Differential Object Marking*
1. Introduction There are a number of languages with overt morphological case-marking which do not mark all their objects uniformly. This paper is going to deal with languages that mark only a subset of their objects overtly and another subset is never marked or is optionally marked morphologically. Following Aissen (2002 and references therein) I will call this phenomenon Differential Object Marking (DOM). DOM, nevertheless, takes different forms in the different languages. The background on which it is based is at least threefold. Aissen (2002) discusses phenomena where DOM stands on semantic and pragmatic grounds. In the languages that are relevant from that aspect it is the pragmatic characteristics of the object that decide whether it is case-marked obligatorily or optionally, or case-marking is excluded. (1)
a. Sinhalese: object case-marking is optional, but only animate referring objects may be case marked b. Hebrew: object case-marking is obligatory, but only definite objects are case-marked c. Romanian: object case-marking is obligatory for animate pronouns and proper nouns, optional for other sets and excluded for yet another.
In Hungarian it is the morphology of the object that determines overt accusative marking. (2) Hungarian Object case-marking is optional in the presence of the 1st or 2nd person possessive suffix, but it is compulsory with the 3rd person possessive suffix.
students in Budapest in 2004. I am grateful to all the participants for their comments – especially my supervisor, Mark Newson, who helped enormously with starting the research, and also gave food for thought for later on. The usual disclaimers apply, of course.
The Even Yearbook 7 (2006), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218–8808, http://seas3.elte.hu/delg/publications/even, © 2006, Gergely Kamper
Differential Object Marking 2
In Chichewa it is the word order that interacts with accusative case-marking. (3) Chichewa Object case-marking is obligatory when the verb and the object are not adjacent, it is optional when the Verb and the object are adjacent (see Baker 1986 for reference). In this paper I will first discuss Aissen’s (2002) views on differential case marking of direct objects with certain pragmatic characteristics and I will also introduce her analysis of the phenomenon within the framework of Optimality Theory. I will then proceed to instances of DOM referred to in (2) and (3), and try to formalize these with the help of similar constraints.
2. Pragmatic DOM (Aissen 2002) 2.1. Varieties of pragmatic DOM Although pragmatically based DOM is surprisingly widespread in disparate languages of the world, it is far from uniform. For example, in Pitjantjatjara, a Pama-Nyungan language from Australia pronouns and proper name objects are always case-marked 1), while inanimate ones do not 2) unless they are objects of worship (5e).
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 3
(5) Malayalam a. avan ku iye a iccu he child.ACC beat-PAST ‘He beat the child.’ b. avan oru patuvine vaa i. he a cow.ACC buy-PAST ‘He bought a cow.’ c. aan tee a vaa i. I coconut buy-PAST ‘I bought some coconut.’ d. avan pustakam vaayiccu he book read-PAST ‘He read the book.’ e. ava ilpatte araadhiccu she statue.ACC worship-PAST ‘She worshipped the statue.’ These sentences exemplify languages in which one semantic feature, namely either definiteness or animacy determines which of the direct objects will be overtly case marked and which will not. As we will see later on, there are several disparate languages where DOM works along these same lines. However, it is not always so. In many other languages both characteristics simultaneously play a role in the shaping of DOM, and the presence or absence of overt case-marking depends on both the animacy and the definiteness of the object. 2.2. Aissen’s OT-model for Differential Object Marking 2.2.1. Markedness Reversal The notion of DOM originates from functional grammarians. formalization of their ideas is shown in (6):
Aissen’s
(6) The higher in prominence a direct object, the more likely it is to be overtly case-marked. The question now arises how the prominence referred to in (6) is determined. Based on (4) and (5) and similar examples from a number of disparate languages the prominence scales in (7) can be established:
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 4
(7) Prominence scales: a. Animacy: Human > Animate > Inanimate b. Definiteness: Personal pronoun > Proper name > Definite NP > Indefinite specific NP > Non-specific NP Empirically, what we see in languages with DOM is that, if a direct object at some point on one of these scales is case-marked, all other direct objects higher on the same scale will be case-marked as well. Lower ranked objects are not necessarily case-marked. As (4) and (5) show some languages use one scale whereas others use the other. There are also languages, like for example Romanian, in which both scales are relevant. However, (6) holds in all cases. Languages differ in which scale they use (or if they use both scales), and also in where the point above which accusative case-marking is optional or obligatory is. One might wonder what motivation lies behind these scales, why they are relevant and why they are ordered the way they are. Aissen’s claim is that intuitively the objects that should primarily be case-marked are the ones that resemble subjects to a greater extent. Although it is not a question of disambiguating between subjects and objects, it can be argued that the prominence scales of (7), which show the order of object markedness, are the exact opposites of the prominence scales relevant for the markedness of subjects. In other words, what is unmarked for an object is exactly what is marked for a subject.1 Comrie (1979:19) summarizes the phenomenon like this: …in natural languages, certain grammatical relations tend to be characterized by certain features, in particular [that] subjects tend to be definite, animate, and topic (thematic); while direct objects tend to be indefinite, inanimate, and rhematic.
This phenomenon has been called Markedness Reversal in the literature (see Aissen 2002 for references). If all this holds then it clearly predicts what is justified by the empirical facts: if for example a definite NP is always casemarked when it appears as a direct object, every NP ranked higher on the definiteness scale (i.e. proper names and personal pronouns in this case) will be case-marked as direct objects as well because the more an object resembles a subject, the more necessary it is for it to be overtly marked as accusative. Let us now turn to the formalization of these observations.
1
A clear distinction must be made here between semantic markedness which is used for establishing the ranking of an NP on the animacy or definiteness scale and morphological markedness, which, on the other hand, is what is or is not expressed by overt case-marking.
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 5
2.3. Harmonic Alignment To derive the proper violable constraints within an Optimality Theoretical framework Aissen (2002) makes use of the notion of Harmonic Alignment. Harmonic Alignment was first introduced for phonology (see Aissen 2002 for references). The basic idea is that there are two scales one of which is binary. This is called Relational Scale. The high-ranking element of the binary scale is associated with the elements of the other scale left to right, and the lowranking element of the binary scale is associated with the elements of the other scale from right to left. This process produces two harmony scales in which the leftmost elements are the most harmonic combinations whereas the rightmost ones are the least harmonic. Let us see how Harmonic Alignment helps us formalize the phenomenon of Differential Object Marking. 2.4. Deriving Constraints 2.4.1. Animacy If we want to express the Markedness Reversal that has been discussed above by using Harmonic Alignment we need to establish two scales that may be aligned with respect to each other. The first one is the Relational Scale, which includes the grammatical functions of subject and object, whereas the members of the animacy scale of (7a) should appear on the second one. The two scales are shown in (8): (8)
a. Relational Scale: Subject > Object b. Animacy Scale: Hum(an) > Anim(ate) > Inan(imate)
If Harmonic Alignment is applied to these scales, that is the higher ranking element of the relational scale is associated with the elements of the animacy scale from left to right and the low-ranking one is associated with the same elements from right to left, the harmony scales of (9) will be achieved: (9)
a. Su/Hum > Su/Anim > Su/Inan b. Obj/Inan > Obj/Anim > Obj/Hum
By reversing these hierarchies we can then derive the markedness constraints that work against the above combinations of grammatical functions and members of the animacy scale. In Aissen’s system the ranking of these constraints is universal as the scales themselves are universal.
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 6
(10) a. *Su/Inan *Su/Anim *Su/Hum b. *Obj/Hum * Obj/Anim *Obj/Inan The harmonic scales in (9) show that an object is least marked when it is inanimate and is most marked when it is human, whereas the constraint ranking in (10) indicates that for this very reason human objects are to be avoided more than animate objects or inanimate ones. 2.4.2. Definiteness We can give the Definition Scale of (7b) the same treatment with respect to the same binary scale as was used in 2.4.1. (11)
a. Relational Scale: Subject > Object b. Definiteness Scale: Personal pronoun > Proper name > Definite NP
Indefinite specific NP > Non-specific NP
By applying Harmonic Alignment again we can establish the harmony scales of (12): (12) a. Su/Pro Su/PN Su/Def Su/Spec Su/NSpec b. Obj/NSpec Obj/Spec Obj/Def Obj/PN Obj/Pro Following the same pattern as we did in connection with the animacy scale the constraint rankings of (13) are established. (13) a. *Su/NSpec *Su/Spec *Su/Def *Su/PN *Su/Pro b. *Obj/Pro *Obj/PN *Obj/Def *Obj/Spec
5. Iconicity and Economy 2.5.1. Forcing overt case-marking Having derived the rankings of these constraints one might be led to believe that in the various languages of the world the more marked of these associations are avoided. However, this is not the case. Languages do not
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 7
avoid any of these combinations, but those with DOM tend to differentiate between them by using morphological case-marking. DOM is realized by the overt accusative marking of marked combinations, which contrasts with the zero case marking of unmarked ones. As Aissen puts it: ‘The morphology of DOM then is privative: zero expression contrasts with audible expression.’ (2002:11). Aissen assumes that inputs do not contain any specification for morphological case, but it is added by GEN. Consequently the absence or presence of morphological case in the candidates cannot be due to inputoutput faithfulness. Thus if we want to formalize the DOM phenomenon in an Optimality Theoretical framework, we need a constraint that penalizes the absence of overt case-specification in these structures. In Aissen’s system this constraint is *Øc. (14)
penalizes the absence of a value for the feature CASE.
We now have a violable constraint that forces overt case-marking given the proper ranking. The next step is to link it to the constraints established in (10) and (13), which characterize the relative markedness of objects. Aissen suggests the tool for doing this should be Local Conjunction (Smolensky 1995). Local Conjunction is the combination of two different constraints into one. To put it simply, this newly formed constraint is violated in a domain if both of the original constraints are violated, and is universally ranked higher than either of its two components. Aissen’s assumption is that the Local Conjunction of *Øc with the hierarchies of (10) and (13) will preserve the rankings of those hierarchies thus yielding the final hierarchies in (15), which characterize the zero case-marking of objects. (15)
Local Conjunction of *Øc with the object related constraint hierarchies Local conjunction of *Øc with hierarchy on object animacy
Local conjunction of *Øc with the hierarchy of object definiteness
These constraints will favour overt case-marking for marked associations and thus we could say that they are iconicity constraints inasmuch as iconicity is something that forces languages to morphologically mark elements that are semantically marked. The ranking in (15) ensures that if an object type is
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 8
case-marked in a language all other object types higher up on the relevant scale will be case-marked as well. 2.5.2. Getting rid of overt case-marking If we left it at that, all objects, marked or unmarked semantically, would be marked morphologically. However, this is exactly what does not happen in languages with DOM. Therefore we must find an economy constraint that penalizes overt case-marking. Aissen suggests that this constraint be
(16)
penalizes a value for the morphological category CASE.
absence of overt case-marking of every object type appearing in the dominated constraints. Thus Differential Object Marking seems to occur due to a tension between iconicity and economy. In every structure there is a constraint that requires overt case-marking of the direct object, whereas another is working against it. In some languages only one of the animacy and definiteness scales plays a role in the realization of DOM, but in others both scales are relevant. In the following section we will see a few examples for all three. 2.6. Pragmatic DOM in the different languages 2.6.1. One Dimensional DOM – animacy or definiteness If we place *STRUCc somewhere into the constraint hierarchy of object animacy we can account for DOM in languages where the presence or absence of overt case-marking of the direct object hinges on the animacy of the object. Where some of the languages relevant here interpolate *STRUCc into this hierarchy is shown in the table in (17).
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 9
(17) .
. . .
. . . . .
. . . .
⇐ *STRUCc (Kalkatungu; no objects case-marked) ⇐ *STRUCc (Yiddish; only some human objects case-marked)
⇐ *STRUCc (Ritharngu; all human and some animate objects case-marked) ⇐ *STRUCc (Malayalam; all animate objects case-marked) ⇐ *STRUCc (Bayungi; all animate and some inanimate objects case-marked) ⇐ *STRUCc (Written Japanese, Dhalandji; all objects case-marked)
As we have seen, Malayalam is a language where DOM is affected by animacy. The examples in (5) are repeated here as (18) for convenience’s sake. (18)
Malayalam a. avan ku iye a iccu he child.ACC beat-PAST ‘He beat the child.’ b. avan oru patuvine vaa i. he a cow.ACC buy-PAST ‘He bought a cow.’ c. aan tee a vaa i. I coconut buy-PAST ‘I bought some coconut.’ d. avan pustakam vaayiccu he book read-PAST He read the book.’ e. ava ilpatte araadhiccu she statue.ACC worship-PAST ‘She worshipped the statue.’
(18) shows us that Malayalam ranks *STRUCc above *Obj/Inan & *Øc.
direct object is inanimate. (18) is also telling in the sense that what is considered animate can vary among cultures. In Malayalam objects of worship are taken to be on a par with animates as far as overt case-marking is concerned. If (17) is correct and the ranking of the locally conjoined constraints therein is indeed universal then it predicts that there will be no languages in which only semantically unmarked objects will be morphologically case-
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 10
marked. That is no language will overtly case-mark inanimate direct objects without case-marking animates and humans as well. Aissen herself notes that she is not aware of the existence of any such language. A problem that arises in (17) is the ranking of *STRUCc in Sinhalese. In this language inanimate objects are never overtly marked, and human objects always receive the accusative. Animate objects are case-marked optionally. Within the framework of Optimality Theory there are many ways to account for optionality, but from the aspect of DOM it will suffice to apply the simplest solution. The claim is that *STRUCc reranks with *Obj/Anim &
it dominates *Obj/Anim & *Øc, and one where it is dominated by it. Either evaluation will result in a different winning candidate. The situation is somewhat simpler when we examine the interpolation of *STRUCc into the constraint hierarchy of object definiteness inasmuch as there is no language dependent variation as to which category a direct object belongs to. (19) .
. .
. .
. .
. .
⇐ *STRUCc (Kalkatungu; no objects case-marked) ⇐ *STRUCc (Catalan; only pronoun objects case-marked) ⇐ *STRUCc (Pitjantjatjara; only pronoun and PN objects case-marked) ⇐ *STRUCc (Hebrew; only pronoun, PN and definite objects case-marked) ⇐ *STRUCc (Turkish; all objects case-marked except non-specific ones) ⇐ *STRUCc (Written Japanese, Dhalandji; all objects case-marked)
In a language where DOM is based on the definiteness of the direct objects the presence or absence of overt accusative marking will hinge on whether the relevant constraint dominates *STRUCc or is dominated by it. An example is Hebrew, where *STRUCc outranks the two universally lowest ranked constraints of the definiteness hierarchy, therefore specific indefinite and non-specific indefinite objects will not be overtly case-marked whereas definite, proper name, and pronoun objects will get overt morphological case-marking.
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 11
(20)
Hebrew a. Ha-seret her’a ’et-ha-milxama. the-movie showed ACC-the-war ‘The movie showed the war.’ b. Ha-seret her’a (*’et-) milxama. the movie showed (*ACC-)war ‘The movie showed a war.’
2.6.2. Two Dimensional DOM Other languages use both the animacy and the definiteness scale to decide whether an object will be overtly case-marked or not. For example in Romanian only objects that are animate referring and specific are overtly casemarked. Aissen (2002) calls this phenomenon two dimensional DOM. Languages belonging to this category include Romanian, Persian, Hindi and 12th century Spanish. In order to achieve a pattern for two dimensional DOM Aissen combines the animacy and the definiteness scales. The result is shown in (21). (21)
(21) predicts that DOM will spread from the top of the table towards the bottom with human pronouns being the most marked objects and inanimate non-specific ones being the least marked ones. Aissen claims that (22) is a true statement as to the structure in (21):
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 12
(22)
If in (21), a dominates b, then: a. if an object of type b may be case-marked, then all objects of type a may be case-marked. b. if an object of type b must be case-marked, then all objects of type a must be case-marked. c. if no object of type a can be case-marked, then no object of type b can be case-marked.
We must note though, that the combinations do not have fixed rankings horizontally. Only a combination higher up vertically outranks the ones below it. Now we need to turn this hierarchy into a ranking of constraints. To derive the relevant constraints Aissen again turns to the tool of Local Conjunction. Firstly, we conjoin the hierarchy on object animacy (10b) to the hierarchy on object definiteness (13b). The resulting hierarchy will correspond to (21). The highest ranked constraint is *Obj/Pro & *Obj/Hum, and the lowest ranked is *Obj/NSpec & *Obj/Inan. A further step is to conjoin these newly formed constraints to *Øc, thus linking it to case-marking phenomena. A language with two dimensional DOM is Hindi. Hindi requires overt case-marking of all human objects except indefinite non-specific ones. Overt marking is optional for inanimates, but only if they are definite. (23) introduces the constraint ranking based on the hierarchy in (21) with respect to the relative ranking of *STRUCc in Hindi. (23)
The constraints above the upper line outrank *STRUCc, which results in the overt case-marking of the object types appearing therein. The constraints below the lower line are dominated by *STRUCc and therefore the object
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 13
types in them will never be case-marked. There are, however, a number of constraints between the two lines as well. With these *STRUCc reranks, which means that the object types of these constraints will be case-marked optionally (cf. Sinhalese in (17)). Two dimensional DOM works very much along the same lines in other languages as well. There is a set of objects with obligatory and another with optional case-marking, whereas a third set is never marked morphologically.
3. Morphological DOM 3.1. Accusative marking in Hungarian Aissen’s system only gives an account of Differential Object Marking in languages in which it is based on semantic or pragmatic grounds, namely on the animacy and/or the definiteness of the object. However, let us take a closer look at data from Hungarian in (24). (24)
Hungarian a. Felvettem a kalapom(at). put-on-PAST-1s the hat-my.(ACC). ‘I put on my hat.’ b. Megetetted a kutyád(at). feed-PAST-2s the dog-your.(ACC) ‘You fed your dog.’ c. Megetettük a fiunk(at) feed-PAST-1pl the son-our.(ACC) ‘We fed our son.’ d. Elvittem Jánosom(at) az állatkertbe. take-PAST-1s John-my.(ACC) the zoo-SUFF ‘I took my John to the zoo.’ e. Téged(et) választalak. you.ACC choose-1s ‘I choose you.’ f. Megcsókolta a *barátn je2 /barátn jét. kiss-PAST-3s the girlfriend-his/girlfriend-his.(ACC) ‘He kissed his girlfriend.’
In Hungarian the potential absence of overt accusative case clearly hinges on the presence or absence of the 1st or 2nd person possessive suffix. Should the direct object bear either of these, overt accusative case-marking is optional.3 2
Grammatical in the sense ‘His girlfriend kissed him’. Note that in a number of cases, although case-marking is clearly optional, most speakers tend to find one of the options degraded, especially with personal pronouns and in the plurals. In 3
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 14
At first glance one might think that this has to do with the definiteness scale, as these objects are clearly placeable thereon. We may even be tempted to establish its own category. However, this clearly does not work. Let us recall (6) repeated here as (25). (25)
The higher in prominence a direct object, the more likely it is to be overtly case-marked.
By (25) then Hungarian direct objects would have to be placed at the bottom of the definiteness scale, as all other objects are obligatorily case-marked. It could hardly be claimed that a nominal like that is placed below non-specific indefinite NPs on a scale like (7b). Note also that first and second person personal pronouns bear the same suffix when in the accusative (independent of the accusative suffix itself), although in these cases this suffix does not mark possession. Therefore we must turn to other solutions if we want to find an explanation as to what happens to accusative case-marking in these sentences. 3.2. The Constraints In these instances of DOM it is an object with a 1st or 2nd person possessive suffix that makes overt case-marking optional. Still going along the same lines as Aissen did with pragmatic DOM, we can then claim that for possessed objects4 it is an unmarked situation to appear in a sentence as direct objects, therefore case-marking has a tendency to ‘forget’ about them. On the other hand the 3rd person possessive suffix does not trigger any such process. We can then establish the following scales5 (following Aissen 1999 and references therein I will call the 1st and 2nd persons local persons): (26)
a. Subject > Object b. Non-Locally possessed (NLPos) > Locally possessed (LPos)
If we apply Harmonic Alignment to these scales, the hierarchies in (27) will be established:
everyday speech téged is clearly preferred over tégedet in sentences like (24e), and fiunkat is much more frequent than fiunk in sentences like (24c). 4 For the sake of simplicity I will call these objects possessed, although note that the relevant suffixes do not always mark possession semantically (cf. 3.1.). 5 Of course such scales stand on much more solid ground if they are based on data from a number of languages. I now understand that a very similar process occurs in Finnish This is certainly an area for further research.
The Even Yearbook 7 (2006), © Gergely Kamper
Differential Object Marking 15
(27)
a. Su/NLPos > Su/LPos b. Obj/LPos > Obj/NLPos
In turn we can derive the constraints in (28): (28)
a. *Su/LPos > *Su/NLPos b. *Obj/NLPos > *Obj/LPos
As what happens on the surface is very similar to what happened in the case of pragmatic DOM we can claim that Aissen’s two basic constraints, *Øc and