Chapter 5From Linguistic Cycles, Elly van Gelderen, in progress
The Dependent Marking Cycles: Case[1]
12 January2010
This chapter on dependent marking and Case has been the hardest to write and is probably the most controversial in the book. As Butt (2006: 5) puts it, “no theory can honestly claim to have `the answer’ as to why case works the way it does cross-linguistically". The chapter is an attempt to deal with Case in the same way as I have with agreement, through Feature Economy. I will argue that Case derives from deixis, actual deixis for the semantic cases and grammaticalized deixis (e.g. [uT] in previous chapters) for the grammatical cases.
Of the three main ways in which languages express grammatical and other relations, word order, head-marking (or agreement), and dependent marking (or Case), the latter two are very prone to grammaticalization. In chapters 2 to 4, we have seen head-marking develop when pronouns are reanalyzed as subject and object markers on the verb. If the subject and object (pro)nominals have special markings for grammatical function, this is often lost in the reanalysis to verbal agreement (since `light’ elements start this cycle). Thus, the increase in head-marking of subjects and objects may result in a loss of or change in dependent marking, as has happened in the modern Romance (see Cennamo 2009) and Germanic languages. The Agreement and Case Cycles are therefore not completely independent of each other, although Case and agreement themselves may be (cf. also Baker 2009). Dependent-marking arises through grammaticalization as well, mainly of deictic markers, as we'll see in this chapter and the next. It is a lot less `well-behaved’ than head-marking.
Dependent marking on a nominal is often referred to as Case (and I will use both Case and dependent marking in this chapter). Dependent marking can be (a)semantic (marking the thematic relations), (b) grammatical (marking the subject and object), and (c) discourse related (marking in/definiteness, topic/focus), and of course they overlap. In much generative work, the grammatical relations are seen as structurally determined, and hence (b) is not specially mentioned. Thus, Chomsky (2002: 113) mentions (a) and (c) in arguing that "[t]he semantics of expressions seems to break up into two parts ...There's the kind that have to do with what are often called thematic relations, such as Patient, Experiencer, etc.; and there's the kind that look discourse related, such as new/old information, specificity, Topic, things like that".
Marking the thematic positions (i.e. (a)) is done through pure merge in e.g. Chinese and English, or through inherent Case and adpositions in e.g. Sanskrit, Latin, Malayalam, Japanese, and Tagalog. Definiteness and specificity (i.e. (c)) can be marked through Case in e.g. Finnish, Turkish, Persian, Japanese, and Limbu (van Driem 1986: 34), through aspect in e.g. Russian (Leiss 1994; 2000;Abraham 1997; Philippi 1997), through position in e.g. Chinese, through a determiner, and through a combination of position and articles in e.g. Arabic, Dutch, and German (Diesing 1992). Grammatical function (i.e. (b)) is most consistently marked by agreement on the verb with some structurally special position. The nominal in this special position may be assigned a structural Case.Agreement is represented in the grammar through phi-features, and they are responsible for the agreement cycle.
In this chapter and the next, I will argue that deictic features are responsible for the Case cycles. I follow Leiss (2000) in considering definiteness on nouns and aspect on verbs as two sides of the same coin and, as before, I adapt Pesetsky & Torrego (e.g. 2001) and Richards (2004; 2008a) in representing dependent marking as [u-T] (nominative, marked on the D) and [u-ASP] (accusative, also marked on D). The choice of [u-T(ense)] and[u-ASP(ect)] points to the connection between nominal and verbal marking. Grammars of specific languages can have an emphasis on nominal marking of aspect and boundedness or on verbal marking. This can be seen in terms of dependent-marking (noun is marked, e.g. Finnish) or head-marking (verb is marked, as in Russian).
The outline is as follows. In the first section, some background on dependent marking is provided. The second section is predominantly a case study from Old English on the change from a grammar where verbal aspect dominates to one where nominal markers do. Section 3 examines the origins of dependent marking on subjects and objects. Another type of marking, differential (subject and object) marking, is discussed in section 4. This kind of marking can be seen as definiteness. Marking on non-core nominals (e.g. location and instrumental) is discussed in section 5. Section 6 returns to a discussion of the cycle and presents a conclusion.
1.Case, its uses, and structure
Case is unlike agreement in that it typically identifies the marked situation. As Comrie (1981: 122) puts it, Case is widespread as an “indication of unnatural combinations of A and P”, i.e. to indicate that the agent is less animate than the patient or the patient more animate than the agent. Case is also used for definiteness and in that function interacts with animacy of course.Nichols (1992: 46-96) claims that head marking and dependent marking “are about equally frequent overall” (1992: 95), but this is not clear. Siewierska & Bakker (2009: 299) say that “case marking of arguments is overall considerably less common cross-linguistically than agreement marking”.
This section first focuses on the types of Caseand how these may be responsible for a variety of functions, e.g. marking the semantic, grammatical, as well as pragmatic roles. It then provides some structural descriptions.
1.1Kinds of Case
As mentioned, three types of information are relevant in a sentence and they are marked by a variety of morphological and syntactic markers that sometimes overlap. In Table 5.1, repeated from Table 1.7, I have given a very simplified picture of the primary functions of these markers. (See Abraham 2007 [2] and Bisang 2006 among others for more detailed views).Prosodic factors are also important but left out here. In the remainder of this section, I discuss the markers of Table 5.1.
SemanticGrammaticalDiscourseAdpositions
Case-inherent
Case-structural Agreement
Aspect
D
"word order" / yes(some) (some)
yesnono
noyesnonoyes no
no(some) yes
no(some)yes
noyesyes
Table 5.1:Morphological and Syntactic Markers
Semantic roles can be divided into core roles (Agent and Theme) and non-core roles (Goals and Locations, etc). The non-core functions are often expressed by means of a preposition or postposition, as in (1), or by means of a specialized case, as in (2) and (3). The specialized Case is also known as inherent Case, and when it is lost it is often renewed by adpositions (as probably occurred in Hindi/Urdu).
(1)Wošehrsejʌnglkojata hẽHindi/Urdu
hecityfromforesttogo-Mbe-3S
'He goes from the city to the forest'.
(2)nagaratvanam gacchatiSanskrit
city-ABLforest-ACCgoes-3S
'He goes from the city to the forest'.
(3)Ayodhya-yamvasa-tiSanskrit
Ayodhya-LOClives-3S
'He lives in Ayodhya'.
I consider quirky Case to be an instance of inherent Case though this is not uncontroversial. I will have little to say about quirky Case per se.
Core grammatical roles are typically not marked by semantic Case but by structural (nominative and accusative) Case and by agreement. I will argue below that `Case’ is the wrong term for structural Case, and this is recognized early on by e.g. Schuchardt, cited in van der Horst 2008: 145), “Der Nominativ ist kein Kasus; … er ist das nackte Nomen”[The nominative is not a case; .. it is the bare noun”][3]. The term `accusative’ is not helpful either. According to the OED, it is a rendering of the Greek aitiatike `of accusing’, but also of to aitiakon `thing directly affected’, and that is the semantic Case of Theme, not the structural position it moves to. Most of the time, the object has a marker of definiteness not thematic role. So, rather than through Case, core nominals receive their thematic interpretation in the VP-shell (as in Hale & Keyser 2002), i.e. they are determined by word order (cf. the Uniformity of Theta Assignment Hypothesis fromM. Baker 1988).
Since the Agent tends to be in the highest structural position, it is also often the subject and the most topical element, especially in languages where all subjects are agents. Van Valin & Foley (1980: 339; see also Andrews 1985: 119) divide languages into Topic-Subject and Agent-Subject. English, Dutch, Finnish, Chinese would be examples of Topic-Subject languages and Dakota and Choctaw as Agent-Subject.Keenan (1976), Schachter (1976), and Mithun (2008b)suggest that not all languages may have the grammatical role of subject. A very helpful observation in this respect is from Donohue (1999: chap 20) who says that in some languages the pivot (or the grammatical role) is directly tied to the semantic role – he mentions Archi and Aceh – but that in others it is tied to pragmatic or syntactic roles. In Tukang Besi, an Austronesian language, the system is mixed, e.g. the addressee of an imperative always has to be a semantic Agent, the pronominal indexing on the verb is tied to the grammatical role (S or A), and cases are tied to the pragmatic roles.
Inherent Case, as in (2) and (3), is connected to a particular theta-role (see e.g. Chomsky 1986a) and contrasts with structural Case which isconnected to a particular position such as the Specifier of a TP for the nominative and possibly to the Specifier of ASPP for the accusative. The structural nominative is responsible for subject agreement on the verb; if such a nominative is lacking, a verb can have default agreement or find a non-nominative to agree with (see Woolford 2006 for more discussion on this).
In Table 5.1, I mark in brackets that adpositions sometimes mark grammatical information. I have Differential Object Marking (DOM) in mind, whichis grammatical or pragmatic. It marks that the object is unusual (or the subject in an ergative language) and in many languages indirect objects are marked this way. Moravcsik (1978: 283) suggests that the accusative is marking the more definite, animate, or affected nominal. According to Malchukov (2008), animacy is redundant for the semantic role but not for the grammatical one. I come back to DOM in section 4. Subjects can also be differentially marked, as e.g. Cennamo (2009) has shown.
Structural Cases such as the nominative and accusative are, in recent minimalism, seen as assigned by finiteness(the T) and transitive verb (little v) respectively. The markings signal specificity, volitionalty, and modality (see Abraham 1997; Diesing 1992; Kiparsky 1998; Leiss 2000; de Hoop 1992, to name but a few). Subjects are typically specific or definite, and as DuBois (1987) points out, in a careful analysis of the ergative Mayan language Sacapultec, new information is often presented as the object of a transitive or the subject of an intransitive[4]. Sacapultec has grammaticalized information structure through its Case system. This situation is not rare, as English, Dutch, Finnish, Turkish, Spanish, Persian, and Urdu/Hindi show. R. Baker (1985: 134) reviews work on the reconstructed accusative marker in Uralic "that [insists] its case function was purely secondary to its main role of marking definiteness".
Thus, the grammatical subject and object positions are connected with discourse informationthrough movement, as well as to their semantic roles through copies in their vP-internal positions. Subjects and objects move from the position relevant for their thematic role to another position. Since Sportiche (1988), floating quantifiers have been used to chronicle the positions an element was copied/moved into, as in (4), which shows the wanderings of the subject through the quantifier that can be left behind:
(4)Those children may (all) those children have (all) those children been (all) those children shouting.
The floating all can appear in any of the positions shown in (4) since the QP subject all those children moved through those. Why subjects and objects need to be expressed is perhaps one of the most puzzling questions. The marking is not one-to-one between nominative case and subject properties such as relativization or control of a reflexive. In many languages, e.g. Icelandic, Gujarati, Bengali, Telegu, and Lezgian, non-nominatives control reflexives (see Newmeyer 2008).
VP adverbials mark the boundaries of the thematic/semantic layer from that of the other layers. Objects that move to the left of these adverbials receive a particular interpretation, e.g. definite or partitive. This marking is different from Differential Object Marking, discussed in section 4, since all objects receive this kind of Case regardless of animacy and definiteness. Meinunger (1995: 92ff) lists some German sentences relevant in this respect which I list in Dutch. In (5a), when the object dat boek`that book’ is inside the VP, either the book was read completely or parts of it were. This would also be the position of indefinite objects. In (5b), however, where the object moves out of the VP (considering the adverb to indicate the left-boundary of the VP), the book has been read completely a number of times. In this position, an indefinite object would be ungrammatical.
(5)a.omdat ik vaak dat boek gelezen hebDutch
because I often that book read have
`because I've read that book often'.
b.omdat ik dat boek vaak gelezen heb
because I that book often read have
The reason for the difference is that a DP that moves to a higher position, as in (5b), moves to one where a certain aspect is checked (as well as specificity and boundedness) and the action must be complete. The DP inside the VP, as in (5a), on the other hand, can be partitive.
Other languages raise objects in a similar way. In Yiddish, a nominal object that moves out of the VP, as in (6a), has to be definite; indefinite nominals in that position result in ungrammaticality, as in (6b):
(6)a.Maks hot dos bukh mistome/ nekhtn/ keyn mol nit geleyentYiddish
Max has the book probably/ yesterday/ no timenot read
‘Max has probably/ never read the book (/yesterday).’
b.*Maks hot a bukh mistome/ nekhtn/ keyn mol nit geleyent
Max has a book probably/ yesterday/ no timenot read
‘Max has probably/never read a book (/yesterday).’
(Diesing 1997: 389-390)
In (14) below, I will make use of this insight that the position of grammatical Case is relevant to specificity and aspect. Chomsky (1995) discusses this point in relation to Object Shift in a number of the Scandinavian (and other Germanic) languages.
Structural Case and specificity are related in other language families, e.g. in Turkish, as Grönbech (1936: 155) argues early on. Grönbech calls the marking by –yiin (7a) an accusative but notes that it is used to prevent the object to become indefinite.This “verschmelzen” `to melt with’ occurs in (7b). So, technically, –yi in (7a) is a definiteness marker not a Case.
(7)a.Ahmet dün akşampasta-yı ye-diTurkish
Ahmetyesterdayeveningcake-DEFeat-PST
‘Yesterday evening, Ahmet ate the cake`.
b.Ahmetdün akşampasta ye-di
Ahmetyesterdayevening cakeeat-PST
‘Yesterday evening, Ahmet ate cake.’ (Kornfilt 2003:127)
See also Enç (1991), de Hoop (1992), and Öztürk (2005).
In short, semantic Case can be marked by adpositions, Case markers, and position; grammatical Case is closely related to specificity/definiteness and aspect in many languages; and `Case’ is perhaps not the most useful term for an element that has moved away from the position where it is marked semantically.
1.2DP, KP, and PP: structures for Case
Modifying work by Bittner & Hale (1996ab), Svenonius (2006), and Asbury (2008), I will suggest that semantic/inherent Case is represented by an (expanded) PPand structural Case by just a DP, not a K(ase)P as Bittner & Hale and others argue. In terms of features, I will suggest that inherent Case is represented by interpretable features such as time and place. Structural uninterpretable Case on subjects is checked by T (as in Pesetsky & Torrego 2001) and on objects by ASP/v. It is grammaticalized deictic marking.
Bittner & Hale (1996a: 6) argue (a) that the nominative is unmarked and therefore just a DP (or NP), (b) that the structural accusative has a KP with an empty K, and (c) that inherent Case has a filled K. Their structure (Bittner & Hale 1996a; 1996b: 537) is as in (8) (I show them without specifier positions).
(8)KP
ei
KDP
ei
DNP
The category K is very similar to P and (9) is therefore the structure that Asbury (2008) argues for. In Finnish, for instance, the addessive -lla of talolla `at (a) house' would be in P, and the genitive -n in D (though they cannot occur together).
(9)PP
ei
PDP
-llaei
DPhi-P
ei
PhiNP
talo-
Svenonius, in various publications, e.g. (2007), argues for a split PP with one P head introducing the Ground and the other small p head introducing the Figure. In other work (e.g. 2006), he presents a more articulated PP for the Ground, as in (10), without the Figure. His (10) can include a DegreeP above PlaceP to accommodate the intensifierright. (Again, I show them without specifier positions).The structure in (10) will be used for complex prepositions and language change when AxPrt, the position for the nominal element in the PP, is reanalyzed as Place.
(10)Path
ei
PathPlaceP
ei
PlaceAxPrtP
onei
AxPrtKP
topof the world
I will use a simple PP, as in (9), when talking about inherent, non-structural Case, but a DP when structural Case is involved. The respective probes for these are V and P for inherent Case and T and v for structural case. I'll now discuss some of the features involved.
Most people ignore feature checking where inherent Case is involved since it is interpretable (Chomsky 1995). Let's first look at the Case in a regular PP. Van Gelderen (2008e; 2009a) uses (11) as a structure for the temporal preposition after.
(11)PP
wp
PDP
after…
[u-phi][3S]
[i-time]/[ACC][u-time]/[u-Case]
Having interpretable features on the P is somewhat similar to Pesetsky & Torrego’s (2006b) assumption that prepositions have [i-T(ense)]. I use [time] in (11) to distinguish prepositional objects from nominatives. The [time] feature is interpretable and is licensing the [u-time]. The [u-phi] feature makes an adposition into a probe and thereby different from an adverb, just like a demonstrative is different from an article.
The structure in (11) could be reanalyzed as inherent Case, as has happened with benefactives, comitatives, and locatives. In those cases, the main verb has to be reanalyzed as licensing a Goal or other theta-role, as in (12). It means adjuncts are made into arguments.
(12)VP
ei
VPP
[u-loc]ei
PDP
[i-loc][3S]
-llatalo`at (a) house’
Structural Case, according to Pesetsky & Torrego (2004; 2006ab), involves an uninterpretable/unvalued T on the nominal which is valued by a finite Tense or transitive v. Pesetsky & Torrego connect Case, finiteness, and agreement by having a tense feature in T (and v) look down the tree for a feature on the DP[5]. My adaptation of this is as in (14), leaving out a separate V(P) and an ASPP. Note that, similar to what Pesetsky & Torrego assume for nominative, I assume for accusative. There are other possibilities, as I discuss in more detail below, e.g. the ASP features may be Num(ber) or Measure.