucially, this means that permission imperatives manipulate the TDL and thus create new items of obligation for the addressee. In my view, this is not correct. While an invitation such as (23) above does indeed create a new obligation, a true weak imperative merely expresses the speaker's indifference/acquiescence to a possible course of action by the addressee.
To paraphrase and expand an example of von Fintel and Iatridou (2017), consider: An invitation imperative cannot be felicitously followed by an expression of indifference, while a weak imperative can: 13 Portner (2007, 373) formalizes the pragmatic function of imperatives as follows: Where ϕimp is an imperative clause and C is a context of the form ⟨CG, Q, T, h⟩ (where CG is the common ground, Q is the question-under-discussion stack, T is the To-Do-List function: a function from individuals to sets of propositions/properties, h is the salient selection function that can be deontic, bouletic or teleological): The main conclusion is that there is a crucial difference between invitation imperatives and permission imperatives: -invitation imperatives create obligations (similarly to order-type imperatives or suggestion-type imperatives), therefore, it is straightforward to assume that they affect the addressee's TDL, -permission (or acquiescence) imperatives do not create obligations, therefore, it is very problematic to simply assume that they affect the addressee's TDL: after the utterance of (26), the addressee is not in any sense obliged to open any window, whereas after the utterance of (25), the addressee is obliged (if only by common courtesy) to take a sandwich. 16 14 See (6) and footnote 4 for a discussion of the discourse particle nyugodtan.
15 (29) is of course perfectly acceptable as a permission; however, in a situation where a permission statement is inadequate and the sentence can only be interpreted as an invitation, it is clearly unacceptable. 16 In other words: in case of an invitation imperative, the speaker wishes that the addressee carry out the action described in the prejacent, so it is the intention of the speaker that ϕimp be added to the TDL. (Since, however, it is merely an invitation but not a command, the addressee has the liberty not the add it to the TDL.) This is different from a permission, where the speaker does not actually wish that the addressee carry out the action, but merely indicates their permission, acquiescence or indifference with regard to that action.
In later work, Portner (2012) addresses this problem by assuming that permissive readings of an imperative p! arise when there is a prior obligation q which is inconsistent with p, and this inconsistency leads to a choice. However, as von Fintel and Iatridou (2017) point out, conflicting obligations do not always give rise to a choice, and Portner (2012) has no convincing account for when exactly they do and when exactly they do not in the case of imperatives. Von Fintel and Iatridou (2017) make a proposal to fix this problem: leaving the larger part of Portner (2007)'s model intact (weak non-modal denotational semantics, compartmentalized To-Do-List in the dynamic pragmatics segment), they suggest a focus on the strength of speaker endorsement behind an imperative.
First it is pointed out that all speech moves, including assertions and questions, display variable speaker endorsement. Thus, the assertion (30a) can be weakened by a number of means to indicate weak speaker endorsement (Malamud &  In (30b-c), the speaker floats the proposition that Tom is here but does not fully commit to it and there is no clear expectation that this proposition should be added to the common ground. Similarly, in so-called conjectural questions such as oare-questions in Romanian (Farkas & Bruce 2010), the speaker indicates that "settling the issue is not necessarily a projected conversational future and therefore that answering the questions is optional" (op.cit., 11): 17 (31) Oare Petru a sosit deja? oare Peter has arrived already 'Has Peter arrived already?' That is, the speaker floats the question but does not insist on it being added to the question-under-discussion stack.
In a similar vein, von Fintel and Iatridou (2017) propose that when it comes to imperatives, it is also possible to indicate weak speaker endorsement: this is how weak (or permission/acquiescence) imperatives come about: the imperative is floated, but it is up to the addressee whether to add it to her To-Do-List. Von Fintel and Iatridou are confident that the rich discourse model built by Farkas and Bruce (2010) for questions and refined by Malamud and Stephenson (2015) for assertions could, with appropriate modifications, accomodate imperatives as well.
Von Fintel and Iatridou (2017) discuss one possible objection to their solution: no matter how weak the speaker endorsement, if the addressee does in the end decide to add the imperative property to her TDL, it becomes an obligation, something which may be considered an unwelcome end result, stemming as it does from the speaker simply having expressed her indifference to the action under consideration. By way of a solution, von Fintel and Iatridou (2017) point out that Portner (2007) in fact assumes a TDL divided into sections such as: -command section: this is where properties received from order-type imperatives are filed -advice section: this is where properties received from a suggestion-type imperatives are filed -etc.
The relationship of the addressee to the properties also varies with each section. In the command section, the addressee is truly obliged to make the property true of herself, whereas in the advice section, the addressees stance is less than a full obligation, more like a commitment. Von Fintel and Iatridou (2017) propose that the TDL has a section for commitments taken on freely by the individual, and this is where the properties denoted by a weak imperative would be filed (if the hearer so decides).
To summarize, we have reviewed three main approaches to modelling weak imperatives. Kaufmann (2012) assumed that weak imperatives are derived from strong imperatives by relativizing their denotationally builtin, originally speaker-oriented modality to the addressee's desires: (32) Go home. [command] ∼ 'in view of the speaker's wishes, the addressee must go home' (33) Go home.
[permission] ∼ 'in view of the addressee's wishes, the addressee must go home' Portner (2007; assumed that imperatives denote a property, and in terms of dynamic pragmatics, the utterance of the imperative adds the task of making this property true of herself to the To-Do-List of the addressee. The To-Do-List has sections based on the nature of the imperative (corresponding to the source of the obligation: order, invitation, advice etc.).
Weak (permission) readings arise when the property denoted by the imperative is inconsistent with a prior prohibition concerning that property. Von Fintel and Iatridou (2017) modulate Portner's (2007; account: they assume that speaker endorsement behind imperatives varies: full endorsement is the default but in the presence of explicit markers or contextual cues, speaker endorsement is perceived to have weakened and this gives rise to weak imperatives.

FCIs in imperatives: previous accounts
While the difference between strong imperatives (expressing deontic necessity) and weak imperatives (expressing deontic possibility) has been a central concern of the formal semantics and pragmatics of imperatives (Lewis 1979;Hausser 1980;Portner 2007, von Fintel andIatridou 2017), the licensing conditions of FCIs in imperatives have so far not been examined in conjunction with these models.
In fact, the scrutiny of FCIs in imperatives was mostly subsumed under the analysis of FCIs in possibility and necessity modal contexts due to the functional similarity of modals and imperatives, under the assumption that FCIs are licensed in permissive imperatives but not in commands (although Giannakidou 2001 does allow FCIs in commands under certain conditions).
In the dependent indefinite analysis (Giannakidou 2001), FC phrases are represented as intensional indefinites, which are licensed only in contexts providing alternatives (worlds or situations), which in turn explains why they are licensed in non-veridical and non-episodic contexts (e.g., modals, generics), and not licensed otherwise (e.g., in episodic declaratives). More formally, FC phrases are represented as: (34) any student = student(x)(w) (or: student(x)(s)) The world/situation and individual variable(s) are to be bound by an appropriate Q-operator (i.e., generic, habitual, modal, intensional) in order for the FC phrase to be licensed. Under this analysis, the universality of FCIs is derived from their intensionality and exhaustive variation: the FCI variable is to be assigned a distinct value in each world or situation under consideration (Dayal's 1997 i-alternatives).
A permissive modality sentence is analyzed as follows in the dependent indefinite framework (Giannakidou 2001 You may borrow any book. w 0 ,g,K = 1 iff ∃w ′ ∈ K, where K is the extended permissive modal base, such that You borrow a book. w ′ ,g = 1.
b. You borrow a book.
c. Values in i-alternatives: Intuitively, the meaning of (35) can be paraphrased as follows (Giannakidou 2001, 711): 'Consider the books that any book can be be assigned as its value in each relevant i-alternative; you are free to borrow one of those books'. Giannakidou (2001) analyses FCIs in permissive imperatives in a parallel fashion, stating that "the quantificational force of a permissive imperative can […] be understood as equivalent to that of permissive modals":  (2012) ascribes a necessity modality to all imperatives). On closer inspection, however, two shortcomings become apparent. At first sight, it may seem straightforward to simply assume that weak (permission/acquiescence) imperatives have the same quantificational force as permission statements and based on this, to use the same semantic model for weak imperatives and permission statements. However, this step immediately begs the question of how to model the semantics of strong imperatives: their quantificational force is certainly different from permission statements and weak imperatives, so this would force us to assume radically different semantic models for strong vs. weak imperatives, an unwelcome situation. 18 Moreover, the account above remains agnostic as to how the illocutionary force of weak imperatives is brought about (Giannakidou 2001, 698). 19 Regarding strong imperatives, Giannakidou (2001, 699-700) argues (contra Strickland 1982 andDayal 1998) that they too allow FCIs under certain carefully designed situations such as: (41) Context: The hotel manager to a candidate cleaning lady who has just asked him which room to clean in order to get the job: Dhen exi simasia, to mono pu thelo na dho ine an kseris na katharizis. Pijene tora, ke katharise opjodhipote dhomatio! ' ? It doesn't really matter, all I want to see is whether you know how to clean. Go now and clean any room (= some room, it doesn't matter which one)!' Giannakidou (2001) does acknowledge the contrast that while FCIs in weak (permission) imperatives are generally acceptable, it is difficult to find cases when an FCI in a strong imperative is acceptable. However, she ascribes this difficulty to the pragmatics of real commands (such as 18 An alternative would be to consider that imperatives are underspecified for modal force and display a systematic ambiguity between possibility and necessity meanings. Such accounts have, in fact, been proposed (cf. Crnic & Trinh 2011;Grosz 2009). However, as pointed out in von Fintel and Iatridou (2017), these models have significant shortcomings of their own (see ibid., 9-11 for a detailed discussion). 19 Following Aloni's (2002; analysis, Kaufmann (2012) argues that imperatives such as 'Pick any flower.' are in fact strong imperatives paraphraseable as 'You must pick a flower but I am indifferent as to which one you pick.' However, as we have seen, similar imperatives in Hungarian can felicitously contain a discourse particle of permission, which is incompatible with a necessity reading. Also, assigning the paraphrase above to FCI-containing imperatives would fail to predict the following contrast: while 'Pick a flower.' is acceptable out of the blue, 'Pick any flower.' is only acceptable if the fact that the addressee is considering picking a flower is common knowledge.

466
Tamás Halm a need for precision) and contends that on the semantic level, FCIs are equally grammatical in weak imperatives and strong imperatives (since both contexts are nonveridical). In other words, FCIs are (semantically) grammatical in both weak imperatives and strong imperatives, however, they are generally (pragmatically) infelicitous in strong imperatives unless an appropriate context renders them felicitous. In the general framework of Giannakidou (2001), FCIs in strong imperatives (and in deontic necessity modal contexts) are indeed predicted to be grammatical as they are nonveridical. This account, however, is problematic on both the descriptive level and on theory-internal grounds.
On the descriptive level, pace Giannakidou (2001), I believe that the case above is clearly not a strong imperative but a weak imperative (permission). The imperative in (41) would be unable to license an FCI if it were uttered out of the blue: its acceptability crucially depends on the carefully designed context where the course of action which the imperative refers to is already being considered by the addressee. It is common knowledge in the contexts that the addressee is already committed to carrying out the course of action; the utterance of the speaker is not a command but merely an expression of the speaker's acquiescence/indifference as to some specific details of this course of action. In (41), the task of cleaning a room has already been communicated to the cleaning lady (quite possibly using a real command-type imperative), and what the imperative in (41) conveys is that the speaker is indifferent as to the exact identity of the room which will be cleaned.
In other words, the context in (41) is, properly understood, not some mitigating factor which renders felicitous an FCI-containing strong imperative, but rather, an indication that the imperative is not a strong imperative in the first place, but a weak imperative. As I will elaborate in more detail in section 6, the common knowledge that the action denoted by the imperative is one already under consideration by the addressee is in fact the differentiating factor between strong and weak imperatives.
Furthermore, note that, as acknowledged by Giannakidou (2001), the FCI in (41) is interpreted existentially. This is a serious theory-internal complication to anyone who would want to analyze (41) as a strong imperative. Since in Giannakidou (2001), FCIs in weak imperatives inherited their (existential) quantificational force from permissive modals, one would expect that FCIs in strong imperatives should likewise inherit their (universal) quantificational force from necessity modals. This, however, is obviously not the case as the FCI in (41) is interpreted existentially.
Finally, it is not even straightforward that the model put forward in Giannakidou (2001) in fact predicts FCIs in strong imperatives to be grammatical. While strong imperatives are indeed non-veridical, one can plausibly argue (see section 6) that they lack a context containing a set relevant alternatives (whereas weak imperatives do not); and this in itself would in fact predict that they are unacceptable with FCIs.
In the universal free choice analysis developed by Menendez-Benito (2010), FCIs in imperatives are not explicitly analyzed. Also working in the propositional alternatives framework, Aloni's (2007) analysis predicts that FCIs are licensed in imperatives in general, with no difference between strong imperatives and weak imperatives, which is clearly not the case.

Reconsidering the basic facts
As we have seen in section 4, the currently available accounts for the behaviour of FCIs in imperatives are unsatisfactory: they fail to explain the contrast between command imperatives and permission imperatives and also fail to derive the modality and/or illocutionary force of imperatives containing FCIs.
In what follows, I will argue for a new approach that addresses these concerns and at the same time, couches the analysis of FCIs in imperatives in the general theory of the semantics and pragmatics of imperatives.
Let us first reconsider the basic facts. As has been pointed out in earlier accounts as well (Strickland 1982), FCIs are acceptable in weak imperatives but unacceptable in strong imperatives (examples below are reproduced from (6) While Giannakidou (2001) and Giannakidou and Quer (2013) contend that FCIs are grammatical (even if pragmatically infelicitous) in command imperatives as well, I believe that this position is both descriptively inaccurate and theoretically problematic, and that there is a fundamental difference between the FCI-licensing capacity of command vs permission imperatives; however, this can only be elucidated within a full semantic and pragmatic account of imperatives.
As we have seen above, strong denotational semantics models of imperatives have serious shortcomings concerning permission imperatives in general and FCIs in imperatives in particular (see section 5) due to the fact that they ascribe an in-built necessity modality to the denotational semantics of imperatives. At first sight, the Portner-von Fintel & Iatridou framework seems to have a better chance of dealing with FCIs due to its very weak, modality-free denotational semantics and a flexible dynamic pragmatics component equipped to deal with different shades and sources of obligations. However, as we have seen above, the treatment of weak imperatives in this framework is problematic. Also, note that in weak denotational semantic approach, an imperative simply denotes a property restricted to the addressee, and then the task of making this property true of herself is added to the To-Do-List of the addressee. As we have seen in sections 3 and 5, FCIs are intensional indefinites which carry a presupposition of exhaustive variation over a set of alternatives. It is unclear where the set of alternatives would come from in the Portner-von Fintel & Iatridou framework. That is, a combination of the weak denotational semantics approach to imperatives and the dependent indefinite analysis of FCIs appears to predict that FCIs in general are unacceptable in FCIs, which is not the case.
At this point, it is useful to consider another set of observations. As we have seen above, Strickland (1982) has noted that FCIs are unacceptable in an out-of-the blue command-type imperative (43), but acceptable in a situation like (44) While (45a) is acceptable without restriction, (45b) would clearly be very infelicitous out of the blue. In fact, a weak imperative like (45b) is only acceptable if the action which the prejacent describes is already on the table: in this particular case, due to the fact that the addressee is visibly tired and in need of rest, and thus the speaker can safely assume that the addressee is already considering stopping. Note that this is very similar to what we have seen above in (27) and (28)  (47) also seems to require a context where the speaker assumes the addressee to be already considering the course of action described in the prejacent.
Finally, recall the purportedly strong imperative discussed by Giannakidou (2001) above in (41), reproduced here: (48) Context: The hotel manager to a candidate cleaning lady who has just asked him which room to clean in order to get the job: Dhen exi simasia, to mono pu thelo na dho ine an kseris na katharizis. Pijene tora, ke katharise opjodhipote dhomatio! ' ? It doesn't really matter, all I want to see is whether you know how to clean. Go now and clean any room (= some room, it doesn't matter which one)!'

Tamás Halm
Again, it is assumed in the context that the action described in the prejacent (that the cleaning lady will be cleaning rooms) is already on the table.
The common thread here is that while a strong imperative is felicitous out of the blue, a weak imperative is only felicitous if it is part of the common ground that the addressee is considering the action which the prejacent describes.
This contrast is difficult to explain in the Portner-von Fintel & Iatridou framework, where imperatives are uniformly supposed to add properties (and thus, obligations to make those properties true) to the To-Do-List of the addressee, and the only difference between strong and weak imperatives lies in the degree of speaker endorsement. Remember, though, that maintaining that weak imperatives create obligations (of sorts) has not been unproblematic in the first place (cf. von Fintel & Iatridou 2017, 29-30).

The List of Actions Under Consideration (LAUC)
In light of the observations above, it seems to me justified (and in fact, natural) to break the assumption that weak imperatives manipulate the TDL of the addressee, just like strong imperatives do. Rather, as (43)-(48) illustrate, weak imperatives affect an altogether separate list, which I will term the List of Actions Under Consideration (LAUC) 20 by the addressee: in order for a weak imperative to be pragmatically felicitous, its prejacent needs to refer to an item which is already on the LAUC of the addressee. I think this modification of the Portner-von Fintel & Iatridou framework is theoretically plausible and empirically justified.
In terms of the effect of the weak imperative on the LAUC, my proposal is that it is essentially similar to that ascribed to permission statements in general (cf. Kamp 1973). That is, a weak imperative lifts a prohibition concerning an action which is on the LAUC of the addressee. More precisely, the weak imperative lifts a prohibition ascribed to the speaker by 20 More precisely, the LAUC is not a list of actions but rather a list of properties which the addressee is considering making true of herself. I believe this slight terminological inaccuracy will not cause any misunderstanding. Note that the To-Do-List itself would be more accurately called To-Make-True-of-Me-List, as pointed out by Portner (2007, 352). In our case too, I think that going with the more precise but also more convoluted option of List-of-Properties-to-Make-True-of-Me-Under-Consideration would be unwise.

Acta Linguistica Academica 66, 2019
Unauthenticated | Downloaded 09/14/23 04:09 AM UTC The semantics of weak imperatives revisited 471 the addressee concerning an action which is on the LAUC of the addressee. Consider: (49) Nyugodtan vedd fel a kék ruhát nyugodtan take:IMP:2SG PRT the blue dress:ACC 'Just take the blue dress.' Note that in weak imperatives (and in permissions in general) i) the speaker presupposes that the addressee is considering the action denoted by the prejacent and also ii) the speaker assumes that the addressee may believe that the course of action denoted by the prejacent is prohibited (discouraged etc.) by the speaker. That is to say, in (49), the speaker presupposes that the addressee is considering taking the blue dress, and the speaker also assumes that the addressee may believe that there is speaker-imposed prohibition on the addressee's taking the blue dress. It is this prohibition that the speaker wishes to lift by uttering the weak imperative: 'as far as I am concerned, you are free to take the blue dress, you are free to delete any prohibition that you may have ascribed to me against your taking the blue dress.' In this framework, weak imperatives have nothing to do with the TDL of the addressee. What they do is to signal that a preexisting prohibition attributed to the speaker (if any) on an action considered by the addressee (i.e., an element of the addressee's LAUC) is to be lifted. After this, concluding the consideration process, the addressee may or may not decide to move this action to her TDL.
Comparing the TDL and the LAUC, what is common is that both of them are addressee-oriented lists of actions (more precisely: properties restricted to the addressee). The big difference is that TDL contains actions which the addressee is already committed to carrying out; whereas the LAUC contains actions which the addressee is only considering carrying out. Formally, I propose that in addition to the common ground, the question-under-discussion stack and the T function of Portner (2007) (which is a function from individuals to sets of properties, assigning to each individual her To-Do-List, for a formal definition see footnote 13), we should assume that the context also includes a function UC ('under consideration'). UC is a function from individuals to properties, assigning to each individual the set of those properties which the individual is considering making true of herself. In the case of the situation where (49) above is uttered:  Furthermore, in the vein of Kamp's (1973) seminal analysis of permission statements, it is straightforward to assume that there is a function P which assigns to each discourse participant the set of actions (or properties) which are prohibited by that discourse participant. That is, if the addressee's taking the blue dress is prohibited by the speaker (based on sufficient authority), this can be represented as: (51) λw. addresse takes the blue dress in w ∈ P (speaker) Now, we are in position to precisely formulate the felicitousness and pragmatic function of a weak imperative: a.
A weak imperative ϕ weak imp can be uttered felicitously if: ϕ weak imp ∈ UC(addressee) b. The utterance of a weak imperative ϕ weak imp has the pragmatic effect of removing ϕ weak imp from the set of properties prohibited by the speaker: I believe that this account for the semantics and pragmatics of weak imperatives is both intuitively appealing and also provides a much better fit to the relevant data than earlier proposals.

The licensing of FCIs in weak imperatives
In terms of the licensing of FCIs in weak imperatives, I claim that the licensing conditions proposed in Giannakidou (2001) are valid, with one 21 To be precise, what is at stake is not whether the action is actually prohibited by the speaker, but whether the addressee believes the action to be prohibited by the speaker. To capture this, we have to define a modified, two-argument version of P above: P (a, b), which denotes the set of actions which, according to b, are prohibited by a. Then, (52b) can be reformulated as: (i) P ′ (speaker,addressee) = P (speaker,addressee)\{ ϕ weak imp } It is of course perfectly possible that ϕ weak imp / ∈ P ′ (speaker,addressee). This is the case when, contrary to the speaker's assumption, the addressee did not assume that the action was prohibited by the speaker. In this case, P ′ = P . Note that a similar situation can arise with strong imperatives: it is perfectly possible that the action which the speaker wishes to add to the addressee's TDL is already there. This is the case when, unbeknownst to the speaker, the addressee was already committed to carrying out the action concerned. In that case, TDL ′ = TDL. The presence of most azonnal 'right now' is a very clear signal that the imperative is to be interpreted as a strong one. This means that the LAUC is not relevant: in terms of pragmatics, the imperative adds a propertyto-be-made-true to the TDL of the addressee. This is fine for (56a): the property λx.take(x, the blue dress) is added to the TDL. However, as far as (56b) is concerned, in the absence of the LAUC, there are no worldproperty pairs which could serve as the i-alternatives, which makes the sentence pragmatically infelicitous.
Importantly, (iii) is not met in imperatives in general: there is no non-veridical operator which would bind the FCI's possible world (and individual) variable. However, this is only a problem if one takes imperatives to denote propositions. Then, the fact the variables of the FCI are not bound would indeed be fatal: a proposition with unbound variables cannot be assigned a truth value and thus the sentence is uninterpretable and ungrammatical. Crucially though, in Portner's proposal, which I follow in this paper (Portner 2004, 238;Portner 2007, 358), imperatives denote not propositions but properties: (57) Sit down! = λwλx : x = addressee.x sits down in w This property returns the value 'true' in the case of those ⟨x, w⟩ pairs where x is the addressee and x sits down in w. That is, the semantic value of an imperative is not a truth value, but rather, a function (a function from possible worlds to functions from individuals to truth values), which returns the value 'true' in the case of those ⟨x, w⟩ pairs where x is the addressee and x sits down in w, and the value 'false' otherwise. So it is not a problem that the world variable on a FCI is not bound in an imperative: imperatives in general are meant to contain an unbound w variable. (Of course, if one follows a framework such as Kaufmann's (2012) where imperatives are analyzed as denoting modal propositions, then the fact that the FCIs are not bound by an operator would be a problem.) To summarize, my main claim is that the semantics of weak imperatives in general and the licensing of FCIs in imperatives in particular can be modelled successfully using Portner's framework for imperatives and Giannakidou's framework for FCIs. On both counts, though, one significant modification is needed. Concerning the pragmatics and semantics of imperatives, a new object, the LAUC needs to be added to our ontology.
Concerning FCI licensing, the innovation is that the licensing conditions are different in case of declaratives (which denote propositions) and imperatives (which denote properties): while (i) the presence of alternatives and (ii) exhaustive variation are needed with declaratives and imperatives alike, (iii) binding by a non-veridical operator is only needed with declaratives. I believe that both of these innovations are significant, but in a sense, they are also natural extensions of the original theories.
It might be useful at this point to contrast my proposal with a recent, alternative proposal (Staraki 2018). Staraki (2018) proposes an account for free-choice in imperatives which combines the strong semantics approach to imperatives (Wilson & Sperber 1988;Kaufmann 2012) with the dependent indefinite approach to FCIs (Giannakidou 2001). Imperatives are taken to denote propositions, and the FCI is assumed to inherit its quantificational force (existential or universal) from the imperative operator binding it. (i.e., the imperative operator is assumed to have a quantificational force, either universal or existential, depending on the context). I believe this proposal has a number of shortcomings. To begin with, by assuming the strong semantics approach to imperatives, the proposal inherits all the problems associated with the strong semantics approach in general (see section 4). In terms of empirical predictions, it falsely predicts that FCI are available in strong and weak imperatives alike. Also, the assumption that the imperative operator has a quantificational force which is variable depending on the context appears unjustified, especially since all the cases discussed by Staraki (2018) where an FCI is interpreted universally can be analyzed as instances of implicit subtrigging, in which case the FCIs is standardly assumed to be bound by the universal quantifier provided by the conditional operator (Quer 1998;2000 andGiannakidou 2001). 22 In light of these problems, I believe that the account proposed in my paper, which is based on the thin semantics approach to imperatives and on the 22 In the subtrigging analysis, an imperative such as Confiscate any stolen guns! is taken to have an implicit relative clause: Confiscate any stolen guns that you find on the premises! This is, in turn, analyzed as construction with an underlying conditional structure: If you find any guns on the premises, confiscate all of them. Along with Quer (1998;2000) and Giannakidou (2001), I assume that in these cases, the FCI is bound by the conditional operator. The fact that in such subtrigged cases, FCIs are acceptable in strong imperatives as well is not surprising: note that the licensing conditions of the FCI here are satisfied not by the strong imperative itself (which, in my proposal, would be unable to satisfy them), but by the conditional. This is just another instance of the well-known phenomenon where subtrigging licenses FCIs in environments which are otherwise hostile to FCIs (cf. Giannakidou 2001, 718-722).

476
Tamás Halm dependent indefinite theory of FCIs, is more appealing theoretically and has better empirical coverage.

Predictions
In this section, I will discuss how the proposal put forward above in sections 6.2 and 6.3 enables us to account for a wide array of empirical facts. Firstly, it becomes straightforward to explain the old observation of why strong imperatives are fine out of the blue and weak imperatives are not. Strong imperatives add (typically new) items to the To-Do-List of the addressee: there is no requirement at all that this item be in any way presupposed or part of the common ground. On the other hand, weak imperatives, rather than adding (possibly new) items to the To-Do-List, lift a prohibition with regard to a certain course of action that is already being considered by the addressee: this course of action (or more precisely, the fact of it being considered by the addressee) naturally has to be part of the common ground. In the absence of this, out of the blue, a weak imperative is infelicitous, since it is not presupposed that its prejacent is being considered by the addressee. 23 Secondly, as we have seen in section 6.3, the fact that FCIs are acceptable in weak imperatives but not acceptable in strong imperatives is also readily explained: with weak imperatives, the LAUC provides a set of i-alternatives and room for exhaustive variation, both of which are missing in the case of a strong imperative. 23 Beáta Gyuris (pc) pointed out to me a possible counterexample where a weak imperative can be uttered even though the prejacent is not under consideration by the addressee. The context is that 'A' is discussing some difficulties she may be facing in the near future with a friend of hers, 'B'. In this case, it is natural for 'B' to say (i) at the end of their discussion of the matter, even if there was no explicit indication that 'A' was considering asking for help: (i) Nyugodtan hívj fel, ha segítségre van szükséged. nyugodtan call:IMP:2SG PRT if help.onto is need:POSS:2SG 'Just call me if you need any help.' Upon consideration, I believe that the felicitousness of (i) does not constitute a counterexample to my claim. That is, it remains the case that (i) is only felicitous as long as 'A' has reason to assume that 'B' is considering the action described in the prejacent. I believe that due to basic social conventions and norms, 'A' and 'B' being friends and just having discussed some difficulties 'B' is facing in the near future; 'A' has strong reason to assume that 'B' is considering asking for help even in the absence of any explicit signals from 'B' to that effect.
Thirdly, introducing the LAUC solves a conundrum which has been dogging weak semantics -strong pragmatics approaches to imperatives for some time (cf. Portner 2007;von Fintel & Iatridou 2017). Namely, by assuming that weak imperatives add properties-to-be-made-true to the addressee's To-Do-List, these models in effect say that weak imperatives create obligations in the same vein as strong imperatives do. This is clearly an unwelcome outcome of the model, and Portner (2007), and von Fintel and Iatridou (2017) do try to square the circle by assuming that the TDL has sections: items originating from strong "order" imperatives are filed under the command section and are true obligations; whereas items originating from weak "acquiescence" imperatives are filed under a section of 'commitments taken freely by the addressse', and are mere commitments (which are very similar to obligations but somehow lesser). Also, it is assumed that in case of strong imperatives, the property-to-be-made-true is automatically added to the TDL, while in case of weak imperatives, the addressee is free to decide whether to add it to the TDL or not. My main criticism of this proposal is that it blurs the difference between weak imperatives (permission/acquiescence imperatives) and certain kinds of strong imperatives such as advices or invitations. Consider the following: (58), (59) and (60) are all strong imperatives: in each case, the effect of the imperative is that the speaker wants the addressee to add the property denoted by the adjacent to her TDL. Naturally, speaker endorsement varies: it is very strong in (58), which can be characterized as an order, meaning that the addressee has little choice but to add the property to her TDL (if the speaker indeed has sufficient authority to utter such an order). (59) is an invitation, meaning weaker endorsement and more latitude for the addressee in terms of whether to add the property to her TDL. The same goes for the advice imperative in (60). Crucially, though, none of these three sentences can be characterized as either permissions, or expressions of acquiescence or indifference. These are strong imperatives, all affecting the TDL. 24,25 24 Recall that in addition to different levels of speaker endorsement, (58), (59) and (60) also differ in what Portner (2007) terms the nature of obligation. Thus, (58) is based on the speaker's authority and expresses deontic necessity, (59) is based on the addressee's desires and expresses bouletic necessity, whereas (60) is based on the addressee's goals and expresses teleological necessity. Clarifying the exact relationship between speaker endorsement and nature of obligation is not trivial. Note on the one hand that it is possible to calibrate speaker endorsement while remaining within a certain kind of imperative (defined by nature of obligation). Consider: (i) Have a piece of fruit.
(ii) Do have a piece of fruit.
(i) and (ii) are both invitation imperatives, but (ii) has stronger speaker endorsement than (i). This would suggest that speaker endorsement and nature of obligation are orthogonal. At the same time, it does feel natural to assume that order imperatives typically carry stronger speaker endorsement than invitation imperatives. This in turn would suggest that there is some correlation between speaker endorsement and nature of obligation. In any case, disentangling the roles of speaker endorsement and nature of obligation in the segmentation of the TDL is beyond the scope of this paper. What matters for our purposes is that speaker endorsement plays no role in separating strong imperatives from weak imperatives. 25 Lars-Olof Delsing (pc) pointed out to me that in Swedish, the non-availability of an overt you subject in imperatives does not correspond to the strong-weak distinction. That is, overt you is available in commands and unavailable in invitations and advices, and in permissions: The semantics of weak imperatives revisited 479 (61), on the other hand, is a weak imperative, affecting the LAUC, but not the TDL. That (61) is a weak imperative can be read off from the presence of the discourse particle nyugodtan 'permissive particle', or from the context (if available). Note that speaker endorsement can be defined for weak imperatives as well, in fact, the existing terminological variation (permissive imperatives, acquiescence imperatives, indifference imperatives) reflects exactly this:  (63) and (64) are all weak imperatives: the effect of the imperative is that the speaker signals to the addressee that any prohibitions ascribed to the speaker concerning the action described by the prejacent can be lifted. Due to strong speaker endorsement, (62) amounts to a real permission: if the speaker indeed has the authority to lift the prohibition, the addressee has little choice but to consider it lifted. (Note that whether the addressee then proceeds to add this property to her TDL is a different question and has nothing to do with the weak imperative and the strength of the endorsement behind it: the effect of the weak imperative only concerns the lifting of the prohibition, and thus, the LAUC; the TDL is out of scope, so to speak.) There is weaker endorsement behind (63), which we can characterise as an acquiescence imperative: stopping short of actually permitting the course of action which the speaker supposes is on the addressee's LAUC, the speaker merely expresses her acquiescence. This means that the ador absence of an overt you subject is not a neat a function of strong vs. weak imperative status. Potsdam (1996) observed that overt you in English imperatives in fact signals strong speaker authority as the source of the imperative (which roughly corresponds to being a command imperative). This means that, rather interestingly, the situation in English appears to be the mirror image of the situation in Swedish: overt you can only be used in the command-subvariety of strong imperatives. The further exploration of this topic is beyond the scope of this paper.

Tamás Halm
Finally, our proposal to differentiate between strong and weak imperatives in terms of the addressee-oriented lists which they affect (TDL vs. LAUC) is supported by the fact the there are languages where the strongweak imperative distinction is encoded by obligatory imperative particles. Poletto and Zanuttini's (2003) study of the Ladin dialect of Val Badia (called Badiot) is a case in point. In Badiot, imperatives obligatorily 29 carry one of four imperative particles: ma, mo, pö, and pa. Out of these, pö and pa are discourse markers found in declaratives as well, whereas ma and mo are unique to imperatives. Poletto and Zanuttini (2003) characterize these two as follows: ma is said to appear in imperatives expressing 'advice and permission' and mo in imperatives expressing an 'order '. 30 vs. the deletion of a prohibition on an action on the LAUC). While Giannakidou and Yoon (2016) tentatively claim that: "the choice of exhaustive vs. non-exhaustive variation affects the interpretation of the imperative -a command in the case of FCI, but a weaker suggestion/invitation in the case of [an RVI]", this is clearly a misunderstanding caused by a confusion of these two very different notions of "strength". Note that in earlier work, Giannakidou (2001) convincingly showed that FCIs are typically easily licensed in weak (permission) imperatives, and claimed that they are very marked (due to pragmatic infelicitousness) in strong (e.g., command) imperatives. Note also that we have marshaled a significant amount of empirical evidence showing that FCIs are not acceptable in strong imperatives cross-linguistically. 29 In the presence of a sentential negative marker, the particles may be omitted 30 Of the other two particles, pö is analyzed as discourse marker ("presuppositional particle") which can appear in declaratives as well, and its function is to signal "that the content of the proposition denoted by the sentence in which it occurs contradicts some proposition which is already in the discourse". As Poletto and Zanuttini (2003) point out, pö is in this respect rather similar to German doch: (i) Al é pö bun s.cl is PÖ good 'Sure it is good!' (contrary to what was said before) In imperatives, Poletto and Zanuttini (2003) claim that pö is restricted to addresseeoriented imperatives, that is, in our terms, weak imperatives. Accordingly, it is grammatical in the first conjunct of imperative and declarative constructions: (ii) Mànge'l pö che spo crësceste eat:it PÖ that then grow:SG2 'Eat it and you will grow.' That is, in imperatives, pö patterns with ma, encoding weak imperative status, and in addition to this, it also signals that the proposition denoted by the imperative is in contradiction with a presupposed proposition.
As far as the particle pa is concerned, Poletto and Zanuttini (2003) propose that in declaratives, its function is to signify verum (or falsum) focus:

Tamás Halm
This is an indication that mo signifies strong imperatives, which are known to be excluded from the first conjunct of an imperative and declarative construction. To conclude, there are strong reasons to believe that what the ma-mo distinction in Badiot encodes is exactly the strong-weak imperative distinction that we are concerned with in this paper. 31 The fact that the strong vs. weak imperative distinction is obligatorily encoded as a clear binary distinction on the overt morphosyntactic level in some languages lends significant support to the model proposed in this paper. Note that this model is binary in the sense that it predicts an imperative to be either strong or weak, with no shades in between, since an imperative either affects the TDL or the LAUC. In contrast to this, the model proposed by von Fintel and Iatridou (2017) is graded. Since the strong-weak imperative distinction is conceived of as a function of speaker endorsement, which is a quasi-continuous (but at the very least non-binary) parameter ranging freely from the very weak to the very strong, the predic-31 Note that Poletto and Zanuttini (2003) described ma-imperatives as expressing 'advice and permission'. Above, I have argued that the particle ma encodes weak imperative status. However, we have seen earlier that certain kinds of strong imperatives were characterized as advice (or suggestion) imperatives. Is this not a contradiction: are advice imperatives to be analyzed as strong imperatives or weak imperatives? I believe this is a misunderstanding caused by imprecise terminology. The communicative function of giving a piece of advice can be exercised by using a strong imperative or a weak imperative alike. That is, it is possible to give an advice out of the blue, using a strong imperative, adding a new task to the TDL:  Poletto & Zanuttini (2003, 4) where a ma-imperative is used to convey a piece of advice are instances of the second mechanism, where a piece of advice is communicated by lifting a presumed prohibition on a course of action already contemplated by the addressee.

Acta Linguistica Academica 66, 2019
Unauthenticated | Downloaded 09/14/23 04:09 AM UTC The semantics of weak imperatives revisited 485 tion is that the weak-strong imperative distinction should also be graded. That is, in addition to the prototypically strong and weak imperatives, we should have the full spectrum: a finely grained scale of strength such as: ⟨prototypically weak < … < weak < … < medium-strength < … strong < … < prototypically strong⟩. However, the evidence from Rhaetoromance, where imperative strength is overtly encoded as a clean binary distinction supports a binary model over a graded model.

Conclusion
To summarize the results of this paper, I have shown based on observations concerning FCIs in imperatives and other independent evidence that the dynamic pragmatic component of the Portner-von Fintel & Iatridou framework for the interpretation of imperatives needs to be significantly revised in order to fully account for the relevant facts. Leaving the denotational semantic component intact, I have argued (pace von Fintel & Iatridou 2017) that in imperatives containing FCIs, and in weak imperatives in general, the pragmatic force of the utterance is not directed at the To-Do-List of the addressee, but rather, at a separate component of the common ground which I termed the List of Actions Under Consideration by the addressee.