Recommended treatment of internal objects (a.k.a. cognate objects)

[amir-zeldes]

I'm not sure if this has been discussed before, but this came up as a question in the context of UD_Ancient_Hebrew-PTNK and I couldn't find a previous issue: What is the recommended treatment of "inner objects", which are common in Semitic languages and variously referred to as "internal objects", "cognate objects" or in Arabic as "maf'uul mut.laq"?

The phenomenon is basically the systematic inclusion of a verbal noun as an extra object of its own cognate verb, in addition to a possible regular object (if the verb is transitive):

• Biblical Hebrew: יֹּ֥אמֶר הַנָּחָ֖שׁ אֶל־הָֽאִשה לֹֽא־מ֖וֹת תְּמֻתֽוּן׃ "the serpent said to the woman, you shall not die a death"
• Modern Hebrew: rutsu ritsa mehira "run a fast run!"
• MS Arabic: s.aama s.auman "he fasted a fast"

In modern Semitic usage for both Arabic and Hebrew, the most frequent contexts in which this is seen is either for emphasis, or when an adjective needs to be used adverbially, and the modified verbal noun serves this purpose (so "run a fast run" is like "run quickly"). In Arabic, if the verbal noun is dropped, you can still get an accusative adjective agreeing with the missing verbal noun and used with adverbial meaning (this is referred to as naa'ib maf'uul mut.laq).

How should this be analyzed in UD? In both Hebrew and Arabic, the accusative marking is explicit wherever the morphology allows it and looks indistinguishable from regular objects. When there is one object, I expect people are tagging this as obj, but when the verb is transitive ("bake something a thorough baking"), this would lead to a double obj analysis, which is usually avoided in UD.

Incidentally this phenomenon can occur marginally in European languages, especially in topicalized negation, which I think is possible in both German and Russian, at least with intransitives:

• Aber Schlafen habe ich nicht geschlafen "but sleep I did not sleep" (=I didn't sleep a sleep, i.e. I did not sleep at all)

[dan-zeman]

It seems straightforward in most cases to analyze them as obj, i.e., as if an intransitive verb is used transitively. The tricky part is when they co-occur with a regular object, as in the bake something a thorough baking example. Here I would ask whether we can define rules to make one of them (preferably the thorough baking one) obl. If we cannot, then one of them has to be iobj, as double obj are not allowed.

[nschneid]

as double obj are not allowed.

Doesn't Hebrew occasionally have two objects with overt accusative marking in causative constructions? IIRC:

הוא האכיל את התינוק את הדייסה He made.eat ACC the baby ACC the cereal

Apologies if I've misremembered the example but I'm pretty sure there is a circumstance with double-ACC, and I'm not sure what other analysis would work than using double obj. (Hebrew has dative prepositions for what in English would be indirect objects.)

[dan-zeman]

It was my understanding from some discussions with @manning that multiple objects should never result in all of them holding the obj deprel. (I suspect that the validator does not enforce this yet.) That would mean that one has to come up with rules how to pick the obj among them, and label the other(s) iobj. Whether datives are or are not core arguments in Hebrew seems to be a separate question; especially if it is a preposition rather than a morphological dative.

The double accusative situation sounds to me quite parallel to the double object situation in English: there are two participants that receive treatment normally used for the prototypical patient, and only one of them is allowed to be obj. On the other hand, the fact that one of them usually has the semantic role of THEME and the other of GOAL is English-specific rather than UD-universal, and it does not have to hold in all languages.

[amir-zeldes]

He made.eat ACC the baby ACC the cereal

This is correct, the Hebrew example contains an unambiguous accusative marker.

Whether datives are or are not core arguments in Hebrew seems to be a separate question; especially if it is a preposition rather than a morphological dative.

Just to clarify the empirical situation: Hebrew has no dative marking, except the preposition "to". The example Nathan gave (and the internal object construction) are both different to that and look similar to each other: they carry acc marking. The 'preposition' in the example is due to the Hebrew acc system, which is consistently split: indefinite objects appear without any mediating marker (classic core object) and definite ones appear with a non-optional preposition "et", which means nothing except "accusative". This is similar to Armenian "z" and other such systems. Nathan's example works with indefinite unmediated objects as well:

הוא האכיל תינוקות דייסה he fed babies porridge (no prepositions used)

Finally regarding "which argument is core", the problem is that both objects are passivizable (but only one of the two!) and the passive verb will agree with whatever was passivized (e.g. singular for the porridge as subject, plural for the babies). The same is true for cognate objects, except that intransitive cognate objects are not passivizable, since the corresponding passive verb form does not exist (so this is blocked; but for a transitive cognate object, either it or the regular object can be passivized).

Not sure what the right answer is in UD, but those are the facts for Hebrew as I understand them.

[amir-zeldes]

Adding some IAHLT people in case you have an opinion: @ivrit @Hilla-Merhav @shirawigi @yifatbm @IsraelLand

[dan-zeman]

OK, passivization works for either object (and I'm now assuming they both are (core) objects, not obliques). Is there any other rule in the grammar where the treatment of the objects will differ?

And what about word order? Can you swap the objects and say he fed porridge babies (still meaning that the babies eat the porridge and not vice versa)?

[IsraelLand]

It's a bit awkward but possible, הוא האכיל דייסה(,) תינוקות he fed porridge(,) babies Normally the definite version (+accusative marker) would be preferred if this fronting is used הוא האכיל דייסה את התינוקות he fed porridge ACC the babies

this is also possible, הוא האכיל את התינוקות את הדייסה he fed ACC the babies ACC the porridge but this seems not to be - הוא האכיל את הדייסה את התינוקות he fed ACC the porridge ACC the babies

instead reverting to the oblique - הוא האכיל את הדייסה לתינוקות he fed ACC the porridge to the babies (and vice versa, *he fed to the babies ACC the porridge)

Please do correct me if you're of a different intuition, in any case, I find these possible in a way the English "the chicken fed, the man" etc. would never be.

[nschneid]

How about:

הוא האכיל תינוקות את הדייסה

הוא האכיל את התינוקות דייסה

הוא האכיל את הדייסה תינוקות

[IsraelLand]

to me, הוא האכיל תינוקות את הדייסה is fine, הוא האכיל את התינוקות דייסה is the most regular/common phrasing, while *הוא האכיל את הדייסה תינוקות is not "grammatical" in the baby-eats-porridge meaning (this is a lot about usage, but I do feel a grammatical core issue to this, probably that babies is usually mediated by the oblique, while porridge is always in the accusative (i.e. "fake" accusative marker/position?).

It's fair to keep in mind the et element in cases like these, with their archaic original oblique-like meaning ללכת איתה to go with.her (it-a) which probably is far fetched but might be some underlying factor in such fronted, irregular word order (for MH) structures

[nschneid]

A summary of judgments with the active form of the verb and two objects—1st/2nd refers to order of the two NPs:

תינוקות(babies)	דייסה(porridge)	Judgment
indef, 1st	indef, 2nd	✓
def, 1st	indef, 2nd	✓ most natural
indef, 1st	def, 2nd	✓
def, 1st	def, 2nd	✓
indef, 2nd	indef, 1st	~ fronted, awkward
def, 2nd	indef, 1st	✓ fronted
indef, 2nd	def, 1st	*
def, 2nd	def, 1st	*

It seems the more natural order is to put the babies first (fed-babies, not fed-porridge) if neither babies nor porridge are oblique. Secondarily, the preferred definiteness is definite babies, indefinite porridge.

[dan-zeman]

So it seems the language still does not treat the two objects as equal. I am not sure what exactly it means for their labeling in UD. But if it holds that both of them cannot be obj when both of them are present, then we can either pick iobj based on word order (the first one, the second one, the one closer to the verb...) or we can do so based on their semantic roles (which sounds evil in the UD context, but in fact we have admitted it for iobj identification at the end of Section 4.1 of de Marneffe et al. 2021). The theme (porridge) would then be obj and the goal (babies) would be iobj. When only one of the objects is present, I would give it obj.

Now, are there examples where it is difficult or impossible to see these two roles, even figuratively? I suspect the "internal objects", i.e., the original topic of this thread, may constitute such an example when they co-occur with a regular direct object (to bake the bread a thorough baking). But even here I can sense something like "to give the bread some treatment/thorough baking", which would mean bread = iobj, thorough baking = obj. On the other hand, the inner object feels like something less important, and giving it the iobj label would even allow to abuse it and read it as inner object rather than indirect object :-) Are there word order preferences comparable to the babies-and-porridge examples above? And can the inner object be passivized?

[amir-zeldes]

It's fair to keep in mind the et element in cases like these, with their archaic original oblique-like meaning ללכת איתה

I think this is unrelated: the form it-a is not actually a presuffixal form of et-, which would be ot-a with "o"; it is the suppletive presuffixal form of pharyngeal 'im "with" (='im-a), and so it is not part of this discussion.

I can sense something like "to give the bread some treatment/thorough baking"

This is just a subjective feeling, but I think this is a transferred reading from the feeling that the English rendition gives. Hebrew does not have a benefactive ditransitive, so this would not be a general sense of that word order. The double object is licensed in the one construction thanks to the cognate object construction (which is much more frequent in Hebrew/Arabic than in English) or the causative morphological form, which licenses two accusatives.

In general I find it a little unfortunate that UD prohibits 2 objects here, because I honestly don't see a huge difference between "make someone eat something" and "feed someone something". In the first case, we get to have two objects, because the default productive English causative happens to use a light verb "make" to carry the first object. In Hebrew, causatives are most often realized through the morphological non-concatenative causative pattern, hi_ _i_, which regularly has both these arguments, marked accusative. Both can be passivized, and no word order is truly ungrammatical, though certainly the unmarked order is to have the causee first (this is usually a given, definite human, so the bias is not surprising).

Even if the semantics of the two objects are not identical, and even though there is a preferred word order, I don't really understand how they are not both direct objects at a syntactic level. Unlike English ditransitives, there is no regular alternation that suggests an indirect object is at play, and there is in general no dative in this language...

I would also find it ironic if we were allowed to call the cognate object of an intransitive verb obj, since it has no competitors, but would then call the cognate object of a transitive verb something else, just because we need to save that label for the regular theme argument. So in sum, I guess my preference would be 1. to allow 2 x obj, or if that is truly ruled out, then 2. to have a non-obj label for both cognate object environments, just so at least they look the same for transitive and intransitive verbs.

[dan-zeman]

I think we need more people from the UD core group to weigh in. I, too, used to find it unfortunate that we cannot have two objs (in fact I wanted to be even more radical and get rid of the iobj label completely). But I'm assuming that those discussions are over now and shouldn't be periodically reopened, they take a lot of energy :-)

If I were to find an advantage on the restriction, then it is its simplicity. True, it may be tricky to find language-specific criteriea to distinguish obj from iobj, but the universal part of the guideline is simple and verifiable. If, on the other hand, UD said that sometimes double objs are allowed, then it should also say how we recognize situations where they are allowed. Describing such situations for all languages may not be easy.

[mr-martian]

A few more examples from Ancient Hebrew that started this discussion:

• transitive, passive: הגד הגד לי כל אשׁר עשׂית = "all that you did has been told to me by telling"
• intransitive? מכל עץ הגן אכל תאכל = "from every tree of the garden you shall eat by eating"
• transitive: הרבה ארבה עצבונך = "I shall increase your pains by increasing"

What I find puzzling about these is that there are examples where the cognate object is morphologically a noun appears in the usual place, after the verb, such as in הבה נלבנה לבנים ונשׂרפה לשׂרפה "let us brick bricks and fire them with fire", but in the examples above, the cognate object is a form of the infinitive and occurs immediately before the verb.

In the noun example, I have them annotated as obj and obl. In the infinitive examples, I was probably somewhat mislead by the typical English translations, which often render them as "surely" or some other, usually adverbial, expression of emphasis and currently have them tagged as ADV with relation advmod (now fully intending to change them when this discussion reaches a conclusion).

[nschneid]

Are the prepositional datives in a canonical argument structure for expressing transfer ("Ruti gave the book to Dani") considered obliques in Hebrew, or are they iobj like prepositional datives in Spanish?

If they were defined as how iobj works in Hebrew, then that would be grounds to eliminate iobj from contention for accusative-marked causative constructions.

[mr-martian]

They are currently considered obliques. iobj is currently unused in both Hebrew-HTB and Ancient Hebrew-PTNK.

[amir-zeldes]

then it should also say how we recognize situations where they are allowed

I agree this is a thorny issue and don't have a good catchall definition. I think the prototypical cases that most motivate two direct object deprels are the ones in accusative marking languages where both are accusative and both are passivizable, which is the case for the Semitic cognate objects.

Then just a clarification about @mr-martian 's examples:

• transitive, passive: הגד הגד לי כל אשׁר עשׂית = "all that you did has been told to me by telling"
• intransitive? מכל עץ הגן אכל תאכל = "from every tree of the garden you shall eat by eating"
• transitive: הרבה ארבה עצבונך = "I shall increase your pains by increasing"

I would remove the word "by" in a literal translation: in order, the first example reads "a telling was-told to me all that you did", and so on (there is no "by" or other indication of obliqueness in the original). I think the pre-poned cognate object is pretty standard in Biblical Hebrew and leads to this 'reduced-masdar' form of the verbal noun, and in modern Hebrew we typically post-pose them and use the full nomen-actionis form, but I don't think that should matter for the deprels - either way you look at it, these things have nominal morphology, the same root as the verb, and nothing licenses them about the verb itself - they are automatically available for any verb.

Finally for the "intransitive?" case, this shows that they are also compatible with optional transitives like "eat", which can appear without an object or with one, and a cognate object is possible for both versions. It's also possible to passivize one of the objects without including the other.

[nschneid]

They are currently considered obliques. iobj is currently unused in both Hebrew-HTB and Ancient Hebrew-PTNK.

OK, then another question: do we know of languages where iobj has been adopted for constructions with two accusatives that are unrelated to transfer/the recipient role, e.g. causatives?

To quote the passage that @dan-zeman alluded to from the 2021 article:

Finally, some verbs in some languages take three or more core arguments, more than one showing behavior that is characteristic of objects (Haspelmath 2015). Prototypically, such ditransitive constructions involve verbs of giving and transfer, and UD analyzes the theme (i.e., the entity that is transferred) as the direct object, and introduces the relation of indirect object (iobj) for the recipient. However, as noted earlier, the iobj relation should only be used if the nominal denoting the recipient is encoded as a core argument.

Elsewhere the article discusses causatives in ergative languages but nothing that I see for nominative-accusative languages.

[sylvainkahane]

I agree with @dan-zeman that iobj must be suppressed. It is an English-centered relation, which is very difficult to use in other languages. And even in English treebanks, it has been used in a very strange way: In the English ditransitive construction, it is the first object that has been labeled iobj, whereas it is the one that can be passived or relativized. If there is a redistribution, there is a redistribution and we don't mind which one was an oblique in the initial construction. Clearly the iobj label in English is used in a semantic way and not a syntactic one. Which will cause confusion for languages which have an alignment of P (patient) with R (recipient) rather than T (theme). (See Haspelmath 2011, On S, A, P, T, and R as comparative concepts for alignment typology.)

[amir-zeldes]

I'm actually not arguing against iobj at all - I think it is very clearly motivated on purely morphosyntactic grounds in languages with clear dative marking (e.g. German, Slavic), where either order is possible (iobj-obj or the opposite) and the only way to get consistent argument mapping is to use two labels. Again, the criterion is not at all semantic in these languages; the extension to English is kind of 'by historical association', and though English no longer has the dative case, there are some syntactic alternation reflexes of iobj that I think make it a good idea. But if we want to discuss iobj as a concept, then I think that deserves a separate issue - for now I would just like to know what to do with cognate objects! And in the best case, it would be nice to have a solution where they are analyzed the same, regardless of whether the normal theme object also appears.

[sylvainkahane]

@amir-zeldes, you are not answering to my objections. We are not discussing cases where we have two complements with clearly different markings (such as different case markers), we are discussing cases where we have two complements with the same marking. (By the way, I don't think that iobj is needed for dative complements, obl could be used as it is used for other non-accusative complements.) When we have two complements that are marked as the standard object complement of a transitive construction (in an accusative language) (let us call it P as it is done in typology), I think we should use two obj relations. If you really want to keep only one obj, you must choose the one which behaves the most like P. This is not was has been done in English treebanks! In "I gave this guy a book", only the first object is relativizable, as far as I know:

*I read the book I gave this guy I know the guy I gave this book

But it is the first object that is analyzed as an iobj!!! If I come back to the discussion about Hebrew or any other language, what would be the strategy if the language has a construction with two complements that behave like P? If there is one that behave more like P than the other, it must be the obj or the iobj?

[nschneid]

Actually "I read the book I gave this guy" sounds fine to me!

In traditional English grammar, the first object is said to be indirect and the second object is said to be direct. Huddleston et al. (2021):

The traditional labels 'direct' and 'indirect' are based on the idea that in clauses describing an action the referent of the direct object is apparently more directly involved in being acted on in the situation than the referent of the indirect object....The indirect object is characteristically associated with the semantic role of recipient. (p. 97)

The main syntactic property distinguishing the two kinds of object is position: in ditransitive clauses...the indirect object precedes the direct object....In addition, the direct object readily undergoes fronting in various non-canonical constructions, whereas the indirect object is quite resistant to it [acceptability judgments vary but fronted indirect object is always less acceptable]. (p. 98)

It is claimed that "He kept the gifts which she had given him" is more readily acceptable than "They interviewed everyone whom she had given gifts". They both sound fine to me, but the former is perhaps easier to process.

Direct/indirect objects also differ with regard to passivization. In American English only the indirect object can be passivized, though there is a rarer option in British English to passivize the direct object: "The key was given me by the boss" (p. 367).

Whether this traditional terminology is really appropriate to extend to all double-object constructions in all languages, I am not so sure.

[sylvainkahane]

@nschneid, you're right. As I'm not a native speaker, I have to verify on corpora, which I ave just done on GUM and EWT ;)

It appears that the first object (iobj) is passived in 76 cases (38 GUM + 38 EWT), while the second object (obj) is passived in only one case, which is quite particular because iobj is relativized.

Conversely, it appears that the second object (obj) is extracted in 19 cases (5 GUM, 14 EWT), while the first object (iobj) can never be. Moreover there are 25 cases of iobj without obj in GUM and 0 in EWT, which is the result, I suppose, of a different analysis in the two treebanks. (Strangely, EWT has also 88 double obj constructions!!!)

Conclusion for English: English has two complements in the ditransitive construction that both behave as P (the object of the transitive construction), none of them behaving more like P than the other. The first complement (R, the recipient) can be passived, but not extracted, while the second complement (T, the theme) can be extracted, but not passived.

General conclusion: I don't think that one complement of the ditransitive construction deserves more the label obj than the other. I think that both must be labeled obj. (We don't mind that the traditional grammar call one indirect object. If we start to take care of traditional grammars for all languages it would never be possible to have a universal annotation scheme.) But of course, there is some advantages to distinguish the two objects, but this must be a sub-relation (for instance, obj:R and obj:T, as proposed by typologists). Same kind of solution for Hebrew. Nothing justify to introduce a "new" relation iobj for English or for Hebrew.

[amir-zeldes]

The first complement (R, the recipient) can be passived, but not extracted, while the second complement (T, the theme) can be extracted, but not passived. ... I think that both must be labeled obj

Well, it seems we now all agree that iobj is a valid label (e.g. in German, Polish), and that there are morphosyntactic differences between the two even in English, so the only question is whether the people working on English feel that at the language specific level, the differences are enough to justify using iobj - which I think both @nschneid and I agree is better, and is also the status quo, which is more comparable across Germanic languages, and does not conflict with current UD guidelines.

But again I would like to suggest that if this is still controversial, we should open an issue for that - for now I would like to resolve the cognate object issue: I see that double object is specifically advised against in the universal guidelines for ditransitives and in prior art such as the programmatic UD journal papers, but I am not sure if this is meant to apply to cognate objects, or, for that matter, the causatives mentioned above, both of which are common in Semitic.

@dan-zeman said above "If, on the other hand, UD said that sometimes double objs are allowed, then it should also say how we recognize situations where they are allowed." - so I would like to know whether "cognate object" and perhaps "causative" with explicit double accusative marking would constitute such situations, and if not, what the alternative should look like. I have pointed out some shortcomings of iobj for this purpose, are there other suggestions? Pinging @manning and @mcdm in case you have some thoughts.

[Stormur]

@amir-zeldes

I'm not sure if this has been discussed before, but this came up as a question in the context of UD_Ancient_Hebrew-PTNK and I couldn't find a previous issue: What is the recommended treatment of "inner objects", which are common in Semitic languages and variously referred to as "internal objects", "cognate objects" or in Arabic as "maf'uul mut.laq"?

The phenomenon is basically the systematic inclusion of a verbal noun as an extra object of its own cognate verb, in addition to a possible regular object (if the verb is transitive):

* Biblical Hebrew: יֹּ֥אמֶר הַנָּחָ֖שׁ אֶל־הָֽאִשָּׁ֑ה לֹֽא־מ֖וֹת תְּמֻתֽוּן׃ "the serpent said to the woman, you shall not die _a death_"

* Modern Hebrew: rutsu ritsa mehira "run a fast run!"

* MS Arabic: s.aama s.auman "he fasted a fast"

In modern Semitic usage for both Arabic and Hebrew, the most frequent contexts in which this is seen is either for emphasis, or when an adjective needs to be used adverbially, and the modified verbal noun serves this purpose (so "run a fast run" is like "run quickly"). In Arabic, if the verbal noun is dropped, you can still get an accusative adjective agreeing with the missing verbal noun and used with adverbial meaning (this is referred to as naa'ib maf'uul mut.laq).

How should this be analyzed in UD? In both Hebrew and Arabic, the accusative marking is explicit wherever the morphology allows it and looks indistinguishable from regular objects. When there is one object, I expect people are tagging this as obj, but when the verb is transitive ("bake something a thorough baking"), this would lead to a double obj analysis, which is usually avoided in UD.

From this description, I get the impression that these "internal objects" are indeed all adverbials, and/or represent some kind of topic-comment structure, especially with regard to the usage that you mention of keeping only the adjective sometimes (so I understand a construction like run a fast (one) with the sense of 'run fast')

With regard to what happens in some European languages:

Incidentally this phenomenon can occur marginally in European languages, especially in topicalized negation, which I think is possible in both German and Russian, at least with intransitives:

* Aber Schlafen habe ich nicht geschlafen "but sleep I did not sleep" (=I didn't sleep a sleep, i.e. I did not sleep at all)

I do not perceive this is an object at all. You separate the two components, the topical position is a clue for this, e.g.:

• Also, schlafen... ich habe nicht geschlafen which then becomes put together with verb inversion: Schlafen habe ich nicht geschlafen 'as for sleeping, I have not slept'.

And I am pretty sure something like (von mir) wurde kein Schlafen geschlafen is unheard of. (besides, the deverbal noun here should be Schlaf).

This appears in Italian too, and there are two major variants (same meaning as the German example):

• Dormire, non ho dormito
• Per dormire / Per aver dormito, non ho dormito

Dormire is the theme, and its oblique function can be highlighted by using a literally purposive per 'for, to' ~ 'as for'.

Another things that such "objects" remind me is the so-called, in Latin, Greek accusative (points b and c), also "accusative of specification". It also appears with morphologically passive, so intransitive, verbs. It has generically the meaning of 'with regard to.../as for':

• femur trāgulā ictus thigh-[N]-SG.ACC dart-ABL hit-PFV.PASS.PTCP-M.SG.NOM 'hit by a dart with regard to his thigh' = '(he) was hit in the thigh by a dart'
  • note that ictus 'hit' cannot refer to femur 'thigh' because of different agreements

In the end, I'd annotate a sentence like the equivalent to

• bake something a thorough baking

as:

• obj(bake,something)
• obl(bake,a thourough baking)

There are other probably more widespread cases of obliques expressed by a direct case, for example temporal arguments. This seems to be an extension of such constructions perhaps licensed by the etymological correlation, as any language has its own constraints about this (e.g. Latin seems to be limited to focusing "parts" of an action). And obl is naturally the nominal counterpart of advmod.

Finally, probably the occurrences of accusatives have to be seen as different phenomena whether they appear with otherwise intransitive or transitive verbs. In the first case, I don't see a problem in a verb acquiring an object (so obj), with the aforementioned exceptional constraints (e.g. it has to be corradical). In the second case, the extra object signals stricter ties with the predicate, but still operates as an oblique (so obl), as the core argument is already "occupied".

An alternative still that comes to my mind is to consider things like

• transitive, passive: הגד הגד לי כל אשׁר עשׂית = "all that you did has been told to me by telling"

as dislocations: a telling has been told me - all that you did. But maybe these examples could need some glosses to be understood better by non-Hebrew speakers :-)

It's also possible to passivize one of the objects without including the other.

Do you have some examples that could clarify this?

---

Secondary objects are a totally different issue in my opinion (and I feel the iobj label is often misunderstood or not well enough defined).

@dan-zeman

(in fact I wanted to be even more radical and get rid of the iobj label completely)

Totally agree, or at least, it should be renamed to sobj = secondoary object. The indirect term is excessively confusing in the light of many established grammatical traditions, and it also overlaps too much with the semantics of oblique.

or are they iobj like prepositional datives in Spanish?

This is for example something that should not happen.

[amir-zeldes]

Thanks for the examples, and for bringing up accusativus graecus!

I get the impression that these "internal objects" are indeed all adverbials

I could see that case being made for accusativus graecus, and Arabic naa'ib maf'uul mutlaq (just the accusative adjective, without the elliptical cognate object it modifies) could be seen as similar. I wouldn't object to those being annotated as adverbs in context, or else something like obl:xyz similarly to English (see discussion here, section 7).

However the core cases discussed here are different, and the strongest evidence of that is that they are passivizable. If we want to annotate them as nsubj:pass when they are subjects, then for me it is quite strange not to annotate them when they appear as accusative objects.

t's also possible to passivize one of the objects without including the other. Do you have some examples that could clarify this?

I mean it's possible to passivize either the cognate object or the traditional theme.

• ha-yeladim katvu et ha-mixtavim ktiva tama "The children wrote the letters a calligraphic writing" (active, transitive, with regular obj and cognate obj with qualifying manner adjective, like the full 'maf'uul mutlaq' - this is the most common type in Arabic AFAIK)
• ha-mixtavim nixtevu al-yedei ha-yeladim "The letters were written by the children" (verb is Masc. Plur. Pass., matching the letters)
• ktiva tama nixteva al-yedei ha-yeladim "A calligraphic writing was written by the children" (verb is Fem. Sing. Pass., matching the verbal noun which is now the passive subject)

It's a little hard to convey the nuance in translation, but these are not odd in Hebrew IMO. The 'preponed' reduced morphological variant (Biblical style, quoted by @mr-martian) is more rigid in my opinion, and I think it is actually grammatical in Biblical Hebrew to passivize it, keep the cognate object as subject in terms of agreement, and STILL use the theme with accusative, despite the passive verb form:

• הֻגֵּד הֻגַּד לַעֲבָדֶיךָ אֵת אֲשֶׁר צִוָּה יְהוָה אֱלֹהֶיךָ אֶת-מֹשֶׁה עַבְדּוֹ (Joshua 9:24)
• A telling has been told to your servants ACC-that which the lord has commanded his servant Moses

What's striking about this example is that the cognate verbal noun is the only NP we can interpret as the subject, since the object clause is explicitly marked as accusative and the verb is morphologically passive.

[mr-martian]

Because we definitely needed more complexity in this question, I just found an example of a cognate object of a copula.

ואברהם היו יהיה לגוי גדול and-Abraham be will-be as-a-people great (Gen 18:18)

[amir-zeldes]

Wow, that one is indeed nasty: because UD guidelines mandate that the verbal copula is a cop dependent, if we wanted the verbal noun derived from it to be obj, that relation would have to be dominated by the predicate:

cop(people, will-be/VERB) obj(people, being/NOUN)

That is definitely not ideal... My other thought for the active case is that we could use dislocated and say that the event itself is an extraposed participant in the predication. For normal cases it would look like this:

• Modern Hebrew: rutsu ritsa mehira "run a fast run!"
• dislocated(rutsu, ritsa) [run --dislocated--> a-run]

But this would destroy the ability to distinguish the passivized case from the active one, and in the passive + transitive case above you would end up with a subjectless sentence:

• הֻגֵּד הֻגַּד לַעֲבָדֶיךָ אֵת אֲשֶׁר צִוָּה יְהוָה אֱלֹהֶיךָ אֶת-מֹשֶׁה עַבְדּוֹ (Joshua 9:24)
• A telling has been told to your servants ACC-that which the lord has commanded his servant Moses
• dislocated(told, telling)
• ccomp(told, commanded)

And no subj of any kind...

Is anybody liking the dislocated idea?

[mr-martian]

I think I like dislocated more than the other suggestions thus far.

[Stormur]

I mean it's possible to passivize either the cognate object or the traditional theme.

* ha-yeladim katvu et ha-mixtavim ktiva tama "The children wrote the letters a calligraphic writing" (active, transitive, with regular obj and cognate obj with qualifying manner adjective, like the full 'maf'uul mutlaq' - this is the most common type in Arabic AFAIK)

* ha-mixtavim nixtevu al-yedei ha-yeladim "The letters were written by the children" (verb is Masc. Plur. Pass., matching the letters)

* ktiva tama nixteva al-yedei ha-yeladim "A calligraphic writing was written by the children" (verb is Fem. Sing. Pass., matching the verbal noun which is now the passive subject)

But can it happen that you have a passivised version of the first "bi-accusatival" sentence where both elements appear? And in which form do they appear? The two passive sentences that you show do not appear to be necessarily derived from the first one, for example, the second could represent the passive form of a sentence without letters. I am trying to compare this with other double accusatives, such as in English or Latin.

In Latin, both accusatives can become subjects, with a very strong preference for the human object, but then the secondary object is kept. By contrast, there are other verbs with apparently two objects, but one of them can never be made a subject if it appears in combination with another. So (adapted from attested examples):

• First type: ei-NOM (ante me) sententiam-ACC rogabantur-PASS 'they were being asked their opinions (before me)'
  • presupposes an active eos-ACC (ante me) sententiam-ACC rogabant-ACT
  • one might much more rarely find something like sententia-NOM (ante me) eos-ACC rogabatur-PASS, but probably an oblique like ab eis 'from them' would have been preferred
• Second type: magnam partem-ACC equitum-GEN flumen-ACC transiecit-ACT' moved the greater part of the cavalry across the river'
  • here the same sentence with flumen 'river' as subject is not possible. The accusative depends also on the trans in the verb, which by itself can be a preposition governing accusative. It has a kind of causative meaning.
  • But then, you could use the same verb only with the object flumen, in which case an alternation like equites-NOM flumen-ACC transiciunt-ACT / flumen-NOM (a equitibus) transicitur-PASS 'the cavalry crosses the river / the river is crossed by the cavalry'
  • So it is possible that flumen by itself can be made a subject, but never in combination with another object.

In English, things looks similar for the first type, right?

Anyway, another concern is that in these two languages (and others), these "true" double-object verbs seem to come up only with verbs of asking, giving, prohibiting (including hiding, as it were), teaching, or similar. It starts looking like a universal tendency indeed. And this is also the case for the example with tell you give. However, all the other examples (write, bake...) seem too general. They appear as different constructions to me.

Is anybody liking the dislocated idea?

No, I also think it does not correctly represent what is happening here. These arguments are well integrated in the clause, they are not "later additions".

[amir-zeldes]

can it happen that you have a passivised version of the first "bi-accusatival" sentence where both elements appear?

I think you can passivize one (making it a subject) and keep the other as accusative, though maybe semantics will make this weird for some specific combinations. What you can't do is make both into subjects, since the verb has to agree morphologically with just one of them, and they can have different gender and number.

These arguments are well integrated in the clause, they are not "later additions".

I'm not necessarily saying I'm fully behind dislocated for these, I'm basically not happy with any of the options :) but the idea was not that they are late additions, and more the fact that that label is used for 'duplicate' realizations of arguments (like "Elephants, they like peanuts"), so you could say this is a duplicate realization of the event argument (in an argument realization event semantics kind of way, where this is the 'e' argument, see Levin & Rappaport-Hovav 2005)

The Latin examples don't seem comparable to me, because they are not cognate objects. In Hebrew if we had the equivalent of "ei sententiam rogabantur", we could STILL add the cognate object and have something like:

• ei sententiam rogationem rogabantur

The thing about cognate objects is that you can always tack them on to an existing structure, and I agree that this is even possible to some extent in select English examples - but it often sounds rather unnatural and hardly ever occurs, while in Middle Eastern languages (at least Hebrew, Arabic, Coptic) it seems fairly common and much less marked.

[Stormur]

The Latin examples don't seem comparable to me, because they are not cognate objects. In Hebrew if we had the equivalent of "ei sententiam rogabantur", we could STILL add the cognate object and have something like:

* ei sententiam **rogationem** rogabantur

The thing about cognate objects is that you can always tack them on to an existing structure, and I agree that this is even possible to some extent in select English examples - but it often sounds rather unnatural and hardly ever occurs, while in Middle Eastern languages (at least Hebrew, Arabic, Coptic) it seems fairly common and much less marked.

Ow, this is interesting. And actually to me looks like another clue in favour of an adverbial treatment. Yes, a kind of internal object expressed as an argument, but not fulfilling the role of a core argument.

Maybe a more similar example in Latin are the (very rare and synchronically not really productive) forms like animum advertere lit. 'turn the own mind/soul (to smth)', so 'to notice, to take care of'. It clearly originates from a construction where animum 'mind/soul' was an effective object, but then it become a transitive construction, so that you can have a sentence like illam rem-ACC animum adverto 'I take care of that thing', with an apparent double object. But it has become so bleached with time that later it appears nearly exclusively univerbated: animadverto (to which we would give the feature Composed=Yes as the only trace of this process). But when written separately, I would probably annotate it as obl, maybe obl:arg to represent the close tie. In a sense, it is not so distant from the accusativus graecus: "I consider that thing, with regard to my mind" = "... with my mind/rationally".

Probably in Hebrew this process is made more general and "easier" exploiting the etymological relation. And as said, it looks very different from the tell-case.

These arguments are well integrated in the clause, they are not "later additions".

I'm not necessarily saying I'm fully behind dislocated for these, I'm basically not happy with any of the options :) but the idea was not that they are late additions, and more the fact that that label is used for 'duplicate' realizations of arguments (like "Elephants, they like peanuts"), so you could say this is a duplicate realization of the event argument (in an argument realization event semantics kind of way, where this is the 'e' argument, see Levin & Rappaport-Hovav 2005)

Haha, no, of course, I was not blaming anyone! :smile: I mean, I see two big cases:

• "duplication", also and especially of the systematical kind, I would treat by means of expl + contextual relation;
• "repeated equivalent argument" with the contextual relation + subtyped dislocation, a construction which in some cases borders with explanatory clauses (for which in Latin we are using conj:expl)

---

Because we definitely needed more complexity in this question, I just found an example of a cognate object of a copula.

ואברהם היו יהיה לגוי גדול and-Abraham be will-be as-a-people great (Gen 18:18)

I have actually not understood what is happening here, can I ask about a clarification? :grimacing:

[mr-martian]

Because we definitely needed more complexity in this question, I just found an example of a cognate object of a copula.

ואברהם היו יהיה לגוי גדול and-Abraham be will-be as-a-people great (Gen 18:18)

I have actually not understood what is happening here, can I ask about a clarification? 😬

This is the Hebrew version of the construction under discussion, where היו is the cognate object or whatever of יהיה. The problem is that this verb, היה, is the copula in Hebrew, so however היו attaches, it should presumably attach to לגוי.

Another idea, for the Hebrew data at least, was to wonder whether it might make sense to treat this as reduplication and so have compound:redup or some such.

[amir-zeldes]

actually to me looks like another clue in favour of an adverbial treatment

Mmm... I suspect that might be because Latin uses accusative to mark adverbials in general (a.k.a. accusativus graecus), but Hebrew doesn't, so I don't have any 'adverby' feeling from these as a speaker. And I believe that Latin accusativus graecus (or Greek for that matter) can't be passivized; these things are morphosyntactically really 'objects', except that they don't saturate the valency of the verb or require it: any verb can add them, and it has no effect on the existing acc object if any.

Another idea, for the Hebrew data at least, was to wonder whether it might make sense to treat this as reduplication and so have compound:redup or some such.

A unique subtype solves the problem of not having a name for these things, but I'd argue against a subtype for compound because this is not really a single 'word' made up of multiple morphemes (at least the modern cognate object is effectively its own noun - in the non-reduced form it can head an NP, get adjectives, etc. etc., and the same is true for Arabic). Second, there is the weird issue with passivization, where I think there is no way to claim that a passive subject forms a compound with its verb, so if we still tag those as nsubj:pass we'd be splitting up the cognate objects into two very different analyses.

[Stormur]

Because we definitely needed more complexity in this question, I just found an example of a cognate object of a copula. ואברהם היו יהיה לגוי גדול and-Abraham be will-be as-a-people great (Gen 18:18)

I have actually not understood what is happening here, can I ask about a clarification? grimacing

This is the Hebrew version of the construction under discussion, where היו is the cognate object or whatever of יהיה. The problem is that this verb, היה, is the copula in Hebrew, so however היו attaches, it should presumably attach to לגוי.

Another idea, for the Hebrew data at least, was to wonder whether it might make sense to treat this as reduplication and so have compound:redup or some such.

Might it be a cleft construction? If so, Abraham would be head with a depending copula and the rest, the highlighted clause, a clausal subject for which there exists the relation csubj:cleft! Are objects involved in this construction?

[Stormur]

actually to me looks like another clue in favour of an adverbial treatment

Mmm... I suspect that might be because Latin uses accusative to mark adverbials in general (a.k.a. accusativus graecus), but Hebrew doesn't, so I don't have any 'adverby' feeling from these as a speaker. And I believe that Latin accusativus graecus (or Greek for that matter) can't be passivized;

The preferred way for Latin to mark adverbials actually passes through ablative and prepositional phrases, and dative to some extent. It is true that many adverbs derived from adjectival/determinal bases have an accusative form, but they act as modifiers, so it is something different from what we are discussing here. The accusativus graecus is a rather marginal and "weird" phenomenon that has never really taken hold in the spoken language, and we find it in literary contexts imitating Greek (where it is probably more productive); in fact, it is something that I have found to befuddle annotators, since it does not correspond to usual schemes. Adverbials for which accusative is standard are just a subclass of temporal and locative complements. Neither of these can be passivised indeed. But neither are they "adverby" in the same sense that a manner adverb is... probably we should better define what we mean by some terms?

these things are morphosyntactically really 'objects', except that they don't saturate the valency of the verb or require it: any verb can add them, and it has no effect on the existing acc object if any.

This sounds exactly as the definition of (non-complement) obliques obl!

What I meant with the example of equites flumen transiecit ~'[he] overthrow the cavaliers the river' = '[he] made the cavaliers cross the river' is that there is the possibility to have a passive sentence where one of the same arguments becomes the subject (transicitur flumen ab equitibus 'the river is crossed by the cavaliers'), but its active counterpart (equites-NOM flumen-ACC transiciunt) is not the first sentence, for which on the contrary flumen cannot be passivised while, being an oblique, stays the same in all versions: equites-NOM flumen-ACC transiciuntur 'the cavaliers are being made to cross the river', vs *flumen-NOM equites-ACC transicitur (not admissible) .

[mr-martian]

With regard to the specific Hebrew verbal form that spawned this discussion, Gesenius' Hebrew Grammar describes the form as a general adverbial clause.

Jeremiah 9:23 בזאת יתהלל המתהלל השכל וידע אותי "in-this boasting the boaster understanding and-knowing me"

Here we have 3 such infinitives, and if we follow Gesenius, they're actually all the same construction (which I guess would be advcl or maybe a subtype) and it just happens that the most common combination is that the main verb and the subordinate verb are the same root.

(Also by my count this is the 9th primary relation proposed in this discussion.)

[amir-zeldes]

Thanks for sharing this interesting example! But I actually don't think this is the construction we've been talking about. This seems to be a cognate subject and not an object ("the boaster shall boast", not "he shall boast a boasting"). The two further objects are indeed reduced masdar forms, but they are not cognates: the boaster shall boast (about) understanding and knowing me.

As for advcl, I don't think cognate objects are oblique, so even if you'd like to consider them clausal, then the label should be ccomp (or when they are passivized, csubj:pass). Maybe the most straight-forward solution is to accept them as normal passive subjects under passivization (since there can only be one in the passive), but subtype them when they are serving as objects, so something like obj:cognate or obj:internal or similar.

[nschneid]

@amir-zeldes is this actually resolved?

[amir-zeldes]

The Hebrew project I work with ended up going with obl:npmod, incl. for cases explicitly marked as accusative and in the case of intransitive verns, even without a second object. For transitives it's tricky because it's hard to tell if a cognate object might be referential or rather filling this empty automatic slot. If they occur as passive subjects (I'm not sure we have such cases in the data), then they would be annotated as nsubj:pass, since the subject is guaranteed to be unique. Not sure if this is exactly 'resolved', but since some policy needed to be adopted, this is what we went with.

[mr-martian]

I'm still not convinced that the construction in the Ancient Hebrew data is nominal and had considered changing the instances in Ancient_Hebrew-PTNK to dep:infabs in the absence of a consensus here (since explicitly marking it as unknown seems at least better than the present advmod).

[amir-zeldes]

@mr-martian I agree it's all murky and non-optimal, but in the end we opted for a nominal interpretation since it allows us to unify the Biblical kind with the modern kind in "ratsti ritsa" for "I ran a run", which is also equivalent to how Arabic does it. I think etymologically, all of these things actually are nominal, and in Arabic the two sort of merge, because the masdar is still very clearly a case-bearing verbal noun. In Hebrew, you can get 'et' on a definite cognate object, which also pleads for a nominal reading, unless you want separate treatment for the type "halox halaxti". But even this one, etymologically speaking, is a noun...

[mr-martian]

Upon further consideration, I think I can accept obl:npmod, and in the copula case I'll do obl:npmod to the predicate and so, indeed, this issue is resolved.