Linguistic Miscellany Thread

[Travis B.]

The present subjunctive is neither disappearing nor archaic in English! Okay, it might be in EngE, but it is not in NAE!

[Linguoboy]

The present subjunctive is neither disappearing nor archaic in English! Okay, it might be in EngE, but it is not in NAE!

Nor is it a present subjunctive! (Pullum calls it "irrealis", Zwicky remains agnostic as to how it should be labeled and just calls it an "underbrush form" that is present in the speech of some speakers and not in others with little consequence for the rest of the system. I've even seen analyses that try to make a non-finite form out of it.)

[Travis B.]

Well yes, it isn't a present subjunctive per se, it is just called one because its forms look similar to the usual present tense forms (even though you could also note that they are identical to the infinite forms too). I have heard people seriously argue that these forms really are non-finite by people here (you know who you are), but I personally find the conciding of these forms with the infinitive verbs to be ignoring that their apparent infiniteness seems to be more just by coicidence tha anything.

[Richard W]

Well yes, it isn't a present subjunctive per se, it is just called one because its forms look similar to the usual present tense forms (even though you could also note that they are identical to the infinite forms too). I have heard people seriously argue that these forms really are non-finite by people here (you know who you are), but I personally find the conciding of these forms with the infinitive verbs to be ignoring that their apparent infiniteness seems to be more just by coicidence tha anything.

There are a few other places that the present subjunctive does turn up, such as implausible but possible protases and in similar concessive clauses ('even though') and very occasionally after 'whether'. What can be odd about these indirect commands is their negative forms and how they transform. The odd negative forms my desire that he leave => my desire that he not leave is parallelled by even though he not know what is best. What I have noticed is that indirect commands of this form have a tendency not to follow the sequence of tenses. This made me think that their mood might be imperative, but that would not explain the negative above.

The past subjunctive is fairly secure - if I were in opposition to if I was is likely to be with us for some time to come, even in the UK.

[Kuchigakatai]

It has auxiliary verbs that express modality at least, much like English... if that counts.

You mean like savoir? (That example is just to explore the bounds.) I'm not yet disputing the ability of French to express the concepts. I'm doubting that they need to be distinguished from ordinary verbs.

That agrees with my feeling that French doesn't really have a special category of modal verbs.

It does look like it wasn't obvious what I meant by "auxiliary verb" and "modal verb".

By the former, I meant the traditional definition of simply any verb that can take a bare infinitive complement right after (which would indeed include savoir: je sais parler espagnol 'I know how to speak Spanish').

By the latter, I meant a semantic category of verbs, not a syntactic definition as you seemed to be pursuing. (In light of priscianic's mention of the sprawling definitions of "modality", I meant specifically the "modality" that I've seen talked about in typological works, involving the deontic and epistemic notions often expressed in English by "can/could/should/must/have to", whether possible/probable/suggested/inferred or certain/necessary/obligatory.)

For example:

Code: Select all

1)  Me alegra         que  haya          llegado Juan.    (Spanish)
    me make.happy.3sg that have.SBJV.3sg arrived Juan
    ‘I'm glad that Juan arrived.’

You can't say this sentence if you don't believe that Juan arrived. You're not saying that you're glad (or would be glad) if Juan arrived in some hypothetical world—you're saying that you're glad that Juan arrived in this world. I'm not sure how much more real you can get—and yet you're required to use subjunctive here.

I'm not sure what you're saying in this example. That sentence can express 100% certainty as you say, yes, but it could also mean 'I'm glad that Juan arrived if it's really true', expressing a bit of a lack of confidence about it being true. In fact, the latter reading would be more common, as there is also the possibility of using the indicative to unambiguously express 100% certainty (me alegra que llegó / ha llegado Juan).

(It's common for English-language grammars of Spanish to have a false, bogus rule that words of emotion like alegrar 'make sb happy' or feliz 'happy' must take the subjunctive, see Palmer (2008) for some amusing yet gory commentary, but I'm going to ignore that. It doesn't seem you were referring to this.)

Your sentence with que la policía los haya interrogado is correct by the way, you weren't misremembering it. At least in my dialect (El Salvador), I can't switch haya interrogado for an indicative.

Relatedly, I love making fun of Spanish for using the present indicative in future-tense conditions (which are typically irrealis semantically) but the subjunctive in future temporal clauses with cuando (when they're realis expressing 100% certainty).

Si vienes, lo terminaremos.
if come.2SG.PRES.INDIC, it finish.FUT.1PL
'If you come by, we'll finish it.' (Context: I'm pretty uncertain you will, but I'm mentioning the result if you do.)

Cuando vengas, lo terminaremos.
when come.2SG.PRES.SBJV, it finish.FUT.1PL
'When you come, we'll finish it.' (Context: I'm certain you will show up, and I'm just re-stating the plan.)


Regarding the discussion, I think your objection to zompist largely boils down again to the usefulness of morphosyntactic labels when discussing specific languages (the Spanish subjunctive_m, to repeat a kind-of-joke from last week) vs. the usefulness of cross-linguistic semantic categories and common phenomena to be able to compare languages (the IRREALIS when morphosyntactically expressed, prototypically actions/events/states of hypothetical worlds).

Well yes, it isn't a present subjunctive per se, it is just called one because its forms look similar to the usual present tense forms (even though you could also note that they are identical to the infinite forms too). I have heard people seriously argue that these forms really are non-finite by people here (you know who you are), but I personally find the conciding of these forms with the infinitive verbs to be ignoring that their apparent infiniteness seems to be more just by coicidence tha anything.

To be fair, verb conjugation is so decayed in English (aside from that of "to be") that it seems a good idea to consider the pros and cons of such analyses if only to conclude rejecting them, including an analysis where English lacks the finite vs. non-finite distinction altogether but rather has contexts where 3SG -s shows up (always after one of the singular pronouns or an NP replaceable by one of them) and others where it doesn't. The distinction makes more sense in languages like Italian where you have limited non-finite systems with only two distinctions (fare, avere fatto; facendo, avendo fatto) vs. a finite system with a bunch of more distinctions each with subject agreement (io faccio, tu fai, lei fa...).

[priscianic]

For example:

Code: Select all

1)  Me alegra         que  haya          llegado Juan.    (Spanish)
    me make.happy.3sg that have.SBJV.3sg arrived Juan
    ‘I'm glad that Juan arrived.’

You can't say this sentence if you don't believe that Juan arrived. You're not saying that you're glad (or would be glad) if Juan arrived in some hypothetical world—you're saying that you're glad that Juan arrived in this world. I'm not sure how much more real you can get—and yet you're required to use subjunctive here.

I'm not sure what you're saying in this example. That sentence can express 100% certainty as you say, yes, but it could also mean 'I'm glad that Juan arrived if it's really true', expressing a bit of a lack of confidence about it being true. In fact, the latter reading would be more common, as there is also the possibility of using the indicative to unambiguously express 100% certainty (me alegra que llegó / ha llegado Juan).

(It's common for English-language grammars of Spanish to have a false, bogus rule that words of emotion like alegrar 'make sb happy' or feliz 'happy' must take the subjunctive, see Palmer (2008) for some amusing yet gory commentary, but I'm going to ignore that. It doesn't seem you were referring to this.)

Hmmm, interesting! It's a common thread in the literature that Spanish/French/Italian can all have subjunctive in the complement of emotive factive predicates, and this is in fact a big puzzle (and one of the counterarguments to the "irrealis" class of analyses for the subjunctive in Romance). Of course, "Spanish" is a lie, and there are a number of different varieties of Spanish and a lot of variation between those varieties, so it's very plausible that what's reported in the literature doesn't accurately represent your Spanish.

So in your Spanish, if there is no factive presupposition when these emotive predicates take subjunctive complements, the following should be possible, felicitous, non-contradictory sentences?

Code: Select all

1)  Me alegra         que  haya          llegado Juan, pero no creo        que  haya     llegado.
    me make.happy.3sg that have.SBJV.3sg arrived Juan  but  no believe.1sg that have.SBJV.3g arrived
    ‘I'm glad that Juan arrived, but I don't think he arrived.’

2)  Lamento    que  María esté        enferma, pero no creo        que  esté        enferma.
    regret.1sg that Maria is.SBJV.3sg sick     but  no believe.1sg that is.SBJV.3sg sick
    ‘I regret that Maria is sick, but I don't believe that she's sick.

The idea being, of course, that if alegrar ‘make glad’ and lamentar ‘regret’ didn't commit you to the truth of their complements, that you could openly deny (your belief in) the truth of their complements.

(If you can have both indicative and subjunctive in these context, then presumably there's some kind of difference between the two. I'm wondering whether that difference is really about the factive presupposition/your level of commitment to the truth of the embedded proposition, or whether it's something else.)

[zompist]

Ser has basically covered what I'd say about subjunctives. As Ser says, it's hard to disentangle "what the subjunctive means" from "what it's come to be used for after thousands of years of development". The prototypical meaning is (I think) still irrealis, but it's not exactly unheard-of for meanings to ramify and even reverse.

The contrast between verbal mood and modality basically boils down to this, from what I can tell: if you can analyze some expression as a quantifier over possible worlds, then it counts as a modal. If you can't, but the expression in question somehow feels "tied to" modality, and it's somehow "verb-y", then it's verbal mood. Again, vague—but I think vagueness is the state of our current understanding of verbal mood, unfortunately.

This works fine for some modals, but not so well for modals of permission (may, might, can), obligation/intent (must, will), or ability/knowledge (can).

There is some link with emotions too-- desire, fear, or benefit. We see this with the Romance subjunctive but also with affixes in Quechua. We could leave these out of mood/modality, though it seems arbitrary.

[zompist]

Necessity corresponds to universal quantification over the modal base: must p can be paraphrased as "for all worlds w in the modal base, p is true in w".

Oh, and I don't quite agree with this. "Must" (as a deductive rather than an obligative) groups nicely with the evidentials, which are mostly differentiated by source of knowledge rather than a particular relationship with possible worlds. In particular "must" implies that you know something by deduction, and in fact it's often accompanied by the grounds for saying it.

[priscianic]

Ser has basically covered what I'd say about subjunctives. As Ser says, it's hard to disentangle "what the subjunctive means" from "what it's come to be used for after thousands of years of development". The prototypical meaning is (I think) still irrealis, but it's not exactly unheard-of for meanings to ramify and even reverse.

I'm not sure we have the same notion of meaning here, if I'm understanding you correctly. Following the norm in formal semantics, I'm understanding "meaning" to be roughly truth-conditional, e.g. what a sentence "means" is a set of conditions that a particular world/situation has to meet in order for that sentence to count as true (plus some restrictions on the discourse context, like presuppositions), and I'm also understanding the "meaning" of a sentence to be compositionally built up from from the meanings of its parts, in a way that we can explicitly formalize (how else can we understand and interpret novel sentences?).

As such, it doesn't really make sense to me for a single, unitary operator to have a "prototypical meaning" in some syntactic contexts, but a completely different meaning (that's in fact completely the opposite of the proposed "prototypical meaning") in other contexts. That's not vagueness, or (discourse) context-sensitivity—that's accidental homophony. And sure, maybe you do want to say that there are actually two subjunctives in Romance, that happen to look exactly the same in all cells of the verbal paradigm, for all verbs—but I think plenty of people won't find that at all satisfying.

So under this kind of view, the puzzle is to figure out a single kind of denotation for a subjunctive operator that is able to account for when exactly it appears and doesn't appear—and this puzzle isn't solved yet.

(It's also a different question where the Romance subjunctive came from, and how that semantic change happened. It may well be true that the subjunctive was originally a "real" irrealis. But that wasn't true in (classical) Latin—the Latin subjunctive shows up in non-irrealis contexts (e.g. result clauses, certain cum clauses), so the proposed "original irrealis meaning" would have to have existed even prior to that. And I'm pretty sure we don't have the relevant kind of data to be making these kinds of claims about e.g. Proto-Italic.)

The contrast between verbal mood and modality basically boils down to this, from what I can tell: if you can analyze some expression as a quantifier over possible worlds, then it counts as a modal. If you can't, but the expression in question somehow feels "tied to" modality, and it's somehow "verb-y", then it's verbal mood. Again, vague—but I think vagueness is the state of our current understanding of verbal mood, unfortunately.

This works fine for some modals, but not so well for modals of permission (may, might, can), obligation/intent (must, will), or ability/knowledge (can).

There is some link with emotions too-- desire, fear, or benefit. We see this with the Romance subjunctive but also with affixes in Quechua. We could leave these out of mood/modality, though it seems arbitrary.

This standard Kratzerian semantics works quite straightforwardly for those readings of modals as well (as explained by her in the foundational papers I linked, as well as in von Fintel and Heim's textbook on intensional semantics). She attributes the variable flavor of (IE) modals to a rampant vagueness/underspecification about the conversational background that supplies the modal base that the modal then quantifies over. So for an epistemic modal (e.g. it must/might be raining), the relevant conversational background is a set of facts/premises that the interlocutors take for granted; for a deontic modal, the relevant conversational background is a set of laws/rules/orders (e.g. you must sign here, you may enter); for a circumstantial/ability modal it's a set of actual circumstances in the world, like a particular agent's abilities (e.g. blueberries can grow here, I have to sneeze).

You then basically intersect all of the propositions provided by the conversational background to create a set of worlds that verify all those propositions (there are some complications to this that Kratzer discusses, but I'll leave them aside). So for an epistemic modal, you end up with a set of worlds in which the relevant premises all hold true; for a deontic modal, you end up with a set of worlds in which all the relevant laws/rules/orders are obeyed; for a circumstantial modal, you end up with a set of worlds in which all the actual relevant circumstances in the actual world hold. And then you quantify over those worlds. So, for instance, for a deontic modal sentence like you must sign here, the resulting meaning can be paraphrased roughly as "in all worlds w in which you fulfill your obligations, you sign here in w".

There is some link with emotions too-- desire, fear, or benefit. We see this with the Romance subjunctive but also with affixes in Quechua. We could leave these out of mood/modality, though it seems arbitrary.

I think this intuitive connection has a lot of merit (though it has been challenged). The standard approach in formal semantics is to say that yes, these things are modal.

It's a longstanding tradition in formal semantics to analyze attitude predicates—from doxastic/belief predicates like believe and think to emotive predicates like want and be glad that—as having meanings that are also modal in roughly the same way, in that these predicates quantify over certain kinds of worlds. This intuition dates back at least to Hintikka 1969 (couldn't find an open-access version; you can find an intro summary in the von Fintel and Heim). In the case of believe, these are an agent's "belief-worlds" (or "doxastic alternatives", if you want to be annoyingly fancy about it), which comprise the set of worlds that are compatible with the agent's beliefs (thinking of it a different way, the set of worlds that the agent considers to be open possibilities for what the actual world could look like). And "x believe p" can be given a meaning like "in all of x's belief worlds w, p is true in w". You can perform a similar move for other attitude predicates—for instance, "want" is associated with a set of "desire-worlds", which consists of the worlds in which an agent's desires are fulfilled, and "x want p" can be given a meaning like "in all of x's desire worlds w, p is true in w".

Necessity corresponds to universal quantification over the modal base: must p can be paraphrased as "for all worlds w in the modal base, p is true in w".

Oh, and I don't quite agree with this. "Must" (as a deductive rather than an obligative) groups nicely with the evidentials, which are mostly differentiated by source of knowledge rather than a particular relationship with possible worlds. In particular "must" implies that you know something by deduction, and in fact it's often accompanied by the grounds for saying it.

I don't see the contradiction you're seeing. Just because "must" (and epistemic modals more generally, actually) has an indirect evidential flavor does not mean that it can't be a modal. For the extreme version of this non-contradiction, see Matthewson 2010 for an argument that all evidentials can be analyzed as epistemic modals.

And it's quite easy to construct a modal analysis, following the same kind of template, that captures the indirect evidential intuition. We just need to have a conversational background that consists of "indirect evidence", and that conversational background determines a set of worlds compatible with that indirect evidence, and we can then quantify over that set of worlds. Et voilà, you have a modal analysis of an evidential. Granted, this kind of analysis is somewhat unsatisfactory (what is "indirect evidence" anyway? how can we isolate a set of "indirect evidence", if it's not indirect evidence for something in particular?).

Von Fintel and Gillies (2010) is an excellent paper that discusses the issue of the evidential nature of must (it also has a really cute title: "Must…stay…strong!"). They provide and argue for a semantics for must as an epistemic necessity modal (i.e. as a universal quantifier over a set of worlds)—most of the paper is actually spent justifying this claim, since a lot of people take must to not be quite as strong as a universal quantifier—and they have a method (actually, they provide two different ones) of ensuring that the modal base only contains "indirect evidence" for the proposition in question.

(I think the paper is quite well-written and beautiful, actually, and I think the core insight is quite right. A lot of people have noted that an assertion of must p feels intuitively "weaker", in some sense, than a plain assertion of p. However, by standard principles of modal logic, if must is a necessity modal, then must p should entail p, which feels like the wrong result. As such, lots of people have argued that must actually isn't a necessity modal. Von Fintel and Gillies argue that people are being too hasty here. They argue that must isn't actually logically weak, and that the inference from must p to p should be considered valid, and that the perceived "weakness" of must is not due to any logical weakness but instead due to an indirect evidential component, and they provide a formal characterization of what exactly this "evidential component" is.)

[zompist]

Ser has basically covered what I'd say about subjunctives. As Ser says, it's hard to disentangle "what the subjunctive means" from "what it's come to be used for after thousands of years of development". The prototypical meaning is (I think) still irrealis, but it's not exactly unheard-of for meanings to ramify and even reverse.

I'm not sure we have the same notion of meaning here, if I'm understanding you correctly. Following the norm in formal semantics, I'm understanding "meaning" to be roughly truth-conditional, e.g. what a sentence "means" is a set of conditions that a particular world/situation has to meet in order for that sentence to count as true (plus some restrictions on the discourse context, like presuppositions), and I'm also understanding the "meaning" of a sentence to be compositionally built up from from the meanings of its parts, in a way that we can explicitly formalize (how else can we understand and interpret novel sentences?).

With "prototypical" I'm referring to prototype theory, well explained in Lakoff's Women, Fire, and Dangerous Things. I think your statement corresponds to what Lakoff calls the "objectivist model" (no relation to the political sense), which he spends a good deal of time refuting. But the real problem comes here:

As such, it doesn't really make sense to me for a single, unitary operator to have a "prototypical meaning" in some syntactic contexts, but a completely different meaning (that's in fact completely the opposite of the proposed "prototypical meaning") in other contexts. That's not vagueness, or (discourse) context-sensitivity—that's accidental homophony.

The problem, I think, is assuming that the Romance subjunctive is a "single unitary operator". Why would it be? At the least it's a radial category, as most common words are. There is absolutely nothing surprising about a word having contradictory meanings, especially given a long historical development. E.g. English "with", which once had the meaning "against", still preserved in certain contexts: "fight with", "contrast with".

As I mentioned earlier, "irrealis" doesn't have a single meaning either, though it has a prototypical one. The fact that we can't find a single meaning doesn't show that there is no meaning; it shows that we're dealing with a radial category, or something even more complicated.

So for an epistemic modal (e.g. it must/might be raining), the relevant conversational background is a set of facts/premises that the interlocutors take for granted; for a deontic modal, the relevant conversational background is a set of laws/rules/orders (e.g. you must sign here, you may enter); for a circumstantial/ability modal it's a set of actual circumstances in the world, like a particular agent's abilities (e.g. blueberries can grow here, I have to sneeze).

This is adding something to the notion of possible worlds, but fine, if you do so I'm sure you can make the modals work.

Oh, and I don't quite agree with this. "Must" (as a deductive rather than an obligative) groups nicely with the evidentials, which are mostly differentiated by source of knowledge rather than a particular relationship with possible worlds. In particular "must" implies that you know something by deduction, and in fact it's often accompanied by the grounds for saying it.

I don't see the contradiction you're seeing. Just because "must" (and epistemic modals more generally, actually) has an indirect evidential flavor does not mean that it can't be a modal. For the extreme version of this non-contradiction, see Matthewson 2010 for an argument that all evidentials can be analyzed as epistemic modals.

I think we're dealing with competing definitions here. I didn't say "must" wasn't a modal; I said it was evidential.

(I think the paper is quite well-written and beautiful, actually, and I think the core insight is quite right. A lot of people have noted that an assertion of must p feels intuitively "weaker", in some sense, than a plain assertion of p. However, by standard principles of modal logic, if must is a necessity modal, then must p should entail p, which feels like the wrong result. As such, lots of people have argued that must actually isn't a necessity modal. Von Fintel and Gillies argue that people are being too hasty here. They argue that must isn't actually logically weak, and that the inference from must p to p should be considered valid, and that the perceived "weakness" of must is not due to any logical weakness but instead due to an indirect evidential component, and they provide a formal characterization of what exactly this "evidential component" is.)

OK, I've read the paper (admittedly, rather quickly). I quite agree that "must" is associated with indirect evidence; that's pretty much what I meant by calling it deductive.

However, I disagree with their statement that "must is never weak". They're cherry-picking examples like logic puzzles, but really, have they never read a detective story? Consider a story that includes this exchange:

Lestrade: We posted agents at every airport; no one was observed fitting the suspect's description.
Watson: There's no other way to come to an island. He must have come by sea.
Holmes: No, gentlemen. He did not arrive in England, because he was here all along.

This is what the paper denies is possible (p. 19): someone directly contradicting a deduction, in a way that shows that a deduction is weaker than direct knowledge. They try to make it sound no one could felicitously deny a deduction, but one can.

The earlier authors they quote have it right, I think: must(p) is weaker than p. In an ideal world perhaps it wouldn't be, because deduction would never be wrong, and people would not make claims to have deduced the truth when there are loopholes they've overlooked. But in ours, deduction is frequently wrong, loopholes cannot be avoided, and direct knowledge is better than indirect.

(Pragmatically, a statement that P is of course itself a claim. Just because we see it's raining doesn't prove that it's raining. There are always weird Gettier counter-examples to any claim. But the issue isn't whether a statement P is always true, only whether it's a stronger claim than must(P).)

I should note, there are evidential systems, e.g. Central Pomo, that contrast visual, aural, and deductive evidentials (and more). Would you really maintain that the deductive evidential is just as strong as the visual? Would you predict that, in a context where a speaker could use any of these, the deductive evidential would be most likely to be used or the most likely to be believed?

[Kuchigakatai]

Hmmm, interesting! It's a common thread in the literature that Spanish/French/Italian can all have subjunctive in the complement of emotive factive predicates, and this is in fact a big puzzle (and one of the counterarguments to the "irrealis" class of analyses for the subjunctive in Romance). Of course, "Spanish" is a lie, and there are a number of different varieties of Spanish and a lot of variation between those varieties, so it's very plausible that what's reported in the literature doesn't accurately represent your Spanish.

I'm not (only) objecting to this from my native sense though. Julia Palmer's point in that paper is that the literature contradicts itself, with a number of linguists saying the indicative is perfectly normally used with words of emotion --she mentions Dwight Bolinger's 1953 article "Verbs of Emotion" (Hispania vol 36) as an early example, Terrell and Hooper's 1974 article "A Semantically Based Analysis of Mood in Spanish" (Hispania vol 57) as the first argumentative piece of scholarship explicitly against the rule, and Stanley Whitley amusingly calling this the "Great Mood Debate" in his book Spanish/English Contrasts (2nd ed., 2002)--, and finds that the culprit of the establishment of the bogus rule appears to be Marathon Ramsey's A Textbook of Modern Spanish (1894), a popular textbook that continued to be in print for over half a century receiving its last revision in 1956.

Unfortunately, Palmer doesn't quote any study with native speakers people from Spain, but only from Puerto Rico and Costa Rica, to address the usual vague objection that Latin Americans use the subjunctive "less often". I wonder whether a difference between (prestigious dialects of) Spain vs. Latin America might be involved after all. Although it's likely some of these objections to the rule considered attestations from Spain...

So in your Spanish, if there is no factive presupposition when these emotive predicates take subjunctive complements, the following should be possible, felicitous, non-contradictory sentences?

Code: Select all

1)  Me alegra         que  haya          llegado Juan, pero no creo        que  haya     llegado.
    me make.happy.3sg that have.SBJV.3sg arrived Juan  but  no believe.1sg that have.SBJV.3g arrived
    ‘I'm glad that Juan arrived, but I don't think he arrived.’

2)  Lamento    que  María esté        enferma, pero no creo        que  esté        enferma.
    regret.1sg that Maria is.SBJV.3sg sick     but  no believe.1sg that is.SBJV.3sg sick
    ‘I regret that Maria is sick, but I don't believe that she's sick.

The idea being, of course, that if alegrar ‘make glad’ and lamentar ‘regret’ didn't commit you to the truth of their complements, that you could openly deny (your belief in) the truth of their complements.

They are impossible, non-felicitous and contradictory. But only because the first clause expresses high confidence (say, 85%~100%), while pero no creo que haya... expresses low confidence (say, 0%~15%). So although I don't accept these two sentences, I do perfectly accept:

Me alegra que haya llegado Juan, aunque no creo totalmente que lo haya hecho.
'I'm glad that Juan arrived, although I don't completely believe he did so.'

Lamento que María esté enferma, si lo que me dices es verdad.
'I feel sad that Maria is sick, if what you tell me is true.'

Me alegra que haya llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
'I'm glad that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'

Lamento que María esté enferma, asumiendo que no te has equivocado.
'I feel sad that Maria is sick, assuming you didn't make a mistake / get confused.'

(If you can have both indicative and subjunctive in these context, then presumably there's some kind of difference between the two. I'm wondering whether that difference is really about the factive presupposition/your level of commitment to the truth of the embedded proposition, or whether it's something else.)

It's what I feel, at least stereotypically in that largely context-less case. Indicative for 100% confidence, and subjunctive for something like 85%~100%. Re-reading Palmer, I see she mentions a number of other things:

In interviews with native speakers [from Costa Rica] in which I included a variety of expressions of emotion, speakers used indicative and subjunctive with Estoy triste que and Me emociona que. The first pair of sentences with estoy triste que are given here:

(1a) Estoy triste que ya no vea (subj) más a mi amigo.
(1b) Estoy triste que ya no veo (indic) más a mi amigo.
“I am sad that I no longer see my friend.”

This difference was commonly described as follows: the sentence with the indicative conveys the finality of no longer seeing the friend (in fact, one participant commented that perhaps the friend had died), while the subjunctive that there is still the possibility of seeing the friend.

[transition sentence]

(2a) Me emociona que todos estén (subj) de acuerdo conmigo.
(2b) Me emociona que todos están (indic) de acuerdo conmigo.
“I am moved that everybody agrees with me.”

The use of the subjunctive in (2b) [sic, meant 2a] seemed to indicate that it was a surprise that everyone was in agreement, but with the indicative the nuance conveyed an expectation that everyone would agree with the speaker. Also interesting was the participant who responded to this sentence in past time: “Me emocionó que todos mis amigos me habían/hubieran dado tantos regalos “I was moved that all of my friends had given me so many gifts,” by saying that the subjunctive indicated that the friends had not really given so many gifts as the speaker had desired.

There is a great deal of data from various studies that shows the use of both moods with expressions of emotion. Moreover, there is no research that proves the prescriptive rule, that is, that the subjunctive is the only mood used with expressions of emotion. It is only within the prescriptive grammar that this claim is made.

As such, it doesn't really make sense to me for a single, unitary operator to have a "prototypical meaning" in some syntactic contexts, but a completely different meaning (that's in fact completely the opposite of the proposed "prototypical meaning") in other contexts. That's not vagueness, or (discourse) context-sensitivity—that's accidental homophony.

Why not? Natlangs extend the meanings that forms represent that way. I don't understand why you and Portner (2018) seem to be trying to find a single definition of "subjunctive" or "irrealis" that may work across languages morphosyntactically, so that when a language like Spanish sometimes expresses (100% certain) factives with the supposedly irrealis form, then the form can't be irrealis but something else, or maybe exhibits homophony. "Unreal" is a fine semantic category, but once you try to peg it to morphosyntax you meet things like the Spanish subjunctive.

(By the way, the RAE in its handbook grammar calls the subjunctive the "non-assertive" mood, if it helps your purpose. Even in el que la policía los haya interrogado tiene lógica, in which I can't switch in an indicative, the subjunctive expresses high but not necessarily full assertive confidence, say 85%~100%, at least in the Spanish I have in my head.)

Relatedly, the Standard Arabic verb tense-aspect-mood form called the "perfect" (e.g. kataba) expresses past tense as the main verb of a clause, typically but not necessarily with perfective aspect, but there's a number of constructions where that same verbal form is used to purely express completive aspect (that the action was achieved/accomplished before) regardless of tense. I've seen Arabic grammars (e.g. Badawi et al.'s Arabic: A Comprehensive Grammar (2002)) struggle to argue that it should be called "the perfective", and I don't understand why they want that.

By the way, I'm not sure if every one of my replies to you has been this argumentative, but if it is the case, I'd like to really thank you for the efforts you've been making writing these posts. They're very interesting and helpful.

[priscianic]

Ser has basically covered what I'd say about subjunctives. As Ser says, it's hard to disentangle "what the subjunctive means" from "what it's come to be used for after thousands of years of development". The prototypical meaning is (I think) still irrealis, but it's not exactly unheard-of for meanings to ramify and even reverse.

I'm not sure we have the same notion of meaning here, if I'm understanding you correctly. Following the norm in formal semantics, I'm understanding "meaning" to be roughly truth-conditional, e.g. what a sentence "means" is a set of conditions that a particular world/situation has to meet in order for that sentence to count as true (plus some restrictions on the discourse context, like presuppositions), and I'm also understanding the "meaning" of a sentence to be compositionally built up from from the meanings of its parts, in a way that we can explicitly formalize (how else can we understand and interpret novel sentences?).

With "prototypical" I'm referring to prototype theory, well explained in Lakoff's Women, Fire, and Dangerous Things. I think your statement corresponds to what Lakoff calls the "objectivist model" (no relation to the political sense), which he spends a good deal of time refuting. But the real problem comes here:

As such, it doesn't really make sense to me for a single, unitary operator to have a "prototypical meaning" in some syntactic contexts, but a completely different meaning (that's in fact completely the opposite of the proposed "prototypical meaning") in other contexts. That's not vagueness, or (discourse) context-sensitivity—that's accidental homophony.

The problem, I think, is assuming that the Romance subjunctive is a "single unitary operator". Why would it be? At the least it's a radial category, as most common words are. There is absolutely nothing surprising about a word having contradictory meanings, especially given a long historical development. E.g. English "with", which once had the meaning "against", still preserved in certain contexts: "fight with", "contrast with".

As I mentioned earlier, "irrealis" doesn't have a single meaning either, though it has a prototypical one. The fact that we can't find a single meaning doesn't show that there is no meaning; it shows that we're dealing with a radial category, or something even more complicated.

I buy that some notion of prototype/"concept"/radial category has to be invoked to account for lexical meaning, but I'm not so convinced that we need those tools for analyzing the meaning of functional items, and I think it's a worthwhile project to figure out if it is in fact possible to give these functional items unitary denotations. If it so happens that we end up showing that this is fundamentally impossible, great! That's a beautiful result—but I'm not fully convinced of that result just yet.

I suspect that this is a foundational philosophical issue about our theoretical commitments that won't get resolved any time soon and probably won't lead to too much useful conversation, so I'm going to stop here.

(I think the paper is quite well-written and beautiful, actually, and I think the core insight is quite right. A lot of people have noted that an assertion of must p feels intuitively "weaker", in some sense, than a plain assertion of p. However, by standard principles of modal logic, if must is a necessity modal, then must p should entail p, which feels like the wrong result. As such, lots of people have argued that must actually isn't a necessity modal. Von Fintel and Gillies argue that people are being too hasty here. They argue that must isn't actually logically weak, and that the inference from must p to p should be considered valid, and that the perceived "weakness" of must is not due to any logical weakness but instead due to an indirect evidential component, and they provide a formal characterization of what exactly this "evidential component" is.)

OK, I've read the paper (admittedly, rather quickly). I quite agree that "must" is associated with indirect evidence; that's pretty much what I meant by calling it deductive.

However, I disagree with their statement that "must is never weak". They're cherry-picking examples like logic puzzles, but really, have they never read a detective story? Consider a story that includes this exchange:

Lestrade: We posted agents at every airport; no one was observed fitting the suspect's description.
Watson: There's no other way to come to an island. He must have come by sea.
Holmes: No, gentlemen. He did not arrive in England, because he was here all along.

This is what the paper denies is possible (p. 19): someone directly contradicting a deduction, in a way that shows that a deduction is weaker than direct knowledge. They try to make it sound no one could felicitously deny a deduction, but one can.

The earlier authors they quote have it right, I think: must(p) is weaker than p. In an ideal world perhaps it wouldn't be, because deduction would never be wrong, and people would not make claims to have deduced the truth when there are loopholes they've overlooked. But in ours, deduction is frequently wrong, loopholes cannot be avoided, and direct knowledge is better than indirect.

(Pragmatically, a statement that P is of course itself a claim. Just because we see it's raining doesn't prove that it's raining. There are always weird Gettier counter-examples to any claim. But the issue isn't whether a statement P is always true, only whether it's a stronger claim than must(P).)

I think you're misunderstanding what "weak" means here—it means logical weakness (i.e. denying that must p entails p), and not "assertive" weakness. It's pretty obvious that "must" is "assertively weak": as you note, must p feels like a "weaker claim" that p. But what's the source of this perceived weakness? One answer is that this is because it's logically weaker than p, and another answer is that this is because must carries an evidential signal.

I'm not sure what your detective example is supposed to show. No one is saying another speaker can't deny a must claim. People deny each other's assertions all the time. If you take out the must in your example, it's still a coherent discourse:

Lestrade: We posted agents at every airport; no one was observed fitting the suspect's description.
Watson: There's no other way to come to an island. He came by sea.
Holmes: No, gentlemen. He did not arrive in England, because he was here all along.

The relevant data would be a speaker trying to assent to not p while at the same time committing themselves to must p. Frr instance, von Fintel and Gillies' example (19):

(19) a. Alex: It must be raining.
b. Billy: [Opens curtains] No it isn’t. You were wrong.
c. Alex: #I was not! Look, I didn’t say it was raining. I only said it must be raining. Stop picking on me!

Or even simpler, Moore's paradox-esque examples like their (16):

(16) a. #It must be raining but perhaps it isn’t raining.
b. #Perhaps it isn’t raining but it must be.

These examples indicate that must p commits the speaker to p in a way that's not immediately deniable (16), and in a way that can't be used to "backtrack" or cover your ass (19). Von Fintel and Gillies argue that examples like these show us that when a speaker commits to must p, they also force themselves to commit to p. A natural explanation for this is if we say that must p entails p—that straightforwardly derives the contradictory feeling of (19) and (16). And if they're right about how the indirect evidential signal works, then you don't need to appeal to logical weakness of this sort to explain the perceived "assertive weakness" of a must claim.

However, if you follow the "mantra", and deny that must p entails p, then you're forced to come up with some other mechanism to derive this kind of data. That might be a solvable problem, given a clever-enough person, but then what's the benefit over the logically strong + indirect evidence kind of position that von Fintel and Gillies espouse?

I should note, there are evidential systems, e.g. Central Pomo, that contrast visual, aural, and deductive evidentials (and more). Would you really maintain that the deductive evidential is just as strong as the visual? Would you predict that, in a context where a speaker could use any of these, the deductive evidential would be most likely to be used or the most likely to be believed?

I think there's reason to believe that not all evidentials behave like logically-strong epistemic necessity modals. For instance, it's possible to felicitously and non-contradictorily assert "EVID p, but not p" for certain evidentials in some languages.

This pattern is particularly common for (some but not all!) reportative evidentials: here's an example from Estonian from AnderBois (2014:238):

Code: Select all

Ta küll   ole-vat aus    mees, aga ta ei  ole üldse  aus.
he surely be -REP honest man   but he NEG be  at.all honest
‘It's certainly been said that he is an honest man, but he's not honest at all’

Not all reportatives behave this way. In St’at’imcets, for example, you can't deny a reportative evidential claim (Matthewson, Davis, and Rullman 2007):

Code: Select all

um’ -en -tsal  -itás   ku7 i      án’was-a    xetspqíqen’ksi táola,...
give-DIR-1s.OBJ-3p.ERG REP DET.PL two   -EXIS hundred        dollar 
"(Reportedly,) they gave me $200,..."

...#t’u7 aoz kw  s  -7um’-en -tsál  -itas   ku  stam’
    but  NEG DET NOM-give-DIR-1s.OBJ-3p.ERG DET what
"...#but they didn't give me anything."

Again, logical strength (i.e. EVID p entailing p) is different from felt "assertive" strength, and is also distinct from deniability by a third party. Logical strength of an evidential would predict the data we get for St’at’imcets reportative ku7 and English must, and it doesn't predict the pattern we get with Estonian -vat (so -vat is very plausibly not logically strong).

[priscianic]

So although I don't accept these two sentences, I do perfectly accept:

Me alegra que haya llegado Juan, aunque no creo totalmente que lo haya hecho.
'I'm glad that Juan arrived, although I don't completely believe he did so.'

Lamento que María esté enferma, si lo que me dices es verdad.
'I feel sad that Maria is sick, if what you tell me is true.'

Me alegra que haya llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
'I'm glad that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'

Lamento que María esté enferma, asumiendo que no te has equivocado.
'I feel sad that Maria is sick, assuming you didn't make a mistake / get confused.'

That's interesting! It would also be worthwhile to compare those sentences to corresponding examples with indicative:

1. Me alegra que ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I'm glad that Juan arrived, although I don't completely believe he did so.'
2. Lamento que María esté enferma, si lo que me dices es verdad.
   'I feel sad that Maria is sick, if what you tell me is true.'
3. Me alegra que haya llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I'm glad that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. Lamento que María esté enferma, asumiendo que no te has equivocado.
   'I feel sad that Maria is sick, assuming you didn't make a mistake / get confused.'

As well as corresponding examples without the emotive predicates:

1. Pienso que ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I think that Juan arrived, although I don't completely believe he did so.'
2. Pienso que María está enferma, si lo que me dices es verdad.
   'I think that Maria is sick, if what you tell me is true.'
3. Pienso que haya llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I think that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. Pienso que María esté enferma, asumiendo que no te has equivocado.
   'I think that Maria is sick, assuming you didn't make a mistake / get confused.'

To ensure that this is really about the subjunctive in these contexts specifically.

(I'm also not assenting to the prescriptive rule that only the subjunctive can appear in these contexts—just to the empirical fact that it can. And these particular contexts have been described as "factive" in the literature (i.e. carrying a truth presupposition), so it's interesting for people building a theory of the subjunctive. And it's also interesting and quite relevant if they aren't actually factive, or if they're "weakly factive", or something like that.)

As such, it doesn't really make sense to me for a single, unitary operator to have a "prototypical meaning" in some syntactic contexts, but a completely different meaning (that's in fact completely the opposite of the proposed "prototypical meaning") in other contexts. That's not vagueness, or (discourse) context-sensitivity—that's accidental homophony.

Why not? Natlangs extend the meanings that forms represent that way. I don't understand why you and Portner (2018) seem to be trying to find a single definition of "subjunctive" or "irrealis" that may work across languages morphosyntactically, so that when a language like Spanish sometimes expresses (100% certain) factives with the supposedly irrealis form, then the form can't be irrealis but something else, or maybe exhibits homophony. "Unreal" is a fine semantic category, but once you try to peg it to morphosyntax you meet things like the Spanish subjunctive.

Again, as in my response to zompist, I'm not so certain that we really want to so readily capitulate to proposing a prototype/radial category kind of analysis for the meaning of functional items, just because the data is sometimes difficult (intuitively, the distribution of most functional items seems to be so regularly and (relatively) easily generalizable that a uniform analysis seems like a reasonable hypothesis to pursue). But I suspect that there are some deeper philosophical divides between us here, and I'm not particularly keen to devote my time to exploring them here (even though it might well be worthwhile). I hope you can forgive me.

By the way, I'm not sure if every one of my replies to you has been this argumentative, but if it is the case, I'd like to really thank you for the efforts you've been making writing these posts. They're very interesting and helpful.

Oh, no worries! You haven't been (overly-)argumentative at all! I really appreciate this. (I'm a conlanger-turned-formal linguist of the (shock and horror) Chomskyan + formal semantic persuasion, and I'm well aware of the wider conlanging community's attitudes towards these things (I used to share those same attitudes), so I'm wading into these waters knowing what lurks beneath :p)

[Kuchigakatai]

That's interesting! It would also be worthwhile to compare those sentences to corresponding examples with indicative:

1. Me alegra que ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I'm glad that Juan arrived, although I don't completely believe he did so.'
2. Lamento que María está enferma, si lo que me dices es verdad.
   'I feel sad that Maria is sick, if what you tell me is true.'
3. Me alegra que ha llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I'm glad that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. Lamento que María está enferma, asumiendo que no te has equivocado.
   'I feel sad that Maria is sick, assuming you didn't make a mistake / get confused.'

(I changed examples 2-4 to the indicative.) With the indicative in each first clause, I find these sentences very awkward unless they're meant as antithetical jokes ("Stalin was the greatest man on Earth, except he actually wasn't").

As well as corresponding examples without the emotive predicates:

1. Pienso que ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I think that Juan arrived, although I don't completely believe he did so.'
2. Pienso que María está enferma, si lo que me dices es verdad.
   'I think that Maria is sick, if what you tell me is true.'
3. Pienso que haya llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I think that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. Pienso que María esté enferma, asumiendo que no te has equivocado.
   'I think that Maria is sick, assuming you didn't make a mistake / get confused.'

(I see examples 1-2 are in the indicative but 3-4 in the subjunctive, so I assume you mean both.) Here, I find the subjunctive entirely ungrammatical after pienso que --it's just bad syntax. With the indicative, I accept examples #1 and #3 (ha llegado x2), but examples #2 and #4 (está x2) seem semantically non-felicitous (although grammatically fine, like "Colourless green ideas...") because the speaker seems to be saying they thought María was sick on no basis until they heard the listener mention it (if what you tell me is true; assuming you aren't wrong). They would be fine if I add something like ahora 'now':

Ahora pienso que María está enferma, si lo que me dices es verdad.
'I am now thinking that Maria is sick, if what you tell me is true.'

Ahora pienso que María está enferma, asumiendo que no te has equivocado.
'I am now thinking that Maria is sick, assuming you aren't wrong.'

But again, this works because pienso can perfectly express non-confident thought. When the speaker says "I think that Juan arrived, although I don't completely believe he did so', he/she is saying he/she thinks that with 60% confidence or maybe 80% or something like that. I don't accept the subjunctive there anyway.



As an aside, regarding the Chomskian proverbial sentence "Colourless green ideas sleep furiously", I once read a story in some blog (no idea which, but probably not LanguageLog) about a linguist travelling by train in the US through the countryside when he saw a big "green tech" advertisement that said "GREEN IDEAS". The linguist was of course very excited to see this, and started talking to the woman next to him, a total stranger, who had also seen the advertisement, to try to say the Chomskian sentence in context, in a damn real-world setting. It seemed to him that he had managed to include "colourless" by using it in the metaphorical sense of "uninspired, boring", therefore turning the use of "green" into irony. But the moment he tried to to use "sleep", the woman asked him what he was thinking he was doing exactly.

It's a very amusing story though, if only because Chomsky in 1955, when he made the sentence, likely didn't foresee the environmentalist movement... Maybe one day there will be metaphorical uses of "colourless" and "sleep furiously" that will allow the sentence to be said in context in a real-world setting...

(I'm also not assenting to the prescriptive rule that only the subjunctive can appear in these contexts—just to the empirical fact that it can. And these particular contexts have been described as "factive" in the literature (i.e. carrying a truth presupposition), so it's interesting for people building a theory of the subjunctive. And it's also interesting and quite relevant if they aren't actually factive, or if they're "weakly factive", or something like that.)

Yeah, that's right. My mention of the bogus rule (that Marathon Ramsey may or may not have made up) was almost an aside, hence why I originally put it in parentheses.

Actually, although it may be interesting for other reasons, I think this phenomenon is probably not all that relevant for people like you with an anti-radial philosophy. It probably doesn't matter if these subjunctives with verbs of emotion aren't actually factive but "weakly factive" by default (and furthermore not factive at all once you start throwing in adverbs like quizá 'maybe': Me alegra que Juan quizá haya venido), because the important thing is that the subjunctive can potentially be used to describe an event the speaker thinks happened with 100% confidence, much in the same way it is valid to use the subjunctive el que la policía haya llegado tiene lógica or cuando vengas, lo terminaremos when the speaker is 100% confident of the past (haya llegado) or future (vengas). Regardless, you still have to contend with these possibly-fully-factive subjunctives.

(I'm a conlanger-turned-formal linguist of the (shock and horror) Chomskyan + formal semantic persuasion, and I'm well aware of the wider conlanging community's attitudes towards these things (I used to share those same attitudes), so I'm wading into these waters knowing what lurks beneath :p)

I see. I don't find much shock and horror in formal syntax tbh. When this discussion began partly from a post of mine, I wasn't questioning the value of formal syntax, but it's true I've only arrived at a higher opinion of it recently in the past three years or so. I was saying that I didn't understand why (English-speaking, formal, post-Chomskian) syntacticians stuck to frameworks, while complaining that many people who like linguistics and maybe conlanging seem to have the wrong idea about what syntax is, and complaining that Syntax courses at English-speaking universities don't include the functional or more informal "framework" for discussing syntax used in so much of the descriptive and typological literature, not even in passing.

E.g. the Wikipedia article on "Syntax" ideally wouldn't begin with "syntax [...] is the set of rules, principles, and processes that govern the structure of sentences (sentence structure) in a given language, usually including word order", having a dedicated section to S-V-O order afterwards and no other specific topic in particular. Way too many people in conlangery at large at least seem to think syntax is word order plus esoteric frameworks and trees. The long list of terms at the end of the article includes things like "gapping" and "subordination", but they're hardly in focus...

Maybe I should edit that Wikipedia article and get rid of its focus on word order, but the thing is it also reflects a widespread attitude among language hobbyists (language learners, conlangers, people who like linguistics but are not academics) towards formal syntax.

[zompist]

The relevant data would be a speaker trying to assent to not p while at the same time committing themselves to must p. Frr instance, von Fintel and Gillies' example (19):

(19) a. Alex: It must be raining.
b. Billy: [Opens curtains] No it isn’t. You were wrong.
c. Alex: #I was not! Look, I didn’t say it was raining. I only said it must be raining. Stop picking on me!

This is where they lost me, so I want to look at it a little more closely. A marking of infelicity is a claim that something is wrong, but what? You say they've proved that you can't assert ¬p and must(p). I say they're trying to argue from a failure of imagination, and that's never a proof of anything. They make Alex talk in a weird way, but don't think about the many ways it's weird (e.g. Alex's wackazoid defensiveness). They're making fun of what they mockingly call the Mantra, but ridicule is not a proof. Quite the opposite, it seems like a defense mechanism.

Alex could have made her point quite reasonably, if pedantically:

Alex: I wasn't wrong. When I said it must be raining, I meant it was a clear deduction from the facts, not that I had seen the rain.

(16) a. #It must be raining but perhaps it isn’t raining.
b. #Perhaps it isn’t raining but it must be.

I'm a little baffled that you think these are identical to

16' a. It's raining, but perhaps it isn't raining.
b. Perhaps it isn't raining, but it's raining.

None of these are impossible, but to me 16' are more anomalous than 16.

Now admittedly the space between must(p) and p is small. Usually we trust our own deductions and therefore think we've actually arrived at p.

But it's common enough that we don't trust our deductions, or we recognize that deduction is not foolproof. That's why "must" is not just an assertion, but also a hedge.

To really explore that small space, you also have to get away from that small class of things that are immediately verifiable. I'd put it to you that statements like 16a sound much more plausible in sentences like this:

Gravity must be mediated by a graviton, though we haven't discovered one yet. But perhaps it isn't.
Life must exist in the Tau Ceti system... but perhaps it doesn't.
God must be about to punish this city full of sin, but perhaps he's not.

These amount to saying "What I know requires a conclusion p. But these are things that are hard to know and I recognize or fear that ¬p is after all not ruled out." They have an air of anxiety, because normally we don't like to confront the fact that our reasoning and beliefs are shaky. But they're certainly not meaningless or impossible.

In my Holmes example, Watson presumably thinks his logic is airtight. But one can easily imagine him reflecting that he's been wrong about such things before: "He must have come by sea.... unless you can think of another possibility?" The problem with the Alex example is not that Alex is entertaining must(p) and ¬p simultaneously, it's that she is acting as if must(p) is not a strong claim at all. Of course it's a strong claim. It's just weaker than P.

You might enjoy, or be troubled by, McCawley's Everything that Linguists have Always Wanted to Know about Logic. It's a book on logic that also emphasizes where language does not work like logic. I think the crux of the disagreement here is that you want "must" to work like a logical operator, and it isn't that simple. Or to put it more palatably, "must" is an asssertion that we have deductive grounds for our belief (or "indirect evidence"). And people can question their own beliefs or the grounds for their beliefs, even if logicians struggle to do so in logic.

Edit: I should add that McCawley has a discussion of modality and possible worlds. I would have to re-read it pretty closely to decide if the approaches presented make sense. So I am not making any strong claims about logic.

[Kuchigakatai]

On a totally unrelated note, etaoin shrdlu was a string of letters used in older printing to fill out lines with a typo, pretty much equivalent to our <asdfjkl;> today. I also can hardly resist reading the first word as if it were Irish, something like [əˈtˠiːnʲ].


Germanic languages apparently have a tendency to not allow combinations of the article, demonstratives and possessive determiners before a noun (*the my book, *a my book, *my this book). But in Romance languages you find attestations of this sort of thing:
- Italian il suo piede 'his/her foot'
- Spanish (a bit archaizing in some dialects but natural in others) esta su muñeca 'this doll of hers', esa su muñeca 'that doll of hers'
- Salvadoran Spanish una su muñeca 'a certain doll of hers, a doll of hers'
- Portuguese qual a cor? 'which colour?'

Or with universal quantifiers:
- French tous les livres 'all books, all the books, every book'
- Portuguese ambos os lados 'both sides' (interestingly considered "vulgar" in some Brazilian dialects, but always prestigious in Portugal!)
- Old Gallo-Romance cadhuna cosa 'each thing' < κατὰ ūnam causam
- Middle French chacune chose 'each thing' < quamque ūnam causam
- Spanish alguna silla 'some chair or other' < aliquam ūnam sellam

[Pabappa]

couldve sworn i'd seen tire con las ambas manos on bathroom paper towel dispensers, but it seems i misread it because the French phrase tirez avec les deux mains is often right above the Spanish. So in French, "les deux" + body part can presumably mean "both (your) ___". That reminds me of both the "my today's breakfast" thing i posted on the other board and the "your both hands" which I posted on wiktionary (or was that here too?) It seems there are quite a lot of options available and that just knowing the basics of grammar wont always tell you the rules for highly specific constructions like this.

[priscianic]

That's interesting! It would also be worthwhile to compare those sentences to corresponding examples with indicative:

1. Me alegra que ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I'm glad that Juan arrived, although I don't completely believe he did so.'
2. Lamento que María está enferma, si lo que me dices es verdad.
   'I feel sad that Maria is sick, if what you tell me is true.'
3. Me alegra que ha llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I'm glad that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. Lamento que María está enferma, asumiendo que no te has equivocado.
   'I feel sad that Maria is sick, assuming you didn't make a mistake / get confused.'

(I changed examples 2-4 to the indicative.) With the indicative in each first clause, I find these sentences very awkward unless they're meant as antithetical jokes ("Stalin was the greatest man on Earth, except he actually wasn't").

As well as corresponding examples without the emotive predicates:

1. Pienso que ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I think that Juan arrived, although I don't completely believe he did so.'
2. Pienso que María está enferma, si lo que me dices es verdad.
   'I think that Maria is sick, if what you tell me is true.'
3. Pienso que haya llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I think that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. Pienso que María esté enferma, asumiendo que no te has equivocado.
   'I think that Maria is sick, assuming you didn't make a mistake / get confused.'

(I see examples 1-2 are in the indicative but 3-4 in the subjunctive, so I assume you mean both.) Here, I find the subjunctive entirely ungrammatical after pienso que --it's just bad syntax. With the indicative, I accept examples #1 and #3 (ha llegado x2), but examples #2 and #4 (está x2) seem semantically non-felicitous (although grammatically fine, like "Colourless green ideas...") because the speaker seems to be saying they thought María was sick on no basis until they heard the listener mention it (if what you tell me is true; assuming you aren't wrong). They would be fine if I add something like ahora 'now':

Ahora pienso que María está enferma, si lo que me dices es verdad.
'I am now thinking that Maria is sick, if what you tell me is true.'

Ahora pienso que María está enferma, asumiendo que no te has equivocado.
'I am now thinking that Maria is sick, assuming you aren't wrong.'

But again, this works because pienso can perfectly express non-confident thought. When the speaker says "I think that Juan arrived, although I don't completely believe he did so', he/she is saying he/she thinks that with 60% confidence or maybe 80% or something like that. I don't accept the subjunctive there anyway.

(Sorry, me forgetting to remove the subjunctives was a mistake in my editing! They're all supposed to be indicative.)

Really cool stuff, thanks for the judgments! The reason I stuck to having pensar in the second set of examples is because I was trying to keep an attitude predicate in there, to be as maximally similar to the original sentences. It's probably also worth checking these sentences with just plain assertions, i.e.:

1. Ha llegado Juan, aunque no creo totalmente que lo haya hecho.
   'I think that Juan arrived, although I don't completely believe he did so.'
2. María está enferma, si lo que me dices es verdad.
   'I think that Maria is sick, if what you tell me is true.'
3. Ha llegado Juan, pero me cuesta creerlo, al menos hasta que lo vea.
   'I think that Juan arrived, but I'm having a hard time believing it, at least until I see him (in front of me).'
4. María está enferma, asumiendo que no te has equivocado.
   'I think that Maria is sick, assuming you didn't make a mistake / get confused.'

The contrast between subjunctive and indicative and subjunctive in the original examples seems clear though, and that's super interesting! Thank you!

E.g. the Wikipedia article on "Syntax" ideally wouldn't begin with "syntax [...] is the set of rules, principles, and processes that govern the structure of sentences (sentence structure) in a given language, usually including word order", having a dedicated section to S-V-O order afterwards and no other specific topic in particular. Way too many people in conlangery at large at least seem to think syntax is word order plus esoteric frameworks and trees. The long list of terms at the end of the article includes things like "gapping" and "subordination", but they're hardly in focus...

Maybe I should edit that Wikipedia article and get rid of its focus on word order, but the thing is it also reflects a widespread attitude among language hobbyists (language learners, conlangers, people who like linguistics but are not academics) towards formal syntax.

Yeah, this is one of my personal bugbears about conlangers and syntax—they think it's all about word order, but we actually don't really think about word order all that much...

[priscianic]

The relevant data would be a speaker trying to assent to not p while at the same time committing themselves to must p. Frr instance, von Fintel and Gillies' example (19):

(19) a. Alex: It must be raining.
b. Billy: [Opens curtains] No it isn’t. You were wrong.
c. Alex: #I was not! Look, I didn’t say it was raining. I only said it must be raining. Stop picking on me!

This is where they lost me, so I want to look at it a little more closely. A marking of infelicity is a claim that something is wrong, but what? You say they've proved that you can't assert ¬p and must(p). I say they're trying to argue from a failure of imagination, and that's never a proof of anything. They make Alex talk in a weird way, but don't think about the many ways it's weird (e.g. Alex's wackazoid defensiveness). They're making fun of what they mockingly call the Mantra, but ridicule is not a proof. Quite the opposite, it seems like a defense mechanism.

Alex could have made her point quite reasonably, if pedantically:

Alex: I wasn't wrong. When I said it must be raining, I meant it was a clear deduction from the facts, not that I had seen the rain.

I think your version of the continuation still sounds quite strange when you put it in context…

(16) a. #It must be raining but perhaps it isn’t raining.
b. #Perhaps it isn’t raining but it must be.

I'm a little baffled that you think these are identical to

16' a. It's raining, but perhaps it isn't raining.
b. Perhaps it isn't raining, but it's raining.

None of these are impossible, but to me 16' are more anomalous than 16.

And these also sound quite strange to me...

For what it's worth, the standard judgment in the literature is that sentences like those are infelicitous—they're known as "epistemic contradictions" (a term which I think goes back to Yalcin 2007), and deriving their infelicity is quite difficult given certain kinds of standard assumptions about the semantics of modals and the pragmatics of assertion. One of Yalcin's points is that, while these epistemic contradictions seem to derive from the same source as Moorean paradoxes (e.g. #It's raining, but I believe it isn't raining), the source of the infelicity can't be the same because some contexts fix Moorean paradoxes but don't fix epistemic contradictions: e.g. Suppose it's raining but I don't believe it's raining… seems like a reasonable thing to say, but #Suppose it's raining but it might not be raining is must stranger. Another puzzle about epistemic contradictions is that, as Beddor and Goldstein (2018) point out, it's perfectly reasonable to personally hold a belief in p, while also acknowledging that that belief might be mistaken and that it might be the case that not p. But asserting both of these things at once, in the same breath, ends up sounding odd (to most people, I guess).

To really explore that small space, you also have to get away from that small class of things that are immediately verifiable. I'd put it to you that statements like 16a sound much more plausible in sentences like this:

Gravity must be mediated by a graviton, though we haven't discovered one yet. But perhaps it isn't.
Life must exist in the Tau Ceti system... but perhaps it doesn't.
God must be about to punish this city full of sin, but perhaps he's not.

These amount to saying "What I know requires a conclusion p. But these are things that are hard to know and I recognize or fear that ¬p is after all not ruled out." They have an air of anxiety, because normally we don't like to confront the fact that our reasoning and beliefs are shaky. But they're certainly not meaningless or impossible.

Yeah, this kind of data is one of the big arguments for the "must is weak" camp—e.g. Lassiter (2016), who argues for a probablistic semantics for must. As you note, it's quite tricky to deal with this for the "must is strong" camp, and you're forced to say something that perhaps isn't so satisfactory. Von Fintel and Gillies (2020) try to address this kind of data, and argue that it involves a change in context—in particular, a change in which possibilities are deemed relevant at a given point in time (in their words, you're "expanding the modal horizon"). Understandably, if you allow the context to so rapidly change like this, you worry that your theory loses any bite at all—or as they put it, you worry about whether they're putting their theory of "pragmatic life support" to shield it from any counterexamples. So they try to support this idea by trying to control for shifts in the "modal horizon" (though some of their data is, as they admit, somewhat shaky).

In my Holmes example, Watson presumably thinks his logic is airtight. But one can easily imagine him reflecting that he's been wrong about such things before: "He must have come by sea.... unless you can think of another possibility?"

Again, here I think the corresponding must-less assertion in a same kind of detective deduction context is just as felicitous, so this isn't an argument against must p entailing p: "He came by sea…unless you can think of another possibility?". (The example gets even better if you add something like "Based on the evidence we have" at the beginning. I don't think this improvement is necessarily an argument against must's strength. Mandelkern (2019) argues that must p is only felicitously assertable if there's an accessible shared argument for p in the context. Of course, plain assertions aren't subject to the same restriction, so you'd ideally control for the context such that there is an accessible shared argument floating around—e.g. by making sure we're in a detective context, or by adding in something like "based on the evidence we have", etc.

I think the crux of the disagreement here is that you want "must" to work like a logical operator, and it isn't that simple. Or to put it more palatably, "must" is an asssertion that we have deductive grounds for our belief (or "indirect evidence"). And people can question their own beliefs or the grounds for their beliefs, even if logicians struggle to do so in logic.

I'm not sure if I understand what the disagreement is then—neither I nor von Fintel and Gillies (nor Lassiter, nor Mandelkern, etc.) are challenging the idea that must indicates deductive grounds/indirect evidence for our belief---I think everyone (now) acknowledges that this is probably true. But the question that people are arguing about in the literature is whether must p entails p, or if it doesn't.

This is also a separate question from the assertability of must p and p. Assertaibility is context-dependent—I might have a certain amount of strength in my belief that p, and that might allow me to assert p in one context but not another. For instance, if we're detectives and we're discussing what we think of the case, I might be able to assert "the suspect came by sea" without being completely sure of it. But if I'm in court, I might not want to straight-out assert that, because I'm not certain enough about it. I might want to hedge, or I might want to explicitly note that I'm making a deduction (e.g. by saying "Based on the evidence I have, the suspect came by sea", or "The suspect must have come by sea"). I don't think anyone is disputing these kinds of facts.

You might enjoy, or be troubled by, McCawley's Everything that Linguists have Always Wanted to Know about Logic. It's a book on logic that also emphasizes where language does not work like logic.
…
Edit: I should add that McCawley has a discussion of modality and possible worlds. I would have to re-read it pretty closely to decide if the approaches presented make sense. So I am not making any strong claims about logic.

I've had this book recommended to me before, but I've never actually read it (unfortunately, it doesn't seem to be obtainable (for free) online, so it might be a while before I get to it…). But I should get on it!

When you say "emphasizes where language does not work like logic", I wonder what kind of logic you mean? I think most semanticists well understand that human language doesn't work on the principles of classical logic—this is why we make use of the tools of modal logic, or sometimes multivalued logics, etc. And it's also pretty clear that the natural language paraphrases we have of certain logical operators (a particularly egregious example is if…then for material implication) don't actually have the same properties as the logical operators they're meant to be paraphrases for—so of course, people have different analyses for these things. In the case of conditionals, a particularly popular one is to say that the if-clause is actually a restrictor of the modal base of a modal (sometimes a covert one)—some of the first proponents of this are Lewis (1975) and Kratzer (1986) (I can't find a version of this online anywhere, unfortunately). The intuition is that the if-clause restricts the modal base to only those possible worlds that verify the if-clause (thinking of it another way, you're just "restricting your attention" to possibilities in which the if-clause holds true).

And of course, acknowledging facts like those isn't as strong of a claim as denying that (certain aspects of) human language can be adequately captured by the tools of any kind of logic; I'm not sure if you're trying to espouse this stronger claim.

[bradrn]

E.g. the Wikipedia article on "Syntax" ideally wouldn't begin with "syntax [...] is the set of rules, principles, and processes that govern the structure of sentences (sentence structure) in a given language, usually including word order", having a dedicated section to S-V-O order afterwards and no other specific topic in particular. Way too many people in conlangery at large at least seem to think syntax is word order plus esoteric frameworks and trees. The long list of terms at the end of the article includes things like "gapping" and "subordination", but they're hardly in focus...

Maybe I should edit that Wikipedia article and get rid of its focus on word order, but the thing is it also reflects a widespread attitude among language hobbyists (language learners, conlangers, people who like linguistics but are not academics) towards formal syntax.

Yeah, this is one of my personal bugbears about conlangers and syntax—they think it's all about word order, but we actually don't really think about word order all that much...

Yet, if I see a ‘syntax’ section in a reference grammar, it’s usually all about word order. So, if that’s not ‘syntax’, then what is syntax actually about, and where does word order fit into it?

To really explore that small space, you also have to get away from that small class of things that are immediately verifiable. I'd put it to you that statements like 16a sound much more plausible in sentences like this:

Gravity must be mediated by a graviton, though we haven't discovered one yet. But perhaps it isn't.
Life must exist in the Tau Ceti system... but perhaps it doesn't.
God must be about to punish this city full of sin, but perhaps he's not.

These amount to saying "What I know requires a conclusion p. But these are things that are hard to know and I recognize or fear that ¬p is after all not ruled out." They have an air of anxiety, because normally we don't like to confront the fact that our reasoning and beliefs are shaky. But they're certainly not meaningless or impossible.

Yeah, this kind of data is one of the big arguments for the "must is weak" camp—e.g. Lassiter (2016), who argues for a probablistic semantics for must. As you note, it's quite tricky to deal with this for the "must is strong" camp, and you're forced to say something that perhaps isn't so satisfactory. Von Fintel and Gillies (2020) try to address this kind of data, and argue that it involves a change in context—in particular, a change in which possibilities are deemed relevant at a given point in time (in their words, you're "expanding the modal horizon"). Understandably, if you allow the context to so rapidly change like this, you worry that your theory loses any bite at all—or as they put it, you worry about whether they're putting their theory of "pragmatic life support" to shield it from any counterexamples. So they try to support this idea by trying to control for shifts in the "modal horizon" (though some of their data is, as they admit, somewhat shaky).

As a non-syntactician, I would think that these sort of sentences would best be analysed by assuming ‘must’ is a radial category. In that case, the prototypical meaning of ‘must’ would be something like ‘X is true in all possible worlds where the premises hold’, but humans have then extended that to just act as a general intensifier. That explanation doesn’t seem at all unreasonable to me (if anything, it seems natural!) — is there any reason it hasn’t been adopted more widely?