Obviation in Hungarian: what is its scope, and is it due to competition?

According to the classical description of obviation, the subject of a subjunctive is disjoint in reference from the attitude-holder subject of the immediately higher clause. Inspired by Ruwet (1984/1991) and Farkas (1988; 1992), I present data from Hungarian where obviation in certain subjunctives is plainly lifted, and data where obviation occurs in indicatives. I argue that obviation is not the result of competition with another construction, and point to promising potential accounts in terms of a clash in semantics or pragmatics. My aim is to contribute desiderata for a theory of obviation and exemptions from obviation in fairly informal terms.


3
Szabolcsi Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1421 arguments, despite the fact that present-day English barely uses a formal subjunctive. 1 For ease of reference, this section retains the example numbering in Ruwet (1991).
Ruwet offers a rich catalog of data, both for vouloir 'want' and for other verbs that select the subjunctive in French: 'fear', 'regret', 'wish', 'demand', 'deserve', 'wait', 'admire', 'notice', etc. Here is my own summary of what I take to be Ruwet's core intuition.
The two coreferential occurrences of the subject in sentences with subjunctive complements "iconically" convey a discontinuity between the will and the actions of a person. b.
If it is mind-boggling how such a discontinuity could exist, disjoint reference arises. c.
Obviation is weakened (the sentence becomes acceptable) when, for some reason or other, that discontinuity makes sense.
b. *Je veux que je parte. I want for me to leave.
Ruwet lists the following kinds of syntactic or lexical modification which, applied to the complement, create a distancing effect that, he claims, makes the coreferential reading of the subjunctive more or less acceptable in French: the use of passive, pouvoir, perfective, coordination [which I think is really a case of contrast], stative predicates, psychological predicates, negation (of vouloir), polar reinforcement, the pourquois veux-tu que construction, and viewing one's action from the perspective of another. The assessment of acceptability of the English counterparts comes from Ruwet (1991).
[39] a. ?Je veux que je sois enterré dans mon village natal. I want for me to be buried in my native village.
[41] a. ?Je veux que je puisse attaquer à l'aube. ?I want for me to be able to attack at dawn.
[46] b. Ah! Je voudrais que je sois déjà parti! Oh! I would like for me to be already gone!
[49] Je veux que tu partes et que je reste. I want for you to go and for me to stay.
[56] a. ?Dites-lui bien que je veux que je reçoive son message dans les plus brefs délais. ??Do be sure to tell him that I want for me to receive his message without delay.
[68] a. Je veux que je sois très amusant ce soir. I want for me to be quite amusing tonight.
[80] b. ?Je ne veux pas que je me trompe de clé (encore). ?I do not want for me to mix up the keys (again).
[82] a. Je veux absolument que j'amuse ces enfants. I absolutely want for me to amuse the children. Farkas (1992) proposes a theory that captures Ruwet's generalizations and also makes crosslinguistic predictions. The theory has two critical components. One is the RESP [onsibility] relation that she had introduced in connection with controller choice, and the other is the infinitive-subjunctive rivalry.

Farkas (1988; 1992)
1 I am not aware of any systematic discussion of the for-infinitive--subjunctive connection. R. Kayne (p.c.) suggests that his 2017 argument that for is (a piece of) a wh-phrase may offer a syntactic starting point. The link to subjunctive may even extend to cases like For him to have said that in public was a mistake, insofar as French usually has subjunctive in subject sentences, no matter what the predicate. Szabolcsi Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1421 (2) RESP(i, s) holds between an individual i and a situation s just in case i brings s about, i.e. just in case s is the result of some act performed by i with the intention of bringing s about. If so, s is called the (possibly) intentional situation and i its initiator. The initiator is similar but not identical to an agent. (Farkas 1988: 36) Farkas (1988) proposes that when there are two possible controllers in an obligatory control configuration, the controller must be chosen so that RESP holds between it and the complement situation, accounting for the contrast between promise (subject control) vs. persuade (object control). With RESP-inducing matrix verbs, whose meanings require that one of the participants be the initiator of the complement situation, the unmarked controller is the initiator participant. The present paper is not concerned with controller choice and merely uses RESP, following Farkas (1992). Farkas (1992) observes that binding theoretic accounts of obviation 2 cannot explain the Ruwet-style cases of exemption from obviation, and the fact that Balkan subjunctives and Hungarian object-controlled subjunctives are not obviative. Her account is in terms of the rivalry between infinitives and subjunctives. She proposes that the infinitive is the preferred expression of canonical control. If a language has an infinitive in the given construction, it is to be chosen. When however the language lacks infinitives in general (as Romanian and Serbo-Croatian do) or in the given construction (as Hungarian does with many object-control verbs, e.g. rábeszél 'persuade'), then the subjunctive option is the only one and there is no obviation.
(3) Canonical control: Both the participant linked to the complement subject and the participant linked to the matrix argument that controls it bear the RESP relation to the complement situation.
(4) Obviation: In subjunctive complements that conform to the canonical control case, the infinitive blocks the subjunctive, if it is available in the language. More formally, "In world-dependent complements [subjunctives] that conform to the canonical control case [have RESP], the form used to mark Su [bject] dependency [infinitive, if there is one] blocks the form used for world-dependency." (Farkas 1992: 102-104)

Probing the predictions in Hungarian
We now probe Farkas's predictions using Hungarian. One prediction is that subjunctive obviation correlates with RESP. Section 4 examines and confirms this.
The other prediction is that obviation is the result of blocking (competition from infinitives). This rests on the assumption that obviation is specific for subjunctives or, more generally, for moods that have competitors. But is it? If not, then its explanation should cover the whole natural class. Starting with Section 5 we start making steps in this direction. The jumping-off point is that obviation-like effects exist in indicatives that do not have any competitors. 2 The binding theoretic account is this. Something in subjunctives causes I to move to C. According to Picallo (1984); Salamanca (1981), the reason is that subjunctives have no independent tense; but Suñer & Padilla-Rivera (1984) and Zaring (1985) show that there is no notion of dependent tense which isolates the right class of clauses in French, Spanish, and Brazilian Portuguese. Kempchinsky (1985) assumes that the subjunctive operator forces I-to-C. In either case, I-to-C extends the domain within which the complement pronominal subject cannot be bound to include the immediately higher clause, and so the unacceptability of a sentence like He i wanted that he i visit MoMA is a Principle B effect, or perhaps an Avoid Pronoun effect since an anaphor would be possible (Bouchard 1984). Avrutin & Babyonyshev (1997) argue, based on Russian, that AgrS and not the pronominal subject is subject to the Principle B effect in subjunctives. See Citko (2012) for a more semantically flavored syntactic account. What is the advantage of using Hungarian in this investigation? One advantage is that the acceptability contrast between RESP and non-RESP subjunctive complements of 'want' is very sharp: the non-RESP cases are perfect. This is useful because, with the exception of [49], the French speakers whose judgments I have received, directly or indirectly with the kind help of Vincent Homer, do not report a strong improvement in the examples where Ruwet claimed there is one. Hungarian has a similar advantage over English, with for NP to VP in the place of the formal subjunctive, although here the improvement seems more pronounced than in French. Thus, the use of Hungarian in the subjunctive domain provides cleaner data in Section 4 and gives us more confidence in the claim that RESP triggers obviation but non-RESP does not.
(Why the strength of these contrasts differs across languages is a big and important question that I have to leave open for the time being.) The other advantage of Hungarian pertains to mood selection. Hungarian is indicative-heavy. Unlike Balkan languages, Hungarian has infinitives alongside subjunctives, but both of these have a much narrower distribution than in Western Romance. Thus we find many indicatives, for example under emotive-factives, that do not have any competitors. If they exhibit effects that belong under the same larger umbrella as classical subjunctive obviation, then competition cannot be the critical factor. Starting with Section 6, a semantic-pragmatic explanation will be sought, inspired by Kaufmann (2019).

Hungarian subjunctives under akar 'want': Obviation iff RESP
We start with the simplest cases. When the verb in the subjunctive complement is agentive, and the sentence is understood to describe "normal circumstances," the subject of the subjunctive is an initiator that bears the RESP relation to the situation, and obviation occurs. When the verb is not agentive, there is no RESP and no obviation. 3 The desideratives kíván 'wish' and szeretne 'would like' behave identically to akar.
English speakers inform me that the judgments for the for NP to VP translations fundamentally replicate the Hungarian judgments, although the contrasts are not as sharp as in Hungarian, as was mentioned in Section 3. Bear in mind that the focus of this paper is Hungarian.
(A) Agentive verbs in complement -obviation (under normal circumstances!) The examples in the first two sets, (A) and (B) are fully glossed, but often only full-sentential translations are given in (C) and (D), because the details of agglutinative morphology play no role and might be distracting. The "non-normal circumstances" fall into two classes. In the first class, the sentence describes an urge, a mistake, or an accident, and obviation vanishes, even if the verb is agentive in the usual grammatical sense. 4 (C) Urges, mistakes, accidents, even if the complement verb is agentive -no obviation (10) Magasságiszonyom van. Nem megyek fel a toronyba, nem akarom, hogy leugorjak. 'I have the fear of heights. I'm not going up the tower, I don't want for me to jump' Fear of heights presents an urge to jump; I do not want to give in to that urge.
(11) Nem megyek el az ülésre. A főnök ostobaságokat fog beszélni, nem akarom, hogy ellentmondjak neki. 'I'm not going to the meeting. The boss will speak nonsense, I don't want for me to contradict him' The use of the subjunctive makes it clear that if I were at the meeting, I would be burning to contradict the boss, but I do not want for that to actually happen. 5 Likewise, there is no obviation when an otherwise perfectly agentive event would obtain by accident or mistake. Note that the subjunctive suffices to indicate that this is so, even without adding an adverb like 'accidentally.' (12) Nem akarom, hogy (véletlenül/tévedésből) az egészséges lábat amputáljam. 'I don't want for me to (accidentally/by mistake) amputate the healthy leg' In the second class, a situation involves a dependence on authorities or the co-operation of others. RESP is likewise absent here, which explains why obviation does not occur.
(D) Dependence on authorities or on the co-operation of others -no obviation (13) (to fairy offering to grant wishes) Azt akarom/kívánom, hogy megöljem a sárkányt és feleségül vegyem a királylányt. 'I want/wish for me to kill the dragon and marry the princess' The "actor negotiates with director" context is an inexhaustible source of examples with agentive verbs that happily occur in the subjunctive under desideratives.
(14) (actor to director) Azt szeretném, hogy táncoljak a következő jelenetben. 'I would like for me [=my character] to dance in the next scene' Finally, contrastive focus on the subjunctive subject pronoun eliminates obviation in all languages that I am aware of. Recall that among Ruwet's examples,[49] was the most acceptable for the French speakers I have received judgments from. 4 The class of urges, mistakes, and accidents distinguishes itself by the way it interacts with polarity, independently of exempting subjunctives from obviation. Positive polarity items (PPIs) do not like to be in the immediate scope of a negation in the same tensed clause. Szabolcsi (2004: fn. 10)  (iv) a. The company wants to harvest new ideas, but doesn't want to spend { any money/ a red cent } on this. b. I don't want to give { any money / a red cent } to the beggar.
Goncharov proposes an account in terms of dynamic epistemic logic.

5
I thank one of the reviewers for this example. The two desires are not two plainly contradictory. Z. Gendler Szabó (p.c.) suggests that the contrast can be described as having a pro tanto desire to contradict but not an all things considered desire to contradict. I place this case in the "need the co-operation of others" class, because the individual corresponding to the subject of the subjunctive obviously cannot, by himself or herself, bring about the desired situation. In the positive cases, other individuals must refrain from performing the action; in the case of focus negation, other individuals must perform it instead. It seems natural to say that these are instances of non-RESP. Z. Gendler Szabó (p.c.) warns that construing RESP as "sole responsibility" is philosophically suspect, similarly to reference to "the cause," as Lewis (1987) observes in connection with causation. But for our purposes it suffices to distinguish between the possibility of sole responsibility and the guaranteed absence of sole responsibility. In the last batch of examples above, either the context or the linguistic form guarantees the absence of sole responsibility.
Unfortunately, subsuming the focus case under non-RESP does not explain why it is so much more felicitous in Western Romance than the other, non-agentive non-RESP cases. So, the role of focus may have to be reassessed.
To summarize, all the cases in which the subjunctive is not obviative are characterized by the fact that the individual corresponding to the subject does not stand in a RESP relation to the complement situation, in the sense that this individual either does not intentionally bring about that situation at all, or he/she needs the authority or cooperation of others to bring it about. With this understanding of non-RESP, Farkas's prediction is exactly borne out. 6 It is worth emphasizing that the non-RESP character of the complements in (C)-(D) is not characterizable in overt syntactic terms. Ruwet's own non-obviative examples involve nonagentive predicates or passives and ability modals; Costantini (2005) offers a like-minded account of Italian, according to which auxiliary constructions (complex tenses, passives, modals) evade obviation. Hungarian does not even have some of those syntactic devices, and (C)-(D) definitely do not use any of them. Out of the blue, these complements are expected to be RESPful. The fact that they are in the subjunctive signals that they are not RESP-ful. As they present an expected initiator as a non-initiator, the interpreter must either reject them as contradictory, or contend that the context is not normal and the expected initiator is indeed not one. Adverbs like 'accidentally' or information about the speech situation may justify that; otherwise the co-operative interpreter must volunteer their own justification. The fact that RESP status is so highly amenable to contextual considerations indicates that its role is eminently semanticpragmatic in nature. 6 Farkas (1988: 36-38) makes a careful distinction between the initiator and the (intentional) agent, but does not consider our "non-normal circumstances." The proper characterization of the cross-linguistically relevant demarcation line between RESP and non-RESP calls for further research, including comparison with Aristotle's notion of involuntary action in Book III of Nicomachean Ethics (I thank J. Knobe for pointing this out).

What is the scope of obviation? From subjunctives to indicatives
We now leave the realm of subjunctives and proceed to a more provocative new question. Is obviation limited to subjunctives? If it occurs elsewhere, is it explained by competition?

Briefly on cross-linguistic differences in mood choice
The reader will recall that, according to Farkas, her theory reproduces Ruwet's generalizations; the difference lies in how they are explained. Although Ruwet's explanation is very informal, it is thoroughly semantic. It does not only refer to semantic notions, it attributes obviation to a semantic clash, and absence of obviation to the absence of such a clash. In contrast, whereas Farkas refers to a more formal and rigorous semantic notion, RESP, her explanation of obviation itself is not semantic but functional. The competition account is especially natural if we combine it with mood choice. Schlenker (2005) argues that the French subjunctive is a typical elsewhere mood: it has many uses and no unitary semantics. In fact, it is semantically vacuous. Adding Farkas's (1992) observation that canonical-control infinitives require RESP, it is now straightforward that (in languages like French) infinitival control blocks subjunctives in RESP cases. The account also predicts that when RESP is absent, both the subjunctive and infinitival control are good (no obviation). Likewise, the absence of obviation is predicted for a language that has no infinitives (Modern Greek, Serbo-Croatian).
In Section 5.2, I point out the presence of obviation in certain indicatives in Hungarian that do not have any kind of competitors, neither infinitival nor subjunctive. Comparable data have not, to my knowledge, been considered in the literature.
Regarding mood choice in French, Schlenker (2005) partially bases his theory on Farkas (2003). Farkas considers two hypotheses, (i) A complement that is true in the world of a context is in the indicative, and (ii) Complements involving an evaluative component are in the subjunctive. But, she says, these do not correctly predict the Romanian, French, and Spanish facts. Therefore she abandons both hypotheses, although they are attractive, and makes an Optimality Theoretic proposal instead. This involves two features: (a) A sentence is +Assert iff its Context Change Potential is assertive, and (b) A sentence is +Decided iff it is decided in the context to which it is added. Thus, +Assert entails +Decided, and if the sentence is presupposed, it is +Decided and -Assert. She proposes two constraints, her [49a,b]: The Romanian ranking is *SUBJ/+Decided ≫ *IND/-Assert, and the French-Spanish one is the opposite. However, Farkas notes that in fact, in French the two constraints are quite close, allowing for variation, but with the subjunctive winning. In Spanish, however, the subjunctive is the only good choice here.
Mood distribution in Hungarian is very different from Western Romance. Unlike Balkan languages, Hungarian has both controlled infinitives and subjunctives; however, their territory is much narrower than in French, Spanish, or Italian. A small sample below illustrates that Hungarian is indicative-heavy. 7 7 When the complement of sajnál 'regret' is infinitival, the matrix verb is effectively a fancy extra-clausal negation, much like fél 'be afraid.' These are one-event verbs in the sense of Landau (2003). An anonymous reviewer points out that remél with infinitival complements can be found in newspapers. I observe that, in contrast to English hope+infinitive, the examples are systematically non-agentive and, in my judgment, very stilted.
The complements of the counterparts of some English subject-control verbs a.
One consequence is that it would be difficult to regard the subjunctive as a default mood in Hungarian; another is that the infinitive rarely competes with another form. Most importantly to us, Hungarian offers new testing ground for attributing obviation to competition.

Obviation-like effects exist in Hungarian with remél 'hope' and sajnál 'regret' that only take indicative complements
The literature typically reserves the term "obviation" for an effect pertaining to the pronominal subject of a subjunctive. I will now point out that a similar effect presents itself in certain indicatives in Hungarian that are complements of verbs like remél 'hope' and sajnál 'regret.' The effect is similar to obviation in subjunctives in that to a significant, although interestingly not full extent it is predictable by the RESP vs. non-RESP distinction, and that all the sentences would be fine if the coreferential/bound subject in the complement were replaced by the name of a different person. E.g. 'I hope that I'm jumping around' is bad, but 'I hope that Mary is jumping around' is impeccable.
From the competition perspective, the interest of these data stems from the fact that remél and sajnál only take indicative complements in Hungarian. In other words, (i) the complements at hand are not subjunctives, and (ii) there is simply no competing shape for the complement in the language. Therefore, whatever will turn out to be the correct explanation for why some of the sentences below are weird or unacceptable, it cannot be blocking by infinitives or by some other mood.
English translations will serve to give a sense of the Hungarian facts. I believe English exhibits similar effects but, as above, the actual claims are about Hungarian.
Two pieces of background information are in order before the examples are presented: First, many Hungarian verbs have prefixless and prefixed versions. Prefixed verbs are typically perfective and, with present tense morphology, futurate. Prefixed verbs will be avoided, because the simultaneous present vs. future distinction is of importance to us. The prefixless present tense form comes close to the English simple present for statives and present progressive for eventives.
Second, as was noted in connection with (16) Examples (a)-(b)-(c) have non-RESP complements, and they are not obviative. Examples (h)-(i)-(j) clearly have RESP complements, and they are obviative. In (i), 'trying to help' ensures that the hope does not concern the success of helping. These contrasts are familiar from subjunctives. But the behavior of (d)-(e)-(f)-(g) is novel.
In (d)-(e), standing straight and standing on one leg are both intentional actions. However, the agent may easily be uncertain whether they are successful in standing straight, whereas they are unlikely to be uncertain whether they are standing on one leg or both. We find that, under normal circumstances, the former is non-obviative and the latter obviative. The observation is reinforced by (f)-(g) that involve classical predicates of direct experience, having vertigo and being cold. The subject is the best and, most likely, the only reliable judge of these, and they are obviative.
What these examples highlight is that, under 'hope,' both the complement subject's intentional action (RESP) and their internal certainty matter, and sometimes plain (un)certainty is decisive. Our task will be to unify RESP with direct perception as a category that violates the uncertainty condition that Anand & Hacquard (2013) attribute to English hope (an emotive doxastic item in their terminology). 10 The default expectations in (d)-(e) can be overridden when the agent is standing in front of a mirror or inside a contraption that may be taking over weight from one of their legs. It is significant that such accidental aspects of the situation matter, but this should not obscure the fact that the agent's own certainty is at stake. The contrastive focus versions of (j), (j') and (j") square with this. Notice that subjunctive obviation exhibits a similar sensitivity to "non-normal circumstances." Recall the actions that would be perfectly RESP-ful if, in the given situation, they were not due to urges, mistakes, accidents, or did not require the authority or co-operation of others. The sentences below are to be read as thought-reports, not as apologetic utterances. Firstpersonal 'I regret that…' may tempt apologetic interpretations, but they should be resisted.

Remél
Switching to Mary and she would eliminate this problem, but it would introduce new ambiguities that can be avoided by sticking with the 1st person.
The reader should bear in mind that, as indicated in the section heading, sajnál doesn't mean exactly the same thing as English regret. For example, at the end of a beautiful concert it is natural to say, Sajnálom, hogy vége van, literally, 'I regret that it is over.' The sentence really means, 'It's a pity it's over' or 'I wish it wasn't over.' This is relevant, because English regret might produce confounding effects that Hungarian sajnál luckily does not.
Sajnál 'regret' being factive, uncertainty is not involved here, and so direct experience is not a discriminating factor. Weirdness (notated, again, with a ?) correlates fairly systematically with RESP. The past tense complements are uniformly good: whatever happened, the individual corresponding to the complement subject can no longer do anything about it. In the episodic present and future cases, they may be able to. Intuitively, the source of weirdness is the fact that the same person is performing or planning to perform an action that she is the sole instigator of and wishes that she were not instigating it.
If the complement subject is merely an agent but not RESPonsible initiator, obviation goes away, much like we saw in connection with 'want'+subjunctive. In what follows, I will assume that the sentences describe "normal circumstances" and not urges or mistakes, for example.
I list the examples in the same present, past, future order as with remél 'hope' above. The translations employ the verb regret, with the caveat noted above.

What is the source of obviation(-like) effects with remél and sajnál?
The explanation why some of the sentences with remél and sajnál are weird or unacceptable cannot be blocking by infinitives or some other form in the complement, since there are none. 11 More precisely, if obviation is explained by competition, competition cannot be restricted to forms that co-exist within a particular language, as is traditionally understood. Going beyond a particular language would make sense in that gradable predicates like 'hope' and 'regret' take subjunctive complements in some languages (e.g. Spanish, see Villalta 2008), and there they may have infinitival competitors. Indeed, Buccola, Križ & Chemla (2018) and Charlow (2019) argue, in connection with other phenomena, that the competitors may co-exist in what they call the language of thought. 12 I will not pursue this approach here.
The hypothesis that I will pursue is related to my reading of Ruwet's intuition: (26) Mind-boggling meanings arise from the combination of the semantics of the matrix verb and the subject's relation to the complement situation. The critical relation has to generalize over responsibility and direct experience (perhaps also other things) and over verbs of the type of want/hope and regret (and perhaps others).

Kaufmann (2019) on directive obviation
For such an approach, we can look to Kaufmann (2019). Kaufmann proposes an explicit semanticpragmatic theory of a related phenomenon: directive obviation. The relevant directives are matrix or complement imperatives with person-number agreement. For Slovenian embedded 11 P. Schlenker (p.c.) notes that NP complements of 'want' also exhibit an obviation effect, although they do not have competitors, and (of course) they are not subjunctives. His observation squares with the general argument here, but I will not specifically pursue NPs.
(i) I want her success / departure.
(ii) I want my success / *departure.
12 "The English sentence John broke all of his arms is odd: it implies that John has more than two arms. The oddity, and this inference, can be explained by competition with the alternative John broke both of his arms, which is a more appropriate description of a situation in which John broke his left arm and his right arm. That the English lexicon contains the terms all and both is crucial to this explanation, and the prediction is that in a language with only a lexical term meaning 'all', and no lexical term meaning 'both', the sentence corresponding to John broke all of his arms ought to not be odd. French is a (not so frequent) case in point; however, the French sentence Jean s'est cassé tous les bras 'John broke all of his arms' is just as odd as its English counterpart (Chemla, 2007). This data point thus indicates that competition goes beyond just the lexicon of a given language. (…) [W]e argue that the level of words is not the right level of analysis for alternatives. Instead, we capitalize on recent conceptual and associated methodological advances within the study of the so-called "language of thought" to reopen the problem from a new perspective." ( (2016) on directive na-subjunctives in Greek; Oikonomou judges the example below to be acceptable when uttered by a person without an alarm clock to her mother, who is known to wake up early. Kaufmann reports that Stegovec shares the same judgment for the Slovenian counterpart: The main idea of the semantic-pragmatic account for directive obviation is that the conventional meaning expressed by directive clauses with the respective subject settings is at odds with the contextual requirements for the felicitous use of a directive." Directive obviation reflects a contradiction that arises, if the director is identical to the instigator, between the director's epistemic uncertainty and public commitment to believing that p will come true once the instigator has received the directive. According to the former belief, it is possible for p to be false, and according to the latter, p is necessarily true, in the same worlds.

(31) Epistemic Uncertainty Condition (EUC)
In uttering a sentence translated as □ R p, speaker S holds possible both p and ¬p: ◊ Bs p ∧ ◊ Bs ¬p (32) Director's Anticipation If director d is publicly committed to believing that instigator a believes that p ∈ ∆ is R-necessary, then d is publicly committed to believing that p will come true: Although Kaufmann does not discuss the weakening of obviation in the absence of control in detail, it is clear that a directive subject that does not stand in the RESP relation to the content p of the directive clause is not an instigator, so Director's Anticipation does not arise, and there is no contradiction. This is supported by Hungarian imperatives which, somewhat similarly to Slovenian, involve a version of the morphological subjunctive, 14  13 Kaufmann (p.c.) kindly provides the following informal explication for how the contradiction arises in the case of a first-person directive: (1p) When a speaker utters a first person directive subjunctive in the commitment case ("I should really write!") and it is assumed that the speaker is the only person who can realize its prejacent p, it becomes mutual joint belief that the speaker believes that (i) the speaker holds possible that p ("I will write") and the speaker holds possible that not p ("I will not write") (ii) if any participant learns that p is optimal and they control p (by assumption: only the speaker themselves) they will carry out p The speaker expresses p is optimal (at-issue meaning of the imperative), so they also become publicly committed to "p is optimal" Therefore, it becomes mutual joint belief that the speaker believes that (iii) p is optimal.
Consequently, the 1st person speaker incurs a conflicting requirement: at all worlds in the context set, their set of doxastic alternatives, (a) is compatible with p and also with not p, (b) entails that every participant to the conversation either [does not know that p is optimal or cannot realize p], or realizes p (c) entails that p is optimal.
Assuming that it is mutual joint belief that the subject referent, i.e. here the speaker, can realize this means that the speaker's doxastic alternatives have to be empty. As this would come about in part thanks to presupposed meaning, the whole thing crashes because it is infelicitous.
14 Hungarian "subjunctives" and "imperatives" both inflect for all six persons and have identical verbal morphology. Traditionally, they are told apart by the order of the verbal prefix, which procliticizes to the verb in the subjunctive but follows in the imperative. But many factors complicate the prefix-verb order.
15 In non-obviative wishes, Modern Greek as well as Hungarian, it does not seem to suffice that the speaker does not bear RESP to the situation. More so than subjunctives under 'want,' which merely carry a realism presupposition, these wishes appeal to some other individual, as in (30) and (34) or to some higher power, as in (35) and (36), as instigators. In the absence of such, the speaker must possess magical powers, or the directive is incoherent:

Kaufmann (2020) on 'want'+subjunctive
If uncertainty on behalf of the director or, more generally, the immediately higher attitudeholder, is a critical factor in directive obviation, then Kaufmann's account promises to extend to the obviative instances of 'want'+subjunctive. 16 The wanter shares the director's uncertainty, and the individual linked to the subjunctive subject, the instigator's certainty. 17 Kaufmann (2020) indeed proposes that her analysis of directive obviation extends to subjunctives under desideratives, as anticipated in Kempchinsky (2009). I will accept this analysis for 'want'+subjunctive, and move directly to the new indicative data introduced in this paper.

The need to go beyond the directive obviation model
Kaufmann's analysis serves as a proof of concept, even as a model, for an account of obviation and exemption from obviation in purely semantic and pragmatic terms, without reference to competition. But there are reasons that compel us to go beyond it. In line with the main focus of this paper, I now consider the case of remél 'hope,' sajnál 'regret,' and some of their relatives. Another issue that calls for comments is directive obviation in imperatives expressing encouragements, which I argue are not necessity modals; the discussions of this is delayed to the Appendix.
Recall that the indicative complements of remél 'hope' and sajnál 'regret' exhibit obviationlike effects without there being any mood competitor in the language; this converges with Stegovec's (2019) observation about Slovenian directives, although the specific moods and the exact meanings are different.
Hoping and regretting clearly do not involve directives, but the general spirit of Kaufmann's contradiction account of obviation may extend to these cases.

'Hope,' uncertainty, and direct experience
As was discussed in Section 5.2.1, it is well-established that the hoper is uncertain whether the complement proposition is true. This parallels the director's uncertainty (EUC) in Kaufmann's analysis. If the director's anticipation (DA) can be successfully replaced by some other consideration that guarantees that, when the hoper is identical to the individual linked to the complement subject, the hoper is certain that the complement proposition is/was/will be true, then Kaufmann's contradiction account can be replicated.
The main interest of the 'hope' data is is that here, not only RESP-ful complements but also ones describing direct experience cause unacceptability. In the case of direct experience, the individual linked to the complement subject is the one and only reliable judge of the truth of the complement; this clearly confirms that a conflict between certainty and uncertainty is at stake.
(37) #Remélem, hogy (nem) szédülök. 'I hope that I (don't) have vertigo' This situation also suggests that RESP itself is relevant in two different ways. The initiator of an intentional situation has an equally first-personal access to it as the experiencer of a direct experience. One might say that in the context of 'hope,' the relevant aspect is "intentional action as direct experience." But in other contexts, 'want' and 'regret' among them, when direct experience does not cause obviation and only RESP does, intentional action as such is relevant.

'Regret' and a counterfactual desire
Kaufmann's approach to directive obviation does not carry over to 'regret' as easily as it does to 'hope.' The regretter does not share the uncertainty of the director, wanter, and hoper. But we find a promising source of a contradiction if we recall the following: Although the counterfactual reasoning ought to be properly formalized, the entailments of the 'regret' sentence have an appropriate whiff of contradiction to them.

Illocutionary relatives (thought relatives)
Z. Gendler Szabó (p.c.) observes that a whole class of felicity condition violations represent clashes similar to the effect with 'regret' right above.
(40) a. #I promise to join but I don't intend to. b. #It is raining but I don't believe it. c. #I don't want to know but who killed the lady? d. #Turn left but it's a bad idea.
As Searle (1969, 62) puts it, [The speaker] could not say without absurdity, e.g. "I promise to do A but I do not intend to do A." The conditions whose violation leads to absurdity above are sincerity conditions in Searle's theory of illocutionary acts. A sincerity condition specifies a psychological state, and the performance of the illocutionary act counts as the expression of the psychological state. Here is a sample: From our perspective, the act aspect of the phenomenon is not critical. (40b) exemplifies Moore's paradox, which can be seen here to violate the sincerity condition of assertion (41b). But Shoemaker (1996, 74-76) argues that asserting (40b) is not needed for (40b) to be paradoxical. "What seems to me too little noticed is that there is something paradoxical or logically peculiar about the idea of someone's believing the propositional content of a Moore-paradoxical sentence, whether or not the person gives linguistic expression to this belief. What really needs to be explained is why someone cannot coherently believe that it is raining and that she doesn't believe that it is, despite the fact that the conjuncts of this belief can both be true… Consider the proposition that is the conjunction of ["It is raining, but I don't believe that it is raining"] and the proposition that the speaker believes this proposition… That is self-contradictory… So it is a feature of the contents of Moore-paradoxical sentences that if they can be believed at all, the subject of such a belief could not, logically, believe that she had it." 18 This qualification is important, because I would like to maintain that the problem with the conjunctions in (40) is grounded in semantics. 19 Szabolcsi Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1421 Let me add that outright negating a sincerity condition is not necessary for absurdity. Calling it into question by adding an epistemic modal will do: (42) a. #I promise to join but I may not intend to. b. #It is raining but it may/might not be raining. c. #I may not want to know but who killed the lady? d. #Turn left but it may not be a good idea.

Self against self: a family of contradictions
Below is a list of the cases discussed above. The English translations are assumed to have the same status as the Hungarian originals.
Moore's paradox #It is raining, but I don't believe that it is raining. b.
Paradoxical promise #I promise to join, but I don't intend to join. c.
'I want for me to visit Mary' e.
'I wish I weren't jumping around' f.
'Hope'+indicative obviation #Remélem, hogy ugrándozok / nem szédülök. 'I hope that I'm jumping around / don't have vertigo' Although I am not in a position to offer a unitary analysis at this point, the items in the list bear, at least, a family resemblance. They are characterized by the following kind of contradiction, or conflict: the speaker or matrix attitude-holder denies, or exhibits uncertainty of, something that she has immediate certainty of or control over. 20 The exact nature of the contradiction, or conflict, seems to correlate with the varied attitudinal contexts. Focusing on immediate certainty or control, our cases involve different sources for it. In (a)-(b), the source is the fact that the speaker performs an illocutionary act, or a "thought act," that carries a sincerity condition that specifies the speaker's own psychological state (self-knowledge, in Shoemaker 1996). In (c, d e), the RESP-ful individual intentionally brings about a situation. In (f), with 'hope,' the source may be that this individual is the direct experiencer of a state. 'Hope' is equally sensitive to direct experience and RESP in the complement. I interpret this as indicating that RESP itself plays a role in two ways. In (c, d, e), initiatorhood is specifically relevant. But in (f), probably all that matters is that engaging in a reliably identified intentional action entails the same kind of immediate certainty as directly experiencing a state; this is contradicted by the uncertainty inherent in hoping.
The similarity in the status of direct experience and RESP-ful action is illustrated by modifying both types of propositions by epistemic modals and attempting to turn them into informationseeking questions: (44) a. #I may have vertigo, I may be cold, I may be itching. b. #I may be jumping around, I may be scratching. 9 What does the internal perspective correlate with in the grammar? Farkas (1992) proposed that infinitives are better suited for the expression of canonical control than are subjunctives. This is correct, even if the unacceptability of 'want'+subjunctive is not due to blocking by the infinitive, but to a semantic-pragmatic clash. The reason why the clash occurs with the subjunctive and not the infinitive is that the two do not mean the same thing. 22 The infinitive has a kind of de se reading that offers an "internal perspective" of the event and is thus well-suited to RESP-ful action, whereas the subjunctive does not offer an "internal perspective." De se readings are typically discussed in connection with DP arguments. Regarding the subject, not only PRO in the infinitive but also the pronominal subject of the subjunctive has a solidly de se reading in the cases that are of interest to us: (46) Azt pro 1sg akarom, hogy pro 1sg jó jegyeket kapjak / pro 1sg egészséges legyek. 'I want for me to get good grades / for me to be healthy' Minden lány i azt akarja, hogy pro i jó jegyeket kapjon / pro i egészséges legyen. 'Every girl wants for her to get good grades / for her to be healthy' There is absolutely no possibility of error through misidentification, which is what we would expect if the subject pronoun were read de re. 23 The attitude-holder is fully aware that his/ her desire pertains to his/her own self. However, Schlenker (2005; argues that in the infinitive, the event argument is also read de se, in contrast to the subjunctive, where the event argument has a de re reading. For concreteness, below is a direct adaptation of Schlenker's (2011Schlenker's ( : 1596 formalization to our cases. The complementizer is a simultaneous λ-abstractor over individuals, events and possible worlds. In infinitival (49), e k on the complementizer directly binds the clause-internal e k . (The event de se reading probably always goes with a subject individual de se one.) By contrast, in subjunctival (50), e m is bound by the existential at the clause edge and is only indirectly identified with e k on the complementizer: e m ≈ e k indicates that the events e m and e k occur at the same time; the interpretation is only individual de se.
Going beyond want, Schlenker (2011Schlenker ( , 1595 comments on remember examples originating with Higginbotham (2000) and goes on to amplify the argument with a direct experience case from French. "Most adults are in a position to utter [our 51b] truly, but very few have such a good memory that they could assert [our 51a], which requires that one actually remembers the event of walking [or, falling] rather than the general fact that one did walk [or, fell] in the past. (51) a. I remember walking to school in the 5th grade / falling downstairs. b.
I remember that I walked to school in the 5th grade / that I fell downstairs.
(…) In a situation in which I see myself in a mirror, realize that this is me, and get the impression that the person I see is shivering, it is possible to use the full complement, as in [our 52b], but it is far less natural to use the infinitive, as in [our 52a]. No such contrast obtains if I have an internal feeling of shivering (if anything, the infinitive is more natural in this case). Importantly, the fact that PRO is read de se is unlikely to explain this contrast: in both cases, I have full knowledge of the identity of the person under discussion. Rather, we appear to obtain a de se / de re contrast, not with respect to the individual argument of the verb, but with respect to its event argument." (52) a. J'ai l'impression de greloter. 'I have the impression to shiver' b. J'ai l'impression que je grelotte. 'I have the impression that I shiver' These observations hold out the hope that grammatical (syntactic and semantic) correlates of the internal perspective can be identified. I admit that this task must be accomplished by future research. I briefly indicate some of the challenges below. 24 On the interpretive side, clauses with RESP and clauses describing direct experience critically require an internal perspective. If the notion of event de se is suitable to capture those, does that approach extend to Searle' sincerity conditions?
On the syntactic side, the examples in the present section might suggest that the internal perspective is available in infinitives and gerunds, but not in finite clauses. Recall, however, that the critical 'hope' and 'regret' complements discussed in the preceding sections are indicatives and yet, they appear to offer an internal perspective.
Likewise, perhaps the most clearly grammaticized case involving an internal perspective requirement, so-called conjunct marking in Kathmandu Newari, obtains in tensed indicative clauses: past and future tense verbs have "conjunct" or "disjunct" endings. Zu (2018) argues in careful detail that conjunct marking requires (i) the binding of the subject by the seat of knowledge (a discourse participant or an attitude-holder), (ii) a RESP relation of the individual linked to the subject to the situation described by the clause with the conjunct-marked verb, and (iii) an internal perspective of the content of that clause. One critical piece of data Zu (2018: 121) discusses for (iii) is our (53): "In the following context, the individual identity holds between baby Shyam, the toy breaker, and adult Shyam, the attitude holder. Shyam is both the seat of knowledge and the responsibility holder for the attitude complement. But the adult Shyam is not reporting his in-body experience of the toy-breaking event, rather, he takes a spectator's perspective as if he is watching someone else breaking the toy. In this context, the use of the conjunct verb is not appropriate." (53) Scenario: Shyam is watching baby videos of himself. In one video, the six-monthold Shyam throws his toy truck to the wall and as a result the toy truck breaks into pieces. He says to his wife, "I really was a trouble maker. I broke a toy just like that." Shyam-a [wã: (s) nheba *tachyan-ā /tachyat-a dhakā:] dhāla. Shyam-ERG he.ERG toy break-PST.CONJ /break-PST.DISJ that said 'Shyam 1 said that he 1 broke a toy' I conclude, therefore, that there is no reliable correlation between the non-finite vs. finite character of the clause and the internal perspective vs. the lack thereof. The absence of a 24 Ruwet (1991: 16) himself grappled with these questions: "All other things being equal, the finite complement, with two occurrences of coreferential subjects, will be adequate if, on the one hand, the relation of self-to-self determined by the meaning of the matrix verb involves an internal differentiation and highlights two distinct facets of the self, and/or introduces a certain distance between self and self-and if, on the other hand, the relation of self-to-self tends to be viewed, in the same context, in the same manner as the relation of self-to-other. Inversely, the infinitival complement, with a single occurrence of the subject, will be appropriate if the relation of self-to-self, as determined by the matrix verb, contrasts with the relation of self-to-other, and if the internal distance between the two instances of the self tends to vanish, or if they are viewed from fundamentally the same point of view." Szabolcsi Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1421 reliable correlation is confirmed, from the other direction, by Hungarian infinitival control clauses that exhibit overt nominative personal pronoun subjects, discussed in detail in Szabolcsi (2009). These invariably have an individual de se reading but, I judge, not an event de se one.
In this respect, infinitival (54) and subjunctive (16) And conversely, Hungarian is a null subject language and the subjunctives and indicatives we have surveyed all have null subjects (unless the subjects are specifically focused). So the non-overtness of the subject is not critical, either. (Some further related issues are discussed in A.3.2-3 in the Appendix.) Getting a grip on grammatical correlations is one of the big tasks for further research.

Conclusion
This paper examined the claim in Farkas (1992) and Schlenker (2005) that obviation in subjunctives under desideratives is due to competition: blocking by an infinitival construction that is better suited to carry the intended meanings. I largely confirmed the semantic description in terms of a RESP relation between the individual linked to the subject and the event in the subjunctive clause and in terms of the absence of an event de se reading that these works offered. However, novel data from Hungarian led me to reject the blocking account. Hungarian exhibits obviation-like effects in indicatives under 'hope' and 'regret' that do not have competitors (either infinitives or subjunctives). I followed Kaufmann (2019;, who derives the obviation effect in directives from a semantic clash that obtains, when the director and the instigator are identical, between the director's Epistemic Uncertainty and the director's Anticipation that the subjunctive proposition will be true in all worlds. But especially the fact that 'regret' does not involve uncertainty led me to place this clash in a wider family, one that includes, among other things, Moore's paradox and negations of the sincerity conditions of illocutionary acts. The clash obtains when an internal perspective of the "obviative clause" is at stake: the speaker or matrix attitude-holder denies, or exhibits uncertainty of, something that she has immediate certainty of or control over. I finally asked whether the internal perspective has an identifiable grammatical correlate, and enumerated a number of descriptive challenges. The conceptual unification and the proper grammatical characterization of obviation effects is a major task for further research. (66) a. #Rögtön vedd el akármelyik almát! (command, #FCI) b. Nyugodtan vedd el akármelyik almát! (encouragement, FCI) The first-person imperative remains obviative in combination with an akár-FCI: (67 #Vegyem el akár-melyik almát! (#imperative w/ FCI, 1sg) take.SUBJ.1SG away any-which apple.ACC I conclude that encouragements are not necessity modals and yet, they exhibit obviation.
M. Kaufmann (p.c.) asks how the possibility modal analysis explains that all imperatives fail the contradictory conjunction test that permissions pass: (68) a.
Hereby, I allow you to leave through the back door, and hereby, I allow you to leave through the front door. b.
OK, you can leave through the back door and you can leave through the front door. c. #(OK,) leave through the back door and leave through the front door.
I propose that a modal force that is "intermediate" between plain possibility and necessity suffices to derive that contradiction. Notice that (70) is as contradictory as (69): Everyone left through the back door and everyone left through the front door.
(70) There is a majority of the people who left through the back door and there is a majority of the people who left through the front door.
We may analogize on (70) with reference to the fact that encouragements are not simple permissions. Quite informally, assume a modal operator , such that p requires for there to be a majority of the permitted or preferred worlds in which p holds (as opposed to ◊p, which only requires for there to be at least one such world). The Fluctuation/Viability constraint excludes cases with no variation, and p allows for variation. On the other hand, the contradiction between Epistemic Uncertainty and Director's Anticipation only obtains if the imperative has the force of necessity.
I conclude that the explanation of directive obviation should have a component that is neutral as to the difference in modal force between commands and encouragements, established above. As a preliminary suggestion, we may invoke a preparatory condition shared by a range of illocutionary acts in Searle (1969 The "not obvious to S" part is similar to the director's Epistemic Uncertainty in Kaufmann's terms, but the "not obvious to H" part is new and seems useful, likewise the "normal course of events" qualification. If H, the potential instigator, is uncertain whether she would do A in the absence of receiving a request, advice, or command, then we should be able to derive that only some S that is not identical to H can be the source of an effective request, advice, or command to H. 27 For this purpose, the specific modal force of the directive will not make a difference. Note that the extension to obviation with want that I adopted from Kaufmann (2000) remains fully in force: it does not involve a modal force issue, nor an addressee (H).

A.2 Subjunctives under 'want' carry a realistic belief presupposition
This section argues that the subjunctive under 'want' carries a presupposition that the infinitive does not. In addition to enriching our understanding of the meaning of the subjunctive, the presence of the extra presupposition may speak against infinitival blocking, depending on the semantic theory applied to competition. Let us start with English. The plain control infinitive and ECM sentences below merely, or primarily, express a desire, or preference. If the desires seem strange, as they do in the cases below, the reaction would be, simply, "Why would you like that?" (72) a. I want to be 10 years old again. b.
I want the Earth to be flat.
In contrast, the for-infinitives that correspond to subjunctives in English signal that the attitudeholder considers the complement situation to be a realistic extension of her belief-worlds. This is quite clear when the complement situations are in fact not realistic. Here the desires are not merely strange but outright crazy, so the reaction would be, "Why do you think that's possible?" (73) a. I want for me to be 10 years old again. b.
I want for the Earth to be flat.
A typical use of such sentences is where the speaker has an action plan for bringing about the described situation. But this does not seem necessary. For example (H. Aparicio, p.c. based on Catalan), the members of the following pair seem to mainly differ as to whether the speaker indicates that she believes that winning is still possible, irrespective of whether she is working on furthering the cause.
(74) a. I want to win.
(regardless of the stage of an election) b.
I want for me to win. (#if the wanter is sane and clearly losing the election) The same contrast holds in Hungarian, modulo the fact that Hungarian does not have ECM, and the for-infinitives are regular subjunctives: The first example merely describes a desire as a matter of psychological fact. The second is an anankastic conditional, where a wants p reports on a preference that the agent assigns a special status to: an action-relevant preference. C&L call this an effective preference. The anankastic conditional as a whole outlines a possible course of action for realizing the preference.
C&L assume that want has an underspecified semantics, and the context selects one of the preference structures that are given as part of the model.
(80) Given a set of worlds W, a preference structure is a pair (P,<), where P⊆℘(W) and < is a strict partial order on P.
Effective preferences have two important inferential properties, Consistency and Realism. Consistency requires that whenever an agent wants two or more things to an equal degree, these things must be compatible with each other and with what the agent believes. Or, to