Dutch causative constructions
Dutch periphrastic causatives consist of an auxiliary predicate (doen or laten), an effected predicate and several nominal slots, as shown in the example below:
(1) De politie deed/liet de auto stoppen. the police did/let the car stop
Causer Auxiliary Causee Effected Predicate Predicate
‘The police stopped the car (let the car stop).’
Most of the corpus-based studies of these constructions (Kemmer and Verhagen 1994; Verhagen and Kemmer 1997; Stukker 2005) suggest that doen is an auxiliary that expresses direct causation. It is used to categorise causative situations in which the Causer uses its own energy to produce the caused event encoded by the Effected Predicate. On the other hand, the auxiliary laten refers to indirect causation, when “some other force besides the initiator is the most immediate source of energy of the effected event” (Verhagen and Kemmer 1997: 67). The semantics of laten also covers situations of letting. In fact, it represents a continuum from coercion to enablement and permission (Verhagen and Kemmer 1997; Speelman and Geeraerts 2009) with some ambiguous cases in between. For example, the construction in (2) suggests two interpretations:
(2) Hij liet iedereen zijn roman lezen.
he let everyone his novel read
‘He had/let everyone read his novel.’
The most typical uses of doen are described as physical and affective causation. The former usually involves an inanimate Causer and Causee and a nonmental observable caused event:
(3) De aardbeving deed de muren trillen.
the earthquake did the walls shake
‘The earthquake made the walls shake.’
Affective causation typically involves an inanimate stimulus (Causer), a human cognizer (Causee) and a mental caused event:
(4) Je kapsel doet me denken aan een vogelnest. your hairstyle does me think to a bird-nest ‘Your hairstyle reminds me of a bird’s nest.’
As far as laten is concerned, its prototype is considered to be inducive causation, with a human Causer affecting a human Causee, normally intentionally and by means of communication (Stukker 2005). An example of inducive causation is given below:
(5) De trainer liet de spelers loopoefeningen doen. the coach let the players run-exercises do ‘The coach had the players do running exercises.’
Therefore, one can expect human Causers to be typical of laten, and nonhuman ones to favour doen. This is also what was found in previous quantitative multivariate studies (Speelman and Geeraerts 2009; Levshina 2011). The inherent semantic classes of the Causee (human being, abstract entity, artifact, etc.) have never shown strong effects in these previous analyses (Levshina 2011). However, the thematic roles of the Causee (quasi-patient or agent) have shown significant effects: laten is favoured by relatively agentive Causees, whereas doen is associated with patient-like affected Causees. The low relevance of the inherent semantic class of the Causee can be explained by the diverse roles of human Causees, who can be both relatively passive experiencers, as in (4), and active agents, as in (5).
As far as the Effected Predicate is concerned, our previous research (Levshina 2011) revealed a few distinctive verb classes at different levels of semantic specificity, from specific verbs to medium-grained semantic classes a la Levin and to highly abstract distinctions. This is in line with the constructionist approach, which claims that both exemplars and generalisations are stored in the speaker’s memory (Langacker 1987; Goldberg 2006: 45-65). On the most lexically-specific level, some verbs, such as denken aan ‘think of ’ are used exclusively with doen, and some others, such as weten ‘know’ and wachten ‘wait’, occur predominantly in the combination with laten. These expressions form low-level constructional pairings with specific meaning. Some exemplars of this type form clusters. For instance, the verbs of perception (zien ‘see’ and horen ‘hear’) are normally used with laten, whereas most predicates that designate internal mental processes - for instance, belief (geloven ‘believe’, vermoeden ‘suppose’), emotion (vrezen ‘fear’) and decision (besluiten ‘decide’) - tend to occur with doen. In addition, verbs of quantitative change along a scale (stijgen ‘rise’, toenemen ‘increase’) also prefer doen. These clusters form the middle level of generalisation. Finally, on the most abstract level, laten is in general preferred by semantically and syntactically transitive verbs (maken ‘make’, doden ‘kill’), whereas doen usually occurs with intransitive verbs with a patient-like first argument (verdwijnen ‘disappear’, smelten ‘melt’).
To summarise, one can expect a high effect of the semantic classes of the Causer and the Effected Predicate slots, and a weak effect of the Causee slot in predicting the choice between the two constructions. As for the Effected Predicate, it will also be interesting to see which level of granularity will be the optimal one in distinguishing between the constructions.
-  Some examples also contain an Affectee, which is the object of the Effected Predicate and theend point of the causation chain, e.g. the window in The strong wind caused the tree to break thewindow. Since Affectees expressed by NPs are infrequent in the corpus, the Affectee slot will notbe considered in our experiments. See also Stukker (2005), who shows that the semantic classesof Affectees are not relevant for the choice between doen and laten.