Basic Valence Orientation and Sint

[chris_notts]

I've been thinking a lot recently about valency alternations in Sint, and I thought maybe some of this might be interesting to other people. Originally I had planned to have a class of ambitransitive / labile verbs, but the issue was that Sint is head-marking with no role marking of core arguments in the clause, with use of grammatical voice to help with reference tracking. But the morphology would have made the ambi-transitives (roots not lexically specified for number of arguments) ambiguous about the number of arguments, which would mess with the reference tracking, which made me decide that zero marked transitivity alternations shouldn't happen.

Basic Valency Orientation

It has been suggested that languages tend to be consistent in their strategy for transitive/intransitive alternations by having a dominant strategy of either:

1. making the intransitive root basic and using causative morphology to create the transitives, e.g. break.intrans -> break.intrans-CAUS. This is the preferred strategy in Salishan languages.
2. making the transitive root basic and using some kind of anti-causative morphology to derive the intransitive. This is the preferred strategy in Spanish, which has a large class of anti-causatives marked by reflexive clitics.

For some papers along these lines, see the following:

Universals of causative and anticausative verb formation and the spontaneity scale
Causatives and anticausatives
Macroscopic and microscopic typology:Basic Valence Orientation, more pertinacious than meets the naked eye

Salish - a radical example of the causative strategy

I don't speak any Salishan language, but since I decided to go in this direction for Sint, I've been reading quite a lot about Salish voice morphology. Ignoring applicatives for now, the basic facts about Salish are that:

1. The language has a very large class of roots corresponding to transitives in languages like English which, when unmarked, are unaccusative (non-control) intransitive verbs. They describe the impact of the event on the patient without including an agent in their default case frame.

2. Almost all transitives require an overt transitivising suffix, although depending on the language the suffix may merge in some cases with a following agreement marker. The suffix often distinguishes degree of control that the actor has over the event. Causative markers may also vary depending on whether the stem they attach to are unaccusative or unergative/transitive.

See the examples below, transcribed from "The Oxford Handbook of Polysynthesis" in an X-SAMPA-ish convention. Note that I think the English glosses in the table are a bit misleading based on reading elsewhere.

k'w@q = it splits = unaccusative intransitive
k'w@q-?@m = split (something) = active intransitive / antipassive
k'w@q-t = split it = control transitive
k'w@q-@xw = have split it = non-control transitive
k'w@q-?@m-sxw = make him split (something) = active intransitive causative
k'w@q-t-@m = it is split = passive
k'w@q-?@m-t = split it for him = control indirective (benefactive applicative)
k'w@q-?@m-@xw = have split for him = non-control indirective (benefactive applicative)

This pattern is followed very consistently, to a degree that Haspelmath comes close to declaring unattested typologically in the papers above. Salish languages have a more limited repertoire of root transitives than most other languages, although there are some roots which only occur with transitive suffixes and which can't be converted into intransitives by taking the transitive marker away.

Some papers on Salish root semantics and transitivity:

Issues in Salish Syntax and Semantics
Situational Aspect and Viewpoint Aspect: From Salish to Japanese
The Function of Salish Applicatives
A Closer Look at Salish Intransitive/Transitive Alternations

Revised Sint transitivity alternations

So I've decided to follow a similar pattern, but with modifications. Every stem in Sint will be transitive or intransitive, but I'm currently thinking that I won't bother with the redundant transitivity marking on lexically transitive roots. This means there are three basic root categories with fuzzy boundaries between them:

Base transitive = interpreted as having two arguments without further overt marking. Mostly limited to roots where it's hard to conceive of the event just happening without an agent to trigger it
Unergative (+control) intransitives = verbs like run, jump, ... normally performed by a controlling animate argument
Unaccusative (-control) intransitives = verbs like break which can happen without participant control

I'm still adjusting the morpheme shapes, but let's say for now that these can be suffixed with at least the following voice morphemes. I've left out the actual agreement suffixes, so all the examples are stem formation only:

-s- = transitiver [+control, +bivalent]

Adds a patient-like argument to the verb. Only compatible with direct causation, cannot be applied directly to unergative verbs to form a causative:

ràk 'break' -> ràk-s -> ràks 'break (it)'
kaas 'run' -> *kaas-s -> *kaats 'make (it/him/her) run'

When attached to an unergative (+control) intransitive, it can form a benefactive applicative:

kaas 'run' -> kaas-s -> kaats 'run to/for (him/her)'

Causative ~ beneficiary applicative alternations are not that uncommon cross-linguistically based on either transitivity or the unergative/unaccusative split.

-nč- = transitiviser [limited control, +bivalent]

Patterns like -s, but marks limited control of the actor over the process. Can be translated as 'manage to', 'accidentally', .... I'm not sure this is the case in Salish, but I think this should probably required with inanimate agents incapable of intent.

Can also feed the benefactive applicative:

kaas 'run' -> kaas-nč -> kaasınč 'run to/for (him/her)'

The beneficiary meaning probably rules out the 'accidentally' meaning of the limited control transitiviser, since it's hard to see how an action can simultaneously be performed for someone's benefit and be unintended.

-m- = anti-passive / applicative [+control, -bivalent, -affected]

On its own, forms an anti-passive with an agent and an unspecified patient. Tends to have atelic / process semantics:

ràk 'break' -> ràk-m -> ràkım 'break (something)'

After elimination of the original patient to form an unergative intransitive, a transitive suffix -s or -nč can be added to promote a dative or beneficiary argument:

ràk 'break' -> ràk-m-s -> ràkıms 'break (something) for (him/her)'

Note also the interesting fact in the Salish examples above that the same suffix partly codes both anti-passive and beneficiary applicative functions, something I already had in Sint before I started digging into Salish.

-ksi- = reflexive / middle [+control, +affected, +univalent]

The transitive suffix -s isn't required before -ksi-, but -nč-ksi- -> -nčki can be added. Reflexives are by default semantically +control. The suffix can be used to convert unaccusatives to unergatives, if the argument is capable of volition:

kalk '(be/become) big' -> kalk-ksi -> kalkıksi 'enlarge/grow oneself (animate)'

If volition isn't possible, then either the bare-root alone or addition of a preverb creates a telic non-control inchoative.

The reflexive, since it creates unergatives, in turn can feed the benefactive meaning of -s:

kalk-ksi-s -> kalkıksis 'grow up for'

Oblique applicative [+/- control, +bivalent]

Attaches to a stem to demote the existing patient (if any), and give a previous oblique control of patient agreement. Typically used with locations, the stimulus in psych verbs, etc. Is compatible with either transitive or intransitive roots. Doesn't affect the control semantics of the stem. I'm not sure what the exact semantic range of promotable obliques should be, and whether I should split it further by adding an instrumental/comitative applicative.

naa 'go' -> naa-ri -> naari `go to'

Summary of mappings:

-s- / -nč-
UNACCUSATIVE -> TRANSITIVE
UNERGATIVE -> DATIVE/BENEFICIARY APPLICATIVE

-m-
UNACCUSATIVE -> ANTIPASSIVE (UNERGATIVE)
TRANSITIVE -> ANTIPASSIVE (UNERGATIVE)

-ksi-
UNACCUSATIVE -> UNERGATIVE (+ CONTROL)
TRANSITIVE -> REFLEXIVE/UNERGATIVE (+ CONTROL)

-ri-
ANY -> TRANSITIVE WITH PROMOTED LOCATIVE/STIMULUS PATIENT

The combinations required to get the benefactive applicative are then simple to derive from the fact that this reading is only available with -s-/nč- when they are applied to unergative (+control) intransitive stems.

Questions:

1. What is the basic valency orientation of other people's conlangs? Are they causativising, anti-causativising, neutral (lots of ambitransitive verbs), ...?
2. Does anyone feel like adding more detail on Salish voice alternations?
3. Sint questions:
(i) Does this description make sense?
(ii) Any comments / feedback?

[Salmoneus]

I've been thinking a lot recently about valency alternations in Sint, and I thought maybe some of this might be interesting to other people. Originally I had planned to have a class of ambitransitive / labile verbs, but the issue was that Sint is head-marking with no role marking of core arguments in the clause, with use of grammatical voice to help with reference tracking. But the morphology would have made the ambi-transitives (roots not lexically specified for number of arguments) ambiguous about the number of arguments, which would mess with the reference tracking, which made me decide that zero marked transitivity alternations shouldn't happen.

I'm not sure I follow. Exactly what problem were you encountering there?

I guess that your verbs only agree with one argument (since otherwise valency would be pretty easy to spot...). At which point, the presence of a second argument (not marked as an oblique in any way) would demonstrate bivalency of the verb. So is it that you need to distinguish between bivalent verbs with an elided argument (i.e. an argument implicitly imported from another clause) and genuine univalents? One option in that case (not to say it's better than what you've done, but to mention for the sake of others with similar problems) is non-canonical switch reference, which could be used to mark that the verb in a second clause did not share an object with the verb in a prior clause (whether there's then an indefinite object or no object will generally not be important, or even meaningful). Another unusual option is the particles some languages have - sorry I can't be more detailed, though iirc soemthing amazonian? - to mark "AARGH WE'RE MISSING SOMETHING!". These particles can be used to mark that, say, a possessor has not been mentioned, or in this case could be used to mark that an object has not been mentioned (i.e. it can't be coreferential with a previous object).

[both these things are things I want to use in a conlang, and have played with, but that have not ended up in any of my substantial things]

Another strategy is to limit what can be exported from certain clauses. For instance, if you DON'T want an object exported from the first clause into the second, you can use voice to downgrade that object to an oblique, if obliques aren't imported. Or indeed to upgrade that object to a subject, if that can prevent it being imported as an object.

But I think I may have misunderstood you?

[akam chinjir]

Looks to me like what you've got makes sense.

It has been suggested that languages tend to be consistent in their strategy for transitive/intransitive alternations by having a dominant strategy of either:

1. making the intransitive root basic and using causative morphology to create the transitives [...]
2. making the transitive root basic and using some kind of anti-causative morphology to derive the intransitive [...]

Of the papers you linked I only looked at the first Haspelmath, but all he's arguing there is that there's a coherent scale, not that languages tend to group at the two ends of the scale. (And that fits what I remember from previous reading, though I'm pretty sure that was reading of earlier Haspelmath.)

In Akiatu, in the terms of Haspelmath's paper, agentive intransitives will be derived; higher on the spontaneity hierarchy it'll be the causatives that are derived. (There'll be few if verbs that are ambitransitive without other meaning shifts. I do have tawaru meaning sing when transitive and laugh when intransitive, though, and there may be other cases like that.)

The normal way to derive an agentive intransitive is through passivisation:

• ana kwitu be/get broken ← kwitu break (transitive)

In some cases there's another possibility, starting with a stative verb:

• titaika taika become/get broken ← titaika be broken (intransitive, stative)

I've also been thinking of generalising the inchoative use of partial reduplication, and one possibility is to allow forms like this:

• kwitu kwitu become/get broken ← kwitu break (transitive)

As for the causatives, they're all analytic, but there's a distinction between monoclausal and biclausal causatives. Monoclausal causatives are supposed to be direct:

• ahjai rawu satisfy ← rawu be content, satisfied
• ahjai nukiwa make sick ← nukiwa be sick
• hwati piwa make eat ← piwa eat

(ahjai do is used with intransitives, hwati give with transitives.)

Haspelmath implies that my third example here couldn't be right, that you can't have direct causatives with transitive verbs. It could be I'm interpreting the direct/indirect distinction wrong. But here's my thinking.

With experience verbs, at least, I don't see a problem, since the subject isn't an agent:

• hwati kiwaita make see, show ← kiwaita see

With other transitives, I take a direct causative to imply that the causee doesn't exercise independent agency. So, e.g., ahjai piwa make eat is appropriate whenever it's entirely up to the causer whether the person eats or not. (Whereas you'd use an indirect causative if the causer is (just) helping or permitting or commanding the eating rather than being fully responsible for it.) I guess I think causative "make" works this way in English, so I'm not too worried that this is a typological faux pas.

Though: I'm now tempted to give the hwati causative a benefactive use as well---so hwati piwa could mean eat on X's behalf, or for the benefit of X as well as make X eat. Apparently that sort of thing is common, and the use of a verb meaning give sort of invites it. And I quite like it. (Though I'd been intending for Akiatu to have no applicatives, oh well.)

(I just did a post about causatives, here.)

[chris_notts]

I'm not sure I follow. Exactly what problem were you encountering there?

I guess that your verbs only agree with one argument (since otherwise valency would be pretty easy to spot...). At which point, the presence of a second argument (not marked as an oblique in any way) would demonstrate bivalency of the verb. So is it that you need to distinguish between bivalent verbs with an elided argument (i.e. an argument implicitly imported from another clause) and genuine univalents?

The verb agrees with both arguments, but there are two problems.

The first is that that only works if all of the patient agreement morphemes are overt. If one is zero, or there is any form of differential patient agreement (conditioned by animacy, definiteness, ...) then there is ambiguity between the a transitive and intransitive verb. That's an issue for reference tracking and semantics, especially if there are any S=P ambitransitives.

For example, if "lower" could mean both "go lower" and "cause to be lower / put lower":

3A-took-0 and 3A-lower-0
= 3A-took-0 and 3A-lower
= He took it and went down
= He took it and lowered it

I think for S=A ambitransitives this is a bit less of an issue since subjects tend to persist, but even there it potentially introduces confusion, e.g. in long distance movement. Sint allows focal fronting, including extraction from complement clauses, but if object marking can be null then it's not clear how many possible sources for the extracted nominal there are:

3A-have pig and goat, pig 3A-want 3A-eat-PURP(-0)
= He has a pig and a goat, and the pig is hungry / wants to eat
= He has a pig and a goat, and he wants to eat the pig

To raise another issue, I had some multi-functional voice morphemes:

1. a causative/comitative voice marker which marks causation with unaccusative verbs but comitative applicative with transitive/unergative verbs
2. an anti-passive / dative applicative

The problem with (1) is that for any S=P ambitransitives, it's not clear whether the causative or comitative meaning is intended:

break.INTRANS-CAUS-0 = break it (make it break)
break.TRANS-CAUSE-0 = break it with (him/her/it)

If break.INTRANS = break.TRANS then these two have an identical surface form.

The problem with (2) is that, for any zero marked beneficiary, it isn't clear whether they exist or not:

give-ANTI
= give-ANTI-0
= be giving (something)
= give to (him/her/it)

The other possible resolution to this is to modify the agreement pattern so that the agreement markers themselves always make it clear whether the clause is transitive or not, and in cases of defective objects (e.g. unanimate / indefinite / ...) overtly mark the demotion via marker like the anti-passive -m. In the case of the causative morpheme, this would create two covert classes of verbs, those which use the causative to derive transitives, and those which have zero derived transitives and use the "causative" to derive a comitative applicative.

One option in that case (not to say it's better than what you've done, but to mention for the sake of others with similar problems) is non-canonical switch reference, which could be used to mark that the verb in a second clause did not share an object with the verb in a prior clause (whether there's then an indefinite object or no object will generally not be important, or even meaningful). Another unusual option is the particles some languages have - sorry I can't be more detailed, though iirc soemthing amazonian? - to mark "AARGH WE'RE MISSING SOMETHING!". These particles can be used to mark that, say, a possessor has not been mentioned, or in this case could be used to mark that an object has not been mentioned (i.e. it can't be coreferential with a previous object).

I decided to avoid switch reference because it doesn't really fit with the feel of the language. Basically every clause is finite and has a form which could be used as a main-clause, and creating dependent verb forms would be a fundamental change.

I'm not sure I've seen the language with the "AARGH WE'RE MISSING SOMETHING!" particles. I guess this is similar to languages that only show object agreement when an overt object NP isn't present in the clause? I do have in Sint a verbal clitic that marks oblique/possessor fronting and extraction, as well as extraction restrictions for core arguments of transitives related to verbal voice marking.

[chris_notts]

Of the papers you linked I only looked at the first Haspelmath, but all he's arguing there is that there's a coherent scale, not that languages tend to group at the two ends of the scale. (And that fits what I remember from previous reading, though I'm pretty sure that was reading of earlier Haspelmath.)

I think he does discuss that the "break" happens at different points for different languages though. The claim that languages tend towards one end or another comes, I think, from Nichols originally. This paper seems to be one source, but unfortunately it's not free to access so I can't read it:

Transitivizing and detransitivizing languages

Some others that cite that one:

Transitivizing-detransitivizing typology and language family history
Lexical Valence Typology project questionnaire
Do languages have a basic valency?

The normal way to derive an agentive intransitive is through passivisation:

• ana kwitu be/get broken ← kwitu break (transitive)

In some cases there's another possibility, starting with a stative verb:

• titaika taika become/get broken ← titaika be broken (intransitive, stative)

One thing I find interesting is whether languages distinguish between the spontaneous decausative ('break'/'get broken' and the suppressed agent form ('be broken'), and if so how productive this distinction is. I guess in Akiatu this distinction is partly lexicalised, e.g. it exists for break but not for many other transitive verbs?

I've also been thinking of generalising the inchoative use of partial reduplication, and one possibility is to allow forms like this:

• kwitu kwitu become/get broken ← kwitu break (transitive)

Cool. I'm not sure I've seen a language that uses reduplication specifically for valency changes before. But I guess that reduplication is associated with progressive, stative, and maybe irrealis senses, all of which are less prototypically transitive.

As for the causatives, they're all analytic, but there's a distinction between monoclausal and biclausal causatives.

But presumably, based on what you said above, you lean towards roots like "break" being inherently causative, and deriving the intransitive forms? So the analytic causative is for the marked part of Haspelmath's spectrum.

Haspelmath implies that my third example here couldn't be right, that you can't have direct causatives with transitive verbs. It could be I'm interpreting the direct/indirect distinction wrong. But here's my thinking.

I have seen that it's very common for verbs of ingestion (eat) and non-control experience/perception to be exceptions to this general rule. In the case of perception verbs, in many languages they're not formally transitive in any case because of either limited control or limited/no affectedness of the object. Causatives of these verbs are often lexicalised, e.g. see -> show, eat -> feed.

With experience verbs, at least, I don't see a problem, since the subject isn't an agent:

• hwati kiwaita make see, show ← kiwaita see

With other transitives, I take a direct causative to imply that the causee doesn't exercise independent agency. So, e.g., ahjai piwa make eat is appropriate whenever it's entirely up to the causer whether the person eats or not. (Whereas you'd use an indirect causative if the causer is (just) helping or permitting or commanding the eating rather than being fully responsible for it.) I guess I think causative "make" works this way in English, so I'm not too worried that this is a typological faux pas.

This makes sense to me.

Though: I'm now tempted to give the hwati causative a benefactive use as well---so hwati piwa could mean eat on X's behalf, or for the benefit of X as well as make X eat. Apparently that sort of thing is common, and the use of a verb meaning give sort of invites it. And I quite like it. (Though I'd been intending for Akiatu to have no applicatives, oh well.)

In his book on applicatives, Peterson claims that causative ~ benefactive is more common than causative ~ comitative polysemy. I find this a bit surprising, since the causative ~ instrumental ~ comitative polysemy seems more natural to me, especially since many morphological causatives cover permission ("let") as well as forcing ("make"). The only difference between those cases is the degree of agency retained by the original agent, and its animacy:

I cut the fabric
-> I made the knife cut the fabric
-> I cut the fabric with the knife (no significant semantic difference except directness of causation, since an inanimate has no volition anyway)

I made her eat / I fed her
-> I let her eat
-> I ate with her (people normally eat together)

But I think perhaps this is frequency driven. If a language has only one applicative, it's likely to include dative/benefactive promotion as part of its range since datives are highly likely to be animate and topical, and languages like making animate and topical referents core arguments of their verbs. So if there are any circumstances where a causative and benefactive look similar, that may get generalised for functional reasons even if an instrumental applicative is a more semantically direct expansion of a causative morpheme.

[chris_notts]

Turns out the Oceanic languages have reduplicative detransitivisation:

Some canonic languages also have detransitivising morphology, although its role is much less significant than the transitivising and causativising morphology discussed above. It takes three forms. Two are straightforward detransitivisation: reduplication, and prefixation of a reflex of one of Proto Oceanic *ma- and *ta-. The third is prefixation of a reflex of Proto Oceanic *paRi- to form a reciprocal verb.

[Curlyjimsam]

The Sint stuff looks really good; this is an important but often neglected area. (If I were to write some kind of LCK-esque guide, I think valency alternations would be something I'd definitely mention, because it's all too easy just to replicate the English system without realising how other languages do things differently.)

I'm not sure Viksen fits very neatly into the causativising/anti-causativising typology, because it doesn't really have many "unaccusative" verbs - states and changes of state are usually expresse with a modifier + wu "to be" or wi "to become".* So "to cool" (intransitive) is if wi cool become and so forth. But transitive verbs of causation can be formed from modifiers with the very productive suffix -xan, e.g. ifxan "to make cool" = "to cool" (trans.). To further complicate things transitives allow discourse-driven agent omission, which may lead to clauses -xan verbs superficially resembling intransitives. The few actual unaccusative verbs like ká "to burn" (intrans.) tend not to have transitive alternants full stop.

* - technically the same verb, wi is the irregular inceptive aspect form of wu. Their relation is clearer in the spelling than in pronunciation, where they are respectively /ji/ and /wu/.

Another interesting thing to consider is "indefinite object deletion" processes such as English Lucy eats cake vs. Lucy eats. The verbs which allow this in English are fairly few in number and it's not clear to what extent this possibility is systematic. (The classic examples contrast eat, where the object is optional, to devour, where it is obligatory, and dine, where no object is possible at all.) It's all too easy to replicate this in a conlang, and it's better to consider other possibilities, e.g. this sort of detransitivisation is possible with all transitive verbs of a given class, or it doesn't happen at all, or it involves particular "add/subtract an object" morphology (which might be distinct from the "add/subtract an agent" morphology), or whatever.

(Viksen handles this sort of thing via antipassive, which is marked with a particle k and is possible with any basically transitive verb.)

[Whimemsz]

I'm not sure I've seen the language with the "AARGH WE'RE MISSING SOMETHING!" particles. I guess this is similar to languages that only show object agreement when an overt object NP isn't present in the clause? I do have in Sint a verbal clitic that marks oblique/possessor fronting and extraction, as well as extraction restrictions for core arguments of transitives related to verbal voice marking.

There's something mentioned in Dixon and Aikhenvald's The Amazonian Languages that might be what Sal's thinking of? In Aryon Rodrigues' chapter on Macro-Jê he writes:

A pervasive inflectional devise in Macro-Jê languages is the marking of the head of a nominal, verbal, or postpositional phrase for textual contiguity (CNT) or non-contiguity (NCNT) of the determiner (or dependent), as in examples (1), (2), and (3) from Panará ..., in (4) and (5) from Timbíra ..., and in (6) and (7) from Ofayé .... When the possessor is stated we find the CNT marker, and when it is absent the NCNT marker.

(1) sɔti j-akoa
animal CNT-mouth
'the mouth of the animal'

(2) sɔti j-õtɔ
animal CNT-tongue
'the tongue of the animal'

(3) s-õtɔ s-akoa amã
NCNT-tongue NCNT-mouth in
'Its tongue is in its mouth'

(4) i tɛ pĩ.co j-ũʔkʰər
1sg ERG.PAST tree.fruit CNT-buy
'I bought fruit'

(5) i tɛ h-ũʔkʰər
1sg ERG.PAST NCNT-buy
'I bought it'

(6) pɨkɨtɨɛn ʃ-ɛnʃɨh
caiman CNT-heart
'the heart of the caiman'

(7) h-ɛnʃɨh
NCNT-heart
'its heart'

This device must be a very old one in the Macro-Jê stock. It is also present in the Amazonian languages of the Tupí family as well as in those of the Carib family and may well be an ancient areal feature.

EDIT:
As for your questions:

1. What is the basic valency orientation of other people's conlangs? Are they causativising, anti-causativising, neutral (lots of ambitransitive verbs), ...?
2. Does anyone feel like adding more detail on Salish voice alternations?
3. Sint questions:
(i) Does this description make sense?
(ii) Any comments / feedback?

1. My main conlang's basic valency overview is written up here (the line numbers referred to are to the lines of this story). Basically all verb roots are inherently either +TR or -TR, with no ambitransitive roots, and there are a number of transitivizing/detransitivizing derivational affixes with varying connotations (some of these are listed here, though I have still never gotten around to finishing the list...).

3. It seems plausible and coherent. Though I've never looked into it closely I always liked what I'd read of the Salishan system so it's neat to see a conlang taking inspiration from that. (On an unrelated note... man I hate the terms "unergative" and "unaccusative" but I realize that's not your doing...)

[chris_notts]

The Sint stuff looks really good; this is an important but often neglected area. (If I were to write some kind of LCK-esque guide, I think valency alternations would be something I'd definitely mention, because it's all too easy just to replicate the English system without realising how other languages do things differently.)

Another interesting thing to consider is "indefinite object deletion" processes such as English Lucy eats cake vs. Lucy eats. The verbs which allow this in English are fairly few in number and it's not clear to what extent this possibility is systematic. (The classic examples contrast eat, where the object is optional, to devour, where it is obligatory, and dine, where no object is possible at all.) It's all too easy to replicate this in a conlang, and it's better to consider other possibilities, e.g. this sort of detransitivisation is possible with all transitive verbs of a given class, or it doesn't happen at all, or it involves particular "add/subtract an object" morphology (which might be distinct from the "add/subtract an agent" morphology), or whatever.

(Viksen handles this sort of thing via antipassive, which is marked with a particle k and is possible with any basically transitive verb.)

I think anything related to the lexicon is easy to miss. I think the voice system of a language will tend to fit with the semantics of the verb roots in the lexicon. For example, if you speak a language with many S=A ambitransitives like English, it's easy to carry that over without thinking about it and realising that, e.g., eat (something) = telic accomplishment, and eat = atelic activity are actually different. Not to say that that's wrong, since lots of languages use the same root for both those things, but not all of them do so ideally it should be a deliberate design decision on the part of the conlang designer.

I'm not sure Viksen fits very neatly into the causativising/anti-causativising typology, because it doesn't really have many "unaccusative" verbs - states and changes of state are usually expresse with a modifier + wu "to be" or wi "to become".* So "to cool" (intransitive) is if wi cool become and so forth. But transitive verbs of causation can be formed from modifiers with the very productive suffix -xan, e.g. ifxan "to make cool" = "to cool" (trans.). To further complicate things transitives allow discourse-driven agent omission, which may lead to clauses -xan verbs superficially resembling intransitives. The few actual unaccusative verbs like ká "to burn" (intrans.) tend not to have transitive alternants full stop.

OK, so the state is lexicalised as an adjective/non-verb? And then intransitive change of state requires a verbal support, and transitive change of state require a derivational suffix. It's interesting that the transitive state change is more tightly integrated than the intransitive state change.

How do you express "burn something"? What about verbs like: freeze, melt, crack, boil, bend, shake, ...? Are all of these intransitive only?

[chris_notts]

There's something mentioned in Dixon and Aikhenvald's The Amazonian Languages that might be what Sal's thinking of? In Aryon Rodrigues' chapter on Macro-Jê he writes:

I've got that book and I completely forgot about the Macro-Jê stuff. I remember reading elsewhere an argument that is has a similar origin to Celtic initial mutations, as an originally phonological change which was grammaticalised as a marker of continuity.

1. My main conlang's basic valency overview is written up here (the line numbers referred to are to the lines of this story). Basically all verb roots are inherently either +TR or -TR, with no ambitransitive roots, and there are a number of transitivizing/detransitivizing derivational affixes with varying connotations (some of these are listed here, though I have still never gotten around to finishing the list...).

Cool. I guess a lot of it is inspired by Algonquian?

One of your examples was interesting because I've been pondering what to do about incorporation and transitivity marking. Your example:

(11b) počóoš-|číípóó|-`f-yóó = počóoščíípóofyóó
flute-play.instrument-detr-3inan.real.impfv
"(S/he) is playing the flute"

It seems like in Salish, when lexical suffixes (noun-like suffixes from a closed class) are added to a verb root, in many languages marking (in)transitivity becomes optional where it would have been obligatory with the bare verb root. I guess this makes sense since the lexical suffixes are derivational in nature, and to the extent that speakers can unambiguously interpret the derived stem as having the correct transitivity class, the overt voice markers become redundant.

3. It seems plausible and coherent. Though I've never looked into it closely I always liked what I'd read of the Salishan system so it's neat to see a conlang taking inspiration from that. (On an unrelated note... man I hate the terms "unergative" and "unaccusative" but I realize that's not your doing...)

Me too! The terms don't even make sense. To me "unaccusative" implies nominative, and the nominative is the typically controlling argument. And "unergative" implies absolutive, which is the typically affected argument. Every single time I want to use them the wrong way round and have to triple check I've got it right. It's a shame they're the established terminology.

What was wrong with going with "agentive" and "patientive"? This would also nicely unify with the terminology often used for split-S and fluid-S languages.

[chris_notts]

Going back to Macro-Jê and marking things as missing...

Sint currently has a rule that in non-local clauses only the lowest ranked core argument can be extracted in focus-type constructions (focus fronting, questions, relative clauses), and if necessary the verb voice is changed to express the required meaning. This is different to topic driven constructions like gapping in coordination, where the highest ranked argument is preferentially gapped (but not that it is not syntactically controlled - high ranked arguments are gapping targets but not obligatory controllers).

For example, adjusting for the recent changes to morphology described above in this thread:

A extraction -> inverse

na kiins na talajiš
na kiins na ta-laj-(r)i-š
art.m man art.m m.inv-see-APP-3f.anim
"the man who saw her"

P extraction -> direct

na kiins na lajik
na kiins 0-laj-(r)i-k
art.m man art.m 3f-see-APP-3m.anim
`the man who she saw'

This makes sense to me intuitively, but it is different to the way familiar languages work, which tend to use their preferred grammatical relation (e.g. subject, absolutive) for both of these classes of construction. I believe though that it may be attested in other symmetrical voice languages. See the examples below:

Movima (see p27)

b. transitive relative clause (direct)
is kaywanra [di’ joy-a-ɬe = is]
ART.PL food REL go-DIR-CO = 3PL.AB
‘the food they took’
[EAO_Vida en el chaco 030]

c. transitive relative clause (inverse)
is rey mowi:maj [di’ manne-kay-a = is]
art.pl mod Movima rel meet-inv-ep = 3pl.ab
‘the Movimas who met them’
[JGD_160808-Fundacion-02 196]

Thus, like in Tagalog, the pivot that shows up in some constructions is not
determined by semantic roles: when the verb is marked as direct, the pivot is the
undergoer, and when the verb is marked as inverse, the pivot is the actor. In this
way, the syntactic effect of Movima direct and inverse marking is comparable
that of undergoer and actor voice marking, respectively, in Tagalog.

The difference here is that in Movima the agreement with the pivot is apparently (I think?) unified (intransitive S and OBV are marked by the same kind of agreement), whereas in Sint they are not (3rd person P is always marked by suffixes, but there are two different sets of A prefixes depending on the ranking A vs P). Sint in some ways has a hybrid hierarchical - accusative system, with the position of agreement in the verb stem being mostly driven by accusative principles, and the shape of some of the agreement morphemes being driven by relative topicality. I'm not sure whether that makes a difference to the plausibility of the system or not, compared to the unified marking of Movima.

This also basically amounts to special marking of extraction of transitive actors, which is of course also attested in languages like Mayan... which are ergative.

In Algonquian, I believe that the direct/inverse alternation is used to maintain referent prominence in coordination chains, so I expect that the the proximate / highest ranked argument is preferentially gapped. But I don't know how the focus type constructions (clefts/focus word order, questions, relative clauses) interact with the direct/inverse marking. Can you enlighten me?

EDIT: More information about Movima direct/inverse:

https://halshs.archives-ouvertes.fr/hal ... 8/document

Interestingly, it does seem to impose the same low ranked argument only restriction to contrastive fronting as well as relative clauses. The basic analysis for Movima is split-ergative, with direct clauses being ergative in agreement and syntax and inverse being the "split" bit.

[Whimemsz]

.

[chris_notts]

I can't speak for every Algonquian language (although I assume they mostly follow similar patterns). But "gapping" in coordination isn't really a thing. For gapping of verbs, verbs are the central and obligatory part of every Algonquian clause with full participant marking, and everything else is secondary (overt NPs are not common in the first place, and clauses with two overt NPs are almost nonexistent); structures equivalent to English "I ate peas and John apples" are not grammatically possible. For gapping of NPs, as noted, NPs are essentially always optional, and a succession of verbs/clauses can have any possible subject and object. (Although, as pointed out in the paper you link to, direct predicates are considerably more common in practice than inverse ones, so there is certainly a preference for maintaining proximates rather than obviatives as subjects through a length of discourse, including through proximate shifts--but this is also partly just the natural result of the fact that obviatives are less topical than proximates and so inherently less likely to be subjects in natural discourse.) Unless you mean something else by "gap" and I've totally misunderstood?

I'm not sure if I was being clear or not. What I meant was, well, in clauses with omitted arguments there are two questions: what can be omitted, and if the reference of an omitted argument restricted in any way?

If I abuse the term "proximate" to mean the highest ranking 3rd person argument, and "obviative" to mean the lowest ranking, even though there is no such overt nominal category in Sint, then my grammar of Sint currently says:

1. That the default word order outside of marked (focus) constructions is VERB PROXIMATE OBVIATIVE
2. That in non-local (3->3) clauses, direct/inverse is more free than in Algonquian and potentially more driven by discourse topicality and syntactic constraints
2. That both NPs may be omitted, but that it is very strongly preferred in non-local clauses that, if one NP is overtly present, this corresponds to the obviative

To take a very silly example (potentially the causative morphology is wrong as I'm still changing it, but it seems to me that probably an unergative verb like bite shouldn't take causative -s to add a "target"):

Inverse = patient omitted
taràjčik na čàmh
3m.inv-bite-appl-3m art.m dog
the dog bit him

Direct = actor omitted
kıràjčik na čàmh
3m-bite-appl-3m art.m dog
he bit the dog

Basically the combination of noun class agreement and constraints on which NPs may be omitted are compensating for a lack of the nominal obviation morphology of Algonquian, which would disambiguate many of these clauses in a different way.

What this omission constraint doesn't mean, though, is that there is any cross-clausal coreference constraint on the proximate. For example:

taràjčik na čàmh, likaas jek
3m.inv-bite-appl-3m art.m dog, away-3m-run =then
the dog bit him, then it/he ran away

The argument of the second verb "run away" could be either the proximate or the obviative of the preceding clause.

In subordinate clauses, raising behaviour depends on the type of verb. Most tightly integrated complements only allow A/S raising, whereas a few of the looser complementation types show free raising. My subordination chapter says:

1. Modal and verb auxiliaries may either be impersonal or show A/S raising
2. Manipulation verbs like want, order, ... again show optional raising of the complement A/S
3. Perception, knowledge, ... matrix verbs allow topical raising from the complement, typically the proximate

For many of these verbs, Sint lacks the control behaviour of English since all complements are finite. The following is ambiguous between a coreference and non-coreference reading:

harsk anaal
3f-want 3f-p-go-opt/purp
`she wanted to come'
`she_i wanted her_j to come'

If raising occurs then only the non-coreferential reading is possible:

harskiš anaal
3f-want-appl-3f 3f-p-go-opt/purp
`she_i wanted her_j to come'

However, at least in Cree and Ojibwe (and probably in Meskwaki, and presumably a number of other Algonquian languages) the default nonlocal word order (with two third person NPs) according to native speakers is V-OBV-PROX (i.e., in direct clauses, VOS, and in inverse clauses, VSO). Deviations from this basic order are either the equivalent of clefting/focus (as in (1) below from Mistissini East Cree [from this paper]), or due to the fact that (for Ojibwe at least) new/indefinite participants (rhemes) are normally introduced preverbally. (In Ojibwe, preverbal NPs are by default interpreted as indefinite and postverbal NPs as definite. Mistissini Cree may work somewhat differently, however, based on the linked paper?) The upshot is that while word order is used for cleft/focus-type meanings, in a sense it doesn't really interact with the direct/inverse system, because the real word order rules are sensitive to topicality and obviation rather than grammatical relations/direction.

(1) uyuuh atimh miyeyimeu uu awaash
this.OBV dog-OBV like-DIR[PROX>OBV]-3 this.PROX child[PROX] (direct OBV-V-PROX = OVS)
"The child_P likes the dog_O" = "It is this dog_O that the child_P likes"

I guess this is where my thinking diverges from Algonquian, primarily because I don't have a category of obviation separate from the verbal voice marking.

Ojibwe also has a cleft construction using the particle mii (which also has a number of other uses), so one possible Ojibwe equivalent to (1) could be (2). (The order in such constructions is mii + focused element + conjunct verb [basically, a subordinate verb--in this case usually relativized/a participle].)

(2) mii ono animoshan [gaa-minwenimaad wa'aw abinoojiinh]
it's this.OBV dog-OBV REL=like-DIR[PROX>OBV]-3.CONJ this.PROX child[PROX]
"It is this dog_O [that the child_P likes]"

Note that a relativized verb or participle like this is what Ojibwe uses for relative clauses. RC verbs can be direct or inverse.

I guess this is because proximates are discourse topics, and therefore mostly not manipulated for local syntactic reasons? And because direct/inverse is highly constrained by inherent features and obviation?

Relative clauses can modify or serve as an NP of any grammatical role in the matrix clause, including obliques, and including both proximate and obviative NPs; and the nominal associated with the RC can be serving any grammatical role within the RC, including obliques and even objects of possessors, as in (5) (from Valentine's grammar, pg. 585).

(5) mii wa nini [dakweman gaa-bkinaagen'jin mbingoo]
it's that.PROX man[PROX] 3-woman-POSSD-OBV PAST<REL>=win-OBV.SUBJ-3.CONJ-OBV.PCPL bingo
"That's the man_P [whose wife_O won at bingo]"

Out of curiosity, is there any marking that the RC head is an oblique or possessor? Or is it just a general "topic" of the RC, and you have to infer its role? In Sint I have a second position oblique extraction clitic which occurs in both focus fronting constructions and RCs but isn't a true applicative (it doesn't allow verb agreement with the extracted oblique, and occurs outside of the verbal word).

Similarly, Wh-questions can occur with both direct (6) and inverse (7, 8) verbs. The question words for "who" and "what" are inflected for plurality and obviation which disambiguates their referent.

(6) awenenan gaa-waabamaad dibikong?
who-OBV PAST<REL>=see-DIR[PROX>OBV]-3.CONJ night-LOC
"Who_O did s/he_P see last night?"

(7) awenenan gaa-waabamigod dibikong?
who-OBV PAST<REL>=see-INV[OBV>PROX]-3.CONJ night-LOC
"Who_O saw him/her_P last night?"

(8) awenen gaa-waabamigod dibikong?
who[PROX] PAST<REL>=see-INV[OBV>PROX]-3.CONJ night-LOC
"S/he/they_O saw who_P last night? = Who_P did s/he/they_O see last night?"

It's interesting, and a bit weird to me, that the proximate can be an interrogative pronoun, since these are almost be definition not a discourse topic. The speaker, if they honestly want to know the answer to the question, doesn't know who or what the interrogative pronoun refers to.

One interesting area where the direct/inverse system does show up is in raising phenomena. In at least some dialects of Ojibwe, with certain types of complement clause, the verb of the matrix clause is inflected as transitive, with the the object equivalent to either the subject of a direct complement verb, or the object of an inverse complement verb (but not for the object of a direct complement verb or the subject of an inverse complement verb), as in (9) (raising of a direct subject) and (10) (raising of an inverse object), but where (11) (raising of a direct object) is ungrammatical (from Rhodes 1994†):

(9) ngikenmaag ninwag gii-baashkzwaawaad Maagiiyan
1-know[TA]-DIR[1>3]-PROX.ANpl man-PROX.ANpl PAST=shoot-DIR[PROX>OBV]-3pl.CONJ-3.CONJ Marge-OBV
"I know that the men_P shot Marge_O" (lit. "I-know-them men_P that-they-shot-OBV Marge_O")

(10) ngikenmaa Maagii gii-baashkzogod ninwan
1-know[TA]-DIR[1>3] Marge[PROX] PAST=shoot-INV[OBV>PROX]-3.CONJ man-OBV
"I know that the men_O shot Marge_P" (lit. "I-know-her Marge_P that-OBV-shot-her man/men_O")

(11) *ngikenmaan Maagiiyan gii-baashkzwaawaad ninwag
1-know[TA]-DIR[1>3]-OBV Marge-OBV PAST=shoot-DIR[PROX>OBV]-3pl.CONJ-3.CONJ man-PROX.ANpl
*"I know that the men_P shot Marge_O" (lit. *"I-know-OBV Marge_O that-they-shot-OBV men_P")

†Note that I don't share all of Rhodes' theoretical conclusions or glossing choices in that paper, and that it unfortunately has some typos in the Ojibwe lines that are likely to confuse. But it does provide a few other syntactic phenomena that are sensitive to directionality (such as the triggering of obviative marking on VIIs).

Interesting that the example is "know". In verbs of knowledge or perception, the A/S of the matrix verb potentially doesn't interact with any of the arguments of the complement at all. Does Ojibwe allow the same pattern with verbs of manipulation, or are those restricted to A/S raising?

[Curlyjimsam]

I'm not sure Viksen fits very neatly into the causativising/anti-causativising typology, because it doesn't really have many "unaccusative" verbs - states and changes of state are usually expresse with a modifier + wu "to be" or wi "to become".* So "to cool" (intransitive) is if wi cool become and so forth. But transitive verbs of causation can be formed from modifiers with the very productive suffix -xan, e.g. ifxan "to make cool" = "to cool" (trans.). To further complicate things transitives allow discourse-driven agent omission, which may lead to clauses -xan verbs superficially resembling intransitives. The few actual unaccusative verbs like ká "to burn" (intrans.) tend not to have transitive alternants full stop.

OK, so the state is lexicalised as an adjective/non-verb? And then intransitive change of state requires a verbal support, and transitive change of state require a derivational suffix. It's interesting that the transitive state change is more tightly integrated than the intransitive state change.

How do you express "burn something"? What about verbs like: freeze, melt, crack, boil, bend, shake, ...? Are all of these intransitive only?

"burn something" would be a separate root (which I don't think I've come up with yet), or else you could get to it by deriving the adjective kánad "burned" and then causativising that as kánadxan. This latter is a bit of a mouthful so is presumably the dispreferred option.

The other verbs act on the pattern of "cool" - you have adjectives meaning "frozen", "melted" etc. and then these can either be used in the "be/become A" intransitive construction or as the basis of derived transitives with the -xan suffix. E.g. izjif "frozen", izjif wi "to become frozen, freeze" (intrans.), izjifxan "to make frozen, to freeze" (trans.).

[mèþru]

I keep reading this thread as "Basic Valence Orientation and Shit"
good stuff here

[akam chinjir]

In amongst the New Year festivities I've managed to think a fair bit more about this, but most of it is too inconclusive and Akiatu-centric to be useful here.

I will say that Akiatu generally likes explicit causatives and inchoatives, and doesn't so much like passives, so its underived break verb may be a bit of an outlier.

Speaking of break words, there are a number of distinctions you could draw related to agency and control, and I'm curious about the different ways these could be marked. Suppose a vase breaks. It could be:

• An agent broke it, intentionally
• An agent broke it, by accident
• Some external force (maybe the wind) broke it
• It just broke, with no apparent external cause

E.g., how common is it to grammaticalise the distinction between the first two of these? Or the second and the third?

I'm not sure I've seen a language that uses reduplication specifically for valency changes before.

I guess one possibility would be to have (literal) cognate objects get reinterpreted.

In his book on applicatives, Peterson claims that causative ~ benefactive is more common than causative ~ comitative polysemy. I find this a bit surprising, since the causative ~ instrumental ~ comitative polysemy seems more natural to me, especially since many morphological causatives cover permission ("let") as well as forcing ("make"). The only difference between those cases is the degree of agency retained by the original agent, and its animacy:

I cut the fabric
-> I made the knife cut the fabric
-> I cut the fabric with the knife (no significant semantic difference except directness of causation, since an inanimate has no volition anyway)

I made her eat / I fed her
-> I let her eat
-> I ate with her (people normally eat together)

How universal is it to allow instruments to become subjects? (Asking both because your first example requires it and because I'm thinking that Akiatu might disallow it without an explicit passive.)

It occurred to me that a possible variable here is the lexical source of the causative. E.g., it's not a surprise that Mandarin has both give causatives and give benefactives, because give verbs make sense as sources for both meanings. (I was also thinking of Mandarin ràng 讓 defer, yield → make, let; as far as I know there are no ràng benefactives, but the leap to benefactives seems easier than the one to comitatives or instrumentals.)

But---The World Lexicon of Grammaticalization gives do, give, and take as common sources for causatives, with give also as a common source for benefactives and take for both comitatives and instrumentals. So I guess that idea's a bust.

[chris_notts]

Speaking of break words, there are a number of distinctions you could draw related to agency and control, and I'm curious about the different ways these could be marked. Suppose a vase breaks. It could be:

• An agent broke it, intentionally
• An agent broke it, by accident
• Some external force (maybe the wind) broke it
• It just broke, with no apparent external cause

E.g., how common is it to grammaticalise the distinction between the first two of these? Or the second and the third?

A pure case of (1) vs (2) is uncommon I think, although Salishan languages seem to do it and so does Sint since I based the stuff above on Salishan. Most Salishan language have a kind of limited control transitiviser which covers meanings like "manage", "accidentally", and even "try to" where the agent is not completely in control. It is probably more commonly a secondary meaning of another marker or category. For example, in languages that mark evidentiality, a first person verb with non-firsthand evidentiality often has a "I wasn't in control / don't remember" type meaning. Miratives and often markers with overtones of unexpectedness might also have this meaning. Detransitive forms like anti-passives or other patient demotions to oblique, which commonly suggest lack of affectedness of the patient, or aspect markings which suggest lower telicity, might also suggest less control of the actor. Compare the following English clauses:

I whacked it
I was whacking away at it <-- to me this has a strong implication of lower control, e.g. I was angry and not aiming very well

(1) vs (4) I think is quite common, and probably almost all languages express this for actions which can spontaneously happen somehow. If that somehow is either a transitiviser (e.g. causative, addition of patient agreement, ...) or detransitiviser (e.g. middle voice) then I guess the pattern is likely to be productive. If it's suppletion then it's probably going to be limited to just the events which are easiest to conceptualise as spontaneous.

(3) is tricky and probably more language specific. I suspect it also depends on the verb. I would bet there are languages where a stative or low patient impact transitive can more acceptably take an inanimate A then a high impact transitive. Especially since in many languages verbs with two arguments but no strongly affected argument are often morphologically intransitive. This is the kind of thing I'm thinking of:

The camera saw him (OK in hypothetical language X)
The gun shot him (not OK in hypothetical language X)

The answer for Sint is:

(1) transitive in -s

ràk-s = break deliberately (animate controller, affected patient)

(2) transitive in -nč

ràk-nč = break with limited control (+/- animate controller, affected patient)

(3) transitive in -nč, impersonal in ine-V-s + instrumental

-nč is more likely with inanimate forces that have energy and are independent of human will, e.g. weather. ràk-nč might be used if the wind broke something during a storm.

The impersonal is more likely with an instrument normally used by a human:

ine-ràk-s ... o na širk = someone cut it with a knife

(4) bare intransitive (for most actions which can be conceptualised as spontaneously happening), ine-V for a root transitive (where the event can't be conceptualised as just happening without an external cause), -ksi reflexive (where an animate affects themselves)

How universal is it to allow instruments to become subjects? (Asking both because your first example requires it and because I'm thinking that Akiatu might disallow it without an explicit passive.)

I guess that's true, a lot of languages do have a problem with inanimates in A roles.

It occurred to me that a possible variable here is the lexical source of the causative. E.g., it's not a surprise that Mandarin has both give causatives and give benefactives, because give verbs make sense as sources for both meanings. (I was also thinking of Mandarin ràng 讓 defer, yield → make, let; as far as I know there are no ràng benefactives, but the leap to benefactives seems easier than the one to comitatives or instrumentals.)

But---The World Lexicon of Grammaticalization gives do, give, and take as common sources for causatives, with give also as a common source for benefactives and take for both comitatives and instrumentals. So I guess that idea's a bust.

Well, it might also contribute to the high frequency of causative ~ benefactive if give is a more frequent diachronic source than take for causatives.

[Salmoneus]

I've been consuming far too much cheese and port in recent days to have been able to concentrate enough for this thread. Even now, it's going to take me a considerable while to get through it...

Can also feed the benefactive applicative:

kaas 'run' -> kaas-nč -> kaasınč 'run to/for (him/her)'

So what's the difference in use between this and the benefactive use of -s?

The beneficiary meaning probably rules out the 'accidentally' meaning of the limited control transitiviser, since it's hard to see how an action can simultaneously be performed for someone's benefit and be unintended.

Conceptually, this is possible, in three different ways, depending how you interpret 'for'.

If your beneficence is centred on intent, then an accidental benefactive occurs when the agent has benefactive intent when performing an action, but the action that occurs is not the projected action. For instance, "But I burnt down the house accidentally for you! (while trying to cook you a special romantic dinner)"

If your beneficence is centred on consequences, then an accidental benefactive occurs when the agent performs an unprojected action that redounds to the benefit of another. For instance, "the sprinter tripped up to the benefit of his rival".

If your beneficence is centred on both, then an accidental benefactive required both intent and consequence: for instance, "he accidentally killed her stalker (while cutting down an overhanging branch for her)".

As a philosophy graduate, I guess I should go one further. If we consider benefactive intent as a kind of virtue, we can distinguish aretaic, consequentialist, but also deontological beneficence. In this third case, accidental benefactives would occur when the accidental action was of a kind inherently tending to the advantage of another, regardless of whether it was beneficial in this particular instance.

This also all assumes that the accident is the action, as it logically should be. But these things can blur, and it could be imagined that these formations could be used when the accident relates to consequences ("I kicked you (and unintentionally knocked you out of the way of the falling piano)") or to subverted inherent form ("I gave you money (not realising it would draw the attention of the police)").

1. What is the basic valency orientation of other people's conlangs? Are they causativising, anti-causativising, neutral (lots of ambitransitive verbs), ...?

*pulls out hair and weeps*

3. Sint questions:
(i) Does this description make sense?

I'm not a linguist, so can't contribute anything useful regarding typology or universal grammar. But conceptually it appears to make sense.

[Salmoneus]

The verb agrees with both arguments, but there are two problems.

The first is that that only works if all of the patient agreement morphemes are overt. If one is zero, or there is any form of differential patient agreement (conditioned by animacy, definiteness, ...) then there is ambiguity between the a transitive and intransitive verb. That's an issue for reference tracking and semantics, especially if there are any S=P ambitransitives.

Fair enough.

I would just say, however, that I think people tend to overestimate the problems caused by potential ambiguity. Many ambiguities can be resolved by context, either through common sense or through codified rules. For example...

For example, if "lower" could mean both "go lower" and "cause to be lower / put lower":

3A-took-0 and 3A-lower-0
= 3A-took-0 and 3A-lower
= He took it and went down
= He took it and lowered it

In this example, the ambiguity is only a problem if you need to track 'his' location (as 'it' goes down either way). Generally it will always be the latter, because lowering is much more common than people going down into places - and from the fact that you begin by talking about 'it', it's probably not as important where 'he' ends up. [eg, he takes the key and either sends it down or takes it down - the emphasis is on the location of the key, and the location of the man is usually not that important, or will quickly become evident from the ongoing narrative]. But you can disambiguate in many ways if necessary - for instance, "he took it and it went down" vs "he took it and they went down", or "he took it and it went down" vs "he took it and he went down with it".

3A-have pig and goat, pig 3A-want 3A-eat-PURP(-0)
= He has a pig and a goat, and the pig is hungry / wants to eat
= He has a pig and a goat, and he wants to eat the pig

Likewise, this is a situation that will rarely cause confusion because of the clear inherent semantics - if you have a wolf and a pig, it's the pig that wants to eat the wolf, not vice versa. Confusion arises when you have two things equally likely to want to eat the other, which is inherently a rare scenario, and you can always just have a rule about restating the subject (or the object) when necessary. [or automatically take the more fronted argument as the subject - here, the pig must want to eat the goat].

break.INTRANS-CAUS-0 = break it (make it break)
break.TRANS-CAUSE-0 = break it with (him/her/it)

[...]

give-ANTI
= give-ANTI-0
= be giving (something)
= give to (him/her/it)

And again, actual confusion here will be very rare, because the classes of things-we-try-to-benefit and things-we-try-to-give-to-people, and the classes of things-we-break and things-we-use-to-break-things don't have a lot of overlap. So "I met a man and I broke it" can mean "I broke (sth) with the man", whereas "I met a vase and I broke it" means "I broke the vase". When both can be the case - "I met a man with a vase and I broke it", it can be resolved by importing the semantically or syntactically prefered item (in this case the man, because it's "man with a vase" instead of "vase with a man"), or just left 'ambiguous', because either way the man's vase ends up broken, and whether he helped break it or not, if it's unimportant enough to be left to pivots and non-overt arguments, is probably unimportant enough to leave to the imagination anyway.



To be clear, I'm not saying there's anything wrong with building a language that's very determined to avoid any hint of ambiguity through systematic, overt morphology. I'm just saying that very often people assume that it's necessary, when very often it probably isn't... (and the tactics, whether syntactic or pragmatic, to defuse the ambiguities that arise, can themselves be very interesting).

[Salmoneus]

Turns out the Oceanic languages have reduplicative detransitivisation:

Some canonic languages also have detransitivising morphology, although its role is much less significant than the transitivising and causativising morphology discussed above. It takes three forms. Two are straightforward detransitivisation: reduplication, and prefixation of a reflex of one of Proto Oceanic *ma- and *ta-. The third is prefixation of a reflex of Proto Oceanic *paRi- to form a reciprocal verb.

Specifically, iirc this is generally to produce unergatives.