Verbal nouns in Georgian

[paulmeurer]

In Georgian, verbal nouns exhibit nominal syntax (as they should; in contrast to inifinitives in other languages).

On the other hand, they are often used more infinitive-like in argument positions that alternatively could be filled by a subclause with a finite verb (often in the Subjunctive), e.g.

მინდა ბიჭების ნახვა / minda bič̣ebis naxva
I-want boys-Gen seeing-VN
“I want the seeing of the boys”

versus

მინდა ბიჭები ვნახო / minda bič̣ebi vnaxo
I-want boys-Nom see-Sub
“I want (that) I see the boys”

In the second case, the analysis is clear: the subclause will get the relation ccomp, and ‘boys’ is obj of ‘see’ (ergative syntax).

In the verbal noun case however, there are two possible analyses: the more superficial nominal one (which is easier to get right with a rule-based parser), where 'boys-Gen' is treated as an nmod:poss of the (nominal) VN, the relation to the main verb being obj, and a verbal analysis, where the relation to the main verb is ccomp, and the pos of the VN is VERB.
I am not sure how to connect the argument of the VN. Should it be obj? Or should it be something like nmod:obj? Or even obl:obj? The rationale would be that all core arguments of the verb are demoted to oblique status (like obl:agent for passives, or, in the Georgian case, obl:iobj, the indirect object in the perfect tense).

In a more complex (slightly contrived) example

წიგნის მიცემა გოგოსთვის ბიჭის მიერ სასურველია / c̣ignis micema gogostvis bič̣is mier sasurvelia
book-Gen giving-VN girl-for boy by is-desirable
“the giving of the book to the girl by the boy is desirable”

there are three core arguments that are demoted to oblique status, which would respectively be annotated as obl:obj, obl:iobj, obl:agent.

Does this make sense?

[dan-zeman]

I think I would favor the nominal analysis. The verbal noun is tagged NOUN (but with the feature VerbForm=Vnoun to distinguish it from ordinary nouns), it is attached as obj to the wanting verb (assuming that verb is not nominalized), and nominal arguments of the verbal noun, if any, are attached to it as nmod. Subtyping the relation as nmod:poss does not seem necessary to me (if it just signals that the argument is in genitive, then it is probably enough that there is Case=Gen in the features).

[paulmeurer]

Ok, good. This is also the easier solution. nmod:poss seems not only unneccessary, but also wrong to me, since the genitive doesn’t mark a possessor in this case. The other demoted arguments (obl:agent, obl:iobj above) should then also be nmod, and a special subtype (e.g., nmod:agent) would be unneccessary I guess. If somebody wants to retrieve the verbal argument structure of such constructioins, the case/postposition marking should suffice to do that.

There is a similar case where an attributively used participle can have an agent; here too, nmod alone will then do.

Future participles (in the adverbial (essive)) case can be used to express a final clause; here, I am not sure what to do. Those constructions look verbal to me, even though also here, the verbal arguments are demoted to obliques, with the same case marking as in the VN case. E.g.:

წერილის დასაწერად დაჯდა / c̣erilis dasac̣erad daǯda.
letter-Gen to-write-FutPart he-sat-down
“He sat down (in order) to write a letter”

If interpreted nominally, the participle would be a simple obl of the main verb. That would somehow obscure the verbal force of the participle, but, on the other hand, be in line with the treatment of the verbal nouns, and the features Vform=Part Tense=Fut would again distinguish them from ordinary adjectives. But at least an obl:final or similar would seem appropriate to me.

[dan-zeman]

There is a similar case where an attributively used participle can have an agent; here too, nmod alone will then do.

I suppose the participle would be treated as an adjective; then we would use obl instead of nmod. Unless it is a nominal where the head noun has been elided and the adjective has been promoted to the head position; then the modifier would modify a nominal and would be labeled nmod.

obl might also be the easiest solution for the future participles, although there may be other justifiable options. (I can imagine that the language-specific guidelines will say that core arguments use case coding A with finite verbs and case coding B with future participles.)

[paulmeurer]

Ah, yes, thanks, I was confused about obl and adjectives. Participles are treated as adjectives (except in periphrastic tense forms). Then obl would work for both cases (attributively used participles etc., and future participles).

[Stormur]

I have two questions:

• can this verbal noun be formed from each verbal root in the same way (barring inflection classes, phonologic adjustments, etc)?
• can we see a pattern in the core arguments of the verbal noun? In particular, I wonder if a genitive marking always corresponds to its subject.

I think that the main issue is the identification of an argument pattern, which might be slightly different between nominal and "finite" forms. For example, let's consider Latin amor'love': it is related to amo 'I love' by means of a not so improductive process. When I look at things like amor matris, where matris is genitive of mater 'mother', I can understand if it is the love by the mother (mater = nsubj as in mater amat) or the love for the mother (mater = obj as in amo matrem) only contextually. That is, the genitive here really is the generic relational case as with all other nouns, no argument is implied. Therefore, this is a very strong argument to treat amor as a NOUN and not as a verbal form (VERB).

The regularity might also be in relation with a construction such as the one with მინდა. If there is a regularity, I would strongly favour the verbal interpretation. In the end, this is exactly what we do with cases such as the "accusative with infinitive" in Latin, where the infinitive is a traditional name for verbal noun, it is a regular and productive form, and the subject is regularly expressed in the accusative.
If there is no regularity, then go for the completely nominal annotation.

Actually, this might be complicated by the orientation of these nominal forms (and of course by ergativity). I discussed Latin, where they are very strongly oriented (passive or active), but it might not be the case for Georgian (this paper by Haspelmath about "passive" participles is very interesting in that regard, and it also cites Caucasian languages which might be areally relevant).

Future participles (in the adverbial (essive)) case can be used to express a final clause; here, I am not sure what to do. Those constructions look verbal to me, even though also here, the verbal arguments are demoted to obliques, with the same case marking as in the VN case. E.g.:

წერილის დასაწერად დაჯდა / c̣erilis dasac̣erad daǯda. letter-Gen to-write-FutPart he-sat-down “He sat down (in order) to write a letter”

(By the way, I think that here we cannot speak of future tense, but of prospective aspect.)

In fact, when I see such pervasive contructions in Georgian, I would strongly lean toward a uniform verbal representation. I do not think it is so strange to see a genitive, i.e. relational case as that of these "implicit core arguments". I can recall very similar constructions for Mongolian and other related languages, it is a widespread thing.

[paulmeurer]

Those are interesting questions.

For the majority of the Georgian verbs, the verbal noun is formed more or less regularly, it is a productive process.

There is a regular pattern in the coding of the core arguments, but it is an ergative pattern: the genitive codes what would be the direct object in the corresponding transitive finite construction, and the subject in an intransitive construction. The transitive subject is marked by a postposition (Gen + მიერ). However, since the verbal noun has no voice opposition, it would not be possible to equate the genitive-marked actant with the object; it could as well be the subject of the passive. So genitive codes S/O in the Dixonian sense, and never A. How would one express that in UD?

Things are a bit more subtle than that, however. There are non-obligatorily transitive verbs like თამაში “to play”, where the genitive is used to mark the transitive object, but only if the subject is overtly marked, otherwise it marks the intransitive subject (but again, this fits the ergative pattern.)

There is an exceptional class of verbs where this marking scheme doesn’t fully apply; those are the so-called inverted verbs, where the subject is marked by the dative and the object by the nominative. They are in some sense weakly transitive, as there are very few subject tests that apply, and translations to other languages can go both ways (ich mag/mir gefällt). Their verbal nouns are often irregular and have a much more nominal feel to them than those of the more regular verbs.

One of them is your example, დედის სიყვარული “amor matris”, which also in Georgian can be interpreted both ways.

So the question is where to draw the line. I would of course like to go for a uniform solution, also because I am using a rule-based parser which should get it right as often as possible. I see strong arguments for the verbal interpretation, based on the regular verb classes, and because it would allow to keep the parallelism between, e.g., მინდა (“I want”) + verb in the subjunctive, and მინდა + VN, with ccomp vs. xcomp. But perhaps cases like დედის სიყვარული don’t warrant a verbal interpretation, and also the choice of relation label for the S/O case is a problem.

The solution I have adopted for now is to analyse the verbal noun nominally, but to use qualified labels, nmod:arg for the S/O case, and nmod:agent for the A case. (Indirect objects, which are unambiguously marked with the postposition -თვის, are coded as nmod:iobj). This coding scheme easily extends to participles. (Future participle is the traditional term; Alice Harris in her Georgian Syntax even proposes a reanalysis as an infinitive, at least when used in the adverbial case).

In the case of დედის სიყვარული, the label would also be nmod:arg, which is vague enough to cover both interpretations.

Since my gut feeling for Georgian only goes so far, I will discuss some of the edge cases with my native speaker friends.

Thanks for the pointer to the Haspelmath article, this looks interesting, but I haven’t had a chance to read it yet.