Ditransitives and Double Accusative Constructions

[EricaBiagetti]

Ditransitives and Double Accusatives in the typological literature:

Ditransitive constructions are constructions consisting of a ditransitive verb, an agent argument (A), a recipient-like argument (R), and a theme-like argument (T) (Malchukov et al. 2010).

cf. English Mary (A) gave John (R) a pen (T).

Three alignment types are attested:

1. Indirect alignment: T = Patient ≠ R
2. Secundative alignment: R = Patient ≠ T
3. Neutral alignment: T = R; in this case we talk of Double Accusative Construction (DAC).

Double Accusatives according to UD Guidelines:

In case a verb requires three core arguments, the UD annotation scheme assigns the role of ‘object’ (label obj) to the noun phrase that is most "directly affected" by the state of affairs brought about by the verb; the additional argument is labeled as ‘indirect object’ (iobj). The UD guidelines further specify that, in languages distinguishing morphological cases, the object is often marked by the accusative, whereas the indirect object takes most commonly the dative.

If the third participant is not a core argument, it can be labelled as:

• Oblique-argument (obl:arg): though syntactically not core, it is a required argument in the valency of the verb
• Oblique (obl): non-core adjunct

Do the UD Guidelines hold for all languages?

UD Guidelines for DACs may be difficult to apply if we cannot rely on speakers' intuition. Take for instance the Vedic Sanskrit verbs yāc- and prach-, both meaning 'ask, ask for, seek'.

• With theses verbs, both R and T arguments can take the accusative, and thus case marking does not help determining which argument should be labeled as obj and which as iobj.
• Passivization does not help either: both R and T can be passivized with prach- (Hettrich, 1994), while yāc- is only attested in the active.

Thus, deciding which argument to label as the direct object based on its similarity to the patient does not seem to be the best option with these verbs.

Solution:

• Since R is always encoded by a nominal in the accusative, whereas T can be encoded in different ways (noun/pronoun in the accusative, infinitival dative, complement clause, direct speech), it seems more safe to label R as obj and T as iobj when they are both realized as nominals in a sentence.
• When only one of the two arguments is expressed, it takes the relation obj.
• Finally, when T is encoded by a subordinate clause or by direct speech, it takes the relation ccomp (complement clause).

What about DACs in other historical languages?

[pkocharov]

In Classical Armenian, the double accusative coding is used in a factitive construction 'X makes Y into Z', cf. arar z=mez (ACC) t῾agawors (ACC) "he made us into kings" (Rev. 1:6). The causee (Y) is an R-like argument that is directly affected by the causer (X) and therefore seems to be most suitable for the obj label, while the nominal part of the factitive predicate should then be labelled as iobj. This distribution agrees with the coding of R as obj in the Vedic example above.

Would it make sense to consider a generalization according to which whenever R is promoted to the DO coding on the case hierarchy (as compared to the coding of a default ditransitive construction), it tends to express the most affected participant and takes the relation obj?

[amir-zeldes]

For double objects of the classic ditransitive kind, including verbs of giving, communication verbs and lexicalized causatives of the type "feed X Y", where both arguments are marked accusative, the standard UD analysis is to make the passivizable object be the obj and the other one be iobj. If both objects are passivizable, some other criteria should be used to determine which one is more 'core', which receives the obj label, and I believe if there are absolutely no morphosyntactic criteria distinguishing them, then the more 'proto-patient'-like object (in the sense of Dowty's "Thematic Proto-roles" paper) should be made obj.

However for 'appointing' verbs, such as "appoint X Y" or "call X Y", the standard UD analysis is different, even if both are marked accusative (e.g. in Classical Arabic for عين 'appoint'). In these cases, UD interprets the verb as implying a secondary predication, in which the subject is shared between the primary predication's object and the secondary predication's subject, which is canonically annotated as xcomp. The secondary predication can then be annotated using the enhanced dependency nsubj:xsubj shown here in blue: