Case frames in Dutch experience predicates
Another example of a set of constructions entertaining horizontal relations of contrast is the argument realisation network in Middle Dutch. Middle Dutch relies on case frames for indicating the agentivity of the participants (Van de Velde 2004). Agentivity can be broken down into features like volition, responsibility, control, animacy, instigation, movement etc. (Lakoff 1977; Dowty 1991; Nrnss 2007; Grimm 2011, among others). Glossing over the details, which are much discussed but are not our immediate concern here, the whole system is then fairly straightforward, and is well-known from other Indo-European languages: the nominative is used for animate participants, exerting volitional control and instigating the action. Accusatives, on the other hand, are used for non-agentive undergoers, who have no control over the action expressed by the predicate. The dative case is used for sentient, volitional, but non-instigating participants (see Nrnss 2007:198 for a similar proposal), and genitives are used for mainly inanimate patients that are not fully affected by the action expressed by the predicate. This yields a cline of cases, as represented in (1), in which dative and genitive occupy the space in the middle, between the highly agentive nominative and the highly patientive accusative.
- (1) Agentive nominative - dative - genitive - accusative non-agentive
These are the prototypical functions of the different cases, from which language users can deviate on semantic grounds. it is, for instance, not uncommon to encounter animate accusatives, and to the extent that animacy is correlated with (or part of) agentivity, this poses a conflict. The accusative then highlights the fact that the undergoer is non-instigating, fully affected etc., glossing over the animacy aspect. in languages where the animacy is given higher priority, this can lead to so-called differential object marking, as in Spanish or Afrikaans, where animate objects can be preceded by a preposition, a and vir, respectively.
An insightful way to account for the assignment of case in Middle Dutch is by taking a constructional approach, in which, unlike in a projectionist approach (see Levin and Rappaport Hovav 2005 for this term), verbs do not assign case automatically according to the valency with which they are registered in the lexicon, but rather select a case frame that contributes its own constructional meaning. The whole system can best be seen at work in verbs that do not straightforwardly map onto the prototypical process in which an actor exerts physical force to affect an undergoer (hit, destroy, break...). This is, for instance, the case with so-called experience processes, expressing mental and sometimes also physical experiences such as amaze, wonder, annoy, forget etc. It is not clear who is the instigator in the case of annoy, and whether the participants involved - the experiencer or the stimulus - are volitionally implicated. Nor is it clear to what extent the annoyee is really affected. A similar case can be made for amaze and forget, for instance. What we see in Middle Dutch then, is that such verbs occur with a wide range of case frames. For the verb wonderen
- (‘amaze’, ‘astonish’, ‘surprise’), the examples in (2) to (5) give an idea of the attested variation.11
- (2) NOMINATIVE-STIMULUS, DATIVE-EXPERIENCER
Sere wonderde Pharaone sine vulmaectheit so scone
much amazed Pharao:DAT his:NOM perfection:NOM so beautiful
‘Pharao was amazed by his very beautiful perfection.’ (MNW, s.v. wonderen)
(3) GENITIVE-STIMULUS, DATIVE-EXPERIENCER
Des wondert mi utermaten
this:GEN amazes me:DAT highly
‘I was highly astonished by this.’ (MNW s.v. wonderen)
(4) NOMINATIVE-EXPERIENCER, ACCUSATIVE-STIMULUS
Die goede man, die in clenen dingen die
the good man who:NOM in little things the:ACC
grootheit van onsen here plach te wonderne wel zere
greatness:ACC of our:DAT lord:DAT used to amaze well very
‘The good man, who used to be highly amazed about the greatness of our lord by little things.’ (MNW, s.v. wonderen)
(5) NOMINATIVE-EXPERIENCER, GENITIVE-STIMULUS
Si wondrens sere algader
they:NOM wonder=this:GEN much altogether
‘They were all highly astonished about this.’ (MNW, s.v. wonderen)
The language user can opt for the strategy of pressing the experience verb wonderen in the canonical transitive mould, as for example in (4). But since the experiencer is a non-prototypical agent, it may be encoded as a dative as well, and since the stimulus as well is a non-prototypical agent or patient, it can get genitive encoding. As is shown in (3), there is no need to have a nominative- marked subject. The so-called impersonal construction has a third person singular verb with obliquely marked participants. This shows that the classical two-way typological distinction between experiencer-subject and experiencer-object verbs 
is a gross oversimplification for Middle Dutch (unless the notion of subject is stretched considerably).
In Van de Velde (2004) it is shown that the distribution of the different constructions over the experiencer verbs is semantically motivated: verbs with an inherently more agentive experiencer (e.g. denken ‘think’) are statistically more likely to go with case frames that acknowledge the agentivity of the experiencer, whereas verbs with an inherently less agentive experiencer (e.g. ontbreken ‘lack’) are more likely to encode the experiencer as an undergoer. Verbs like wonderen take a middle position, and excel in the use of the impersonal construction exemplified in (3). Focusing on the cases where the verbs come with two arguments, thus ignoring cases like (6) and (7) without an explicit stimulus and without an experiencer respectively, the different case frames can be ordered on a cline from agentive to non-agentive experiencers, see Table 2. Experiencers are seen as maximally agentive when they are the subject of a regular transitive nominative-accusative construction, as this is the construction that is used for straightforwardly transitive verbs like break, destroy etc. Marking the stimulus with genitive case decreases the agentivity of the experiencer as the stimulus is now represented as not fully affected, and thus somehow escapes the control of the experiencer. The next level is the so-called “impersonal construction” with a dative and a genitive argument, representing a twofold deviance from the transitive frame. The experiencer can be represented as even less agentive, when the stimulus is in the nominative, marking the latter as the starting point of the causal chain.
(6) NOMINATIVE-EXPERIENCER, NO STIMULUS
Alle diet sagen an, wonderden wijf ende man
all who=it saw to amazed woman and man
‘All who saw it, both women and men were amazed.’ (MNW, s.v. wonderen)
(7) GENITIVE-STIMULUS, NO EXPERIENCER
Waer dat zake dat zijns ontbrake
were it case that he:GEN lack
‘If it is the case that he is lacking (=he is dead).’ (MNW, s.v. ontbreken)
Table 2: Case frames ordered on a cline of experiencer agentivity
Experiencer-nominative; Stimulus-accusative Experiencer-nominative; Stimulus-genitive Experiencer-dative; Stimulus-genitive Experiencer-dative; Stimulus-nominative
The graph in Figure 6 (based on Van de Velde 2004, with data from the citation corpus of MNW) shows that the case frames are non-randomly distributed over different types of verbs: inherently more “agentive-experiencer” verbs select constructions higher on the cline in Table 2, and vice versa. The correlation is statistically significant.
Figure 6: Distribution of constructions over various types of experiencer verbs
The different case frames combining with the experience predicates are horizontally related in the constructional network: differences in meaning are correlated with differences in form, in such a way that we get a cline of closelyrelated constructions, which are distinguished from one another in the values they have for a set of features. For constructions with experience predicates, the cline goes from “agentive experiencer” to “non-agentive experiencer”, and the constructions that form aggregate points on a cline do not entertain a hierarchical relation to each other, but are related in a horizontal way. The case frames can be seen as a cluster of constructions at a certain horizontal level in a constructional network (see also Trousdale 2008: 308, Figure 3 for a visualisation).
-  In fact, the use of case-marking in argument realisation is derivative from their earlier function in spatial marking (see also the localist hypothesis). With the exception of the nominative,which probably is merely the absence of case (Van der Horst 2008:145, referring to Schuchardt),the cases also have or had a role in the expression of essive relations. Accusative could be usedto express adessive (cf. Latin Romam ire) and the genitive, as its name suggests, expresses theabessive, or “source”, whether the source of possession, the source of an experience etc.
-  The examples under (2)-(5) do not exhaust all the possibilities. Apart from case-based argument realisation, there were also voice-based strategies (see Van de Velde 2004 and below). Fora resolution of the difficulties posed by case syncretism etc., see Van de Velde (2004: 73-76).
-  I will not dwell on the applicability of the notion subject in non-canonically case-markedpredicates in Middle Dutch. On this topic, see Weerman (1988), Eythorsson and Barddal (2005)and Barddal and Eythorsson (2012).
-  Kendall’s Tau-b, a test for association, indicates a correlation of 0.69 (Asymptotic StandardError = 0.02). Chi-Square: p < 0.0001.