1. Introduction
This article investigates the role interpretation of an action as intentional or accidental plays in the grammar.Footnote 1 While the effects of interpreting an action as intentional or accidental have been discussed in the literature with respect to a number of phenomena, to the best of my knowledge only a handful of studies have tried to make a connection between different phenomena sensitive to an action’s intentionality.Footnote 2
My goal is to contribute to the project of constructing a unified analysis of the intentionality effects observed in the literature. To that end, I take a close look at two concrete phenomena that have been observed to exhibit intentionality effects but have not been previously unified. The intentionality effect of interest is a situation where a particular grammatical restriction is lifted or weakened when an action is interpreted as accidental, as opposed to intentional.Footnote 3
The first phenomenon we look at pertains to so-called subject obviation, whereby the subject of the subjunctive clause cannot co-refer with the subject of the matrix clause, see (1a). However, this restriction is relaxed when the action described in the subjunctive clause is interpreted as accidental, as is the case with predicates like se tromper de ‘to be mistaken about’ in (1b):
(Ruwet Reference Ruwet1984, cited in Szabolcsi Reference Szabolcsi2010: 2)
Our second phenomenon concerns the restriction on the use of the perfective aspect in Slavic languages. We will look at two sub-types of this restriction. First, we will see that negative imperatives in Slavic cannot be used with perfective verbs, see (2a). However, this aspectual restriction is lifted when the action expressed by the verb is interpreted as accidental, see (2b) (e.g., Bogusławski Reference Bogusławski1985).
Second, we will see that the same aspectual restriction exists in the complement of a negated desire predicate, as illustrated in (3a). This restriction is also lifted when an action is interpreted as accidental, see (3b).
What unifies desire statements (as with the obviation in (1) and the second sub-type of the aspectual restriction in (3)) and imperatives (the first sub-type of the aspectual restriction in (2)) is that they represent one class of modality, which we will refer to as decisive modality. To account for the intentionality effect in constructions with decisive modality, we will build on Kaufmann’s (Reference Kaufmann2019) semantic-pragmatic analysis. This analysis is formalized using multi-agent modal logic and derives obviation as inconsistent requirements on the context in which sentences in question are uttered. I propose a modification of Kaufmann’s analysis of directive obviation, introducing a new pragmatic condition on constructions with decisive modality.
It must be stressed at the outset that the goal of this article is to develop a unified account for the intentionality effect in sentences with want and negative imperatives, rather than advocate for a particular approach to subjunctives, imperatives, or aspect. Therefore, I will be using a specific set of theoretical assumptions without devoting much space to defending them against existing alternatives. Any theory of subjunctives, imperatives, aspect, or any other phenomena touched by intentionality effects (see more on this in section 6.2) should provide an account for these effects. Moreover, only those theories that do not appeal to language- and construction-specific constraints to account for intentionality effects can be used as a basis for a unified analysis. The semantic-pragmatic account adopted here meets this requirement.
Looked at differently, the need to unify intentionality effects across different grammatical phenomena restricts possible accounts of any particular phenomenon. As we will see in section 5, many current proposals are phenomenon-specific, which does not allow them to be transferred to other grammatical domains where the intentionality effect is also attested. To avoid future limitations along these lines, the article concludes with a general discussion of intentionality in grammar and presents some other phenomena showing sensitivity to the interpretation of an action as intentional versus accidental.
The flow of the article is as follows: In section 2, we look at subject obviation and aspectual restriction in more detail, focusing on those contexts that relax the restrictions in question. Section 3 discusses directive obviation and the analysis thereof put forward in Kaufmann Reference Kaufmann(2019). Section 4 presents the modification of this analysis and demonstrates how it can explain the weakening of subject obviation and aspectual restriction in accidental contexts. In section 5, we see the most prominent accounts of the weakening of subject obviation and aspectual restriction and their limitations. Section 6 contains some general remarks about the notion of intentional and accidental actions and reviews other grammatical phenomena that are sensitive to the interpretation of an action as intentional versus accidental. Section 7 concludes.
2. Two phenomena in detail
In this section, we will discuss two phenomena exhibiting the intentionality effect. The first set of observations concerns co-reference and what is known in the literature as subject obviation, subjunctive obviation, or disjoint reference effect. The crucial observation for our purposes is that subject obviation is relaxed or weakened when the action expressed by the verb in the embedded clause is interpreted as accidental. The second set of observations pertains to aspect choice in Slavic languages. Here, we will look at two constructions: root imperatives and desire sentences with infinitival complements. Both constructions are subject to the aspectual restriction disallowing the perfective aspect in negative sentences. As is the case with subject obviation, the aspectual restriction is lifted when the action is interpreted as accidental.
2.1 Intentionality effect and subject obviation
In many European languages, including Hungarian, Romance languages, and Slavic languages, the subject of the subjunctive clause cannot co-refer with the subject of the matrix clause. We will refer to this restriction as subject obviation. Subject obviation is illustrated for French in (4) and (5), where subjunctive constructions are contrasted with infinitives. In (4), repeated from (1), subject obviation is illustrated for 1sg. Example (5) shows subject obviation with 3sg where il ‘he’ in the subjunctive clause cannot co-refer with Pierre in the matrix clause. Subject obviation is well-documented and has been studied extensively, especially in Romance languages (e.g., Bouchard Reference Bouchard1983; Ruwet Reference Ruwet1984, Reference Ruwet1991; Picallo Reference Picallo1985; Kempchinsky Reference Kempchinsky1986, Reference Kempchinsky2009; Farkas Reference Farkas, Ivan A. Sag and Szabolcsi1992; Schlenker Reference Schlenker2005, Reference Schlenker, Claudia, Klaus and Paul2011; Costantini Reference Costantini2006; Szabolcsi Reference Szabolcsi2010, Reference Szabolcsi2021; Stegovec Reference Stegovec2019).
A number of accounts have been proposed to explain subject obviation, both syntactic (see e.g. Kempchinsky Reference Kempchinsky1986) and semantic-pragmatic (see e.g., Schlenker Reference Schlenker2005, Reference Schlenker, Claudia, Klaus and Paul2011; Costantini Reference Costantini2006; Farkas Reference Farkas, Ivan A. Sag and Szabolcsi1992), the most notable of which will be discussed in section 5.
A noticeable property of subject obviation is that it is weakened when the action expressed by the embedded clause does not (fully) depend on the will of the agent.Footnote 4 This lack of control can be conveyed by using a passive construction, as in (6a), or a modal as in (6b). It can also be expressed by mental state predicates like se tromper de ‘to be mistaken about’, see (6c), or experiencer object verbs like amuser ‘to amuse’, see (6d) (see e.g., Ruwet Reference Ruwet1984, Reference Ruwet1991; Zaring Reference Zaring1985; Farkas Reference Farkas, Ivan A. Sag and Szabolcsi1992).
(Ruwet Reference Ruwet1984, cited in Szabolcsi Reference Szabolcsi2010: 2)
Subject obviation is not restricted to French or Romance languages. It is also found in Slavic languages, see (7a) for Russian and (7b) for Polish. It is worth noting here that subject obviation is found in both positive and negative sentences and when the embedded predicate has either the perfective or imperfective morphology. This point will become important in the next section where we discuss the aspectual restriction in Slavic.
As in French and other Romance languages, subject obviation in Slavic languages is weakened when the action is non-intentional, see (8) for Russian and (9) for Polish. Again, notice that in these examples, perfective is acceptable.Footnote 5
(Szabolcsi Reference Szabolcsi2010: 9)
Subject obviation and its weakening in non-intentional contexts are also found in Hungarian, as illustrated in (10).
(Szabolcsi Reference Szabolcsi2010: 3–4)
Szabolcsi Reference Szabolcsi(2021) points out that the Hungarian data is important because they demonstrate that subject obviation is not limited to subjunctive clauses. In particular, Szabolcsi Reference Szabolcsi(2021) discusses two verbs, remélni ‘hope’ and sajnálni ‘regret’, which can only take indicative complements in Hungarian. Similar to what is found in subjunctives, these verbs trigger the obviation effect in intentional contexts, see (11a,b) and (12a,b), but not in non-intentional contexts, see (11c,d) and (12c,d).Footnote 6
(Szabolcsi Reference Szabolcsi2021: ex.22)
(Szabolcsi Reference Szabolcsi2021: ex.25)
The Hungarian data show us that there is no necessary connection between obviation (and its weakening) and the subjunctive mood. This is important for constructing a unified analysis of subject obviation and aspectual restriction, since the latter is found in indicative clauses.Footnote 7
2.2 Intentionality effect and aspectual restriction
2.2.1 Imperatives
Across Slavic languages, positive imperatives can take verbs in either imperfective or perfective with minimal interpretative differences, as shown in (13).Footnote 8 Negative imperatives, on the other hand, are well-formed only with the imperfective, see (14). This aspectual restriction on imperatives has been amply discussed in the literature (e.g., Forsyth Reference Forsyth1970, Bogusławski Reference Bogusławski1985, Zaliznjak Reference Zaliznjak2006, Paducheva Reference Paducheva2013, Despić Reference Despić, Kosta and Radeva-Bork2020).Footnote 9,Footnote 10
(Despić Reference Despić2016: ex.5)
(Despić Reference Despić2016: ex.5)
The property that interests us here is that the aspectual restriction is lifted when the action expressed by the verb is interpreted as accidental. Accidentality can be part of the lexical meaning of the verb, as with unaccusatives, such as fall, and mental state verbs, such as forget, lose, and mix up, see (15). These verbs are low on the agentivity scale (in the sense of Dowty Reference Dowty1991) and thus normally do not express intentional actions.
(Despić Reference Despić2016: ex.8)
Accidental interpretation can also be triggered by the context, as shown in (16). Notice that in these examples, a regular agentive transitive verb is used and adverbials marking accidentality are optional.Footnote 11
The pattern we see here is similar to the intentionality effect with subject obviation. In both cases, we have a particular grammatical restriction (co-reference or aspect choice) which is relaxed when the agent lacks full control over the action or the action is unintentional. In both cases, weakening is observed with a range of predicates including unaccusatives, mental state verbs, and agentive transitive verbs. The difference between subject obviation and aspectual restriction is that the latter is present only in negative sentences.
2.2.2 Desire-infinitives
In the previous section we saw the aspectual restriction with imperatives. We will now see that the aspectual restriction and its weakening is also found in desire sentences with infinitives. As illustrated in (17) for Russian, infinitival desire statements display the aspectual restriction that we have already observed with imperatives: in positive sentences, both perfective and imperfective are acceptable, whereas negative sentences only allow verbs in the imperfective.
As is the case with imperatives, the aspectual restriction in negative infinitival desire statements can be lifted when the action is interpreted as accidental. For example, pfv becomes available with accidental actions, such as ‘run late’ or ‘mix up the keys’, see (18a,b), as well as with unaccusatives, see (18c).
To the best of my knowledge, the intentionality effect in infinitival desire statements in Slavic is not well studied, apart from some mention in Zaliznjak Reference Zaliznjak(2006). More research in this area is needed to explore (among other things) cross-Slavic variation of the phenomenon.Footnote 12
The important observation for us here is that in Slavic languages which show the aspectual restriction and its weakening in non-intentional contexts in imperatives, the same restriction and the same weakening occur in infinitival desire statements. This observation is important because it shows that the aspectual restriction and obviation do not depend on each other.
To illustrate, let us first recall from the previous section that subjunctive clauses in Slavic exhibit subject obviation which is weakened in non-intentional contexts, see (19) repeated from above. Moreover, the obviation effect is found with both the perfective and the imperfective, that is to say, no aspectual restriction is observed.
2.2.3 Combining the facts about aspectual restriction
We have seen that the aspectual restriction is found in imperatives, which are comparable to the disjoint-referent situation. This is because in imperatives, the individual who gives the order is normally different from the individual who performs the action. We also saw that the aspectual restriction is found in desire-infinitives, in which PRO is obligatorily co-referential with the subject of the matrix clause (e.g., Morgan Reference Morgan1970, Chierchia Reference Chierchia and van Emde Boas1989). The latter constructions are important because they show us that aspectual restriction and subject obviation are independent of each other, demonstrating that the intentionality effect is a grammatical phenomenon in its own right and has to be studied as such.
We can summarize our observations for negatives in Russian in table 1, where the ban on pfv does not lead to the impossibility of the matrix subject/director and the embedded subject/instigator referring to the same individual or vice versa. From this we can conclude that the aspectual restriction and the obviation effect are independent restrictions. The fact that both are weakened in the same contexts indicates that there is an underlying mechanism to which both restrictions are sensitive, which constitutes the topic of this article.
Aspectual restriction and obviation in negative utterances in Russian
To summarize, in this section we saw two sets of data that demonstrate the intentionality effect. The first set concerns the weakening of subject obviation in non-intentional contexts. The second set pertains to the lifting of the aspectual restriction in non-intentional contexts. The restrictions that the intentionality effect acts on (co-reference and aspect choice) are shown to be independent of each other. It is important to note here that the two phenomena discussed in this section represent a particular type of modality, which we will refer to as decisive modality (following Kaufmann and Kaufmann Reference Kaufmann and Kaufmann2012, among others). This type of modality covers imperatives, desire statements, and deontic modals.Footnote 13 The data discussed in section 2 show clear parallelism with regard to the intentionality effect, calling for a unified analysis. The goal of the following sections is to develop such an analysis.
3. Directive obviation at the semantic-pragmatic interface
In this section, we look at Kaufmann’s (Reference Kaufmann2019) semantic-pragmatic analysis of an infrequently discussed phenomenon she calls directive obviation. We will use this discussion as a foundation for our explanation of the intentionality effect in the next section.
3.1 Directive obviation
Directive obviation in its simplest configuration can be illustrated by the inability of first-person exclusive forms to be subjects of regular root imperatives or subjunctives used for directives. As shown in (20), Greek na-subjunctives can be used as directive speech acts. In this case, we can say that the speaker plays the role of a director while the addressee is an instigator of the action (using the terminology from Kaufmann Reference Kaufmann2019).
(Oikonomou Reference Oikonomou2016: 73)
The picture with first-person forms is different. For example, (21) cannot be used as a directive for the speaker to send an email to Anna. This is because in this case, both the director and instigator are the same individual – the speaker.Footnote 14 The ban on having the director and instigator of the directive speech act co-refer is what is known as directive obviation.
(Oikonomou Reference Oikonomou2016: 168)
This restriction, however, is not morphosyntactic. As shown below, when the instigator is understood as not being in full control of their action, the obviation disappears (see e.g., Oikonomou Reference Oikonomou2016; Kaufmann Reference Kaufmann2019). The example in (22) using Greek na-subjunctives is acceptable in a context where the speaker does not have an alarm clock and indirectly asks her mother, who usually wakes up before 6 a.m., to wake her up.
(Oikonomou Reference Oikonomou2016: 168)
Directive obviation has been attested in other languages. For example, in Slovenian, directives can be expressed using regular imperative morphology with second- and first-person inclusive or by naj-subjunctives with third person. Crucially, first-person exclusive forms are unavailable (see Stegovec and Kaufmann Reference Stegovec, Kaufmann, Eva Csipak and Zeijlstra2015; Stegovec Reference Stegovec2017, Reference Stegovec2019), among others). This paradigm is illustrated in (23). As with the Greek example in (22), the obviation in Slovenian disappears once the instigator’s action is interpreted as non-intentional (see Kaufmann Reference Kaufmann2019).
(Kaufmann Reference Kaufmann2019: 642–3)
Directive obviation has also been studied in more complex constructions, such as interrogative imperatives and embedded imperatives (see Kaufmann Reference Kaufmann2019 for discussion and references). For the purpose of this article, it suffices to point out that the phenomenon is widespread and that there are convincing arguments that it is grammatical in nature. For instance, the following examples show that imperatives can be embedded in Slovenian, unless the subject of the matrix and the imperative are co-referential. Compare (24a) with (24b):
(Sheppard and Golden 2002, cited in Kaufmann Reference Kaufmann2019: 439)
The ungrammaticality of (24b), however, cannot be attributed to the oddness of reminding oneself of what one has to do, for such a reminder (and a corresponding report) is perfectly expressible using deontic modals instead of imperative morphology, as in (24c). This demonstrates that the phenomenon is grammatical in nature.
3.2 Semantic-pragmatic account of directive obviation
Kaufmann Reference Kaufmann(2019) builds her account of directive obviation from earlier work (Schwager Reference Schwager2006; Kaufmann Reference Kaufmann2012; Kaufmann and Kaufmann Reference Kaufmann and Kaufmann2012, et seq.) where directive speech acts are carried out with modalized sentences (i.e., Open the door! ≈ You must/should open the door) that come with a set of pragmatic presuppositions restricting the context in which directives can be felicitously used. These restrictions derive the non-assertive character of imperatives and non-canonical directives. In particular, there are two general conditions on the use of imperatives: (a) a director d of the imperative presents himself as uncertain about the course of events (Epistemic Uncertainty Condition, EUC) and (b) d presents himself as an authority on what follows from the modal statement (a combination of Epistemic Authority Condition [EAC] and Decisive Modality [DM]).
The account is formalized using multi-agent modal logic with two unary operators 
$\Box$ and 
$\Diamond$ indexed for epistemic and prioritizing relations with respect to different individuals.Footnote 15 Interpretation proceeds with respect to a frame F = < W, D, B, R>, where W is a set of possible worlds, D is a set of individuals (including S(peaker) and A(ddressee)), B maps any x in D to x’s belief relation B
$_x \subseteq W \times W$ (serial, transitive, shift-reflexive), R 
$\subseteq W \times W$ is the salient prioritizing modal flavour.Footnote 16 For convenience, two additional modal operators based on B are defined:Footnote 17
(Stalnaker Reference Stalnaker2002)
Imperatives and sentences with decisive modality are translated as follows:
More definitions are needed before we can spell out the felicity conditions for imperatives:
The felicity conditions for imperatives are defined as follows:
According to this system,“[i]mperative marking triggers DM, EAC, and EUC as presuppositions, which means that the speaker takes them to be entailed by the context set by the time the content of this utterance is used to update the context set” (Kaufmann Reference Kaufmann2019: 654). As we will see in section 4, the inclusion of the Control Condition (i.e., CONTROL(a,q)) in DM is problematic when accounting for imperatives with accidental actions; I will propose a modification to overcome this issue and extend Kaufmann’s analysis to aspectual data.
Let us see now how directive obviation is derived in the system. According to Kaufmann Reference Kaufmann(2019), directive obviation (e.g., universal lack of designated first-person exclusive imperative forms) is derived as a case of inherent contradiction between the felicity conditions of imperatives in cases in which the director d is identical to the instigator a. To see how the contradiction arises, we first derive the principle of Director’s Anticipation as follows:
Let us now consider what happens when the root imperative ‘p!’ with 1sg exclusive morphology is uttered. In unmarked cases, when the speaker S utters ‘p!’, S is the director d of the directive speech act. The instigator a of the directive speech act is the grammatical subject. In our case (1sg exclusive morphology), the instigator is also S. The identity between the director and instigator results in contradiction as shown below:
As desired the contradiction arises because the director (the speaker) is identical to the instigator. The first conjunct in the contradictory statement at step 6, namely 
$\Box^{PB_d}p$, is derived from Director’s Anticipation, whose antecedent is introduced by the speaker’s public commitment that 
$\Box^Rp$. If the director and the instigator do not co-refer (as is the case with canonical second-person imperatives), the speaker’s public commitment that 
$\Box^Rp$ does not entail 
$\Box^{PB_d}\Box^{B_a}\Box^Rp$ and the contradiction does not arise.
Kaufmann Reference Kaufmann(2019) proposes that cases of embedded directive obviation are explained using the same mechanism because the semantics of embedding predicates (say, tell, order...) require that the matrix subject be the director and that the felicity conditions must be satisfied at the level of the reported speech act. It is also suggested that the same mechanism is at work in subject obviation, given the similarity between imperatives and desire statements, although no details are provided.
4. Proposal for intentionality effect
In this section, we discuss why Kaufmann’s system, in its current state, is problematic for capturing imperatives with accidental actions. I put forward a modification of the system and show that with this modification, the system can be used to explicate subject obviation, aspectual restriction, and their weakening in accidental contexts.
4.1 Intentionality Effect Condition (IEC)
Kaufmann (Reference Kaufmann2012, Reference Kaufmann2016) makes a distinction between imperatives used for commands, see (37a), and imperatives used for wishes, see (37b,c).Footnote 18 The latter can be used “only if the context cannot be construed as practical for the addressee: either because the addressee is not actually present, or because the prejacent is already settled” (Kaufmann Reference Kaufmann2016: 341). The example in (37b) illustrates a wish with an absent addressee. The example in (37c) shows a situation where the prejacent is settled, that is to say – “the prejacent describes a state of affairs that is fully determined by events before the utterance time” (Kaufmann Reference Kaufmann2016: 328).
(Culicover and Jackendoff Reference Culicover and Jackendoff1997, as cited in Kaufmann Reference Kaufmann2019: 650)
As pointed out to me by a reviewer, the problem arises with imperatives with accidental actions, such as (38). According to Kaufmann Reference Kaufmann(2016), the example in (38) cannot be used for a wish because it can be uttered in a practical context (with a present addressee and non-settled prejacent). However, it cannot be used for a command either because the instigator does not control the action which violates DM.
One solution would be to expand the definition of imperatives used for wishes by including accidental actions. But, as pointed out to me by a reviewer, this kind of solution would go against linguistic evidence. The distinction between commands and wishes as defined in Kaufmann Reference Kaufmann(2016) is morphologically encoded in some languages. For example, in Greek, imperatives can be used for directive speech acts but not for wishes. Examples like (38), however, can be expressed using imperative morphology (e.g., Oikonomou Reference Oikonomou2016).
The solution I propose here draws a line between commands, practical wishes, or warnings, on the one hand, and non-practical wishes on the other. Non-practical wishes are examples like (37b,c), which are referred to simply as ‘wishes’ by Kaufmann (Reference Kaufmann2012, Reference Kaufmann2016). Practical wishes or warnings are examples like (38), that is to say, imperatives with accidental actions used for wishes or warnings in practical contexts. To illustrate the difference between practical and non-practical wishes, let us consider the example in (39), suggested to me by an anonymous reviewer. This example can be used in two different contexts: (i) as a silent wish while watching somebody from afar (in which case it is a non-practical wish) and (ii) as a warning or practical wish for a particular addressee to behave in a more cautious way to avoid falling down.Footnote 19
To implement this solution, I propose a modification of Kaufmann’s system. In particular, I propose to extract the Control Condition from the definition of the decision problem and regard it as a default interpretation. That is to say, if there are no indications that the action is accidental, the action is taken to be intentional. I call this condition an Intentionality Effect Condition (IEC).
In Kaufmann’s system, we have the decision problem defined as in (40), repeated from (30):
This definition leads to the following requirement on the context in which imperative morphology can be felicitously used:
I propose to redefine the decision problem as in (42) and add the Intentionality Effect Condition (IEC) in (43), which states that the Control Condition, reformulated as the speaker’s public commitment (i.e., 
$\Box^{PB_S}(TRY(a,p) \rightarrow p)$), obtains only if there are no indications that the speaker publicly considers it possible that the agent of the action is not in (full) control (i.e., there are no indications that 
$\Diamond^{PB_S}\Diamond^{B_S}$(TRY(a,p) 
$\wedge\ \neg$ p).Footnote 20 Note also that IEC in (43) is formulated as a condition on use for sentences with decisive modality.
The addition of the conditional in IEC allows us to have a configuration in which the imperative is uttered but the Control Condition is not met. This will happen in cases where there are indications that the action described by p is accidental. That is to say, IEC is satisfied in the two cases shown in (44), assuming that we are interested in situations where the imperative (or more generally, any sentence with decisive modality) is uttered and accidentality excludes (full) control.
Accidentality in IEC is formulated as a condition on a belief state. We will discuss the philosophical origin of this notion of accidentality in section 6.1. An advantage of IEC (compared to other attempts to address the intentionality effect in grammar discussed in section 5) is that it does not strive to define intentional or controllable actions. Rather, it takes intentional actions to be any action not marked as accidental (where accidental means non-controlled and unintentional actions). Intuitively, if the action is intentionally initiated and is fully controlled by the agent, there is a belief that if the agent of the action tries to bring it about that p, then the state of affairs described by p obtains. On the other hand, the action is accidental if it is possible that the agent tries to bring it about that p, but 
$\neg p$ obtains. This is shown in (45).
Consider, for example, the verb win in John won the lottery. In this case, win describes a prototypical accidental action. According to (45b), this amounts to saying that it is epistemically possible that John tries to win the lottery (goes to a convenience store, buys a ticket, etc.), but the desired state of affairs does not materialize. By contrast, for John lifted his finger, it is normally believed that if John tries to lift his finger, he will succeed.
There is another aspect of IEC which requires clarification, namely ‘indications of accidentality’. What are these indications? Do they vary from language to language? Are they all lexical or they can be contextual? We already saw answers to some of these questions when we discussed obviation and the aspectual restriction in section 2; this will be further discussed in section 6.2.
Let me now demonstrate that the modification introduced above successfully derives directive obviation. It was mentioned above that IEC is satisfied in two (relevant) cases. Case #1 is realized when there are no indications that the speaker publicly considers it possible that the action is accidental (i.e., there are no indications that 
$\Diamond^{PB_S}\Diamond^{B_S}(TRY(a,p)\wedge \neg p)$) and the Control Condition obtains, see (46). As shown in (47), in this case, Director’s Anticipation is derived, and in cases where the director and the instigator refer to the same individual, this leads to directive obviation, as desired. This result is identical to what we saw in section 3.2.
Unlike in Kaufmann’s system, our modified system has the second case where IEC is satisfied. Case #2 obtains when there are indications that 
$\Diamond^{PB_S}\Diamond^{B_S}(TRY(a,p)\wedge \neg p)$ and the Control Condition does not hold, see (48). This second case ensures that imperative morphology can be used for practical wishes and warnings. Additionally, in this case, no Director’s Anticipation is derived. This predicts that in non-intentional contexts, co-reference becomes possible. As we will see shortly, this property of IEC will allow us to account for weakening of subject obviation and aspectual restriction.
As we see, the modification of Kaufmann’s system in terms of IEC successfully derives directive obviation in intentional contexts. In addition, it avoids the problem faced by the account in Kaufmann (Reference Kaufmann2012, Reference Kaufmann2016, Reference Kaufmann2019) allowing imperatives to be felicitously used in non-intentional contexts for practical wishes or warnings. The modification also provides room for modelling the intentionality effect, to which we now turn our attention.
4.2 Explaining subject obviation and its weakening
Much research has pointed out the connection between directive obviation and subject obviation (Kempchinsky Reference Kempchinsky2009; Oikonomou Reference Oikonomou2016; Kaufmann Reference Kaufmann2019; Stegovec Reference Stegovec2019, among others). Assuming that desires can be modelled using decisive modality, the connection is straightforward. Recall from section 2.1 that in many European languages, the subject of a subjunctive clause and the subject of the matrix clause cannot refer to the same individual, unless the action is interpreted as accidental, see (49), partially repeated from (5b) and (6d).
Subject obviation and its weakening are accounted for similarly to directive obviation. That is to say, the identity between the attitude holder and the agent of the action in the subjunctive clause results in inconsistent requirements. In this case, the inconsistency lies in the speaker’s belief set rather than the context set.
Let us see how this works. Heim Reference Heim(1992) shows that desire statements carry an uncertainty presupposition; in other words, the attitude holder considers both p and 
$\neg p$ epistemically possible, see (50). This presupposition of want is similar to EUC for imperatives.
As we saw above, in intentional cases where the director and instigator co-refer, EUC comes in conflict with Director’s Anticipation, which explains directive obviation. An equivalent of Director’s Anticipation in case of want sentences is Attitude Holder’s Anticipation (AHA), see (51). AHA is based on the principle of rationality, a standard principle in multi-agent epistemic logic which assumes all agents to be rational. According to AHA, if the attitude holder ah believes that the instigator a believes that p is the optimal solution to a decision problem (i.e., R-necessary), ah believes that p will be realized (i.e., a will act rationally).
AHA is derived the same manner as Director’s Anticipation. The only modification we need to make is to generalize the conditions on decisive modality (including IEC) from the conversation participants to all rational agents. With this generalization, it is important to be careful with the clause in (31c), repeated in (52c). In imperatives, ‘learning’ that q is optimal (i.e., R-necessary) is the event of the imperative being uttered (i.e., the speaker’s public commitment, see step 2 of the proofs in (35) and (47)). In case of attitude predicates, ‘learning’ consists of the attitude holder becoming aware that q is R-necessary, which is obtained only if ah and the instigator are the same individuals. The derivation of AHA in intentional contexts is shown in (53).
Subject obviation is the result of inconsistent beliefs of the speaker, who, on the one hand, by uttering 
$want_{ah}\ p$ commits herself to believing that 
$\Box^{B_{ah}}\Box^R p$ and by AHA, that 
$\Box^{B_{ah}}p$. On the other hand, the speaker holds to the uncertainty presupposition that 
$\neg\Box^{B_{ah}}p$. This inconsistency can be avoided in two ways. First, AHA is not derived because the attitude holder is different from the instigator (step 2 in (53) is invalid). These are non-obviative cases. Second, AHA is not derived because IEC with accidental actions disallows the step from 4 to 5 in (53), as we saw above for imperatives. These are weakening cases.Footnote 21
Before moving to the aspectual restriction, I would like to mention a number of points that I consider to be important but do not have the space to develop here. First, the account of subject obviation presented above does not directly depend on the presence of the subjunctive in the embedded clause, but rather on factors such as the presence of the uncertainty condition and the possibility of deriving AHA. This is supported by Hungarian data (Szabolcsi Reference Szabolcsi2021). My second point concerns the fact that subject obviation is less sensitive than imperatives to contextual cues that the action in the prejacent is interpreted as accidental. This fact can be linked to the idea that subject obviation is not dependent on the contextual information that participants to conversation share. Rather, the intentional/accidental distinction comes from linguistic cues (passives, unaccusatives, etc.) and encyclopedic knowledge about some actions (win, lose, etc.). A detailed exploration of these points is beyond the scope of this article.
4.3 Explaining aspectual restriction and its weakening
To explain the intentionality effect with aspectual restriction in Slavic, I will look more closely at the meaning contribution of the perfective and imperfective aspects in these languages. Following recent work by Grønn (Reference Grønn2003), Zinova and Filip (Reference Zinova and Filip2014), among others, I assume that the perfective aspect in Slavic asserts that the action has achieved its endpoint and has an implicature that the action has begun, see (54). The imperfective aspect asserts that the action has started and generates no relevant implicatures, see (55).
To formalize these intuitions, let us assume the semantics for Slavic aspect as in (56) (e.g., Altshuler Reference Altshuler2012, Reference Altshuler2016). These are simplified denotations that ignore interaction with time and discourse as well as the diversity of perfective prefixation in Slavic, but they suffice for the purpose of this article as they emphasize the main relevant difference between pfv and ipfv. Both pfv and ipfv are functions from a set of events provided by VP to a set of states of these events. pfv returns a set of final states, whereas ipfv returns a set of beginning states.Footnote 22
Given these denotations, the truth-conditions of simple negative sentences are as shown in (57) and (58). As beginning and finishing the event stand in an asymmetrical entailment relation, pfv, but not ipfv, has enriched truth-conditions, as shown in (58c).Footnote 23
We will see shortly that the enriched meaning of sentences with pfv is responsible for the aspectual restriction in Slavic imperatives and desire statements when the Control Condition is present. In order to see this, let us abbreviate the sentences above as follows:
Using the abbreviations above, imperatives will look as follows (where S = the speaker and the director, u is the addressee):Footnote 24
The derivation of the aspectual restriction and its weakening is similar to the derivation of directive obviation in that they arise from contradictory pragmatic requirements. However, the difference lies in that contradictory requirements are introduced by EUC and the aspectual inference of pfv, rather than EUC and Director’s Anticipation. To see how this works, let us restate IEC in terms of aspect, see (62), which is a sub-case of our IEC in (43).Footnote 25
Let us first consider negative imperatives with pfv, as in (61a). Recall that in intentional contexts, these give rise to the aspectual restriction, as in (14), which is lifted in accidental contexts, as in (16). According to our system, the aspectual inference of pfv, combined with IEC in intentional contexts, derives that the speaker is publicly committed to believing that FIN(a,p). On the other hand, EUC requires that the speaker consider 
$\neg$FIN(a,p) epistemically possible. Since these requirements are inconsistent, see (63), we obtain the aspectual restriction.Footnote 26
In accidental contexts, IEC returns 
$\Diamond^{PB_S}(BEG(a,p) \wedge \neg FIN(a,p))$ which is compatible with the aspectual inference of pfv as well as EUC. Thus, we obtain weakening of the aspectual restriction in accidental contexts.
The situation with negative imperatives with ipfv, as in (61d), is simpler. We know that they do not give rise to the aspectual restriction. According to our system, this is because ipfv does not have an implicature and thus does not interact with IEC in either intentional or accidental contexts.Footnote 27
It is easy to see that an identical explanation applies to infinitival desire under the assumptions spelled out for subject obviation in section 4.2.
To summarize, our account of subject obviation, aspectual restriction, and their weakening in accidental contexts revolves around the possibility of deriving Director’s Anticipation (or a similar principle for the attitude holder in desire statements) or its equivalent in the aspectual restriction via the obligatory inference of pfv. When these principles are derived, we obtain inherently contradictory requirements on the context set or speaker’s beliefs. These principles are not derived in two cases: (i) when the director (or attitude holder) and the instigator are not co-referential and (ii) when the action is not fully controlled by the instigator. In these cases, we find the weakening effect.
5. Previous linguistic accounts
There have been a number of previous accounts of the weakening effect in subject obviation and aspectual restriction. Although Farkas Reference Farkas(1988) did not directly aim at explaining subject obviation, she was one of the first to propose a way to capture intentional/accidental distinction in the grammar by using a RESP(onsibility) relation. This relation (or a related operator) was later used in Farkas Reference Farkas, Ivan A. Sag and Szabolcsi(1992) to account for subject obviation and its weakening and in Szabolcsi Reference Szabolcsi(2010) to explain subject obviation in combination with the distribution of Positive Polarity Items (PPIs) (see section 6.2). RESP was considered and rejected by Schlenker (Reference Schlenker2005, Reference Schlenker, Claudia, Klaus and Paul2011) when analyzing subject obviation in French, and used by Grano Reference Grano(2017) to provide the semantics for intend as an attitude predicate.
Farkas’s (Reference Farkas1988) goal was to explain controller choice in sentences like (64), that is to say the fact that in (64a) the implicit argument PRO in the embedded clause refers to John, whereas in (64b), PRO refers to Mary. For this purpose, she postulated a new semantic relation RESP defined as in (65):
(Farkas Reference Farkas1988: 36)
Farkas Reference Farkas(1988) argues that RESP is not reducible to the traditional thematic role of agent. One argument for this is that the initiator i of RESP does not have to be a participant in the situation s, which i brings about. For instance, in (66), John is the initiator of the situation described by the embedded clause, but not a participant in it. An agent cannot be a non-participant.
According to Farkas, convince and promise are RESP-inducing in that they have RESP among their satisfaction conditions. The difference between (64a) and (64b) is that convince requires the first argument of RESP to be associated with the direct object in the matrix clause, whereas promise requires the first argument of RESP to be associated with the matrix subject. In the unmarked situation (canonical control), the controller choice falls on the matrix argument that bears RESP to s when the controlled argument is also in RESP with respect to s. The important contribution of Farkas’s work is that it demonstrated the need to distinguish intentionality from agentivity, a closely related but distinct grammatical concept.
To account for weakening of subject obviation, Farkas Reference Farkas, Ivan A. Sag and Szabolcsi(1992) applies RESP by suggesting that in order for the obviation effect to emerge, both the subject of the matrix clause and the subject of the embedded clause have to stand in the RESP relation to the situation described by the complement. If either RESP is deficient, the obviation effect is weakened, as we saw in section 2.1.
The RESP-account has been criticized regarding its direct application to subject obviation. For example, Schlenker Reference Schlenker(2005) provides the examples in (67) to question whether RESP is in fact the right kind of relation, pointing out that they do not make it obvious in what sense Jean is responsible for being handicapped. Nonetheless, the obviation effect holds, compare (67a) with (67b). Schlenker (Reference Schlenker2005, Reference Schlenker, Claudia, Klaus and Paul2011) proposes an alternative account for subject obviation and its weakening in terms of event-De-Se construal. This account, however, cannot be straightforwardly extended to the aspectual restriction and its weakening (nor to other phenomena that exhibit the intentionality effect, see section 6.2), so we set it aside and refer the interested reader to the cited papers for details.Footnote 28
(Schlenker Reference Schlenker2005: 294)
More recent accounts notice the parallelism between subject obviation and its weakening and other similar phenomena. For example, Zu Reference Zu(2018), building on Szabolcsi Reference Szabolcsi(2010), proposes a unified syntactic account for subject obviation, distribution of PPIs, and verbal inflection in Newari. Specifically, Zu Reference Zu(2018) postulates a Sentience projection (SenP) at the left periphery of attitude complements, whose specifier hosts a perspectival expression (building on ideas in Speas and Tenny Reference Speas, Tenny and A. DiSciullo2003). This Sen-head comes with a different featural mark-up in what Zu calls canonical control (RESP), non-canonical control (non-RESP), and non-control, see (68). In canonical control, Sen comes with an unvalued ϕ-feature, which triggers domain suspension as in Bobaljik and Wurmbrand Reference Bobaljik, Wurmbrand, Ora Matushansky and Marantz(2013), making the searchable domain of canonical control as large as the matrix CP, see (68a). In non-canonical control, Sen comes only with an unvalued feature [uD], valued by PRO via Spec-head agreement, and the domain is closed off at SenP, see (68b). In non-control cases, Sen has no uninterpretable features and thus can host any perspectival expression, see (68c).
For subject obviation, Zu Reference Zu(2018) proposes that subjunctive sentences are ambiguous between non-canonical control, as in (68b), and non-control, as in (68c). The former is associated with no-RESP interpretation and allows the matrix and embedded subjects to co-refer (weakening of obviation). The latter represents core cases of subject obviation (i.e., no co-reference between the matrix and embedded subjects).
The proposal in Zu Reference Zu(2018) is tailored to account for the co-reference phenomena. It is hard to see how it can be extended to the aspectual restriction and its weakening which, as we saw, are independent from obviation.
To the best of my knowledge there are no fully developed formal semantic/pragmatic accounts of aspectual restriction weakening in Slavic languages, although the literature mentioned in section 2.2 discusses this effect in various descriptive terms.Footnote 29
One fully fledged syntactic account of the aspectual restriction in Slavic and exceptions to it is Despić Reference Despić, Kosta and Radeva-Bork(2020). His account uses the following four assumptions: First, there is an Agree-relation between the imperative (imp) and the inflection on the verb (in Asp). Second, the imperfective aspect is located above vP, whereas the perfective aspect is below vP (e.g., Svenonius Reference Svenonius and Peter Svenonius2004). Third, imp cannot scope below negation (Han Reference Han1999; Zeijlstra Reference Zeijlstra2006, among others). Fourth, the Phase Impenetrability Condition (PIC) is as defined in Chomsky Reference Chomsky and M. Kenstowicz(2001), that is to say, the (complement of the) lower phase becomes unavailable for syntactic operations as soon as the higher phase head is merged. Provided these assumptions, the derivation of positive imperatives looks as in (69), where imp can Agree with both Asp1 and Asp2.
In negative imperatives, on the other hand, only the imperfective can Agree with imp, see (70a). This is because NegP is merged and imp must be located in C (next highest functional projection) in order to out-scope negation. However, C is a phase head, and as soon as C is merged, the complement of vP (including the perfective aspect Asp2) is unavailable for Agree, see (70b).
According to this system, the exceptions with accidental actions are explained as follows: in accidental configurations, vP is a weak phase (or a non-phase), thus PIC does not preclude Agree between imp and Asp2-PFV, see (71):
One immediate problem with the system above is that it can only account for exceptions to the aspectual restriction with unaccusative verbs and predicates like forget. It is plausible to claim that with these verbs, vP is a weak phase (or a non-phase). It is much less plausible to claim the same for agentive transitive verbs like open and tell when they are used in non-intentional contexts. Even more problematic is that the syntactic account cannot explain the identical aspectual restriction with infinitival desire statements (and exceptions to it). This is because the syntactic analysis crucially depends on the requirement that imp has to out-scope negation and thus be located in C in negative imperatives. No such requirement is necessary for want. Furthermore, the syntactic account for the aspectual restriction in Slavic developed in Despić Reference Despić, Kosta and Radeva-Bork(2020) cannot extend to subject obviation or any other phenomena exhibiting the intentionality effect.Footnote 30
To summarize, we have seen that the dependence of grammatical phenomena on the interpretation of the action has not gone unnoticed. Both semantic and syntactic accounts have been put forward to explain this curious effect. Moreover, recent accounts recognize the need to construct a unified analysis. The present article can be viewed as part of this joint effort by combining two phenomena that have not previously been treated together.
6. General notes on intentional and accidental actions
This section contains some general notes regarding the notion of intentionality and its presence in the grammar. I start with some cursory philosophical remarks about intentions (section 6.1) and then present a list of additional linguistic phenomena that are sensitive to the interpretation of an action (section 6.2). The goal of this section is to attract interest to the topic and provide some preliminary resources to interested scholars.
6.1 Philosophical remarks
Here I discuss the development of concepts related to intentions and intentional actions in philosophy, with the goal of providing some preliminary background to the idea of capturing intentions in terms of beliefs used in this article.
Beginning in Ancient Greece, Aristotle defined an action as voluntary if its outcome is desirable, the action is “in a man’s own power”, and it is done “with knowledge, i.e. not in ignorance either of the person acted on or of the instrument used or of the end that will be attained” (Aristotle, Ethica Nicomachea, V, 10, in Aristotle 1941). Since Aristotle, philosophical reflections on the nature of intentions and intentional actions have revolved around two key components: desire and belief (Anscombe Reference Anscombe1957; Davidson Reference Davidson1963, Reference Davidson1980, Reference Davidson2001; Bratman Reference Bratman1987, Reference Bratman1999; Velleman Reference Velleman1989; Raz Reference Raz2011, among others). If the agent of the action does not desire to bring about a particular outcome and/or the outcome is not foreseen, the action has been considered to be unintentional/involuntary.
In the recent literature, a third component – control, “man’s own power” – is added (e.g., Mele and Moser Reference Mele and Moser1994). When an action is beyond one’s abilities or is performed under coercion, such an action is not voluntary or intentional (in an everyday sense), although the outcome may be highly desirable and foreseen.
However, the presence of control (in addition to desire/motive and belief/foreseeing) does not necessarily guarantee that the success of one’s action is immune to luck (Pritchard Reference Pritchard2005, Reference Pritchard2016; Horst Reference Horst2015). Borrowing an example from Pritchard Reference Pritchard(2016), imagine a skillful archer who confidently hits the target, but, unbeknownst to her, most targets on the field are fitted with an invisible force field that repels arrows, and she just happened to choose the one that is not. Intuitively, the archer’s success is still accidental, although all three ingredients – desire, belief, and control (or skillfulness) – are present. It is this notion of epistemic luck that has been exploited in this article.
6.2 Other linguistic phenomena with intentionality effect
Though this article has focused on co-reference and aspect choice, these are not the only grammatical domains sensitive to the interpretation of an action as intentional versus accidental, and similar phenomena have been observed in other languages than those we have looked at. By putting these observations side by side, I want to underscore the universal nature of the intentionality effect.
The first two phenomena I mention here concern the polarity system. Szabolcsi (Reference Szabolcsi2004, Reference Szabolcsi2010) observes that PPI anti-licensing in the infinitival complement of not want is sensitive to the interpretation of the action in the complement clause. When the action is intentional, a PPI cannot be interpreted under negation, whereas with accidental actions, a PPI can be interpreted under negation. It has also been shown that strong Negative Polarity Items (NPIs) are also sensitive to the interpretation of an action as intentional versus accidental, but that they show a mirror pattern: strong NPIs under not want are fully acceptable with intentional actions, but are degraded with accidental actions (Goncharov, Reference Goncharov2020a). A related observation is made about Free Choice Items (FCIs) (polarity sensitive expressions similar to English whatever). Their acceptability is also affected by the interpretation of an action as intentional versus accidental (Choi and Romero Reference Choi and Romero2008; Alonso-Ovalle and Menendez-Benito Reference Alonso-Ovalle and Menendez-Benito2017 among others).
Another phenomenon already mentioned above concerns verbal inflection in Newari (Sino-Tibetan). In Newari, the so-called disjunct inflection with co-referential subjects is normally ungrammatical, but it becomes possible when the action is accidental or non-controlled (e.g., Hale Reference Hale and Ronald L Trail1980; Wechsler Reference Wechsler2018; Zu Reference Zu2018).
Additionally, interpretation of an action as intentional versus accidental affects case marking in Hindi/Urdu and Central Pomo (Tuite et al. Reference Tuite, Agha and Graczyk1985, Mithun Reference Mithun1991) and can be signalled by a specialized ‘out-of-control’ circumfix in Salish (Demirdache Reference Demirdache, Mendikoetxea and Uribe-Etxebarria1997, Davis et al. Reference Davis, Matthewson, Rullmann, Lotte Hogeweg, de Hoop and Malchukov2009).
These phenomena demonstrate the range and diversity of grammatical domains affected by the interpretation of an action as intentional versus accidental. Little research has been carried out on these phenomena, and then only focused on individual cases or a subset of cases. No overarching theory of the linguistic underpinnings of the intentionality effect has yet been formulated.
7. Conclusion
Natural languages have developed multiple ways to mark the presence or absence of intentions, ranging from the availability of a co-referential interpretation in subject obviation configurations in many European languages and aspect choice in Slavic (explored in this article) to a dedicated ‘out-of-control’ morpheme in Salish and different case marking in Hindi/Urdu and Central Pomo. These observations call for an overarching theory of the intentionality effect in the grammar. This article represents one step towards such a theory. It focuses on two concrete phenomena – weakening of subject obviation and the aspectual restriction – and proposes a unifying semantic-pragmatic account for these phenomena.
Acknowledgements
I would like to express my gratitude to my anonymous reviewers, whose thorough comments and challenging questions helped to bring the article to its current state. Some of their suggestions were incorporated into the analysis, which is acknowledged in the text, but oftentimes I wished I could make a more proper attribution of their ideas. I am also grateful to Magda Kaufmann, Anna Szabolcsi, and Hedde Zeijlstra for discussing this topic with me at different stages. This research was partly supported by ZE 1040/13-1 DFG research grant.
Open access
