Syncretisms in case systems, the division of labour between adpositions and case morphemes, stacking of case morphemes and similar phenomena led Caha (2009) to a finely articulated functional sequence for case. This underlying representation together with the spellout algorithm of Nanosyntax enabled a simple explanation for a wide variety of cross-linguistic phenomena, a success which spawned a wave of interest in the topic (e.g. Jayaseelan 2015; Harðarson 2016; McFadden 2016; Smith et al. 2016; Zompi 2017). From the outset however, one paradox stuck out: a number of languages point to a hierarchy including ‘DAT > GEN > ACC’, while some Germanic languages such as Icelandic fairly systematically point to ‘GEN > DAT > ACC’ (Caha 2009: 273ff).

The following Czech paradigms illustrate one kind of facts that led Caha (2009) to ‘DAT > GEN > ACC’:

(1)   two, MASC man, SG
  NOM dv-a muž
  ACC dv-a muž-e
  GEN dv-ou muž-e
  LOC dv-ou muž-i
  DAT dv-ěma muž-i
  INS dv-ěma muž-em

Hierarchies are inferred from such paradigms given structural contiguity as a requisite for syncretism, a theorem falling out from an independently developed spellout mechanism in the Nanosyntax approach of Starke (2002; 2009); Caha (2009); Taraldsen (2009); Pantcheva (2011); De Clercq (2013); Márkus (2015); Vanden Wyngaerd (2016) among others – but see also e.g. Bobaljik (2007); Burzio (2007) for different approaches. This adjacency, or rather the impossibility of syncretism under non-adjacency has come to be called *ABA, a fitting name devised by Bobaljik (2007).

The problematic Icelandic facts are illustrated by:

(2) NOM hlutur faðir hreyfing
  ACC hlut föður hreyfingu
  GEN hlutar föðurs hreyfingar
  DAT hlut föður hreyfingu

Under Caha’s DAT > GEN > ACC hierarchy, the Icelandic cases come out as a *ABA violation, e.g. for hreyfing ‘movement’:

(3) ACC u
  GEN ar
  DAT u

The same *ABA pattern is attested in a number of Germanic languages, for instance Middle English (see also Harðarson 2016 for other Germanic languages):

(4) NOM namen
  ACC namen
  GEN namene
  DAT namen

The Middle English pattern is particularly interesting in that it shows a classical ACC ≠ (GEN = DAT) pattern elsewhere, e.g. in the singular of the same declension:

(5) NOM name
  ACC name
  GEN namen
  DAT namen

What has not been noted is that the same paradox is found in English and in Spanish, depending on what one chooses to call “dative” and “accusative” respectively. English dative arguments surface either as prepositonal phrases, or as noun phrases (in the so-called “dative shift” construction):

(6) a. Sally sent a toy to Justine
  b. Sally sent Justine a toy

In the former case, the case marking is the preposition ‘to’ (I am adopting the tradition – formalised in Caha’s system – which views prepositions such as to, of, by, as realising the same features realised by case affixes in other languages). In the shifted dative there is no case-marking (or put differently there is a zero case marking).

If we choose to call “dative” the latter case, English has an *ABA violation given Caha’s hierarchy, since now the accusative and the dative are syncretic (∅) across the genitive, which is expressed by the prepostion of:

(7) ACC
  GEN of

If on the other hand, we call “dative” the unshifted (6a), there is no *ABA violation in English:

(8) ACC
  GEN of
  DAT to

When comparing “dative” across languages, it is therefore crucial to decide which form we pair with which. Does a Czech “dative” correspond to the English PP form or to the English shifted (bare) dative? What about the Icelandic dative? And how can we tell?

Spanish shows the same problem, but for accusatives. The Spanish accusative receives two possible realisations (data from Rodríguez-Mondoñedo 2006):

    1. (9)
    1. a.
    1. María
    2. Mary
    1. quiere
    2. wants
    1. a
    2. PREP
    1. un
    2. a
    1. abogado
    2. lawyer
    1. [+animate, +specific]
    1. Mary wants a (specific) lawyer
    1. b.
    1. María
    2. Mary
    1. quiere
    2. wants
    1. un
    2. a
    1. abogado
    2. lawyer
    1. [+animate, -specific]
    1. Mary wants a lawyer (any lawyer)

Again, if one chooses to call “accusative” the former, i.e. (9a), Spanish also exhibits an *ABA violation: the dative is syncretic with the accusative in using the preposition a, while the genitive is distinct, using de:1

(10) ACC a
  GEN de
  DAT a

If on the other hand, we call “accusative” (9b), there is no *ABA violation in Spanish either:

(11) ACC
  GEN de
  DAT a

To resolve apparent violation of the *ABA generalisation such as the one in Icelandic, Harðarson (2016) proposes that each language can choose a (slightly) different hierarchy. The English and Spanish facts suggest that the spirit of this approach is on the right track: once the correct cases are compared to each other, Icelandic and Czech will turn out to use different parts of a single underlying hierarchy, creating the surface illusion of two different hierarchies, similarly to Harðarson (2016) but without giving in to postulating different hierarchies for different languages.

The core of the issues above is that our current taxonomy of case leads us to give a single name to two distinct entities, potentially leading to spurious generalisations. We need to be able to distinguish the two “dative” realisations in English, and the two “accusative” realisations in Spanish. Only then can we ask the questions we should have been asking all along: what are the syncretisms of the bare shifted dative, are they different from the syncretisms of the PP dative? Is there a systematic pattern to that across languages? And in languages which have only a single realisation of the dative, which one of the two does it correspond to? Is that stable cross-linguistically in languages that have a single realisation of the dative? Mutatis mutandis for the accusative.

Let us thus revise the traditional taxonomy of case and distinguish the case of the shifted dative from the case of the base (unshifted) PP dative. I will call the former case an S-dative (SDat), standing for “shifted”, but also “structural”, or “smaller”. The PP version I will call the B-dative (BDat), for “base” dative (unshifted), but also “bigger”.2

The Spanish alternation should similarly be viewed as two distinct cases, both related to the theme argument (very much in the same way as an accusative theme and a nominative passivised theme are two distinct cases relating to the same theme argument). The Spanish accusative alternation also involves a bigger PP versus a smaller noun phrase, so I’ll adopt the same naming convention: B-accusative (BAcc) for the PP version, and S-accusative (SAcc) for the zero-case noun phrase realisation.

What is the case hierarchy of SDat/BDat/SAcc/BAcc then, and how does it relate to nominative and genitive case? The syncretism pattern relevant to our puzzle is straightforward. The English SDat is syncretic with the nominative and with the accusative in being case-less, here symbolised by ∅. At this stage, we don’t know whether the English accusative is SAcc or BAcc, and since it doesn’t matter for the argument, I will for the moment simply label it ENG-Acc. The first step is thus the observation that BDat = ‘to’ contrasts with SDat = ENG-Acc = Nom = ∅:

(12) BDat ≠ {SDat = ENG-Acc = Nom}

It is uncontroversial that the nominative is the bottom of the Nom/Acc/Dat/Gen hierarchy (cf. Caha 2009; McFadden 2016; Smith et al. 2016, etc.), this entails that SDat and the English accusative are adjacent to the bottom of the hierarchy, while BDat is higher (the “|” symbol indicates lack of ordering):

(13) BDat > {SDat | ENG-Acc} > Nom

Where is the genitive in this hierarchy? Given that Nom is the lowest, and Nom = ENG-Acc = SDat, the genitive must necessarily be above those, leaving only two options:

(14) a. BDat > Gen > {SDat | ENG-Acc} > Nom
  b. Gen > BDat > {SDat | ENG-Acc} > Nom

Whichever of these two options is correct, one crucial result follows: there exists a “dative” below Gen. On the (14a) option SDat is below Gen, on the (14b) option both SDat and BDat are below Gen. Given Caha’s results that Gen is sometimes sandwiched below a Dat and above an Acc, (14b) is ruled out, leaving us with (14a).

So far, the English lexicalisation of case thus looks like:

(15) NOM
  GEN of
  BDAT to

The reason for the (Dat = Acc) ≠ Gen pattern with English shifted datives is that we are looking at the Gen > SDat | Acc portion of the hierarchy, and hence the syncretism of an accusative with SDat against Gen is expected.

In Spanish, the situation is reversed. Taking the PP dative to be BDat, we have BDat = BAcc realised as a, and SAcc = Nom realised as ∅:

(16) {BDat = BAcc} ≠ {SAcc = Nom}

The Spanish genitive (de) is different from both of these blocks, and hence must be either between them or above them. Since we have already concluded that Gen is below BDat, Gen cannot be above both blocks, and hence we get:

(17) {BDat | BAcc} > Gen > {SAcc | Nom}

Given that Nom is the lowest, this translates into:

(18) {BDat | BAcc} > Gen > SAcc > Nom

The reason we find the apparently anomalous (Dat = Acc) ≠ Gen pattern in Spanish is now the reverse of English: the syncretism is (BDat = BAcc) ≠ Gen, so we are looking at the higher part of the case hierarchy, where BDat & BAcc are indeed adjacent “outside” of Gen. The syncretism is thus expected and does not violate the *ABA generalisation.

Let us put together the portions of the hierarchy revealed by English and by Spanish:

(19) a. EN: BDat > Gen > {SDat | ENG-Acc} > Nom
  b. SP: {BDat | BAcc} > Gen > SAcc > Nom

Spanish shows that the accusative below Gen is SAcc, not BAcc, and hence ENG-Acc can be substituted by SAcc:

(20) a. EN: BDat > Gen > {SDat | SAcc} > Nom
  b. SP: {BDat | BAcc} > Gen > SAcc > Nom

Putting the two together, we get:

(21) {BDat | BAcc} > Gen > {SDat | SAcc} > Nom

The relative order of BDat/BAcc and SDat/SAcc is immaterial for our puzzle about (DAT = ACC) ≠ GEN. For completeness, I’ll note that ongoing work suggests that the order is:

(22) BDat > BAcc > Gen > … > SDat > SAcc > Nom

Let us turn to our initial puzzle: why do the syncretisms of e.g. Czech seem to contradict the syncretisms of e.g. Icelandic?

Czech has a single realisation of the dative and accusative case. Barring strong evidence to the contrary, we do not want to postulate a third flavor of dative, or of the accusative, and hence Czech datives are either SDat or BDat, and Czech accusatives are either BAcc or SAcc. Which is it?

The syncretism patterns of Czech require the Czech accusative and the Czech dative to be on different sides of the genitive. This leaves only two options for the single Acc and Dat in Czech (leaving out irrelevant portions of the sequence for clarity):

(23) a. BDat > Gen > SAcc > Nom
  b. BAcc > Gen > SDat > Nom

The Czech accusative can be syncretic with the nominative while contrasting with both the dative and the genitive, as (1) above or the following (adapted from Caha 2009: ch. 3, ex. (50)):

(24)   chicken, PL
  NOM kuřat-a
  ACC kuřat-a
  GEN kuřat-∅
  DAT kuřat-ům

This rules out (23b) and leads us to conclude that the Czech dative is BDat (on a par with the English PP dative), and that the Czech accusative is SAcc, just like the English accusative. In fact, BDat + SAcc is a pattern which turns out to be common across Germanic, Romance and Slavic.

How about the (single) Icelandic dative? Is it an SDat or a BDat? And the (single) Icelandic accusative, is it an SAcc or a BAcc? The syncretism patterns in Icelandic require the Icelandic accusative and dative to be adjacent, “outside” of Gen. This entails that they are either both B-type (BDat & BAcc) or both S-type (SDat & SAcc), yielding the two options:

(25) a. {BAcc | BDat} > Gen > Nom
  b. Gen > {SDat | SAcc} > Nom

As noted by Harðarson (2016), Icelandic allows the syncretism {Nom = Acc = Dat} ≠ Gen. This is compatible with (25b) but not with (25a). We thus conclude that Icelandic uses the lower part of the hierarchy, with the Icelandic dative realising SDat, and the Icelandic accusative realising SAcc: Gen > {SDat | SAcc} > Nom. Icelandic uses the English dative shift pattern as its default pattern, so to speak.

We now hold the real difference between Icelandic and Czech: the Icelandic “dative” is a shifted, small, structural dative (similar to the English shifted dative) whereas the Czech dative is a big, base dative (similar to the English PP dative). The appearance of an ordering paradox between Icelandic and Czech came from mistakenly assuming that “dative” means the same in Czech and in Icelandic.

The icing on the cake of this solution is that its ingredients are independently known. Not only do we independently know that cases such as those of recipients and benefactors alternate between two morphological realisation in many languages, but it has also long been noted in the literature that the Icelandic dative is special in being more “structural”:

“Unlike German, where dative objects are oblique and behave syntactically much like PPs (Vogel & Steinbach 1998), Icelandic dative case is structural, and dative-marked objects pattern with ordinary accusative objects for various phenomena such as control, binding, secondary predication, promotion under passive, and so on (Maling 2001).” (Svenonius 2005)

The (DAT = ACC) ≠ GEN pattern is thus compatible with Caha’s strong claim: there is a single universal case hierarchy, and case syncretisms are structurally contiguous in that hierarchy.3