1. INTRODUCTION
Case is perhaps the most intriguing and also the most widely discussed of all grammatical categories. It is fair to say that case, more than any other category of language, highlights the fact that our understanding of language externalization and variation is limited. Many languages do not, in fact, have any overt case marking. The figures in (1) are drawn from Iggesen (Reference Iggesen2011b).
(1)
Footnote 1 Iggesen's classification applies to ‘regular’ or ‘general’ noun-case distinctions. Thus, English counts as a two-case language (Nom and Gen), rather than a three-case language even though it has a three-case distinction in parts of its pronominal system. Actually, it would be more logical to count only marked cases; that is, to refer to languages as zero-case languages (no marked case), one-case languages (one marked case, two distinct terms), and so on. I will return to this issue in relation to nominative case in Section 2.
Blake (Reference Blake2001:156) has suggested that there is a morphological case hierarchy or scale, such that languages ‘pick’ their cases in the preference order in (2).Footnote 2
(2) Nom > Acc/Erg > Gen > Dat > Loc > Abl/Inst > others (Part, Com, Purp, Perl, . . .)
The following tendencies can thus be discerned:
(3) Common typological case generalizations
a. If a language has cases, one of them is morphologically and/or functionally unmarked in relation to the other(s) and is thus likely to be called ‘nominative’ (or ‘direct’) by linguists.
b. The second case (the first marked one) is likely to distinguish objects (Acc) from subjects or to specifically distinguish agentive subjects (Erg) from other core arguments.Footnote 3
c. A third case is likely to be an adnominal ‘possessor case’ (Gen).
d. A fourth case is likely to be an ‘additional core argument case’ (Dat, etc.).
e. A fifth case is likely to mark spatial relations (Loc, etc.).
f. Additional cases make more fine grained distinctions between arguments or NPs in general (Abl, Inst, Part, Com, etc.).
Presumably, these are third-factor effects in the sense of Chomsky (Reference Chomsky2005): natural options allowed for by underspecification of Universal Grammar (UG).Footnote 4
Blake (Reference Blake2001:156ff.) discusses a number of variously rich case systems, including the following ones (disregarding the vocative and some vestigial cases as well as complex locative case systems):
(4)
However, this gives an idealized picture of order in a complex ‘case world’. First, many languages do not ‘fit’ into the Blake hierarchy. Faroese, for instance, is losing Gen but keeping Nom, Acc and Dat (Thráinsson et al. Reference Thráinsson, Petersen, Lon Jacobsen and Hansen2004:248ff.), which is the opposite of the development in Greek. Gaelic has Dat and Gen but no Acc, Hungarian has no Gen but Dat, Inst, etc., and Finnish has a general Gen but no inflectional Acc except for pronouns, also having Part, ‘instead of’ the higher ranked Dat, Loc, Abl and Inst. For more examples of ‘deviant’ systems, see Blake Reference Blake2001 and Malchukov & Spencer Reference Malchukov and Spencer2009a, Reference Malchukov and Spencerb. Obviously, also, notions like ‘dative case’ are not usually ‘constant’ across languages with variably rich case inventories: ‘the dative’ is something different in Faroese, Turkish, and Tamil.
Second, languages with identical case inventories can show various splits and asymmetries across grammatical categories, such as Person, Tense and Aspect, or across lexical categories, such as nouns and pronouns (see Iggesen Reference Iggesen2011a). A familiar split or asymmetry of this sort is seen in a number of European languages, including English, with a general two-case system (Nom, Gen) for nouns (and NPs) and a three-case system for central parts of their pronominal systems (Nom, Acc, Gen). Even ‘simple’ systems of this sort show considerable complexities and cross-linguistic variation (see Quinn Reference Quinn2005 on English). And there are many more types of split case systems, including split ergative systems of various sorts.
Third, languages with the same number of cases can distribute their cases quite differently across semantic and grammatical roles – call this case mismatches. Case mismatches are most easily demonstrated for two-case languages, where various patterns can be discerned (see e.g. Blake Reference Blake2001, Arkadiev Reference Arkadiev2009, Malchukov & Spencer Reference Malchukov and Spencer2009b, Iggesen Reference Iggesen2011a). Thus, the oblique (or the ‘marked’/‘indirect’) case in nominative–oblique or morphologically unmarked–marked systems may mark various relations, as sketched in (5) for a few languages (disregarding ergative systems, which add another dimension to this case mismatching picture).Footnote 5
(5)
Developing some understanding of all this variation is a major challenge. As suggested by the presentation in (3) above, it presumably follows from general third-factor effects and principles. That is, case is not provided by or part of Universal Grammar (UG), a conclusion supported by numerous facts, for example extensive case mismatches and also the simple fact that having no case marking at all is highly common, not only in ‘usual’ languages, but even more so in creoles and sign languages (see Sandler & Lillo-Martin Reference Sandler and Lillo-Martin2006:23; see also Meir Reference Meir2003).Footnote 6 It is thus unsurprising that UG-based approaches to case variation have never been successfully developed.
Saying that language variation follows from underspecification of UG and third-factor effects makes a very general statement. It does not release linguistics from developing some coherent understanding of variation. As in any other scientific endeavour, developing such understanding requires not only broad overviews but also detailed analyses of specific data. In the following, I will look into some of the details of Icelandic case assignment, comparing it with case assignment in some other languages, above all German, Icelandic and German having an identical case inventory: Nom, Acc, Gen, and Dat. Even though the Icelandic/German four-case system is an offspring of the Proto-Indo-European eight-case system (Nom, Acc, Gen, Dat, Loc, Inst, Abl, and Voc), I will be focusing on case growth, setting case decline aside.Footnote 7
Section 2 discusses the nature of nominative case and its correlation with verb agreement, Section 3 discusses the non-nominative argument cases and argument case growth (case star augmentation), Section 4 describes non-core case of various sorts, Section 5 describes and analyzes case expansion from one subsystem of case marking to another, and Section 6 concludes the paper.
The following journey is part of a much longer voyage (Sigurðsson Reference Sigurðsson2004, Reference Sigurðsson and Boeckx2006a, Reference Sigurðsson, Hartmann and Molnarfib, Reference Sigurðsson2011b, Reference Sigurðssonc, Reference Sigurðsson2012, etc.), aiming at showing that internal syntax is even smaller or narrower and that the externalization component of language is thus much broader and richer (‘more syntactic’) than commonly assumed in generative approaches. As it turns out, however, case variation can, to a large extent, be analyzed in terms of only two atomic ‘ingredients’: event licensing of NPs and variable PF marking of the licensing relation.
2. NOMINATIVE
All languages ‘begin life’ without any case marking, a conjecture that is not only reasonable on general conceptual grounds but also strongly supported by the fact that all known sign languages are caseless or next to caseless, sign languages generally being very young (see e.g. Sandler & Lillo-Martin Reference Sandler and Lillo-Martin2006). If so, original or non-borrowed case systems plausibly always come into being by a simple split in two, with a marked or an oblique case and an unmarked one, commonly referred to as ‘nominative’, some such systems further developing or growing by adding further cases over time (and then potentially declining, then even expanding again in new directions, see Kulikov Reference Kulikov2009). As mentioned above, it would actually be more logical to count only marked cases, languages, then, generally being ‘born’ as zero-case languages, then going through a stage with one marked case, with either a marked core case (Acc or Erg) or a marked borderline case (Loc, Inst, etc.), then through a stage with two marked cases (e.g. Acc and Gen, Acc and Loc), and so on. On this view, nominative is always a non-case.Footnote 8
However, unlike German and most other modern European case languages, but like Latin, Icelandic has a split marked nominative system, in the morphological sense (rather different, it seems, from the African systems described in König Reference König2009). That is, it has a morphologically marked nominative form in certain declensions but an unmarked nominative form in other declensions. Icelandic has many different noun declensions – a comprehensive description minimally requires around 70 distinct paradigms (even if much morphophonological variation is disregarded).Footnote 9 Simplifying, we can say that the so-called strong masculine declensions usually have marked nominative forms in the singular (mostly in -ur), while most other declensions have unmarked or not clearly marked singular nominatives. A few of the most regular patterns are exemplified in (6); the strong–weak dichotomy is a formal one (pure vowel endings in all weak singular forms), with no semantic or syntactic import or correlates.
(6)
This is the traditional analysis of the case endings. Alternatively, one could analyze the Nom.Sg forms of all weak nouns as the basic stem form, other forms being derived by vowel truncation, deleting the first of two unstressed vowels, a process that is in any case operative elsewhere in the language (/penni-a/, /vika-u/ > penn-a, vik-u, etc.). Be that as it may, it is clear that Nom is marked to a variable degree, depending on both number and declension class.Footnote 10 Despite this morphological split, all nominatives show parallel behavior with respect to syntactic distribution and agreement processes. That is, the marking of marked nominative forms is not syntactically triggered, instead being a purely morphological (PF) phenomenon.
Case researchers commonly make no clear distinctions between morphological case terminology, functional case terminology and alignment case terminology. Morphologically, there is a distinction to be drawn between marked and unmarked case forms (Dixon Reference Dixon1979). Functionally, there is a distinction between general/common and less general/common, often referred to as marked vs. unmarked (or default) as well (see Schütze Reference Schütze2001). In contrast, traditional case terms, Nom, Acc, Erg, etc., are commonly used as alignment terms, indicating which case usually aligns with which thematic role (Agent, Patient, etc.) or grammatical function (subject, direct object, etc.). This is simply illustrated in (7).
(7)
Commonly, the term ‘nominative’ is used to designate (i) the morphologically unmarked case form, (ii) the functionally general or default case, and (iii) the case of subjects (in non-ergative systems), but there are obviously many mismatches between morphological, functional and alignment mappings.
Mainstream generative approaches to case (Chomsky Reference Chomsky1980, etc.) take it that Nom is a syntactically active case, ‘assigned a value under agreement’ with the Tense head (T) in finite clauses (Chomsky Reference Chomsky2001:6). The assumption that Nom is the ‘responsibility’ of the finite, φ-complete, T has commonly been taken to account for finite verb agreement in regular Nom-Acc languages, such as English and German. However, as we have seen, it is more coherent to assume that Nom in Nom-Acc systems is simply the unmarked opposite of a marked Acc or Obl case. If so, verb agreement (in case + agreement languages) enters non-null-subject (PF) grammars as an extra alignment strategy, for example along the path sketched in (8).Footnote 11
(8)
As suggested by this presentation, agreement is commonly redundant in non-null-subject systems. When such a system develops marked nominatives, a number of patterns may emerge, including patterns with two marked cases. Consider the Icelandic examples in (9) (agr = overt agreement).
(9)
Here, the verb form does not yield any clear alignment information (as both the subject and the object are in the third person singular), but it does in many other cases, for example when the subject and the object are in different numbers. In many such cases, nominative alignment (with A/S) is doubly marked, with a marked Nom + verb agreement. If the object carries a marked Acc, this yields a threefold alignment marking (as in (9a)), where single marking (only marked Acc, only marked Nom, or only verb agreement) would be sufficient to disambiguate the NP-predicate relations.
Nom is a non-case.Footnote 12 That is, whenever grammar gives no specific case instructions to the morphological case component, the NP in question will show up in Nom, regardless of the overt shape of the nominative elements expressed. Finite verb agreement, in turn, arises in morphology whenever the verb successfully probes a non-cased NP (be it morphologically marked or not), given, of course, that the language in question has finite verb agreement. Thus, we have to sharply distinguish between grammar instructions leading to case marking and case realization or externalization in PF. As we will see, actually, case morphology often behaves like a virus in morphological systems like the Icelandic one, without any clear semantic or syntactic import or correlates. Marked nominatives are just one example of such ‘PF viruses’.Footnote 13
3. THE NON-NOMINATIVE ARGUMENT CASES
Blake's hierarchy in (2) and the related generalizations in (3) would seem to suggest that the first cases to get marked in the historical development of case systems are the core cases: First Acc or Erg, then Gen, then Dat, and so on. However, as the hierarchy is mostly based on observations on established ‘dinosauric’ case languages (with perhaps a 100,000 years old history), it is unclear whether it reflects usual historical developments of case systems (for some observations and speculations on the development of case systems, see Blake Reference Blake2001, Kulikov Reference Kulikov2009).Footnote 14 It is quite possible that a language develops some borderline or peripheral cases before developing Acc or Erg; 24 of the 266 languages in Iggesen Reference Iggesen2011b have exclusively borderline case marking, including Plains Cree ‘whose only case-inflecting device is the locative suffix -ehk’ (Iggesen Reference Iggesen2011b). For expository purposes, however, I will assume that Blake's hierarchy and the generalizations in (3) reflect at least a common case growth pattern over time.
In ‘Derivation by phase’, Chomsky (Reference Chomsky2001) suggests that Acc in regular accusative systems is the responsibility of φ-complete v, designated as v*.Footnote 15 In contrast, he analyzes defective v as not assigning or licensing any case value, thereby rendering the underlying object in defective vP types accessible to Nom (as in canonical passives, unaccusatives, and anticausatives). As illustrated in (10), this yields Burzio's Generalization (for English), that is to say, the Acc-to-Nom conversion typical of defective predicate types.
(10)
Case star augmentation, then, yielding v* in addition to plain v, is involved when a language develops from a no-case language to an accusative language, distinguishing Acc NPs from non-cased (Nom) NPs. This development is a PF change, not applying in caseless languages. In a language where it has taken place, active Acc assignment in the v-system is based on the structure in (11), whereas the corresponding unaccusative and passive structures yield Nom, as sketched in (12); the arrows connecting v*-V and NP1 indicate a matching relation that gets interpreted in terms of non-nominative case in PF morphology, here Acc (as indicated on the right hand side).Footnote 16
(11)
(12)
I assume that all predicates are embedded under some Voice head, passives being embedded under Voice/pass, while unaccusatives as well as anticausatives are embedded under expletive Voice, Voiceexpl (even though unaccusatives and anticausatives have different vP-internal structures). Agentive or active Voice, Voice/ag, in turn, introduces the subject in active transitive structures, such as (11) (see Kratzer Reference Kratzer, Rooryck and Zaring1996 and much related work).
In ergative systems, PF case star augmentation applies to Voice/ag, yielding a case licensing Voice*/ag, as sketched in (13).
(13)
Ergativity and accusativity are thus two sides of the same coin: while ergative case is directly licensed by Voice*/ag, accusative is licensed by a v* under c-command of Voice/ag.Footnote 17
The case decision for a non-nominative NP can be described in three steps. First, syntax transfers structures such as (11) and (13) to deep PF, without any case stars (hence all case marking is out of semantic/syntactic reach, even when it has semantic correlates, see Section 5). Thus, the syntactic message or information to PF is purely structural (and ‘lexical’), without any inflectional instructions: [Voice (↔ NP) ↔ v/V ↔ NP] type structures, where the arrows denote matching relations (and where V and NP contain abstract ‘lexical’ roots). This first (plausibly universal) step is not shown in (11)–(13). Second, language-specific PF assigns a particular type of case star to one or more licensing heads (v, Voice, etc.), depending on the syntactic structure (and on lexically related case star selection, see Section 5). This yields the input structures in (11) and (13) (i.e., the structures to the left of the horizontal arrows). Third, PF interprets the particular case star as one of the morphological cases of the language in question, again depending on the transferred syntactic information (and on deep PF properties, most importantly lexical categories). The right hand horizontal arrows in (11)–(13) indicate this third and most shallow step in this case-marking description. Note that I say ‘description’. It seems reasonable to assume that steps two and three are instantaneously intertwined or just a single step in the actual morphological derivation.
On this approach, case is absent from Universal Grammar (and hence the previously mentioned fact that 100 out of 266 languages in Iggesen (Reference Iggesen2011b) lack case marking is unsurprising). Note that the approach is fully compatible with the Inclusiveness Condition (obviously, the Inclusiveness Condition does not apply to PF, see Chomsky Reference Chomsky1995:228, 381, n. 10). These claims and assumptions will become more transparent and understandable as we proceed.
In addition to either Acc or Erg, languages with two marked cases (three-case languages) most commonly have adnominal Gen or benefacive/recipient Dat. These cases do not usually involve Voice or v, the core heads of the verbal system. Rather, the canonical Gen is adnominal, and indirect object Dat is commonly taken to be introduced by an applicative head, Appl (Marantz Reference Marantz and Mchombo1993, McGinnis Reference McGinnis2001, Cuervo Reference Cuervo2003, Pylkkänen Reference Pylkkänen2008, Schäfer Reference Schäfer2008, Wood Reference Wood2012b). The Appl enriched structure of an ordinary ditransitive clause like Icelandic Hún gaf mér bókina ‘she.nom gave me.dat the book.acc’ is illustrated in (14).
(14)
Arguments, then, are event licensed by specialized heads (Pylkkänen Reference Pylkkänen2008, Sigurðsson Reference Sigurðsson2012): agentive or active subjects by Voice/ag, indirect objects by Appl, and direct objects by v-V. The TP structures of regular transitive and ditransitive clauses are sketched in (15) and (16).
(15)
(16)
DO, IO and SU are just convenient expository labels (instead of NP1, NP2, NP3 or θ1, θ2, θ3). The derivation of (16) starts out by introducing DO (NP1/θ1) and event licensing it by v-V, then introducing IO and event licensing it by Appl, then introducing SU and event licensing it by Voice/ag.Footnote 18
In caseless languages none of the event licensing heads are augmented by a case star, whereas languages like German, Icelandic and Turkish have both case licensing v* heads and Appl* heads (licensing Acc vs. Dat). Ergative systems, in turn, have a case star augmented Voice*/ag, as discussed above. The generalization that emerges from these observations is stated in (17).
- (17)
a. NPs are event licensed by heads: Voice, Appl, v, n, a, p, . . .Footnote 19
b. Event licensers of NPs may be case star augmented
c. The augmented case star may be variably marked: *, *+, *++
d. Nom = a non-case (regardless of its overt marking)
The ‘plus’ notation on case stars may be interpreted as ‘more marked’. Thus, *+ is more marked than *; *++, in turn, is still more marked than *+.Footnote 20 As we will see, a morphological case may be variably marked depending on which structure it is licensed in; that is to say, the markedness of some particular case, say Dat, depends on whether it is an Appl-case (unmarked as compared to Acc), a v-case (more marked than Acc), and so on. As event licensers and case stars are the basic ‘ingredients’ in case systems, I refer to (17) as the Case Ingredients Generalization.Footnote 21
It does not come as a surprise, of course, that argument-introducing heads may also license some overt case marking. In the often cited words of Blake (Reference Blake2001:1), case is ‘a system of marking dependent nouns for the type of relationship they bear to their head’. The variable markedness generalization in (17c) is less expected, but it is corroborated by many facts, one being that indirect objects are variably marked both within and across languages. Thus, while Icelandic has more than 280 ditransitive verbs that take a dative indirect object (mostly Nom-Dat-Acc), it also has around 60 ditransitive verbs that take an accusative indirect (or ‘first’) object, mostly Nom-Acc-Dat verbs (Jónsson Reference Jónsson2000; Thráinsson Reference Thráinsson2007:173). Two examples are given in (18); as indicated, the ‘first’ object is optional, as indirect objects commonly are.
(18)
As suggested by these examples, accusative indirect objects commonly resemble Latin ablatives of separation. Presumably, they came into being as Acc ‘invaded’ the Appl-system from the v-system (perhaps in tandem with or as a consequence of the Germanic ‘Abl-death’). As these ‘ablative accusatives’ are clearly more marked in the Icelandic Appl-system than are regular dative indirect objects, we need to distinguish between Appl*, licensing dative indirect objects and Appl*+, licensing the ‘ablative accusatives’. Languages like English and Swedish (see Holmberg & Platzack Reference Holmberg and Platzack1995:197), with only accusative indirect objects, have only one Appl head, Appl*, whereas Gaelic has lost its Nom–Acc distinction, thus having only nominative indirect objects, licensed by plain Appl. Thus, ‘marked’ and ‘more marked’ are relative notions, depending not only on event licensers but also on languages.Footnote 22
Variable case marking also applies to direct objects in many languages (see e.g. Malchukov & de Swart Reference Andrej and de Swart2009); thus, Russian has Acc, Gen, Dat, and Inst direct objects (see Richardson Reference Richardson2007, Bailyn Reference Bailyn2012), German has Acc, Gen, and Dat direct objects, Icelandic has Acc, Gen, Dat, and Nom direct objects, and so on.Footnote 23 Some of this variation is language-internally regular to some (variable) extent, but it is typically unpredictable cross-linguistically. Thus, ‘Maling (1996) [an unpublished work] contains a list of more than 750 [Icelandic] verbs which in at least one sense occur with a dative object . . . [whereas the] corresponding number of verbs for German is approximately 140, and, for Russian fewer than 60’ (Maling Reference Maling2002:31). There is nothing (synchronically) in the case systems of these languages that would lead one to expect these dramatic differences – recall that Icelandic and German have identical four-case inventories: Nom, Acc, Gen, and Dat.
The Icelandic verbs in (19) take a dative direct object:
(19) Some Dat-taking Icelandic verbs
ausa ‘scoop’, beina ‘direct’, bjarga ‘rescue’, bjóða ‘invite’, bylta ‘overturn’, dreifa ‘spread’, fagna ‘welcome’, fleygja ‘throw away’, fleyta ‘float’, fljúga ‘fly’ (e.g. an aeroplain), giftast ‘marry’, gleyma ‘forget’, heilsa ‘greet’, hella ‘pour’, henda ‘throw, throw away’, hjúkra ‘nurse’, hlífa ‘protect, spare’, launa ‘pay, reward’, misþyrma ‘torture’, ráða ‘decide’, ríða ‘ride’ (e.g. a horse), róa ‘row’, sigla ‘sail’, snúa ‘turn’, stjórna ‘control, govern, rule’, sökkva ‘sink’, tortíma ‘exterminate’, ýta ‘push, shift’, þjóna ‘serve’, þóknast ‘please’
Corresponding verbs in German all take an accusative object (auslöffeln ‘scoop’, retten ‘rescue’, einladen ‘invite’, etc.).
The Icelandic verbs in (20) all take a genitive direct object (some of them can alternatively take a PP complement).
(20) Some Gen-taking Icelandic verbs
afla ‘procure’, biðja ‘ask for’, bíða ‘wait for’, geta ‘mention’, gæta ‘heed; take care, look after’, krefjast ‘demand’, leita ‘look for, search for’, minnast ‘(be able to) remember’, neyta ‘consume’, njóta ‘enjoy’, óska ‘wish for’, sakna ‘miss’, æskja ‘wish for’, þarfnast ‘need’
Corresponding verbs in German (bitten‘ask for’, erwähnen‘mention’, etc.) take either an accusative object or a PP complement.
The examples in (21) and (22) illustrate this Icelandic–German dichotomy with respect to direct object case marking.
(21)
(22)
Importantly, there are no semantic differences between Icelandic Dat/Gen objects in examples like (21a) and (22a) and corresponding Acc objects in German examples like (21b) and (22b).
Case mismatches of this sort are not only found across related languages but also within individual languages. Consider the Icelandic direct object case marking contrasts in (23)–(24) (showing just a few samples of such contrasts); the verb pairs either have similar (sometimes identical) or opposite meanings.
(23)
(24)
These facts highlight two important albeit often neglected or ‘downgraded’ aspects of case variation. First, it is cross-linguistically unpredictable, even for closely related languages with identical case inventories; that is, there are no straightforward universal generalizations or ‘truths’ behind case variation. Second, also language internally there are case irregularities and mismatches; that is, even internally to individual languages variable case marking cannot be fully described in terms of regularities. This is in fact made quite obvious by historical sporadic case marking changes in otherwise stable case systems. Icelandic witnesses many such minor shifts in case use without any concomitant semantic changes (see Barðdal Reference Barðdal2001, Jónsson & Eythórsson Reference Jónsson and Eythórsson2005, Thráinsson Reference Thráinsson2007). When a verb starts, say, taking a dative object instead of an accusative one, without any meaning shift or system shift, then it cannot be the case that both historical stages (V-Acc vs. V-Dat) represent ‘the regular system’.
Languages are of course full of forms that correlate with some semantics. Such form–semantics correlations are typically significant but incomplete and variably regular and cannot be explained as a result of a direct interpretative connection between PF morphology and semantics (a widely held misconception). Rather, they stem from the fact that semantic interpretation and morphological interpretation is read off at the interfaces from a common underlying syntactic structure (on the minimalist single cycle approach to the derivation up to transfer). Thus, there are indirect but no direct or causal correlations between semantics and morphology.
‘[C]ommunication is a more-or-less affair, in which the speaker produces external events and hearers seek to match them as best they can to their own internal resources’ (Berwick & Chomsky Reference Berwick, Chomsky, Di Sciullo and Boeckx2011:40). Understanding external language (E-language) is possible because language users interpret or decode it in relation to their own internal language (I-language), and not because of any exact mappings between I- and E-language. Seemingly meaningful morphemes and items like, say, English -s, -ed, and apple, are grammatical coins: By using them, the speaker ‘pays’ for accessing and activating someone else's I-language, ‘buying’ understanding or reaction. The coins are valid currency, not because they are part of anybody's I-language but because they relate to and activate I-languages: You need a pen or an apple and you pay with a coin because it gets you what you desire (and not because the coin is a pen or an apple). This understanding does not, of course, release linguists from the duty – or deprive them of the joy – of carefully analyzing form–semantics correlations in individual constructions and languages, but it ought to save them from the pitfall of analyzing such (more or less inaccurate) relations as being part of I-language.
Let me emphasize that developing an analysis of exactly how some grammatical coin, like, say, a dative case marker (or ‘simply’ a lexical item like pen or apple), can gain currency in some linguistic community is a worthy undertaking. However, the complexity of the issue cannot be overstated. Understanding it requires a theory of at least three subsystems, each of which is largely beyond present day limits of science. First, it requires a coherent understanding of the structure and nature of I-language, the linguistic thought system (there biolinguistics actually seems to be making some encouraging progress, see e.g. Berwick & Chomsky Reference Berwick, Chomsky, Di Sciullo and Boeckx2011, Hinzen Reference Hinzen2013). Second, it requires an analysis of how one individual I-language can relate to another I-language via the grammar of external language (E-language) and body expressions (acoustic, facial, manual, etc.; see the double transfer approach in Sigurðsson Reference Sigurðsson2011c). Third, it also requires a theory of how E-grammar and body expressions can establish themselves as a part of the ‘E-language contract’ in a community.
Regardless of how we deal with these fundamental issues, the study of language and thought is subject to the dilemma of Plato's cave allegory: We can only access and study I-language via E-language, but E-language is merely a fluctuating shadow of I-language, the relation between the two being fundamentally and inescapably non-isomorphic.Footnote 24 It follows that a dative case marker, for instance, cannot really be a marker of dative case, paradoxical and distressful as that may seem. Not any more than a coin is a pen or an apple.
4. BEYOND THE V/N/APPL-SYSTEMS
The Icelandic cases are found widely outside of the core argument system. The overview in (25)–(28) is a very brief non-exhaustive sketch of the distribution of the non-nominative cases beyond core arguments (disregarding dislocated NPs and other case agreement phenomena).
(25) Accusative
a. Complements of certain prepositions
b. Temporal and spatial adverbial NPs
(26) Dative
a. Complements of most prepositions (including agentive NPs in af- ‘by’ phrases)
b. A few adnominal NPs (/she looked in eyes me/; /to defence me/ = ‘to my defence’)
c. Complements of certain adjectives (/she was me kind/)
d. Comparative and iterative/frequentative adverbial NPs
(27) Genitive
a. Complements of some prepositions
b. Most adnominal NPs, reflecting an array of semantic/syntactic relationsFootnote 25
c. Complements of a handful of adjectives (/she was not worthy his/ = ‘worthy of him’)
d. Some partitive NPs (/most their/ = ‘most of them’)
e. NPs in various other adverbial or adverbial-like functions (see Kress Reference Kress1982:228ff.)
Only a few illustrative examples follow. The accusative adverbial NPs in (28a–b) are temporal, while the one in (28c) is spatial.Footnote 26 As indicated to the right, other cases are excluded.
(28)
The datives in (29) are iterative/frequentative, while the ones in (30) are comparative.
(29)
(30)
The adjectival complements in (31) are dative (not uncommon) and genitive (rare).
(31)
These few examples are only meant to give an initial idea about the nature of some of the Icelandic borderline case phenomena. Similar (but only partly identical or overlapping) facts are found in many other case languages. I will not develop any detailed analysis of these facts here – they are lexically restricted with idiomatic or semi-idiomatic properties. What these facts serve to show, however, is that morphological case can behave very much like a virus, penetrating every possible corner of a language, not tolerating a single NP without some morphological case marking. As previously mentioned, marked nominatives, case agreement and case percolation into PRO infinitives also illustrate this tendency to case mark nominal categories even when they are not subject to any independent case instructions.
This virus-like behavior of morphological case is perhaps most clearly demonstrated by adpositional case. Icelandic prepositions fall into four classes with regard to case marking; simple prepositions are listed in (32), with only their most central translations.Footnote 27
- (32)
a. Dative
að ‘towards, up to’, af ‘off, from’, andspænis ‘opposite, facing’, ásamt ‘along with’, frá ‘from’, gegn ‘against’, gegnt ‘opposite, facing’, handa ‘for’, hjá ‘at, with, beside’, meðfram ‘along’, móti ‘against, towards’, undan ‘from under(neath)’, úr ‘out of, from’
b. Genitive
b1. auk ‘in addition to’, án ‘without’, meðal ‘among’, milli ‘(in) between’, til ‘to, towards’, vegna ‘because of’
b2. austan ‘east of’, norðan ‘north of’, sunnan ‘south of’, vestan ‘west of’, handan ‘beyond, on the other side of’, innan ‘within, inside of’, neðan ‘under, below, beneath’, ofan ‘above’, utan ‘outside (of)’
c. Accusative
gegnum ‘through’, kringum ‘around’, um ‘about, around’, umfram ‘beyond, besides’, umhverfis ‘around, nearby’
d. Variably dative or accusative
d1. á ‘on, in, at’, í ‘in, on, at’, undir ‘under, underneath’, yfir ‘above, across, over’
d2. fyrir ‘for, before’, eftir ‘after, by’, með ‘with’, við ‘with, to, at’
All prepositions containing the string /um/ take an accusative complement and any single-word preposition/adverb containing /an/ assigns genitive. Beyond these bleeding phonological relations, dative is the unmarked, general prepositional case.
The variation in (32d) is to an extent semantically predictable and similar phenomena are found in other Indo-European languages (see Libert Reference Libert, Amberber and Collins2002). Generally, however, the case licensing properties of prepositions are not decided by their semantics. Thus að ‘towards, up to’ and af ‘off, from’ have roughly opposite meanings, and so do gegn ‘against’ and handa ‘for’ as well as, e.g. frá ‘from’ and hjá ‘at, with, beside’, and, yet, all these prepositions obligatorily take a dative complement. Similarly auk ‘in addition to’ and án ‘without’ have roughly opposite meanings, both nevertheless taking genitive complements (án variably took Acc, Dat or Gen in Old Norse).
Again, a comparison between Icelandic and German is instructive (inasmuch as such a comparison is possible and plausible). There are a number of roughly synonymous prepositions that assign the same case in both languages, such as dative-assigning German bei and Icelandic hjá, both meaning ‘by, at’, accusative assigning German durch and Icelandic gegnum meaning ‘through’, and so on. However, there are also many prepositional case mismatches between the languages. Recall that they have identical case inventories, and, yet, we find differences like the ones listed in (33).
(33)
The case literature contains numerous suggestions that inherently case-marked NPs are actually part of larger structures, PPs or KPs (with a silent K(ase) head that is responsible for the case assignment, functioning much like an overt adposition). See, for instance, Emonds Reference Emonds1987, Bittner & Hale Reference Bittner and Hale1996, Bayer, Bader & Meng Reference Bayer, Bader and Meng2001, McFadden Reference McFadden2004, and Asbury Reference Asbury2010 for a number of different versions of this basic idea. Demotion (from a case marked NP to an adpositional complement) and patterns like Dative Shift (‘give the book to X’, vs. ‘give X the book’) are commonly taken to lend support to this view.
There are four basic combinations of case markers and adpositions, as simply sketched in (34), where NP0 and NPC stand for a non-case-marked NP vs. case-marked NP, and where P denotes ‘adposition’.
(34)
Some so-called case markers can be successfully analyzed as adpositions in some constructions in some languages; this seems for instance to be true of various locative markers in Finno-Ugric languages, for instance the inessive and the terminative markers in the Hungarian examples in (35) (from Asbury Reference Asbury2010:5).
(35)
However, even though it seems profitable to analyze the Hungarian elements ben and ig in (35a–b) as postpositions rather than genuine case makers, and to analyze the adverbial NPs in constructions like the ones in (34a–b) as PPs or KPs with a zero head (see McFadden Reference McFadden2004, inter alia), there is no gain in analyzing inherently case marked NPs in the Latin/Icelandic/German type of languages as always being PPs or KPs. If, for example, plain datives (indirect objects, direct objects, quirky subjects, etc.) are analyzed as K + NP, we are forced to either assume an empty K even in the presence of an overt P, as in (34d), or to come up with some plausible account of why some K heads must and some must not be spelled out as overt Ps.
Adverbial NPs like the ones in (34a–b) do have similar distributional and semantic properties as certain PPs, but inherently case marked objects commonly have properties that are quite distinct from those of PPs. Thus, Icelandic dative and genitive objects raise to subject in numerous construction types (passives, etc.), whereas full-fledged PPs never do, as illustrated in (36) vs. (37).
(36)
(37)
As suggested in Sigurðsson Reference Sigurðsson2011a, an overt preposition arguably blocks Tφ from probing prepositional objects (in Icelandic). That is: Prepositions introduce both structural and semantic information that is otherwise absent.
On a general PP or KP analysis of inherently case marked NPs, one would presumably expect the P + NPC type in (34d) to be exceptional in case languages, as compared to the P + NP0 type in (34c). The opposite is true: Typological research suggests that most morphological case languages do indeed have case-marked complements of adpositions, and, conversely, that non-case-marked complements of adpositions are highly exceptional in case languages (see Libert Reference Libert, Amberber and Collins2002).
Observations of this sort indicate that morphological marking, even in the simplest and most regular morphological systems, is an unnecessary ‘quirk’ that languages really can do without – suggesting, in turn, that morphological variation is a sociobiological phenomenon rather than a strictly linguistic one. In other words, morphological marking is plausibly not part of or ‘predicted’ by the narrow innate and internal language faculty (Universal Grammar), instead arising in the (PF) externalization process (controlled and affected by biological and social factors that are not specific to the language faculty).Footnote 28
5. VIRUSES AND STAR WARS
Recall from Section 3 that case star augmentation, yielding v* in addition to plain v and Voice/ag* in addition to plain Voice/ag is involved when a language develops from a no-case language to either an accusative or an ergative language, distinguishing Acc or Erg NPs from non-cased (Nom) NPs. Case star augmentation (and case star deletion) is a general phenomenon in case systems.
Recall also the Case Ingredients Generalization in (17), repeated here.
- (17)
a. NPs are event licensed by heads: Voice, Appl, v, n, a, p, . . .
b. Event licensers of NPs may be case star augmented
c. The augmented case star may be variably marked: *, *+, *++
d. Nom = syntactically a non-case (regardless of its overt marking)
Accordingly, little v, for instance, comes in several flavors in individual case languages, as v, v*, v*+, v*++ (Sigurðsson Reference Sigurðsson2012). Thus, the v-case system in nominative-accusative/genitive/dative languages can be described as in (38), where the arrow reads as ‘yields’ (in PF morphology).Footnote 29
(38)
The case star approach has one thing in common with KP approaches, namely that it explicitly states that a cased NP is licensed in a more marked or informative structure than is a non-cased (Nom) NP, the more so the more oblique the case. Importantly, however, the augmented markedness or complexity is morphological and not syntactic. That is, syntactically, Icelandic is just like caseless languages, Chinese, Thai, Vietnamese, etc., in having only a general object-licensing v, the v*-flavors being activated in the externalization process, as (abstract) language-specific PF interpretations. In other words, as discussed in Section 3 above, case instructions (e.g. the ones in (38)) are PF externalization instructions (or interpretations).
Presumably, both Dat and Gen always ‘begin life’ as non-v cases, Gen as an adnominal n*-case and Dat as an Appl*-case. Subsequent introduction of both Dat and Gen into the v-system can be seen as the result of two ‘virus invasions’ or ‘star attacks’. Gen is commonly more oblique or peripheral than Dat within the v-system (see Haspelmath & Michaelis Reference Haspelmath, Michaelis, Crobett and Noonan2008), thus being licensed by a more marked v type head than both the other v-cases, as indicated in (38). That suggests, in turn, that Dat invaded the v-system before Gen did. Thus, nominative-accusative /n-genitive/Appl-dative/v-dative/v-genitive case systems might (for instance) come into being along the lines sketched in (39).
(39)
Similarly, Dat can invade the n-system, yielding adnominal datives, such datives being more marked than Gen within the n-system, thus licensed by n*+.Footnote 30 Also, as mentioned in Section 3, Acc has invaded the Appl-system from the v-system (presumably at least partly as a consequence of the loss of ablative in the prehistory of Germanic), thus being more marked than Dat within the Appl-system (licensed by Appl*+), even though it is less marked than Dat within the v-system.Footnote 31 In addition, both the adjectival (a) and the prepositional (p) subsystems, discussed in Section 4 above, have developed case licensing a- and p-heads.Footnote 32 Thus, (much of) the Icelandic case system can be described as in (40) (where Nom is disregarded).
(40)
The five subsystems thus make use of only three ‘common’ cases to mark 12 distinct relations. Other languages sometimes take different routes, introducing specific cases for (or limiting them to) some of the subsystems (‘dative’ for instance being confined to prepositional complements in Scottish Gaelic). Subsystem specificity of this sort can yield systems with many more cases than the Icelandic one (an interesting factor that is nevertheless largely disregarded here).
A parallel sketch for English is given in (41).
(41)
The cases are evidently not unitary building blocks. That is, applicative dative is not the ‘same’ case as verbal dative, adnominal genitive is not the ‘same’ case as verbal genitive, and so on. Not surprisingly, the traditional view that individual cases are primitives yields recalcitrant puzzles. Thus, it has been a well-known but mysterious fact (see Holmberg Reference Holmberg1991:143; see also Thráinsson Reference Thráinsson2007:173ff. and the references there) that Icelandic has genitive direct objects but no genitive indirect objects. The simple account or description under the present approach is that Gen has invaded the v-system as opposed to the Appl-system.Footnote 34 Another fact illustrating that the cases are not unitary is that Appl-datives differ from v-datives in being retained in the anticausative -st-construction (see below).
So-called ‘inherent’ case marking of objects is structural rather than lexical (pace Woolford (Reference Woolford2006) and many others). That is, the case is not the ‘direct responsibility’ of the lexical V root, but of the particular v type that combines with V to make up a full-fledged verb (and similar considerations apply to the other lexical categories). The v-selection of V, however, is commonly idiomatic or ‘fossilized’, such that a particular V may only select one particular v type, seemingly yielding a ‘lexical’ case marking relation. As one would expect, though, many Vs may select more than one v type. Thus, to mention only one type of such variation, some Icelandic verbs make a distinction between dative objects (commonly benefactive) and accusative objects (commonly affected), yielding minimal pairs like ‘comb the child.dat’ vs. ‘comb the hair.acc’ (for more observations of this sort for Icelandic, see Barðdal Reference Barðdal2001; Maling Reference Maling2002; Svenonius Reference Svenonius2002; Jónsson Reference Jónsson, Brandner and Zinsmeister2003, Reference Jónsson and Þráinsson2005; Thráinsson Reference Thráinsson2007; more generally, Malchukov & de Swart Reference Andrej and de Swart2009 and the references there).
Actually, the so-called ‘inherent’ or ‘lexical’ cases may be erased in certain Voices. Thus, the expletive Voice head in the anticausative -st-construction triggers deletion of v-datives, as opposed to Appl-datives. This is illustrated in (42)–(43).
(42)
(43)
The generalization behind these facts is that Voiceexpl erases all v-case stars as opposed to the Appl-case star, the latter triggering applicative dative marking (for a more detailed discussion of these and related facts, see Sigurðsson Reference Sigurðsson2011a, Reference Sigurðsson2012; Wood Reference Wood2012b).
The ‘final’ case of an argument, thus, is not only decided by the initial case licenser, such as v* and v*+, but also by Voice type, a fact which shows, again, that the individual morphological cases are not simple unitary building blocks, neither across languages nor language-internally. Rather, they are morphological entities that interpret a variety of underlying syntactic structures, with extensive (but not unlimited) variation both internal to and across languages.
6. CONCLUDING REMARKS
The Blake (Reference Blake2001) hierarchy in (2) represents a long tradition in the study of language, where overt markers are taken to be largely unitary and uniform building elements, both language-internally and across languages, each with canonical functional properties. This way of thinking of easily observable variation in some surface form as being linked in some more or less one-to-one–like fashion to a deeper and a more general system has been widely abandoned in the natural sciences, but it is still the prevailing view in linguistics. Call it the animistic view (avoiding the term ‘realism’): it is animistic in the sense that it ascribes some deeper sense to observable objects, such as case markers.
If the animistic view of case could be upheld – even only some very weak version of it – we would expect a parametric approach to case variation to be basically successful. However, no such approach has ever been developed. Symptomatically, Baker (Reference Baker2001) mentions only one putative case parameter in his ambitious attempt to construct a ‘periodic table of languages’, the ‘ergative case parameter’.
However, recall the Case Ingredients Generalization in (17), repeated here:
- (17)
a. NPs are event licensed by heads: Voice, Appl, v, n, a, p, . . .
b. Event licensers of NPs may be case star augmented
c. The augmented case star may be variably marked: *, *+, *++
d. Nom = syntactically a non-case (regardless of its overt marking)
In view of the extensive case variation observed within and across languages, the ingredients of case systems are amazingly limited: only a number of event licensing heads and a few variably marked case stars. These atomic elements readily lend themselves to relatively simple ‘parametric’ statements, as we already saw in Section 3 above in relation to ergativity and accusativity, the former arising as a consequence of case star augmentation of Voice/ag itself, the latter as a consequence of case star augmentation of little v c-commanded by Voice/ag. Similarly, as we saw, case star augmentation of Appl yields specific case marking of indirect objects, invasion of n*-case into the v-system yields a grammar with some genitive objects, and so on.Footnote 35
However, statements of this sort are not UG anchored. Markedness is a general third-factor phenomenon and syntactic heads are arguably materialized in individual I-language (out of general ‘concept material’) rather than pre-stored in UG (Sigurðsson Reference Sigurðsson2011b, Reference Sigurðssonc). That is, the statements in question are third-factor externalization generalizations and not UG parameters in the traditional Principles and Parameters sense. In view of the fact that UG parametric approaches cannot be upheld (Berwick & Chomsky Reference Berwick, Chomsky, Di Sciullo and Boeckx2011; Boeckx Reference Boeckx, Di Sciullo and Boeckx2011; Sigurðsson Reference Sigurðsson2004, Reference Sigurðsson2011b, Reference Sigurðssonc, Reference Sigurðsson2012), it is interesting, even uplifting, that language variation, nevertheless, can be systematically analyzed. It raises some hope that the incredibly complex externalization component of language can be successfully studied.
ACKNOWLEDGEMENTS
This article is a ‘follow up’ of sorts to the paper I presented at the Brussels Case at the Interfaces conference (BCGL 5) in December 2010 (see Sigurðsson Reference Sigurðsson2012). I would like to thank the organizers of the conference for their hospitality and the audience for helpful comments and suggestions. Many thanks also to Thórhallur Eythórsson, Martin Ringmar, Jim Wood, and, not least, Jeff Parrott for valuable comments and discussions. Thanks also to three reviewers for critical remarks that helped me clarify some moot issues and improve my presentation.
