Syntactic alternations of Dutch psych verbs

Psych verbs, i.e. verbs expressing some mental state or event, are known for their wide range of syntactic variation in and between languages (Croft 1993; Kitis 2009; Verhoeven 2010), and Dutch is no exception. Particularly interesting about the Dutch psych verbs is that two argument realizations may be possible for a single verb. Below, the verb ergeren (‘to annoy’) is shown to realize its experiencer, i.e. the participant that experiences the mental state, in both object (1) and subject (2) position. (1) Transitive argument construction: Maar iets ergert medefirmant Melkert (ConDiv corpus) But something annoys business_partner Melkert ‘But something annoys business partner Melkert.’ (2) Reflexive argument construction: Hij ergert zich aan de besluiteloosheid van het kabinet (ConDiv corpus) He annoys himself to the indecisiveness of the cabinet ‘The indecisiveness of the cabinet annoys him.’ Drawing from data of the ConDiv corpus of written Dutch (Grondelaers et al. 2000) and the Corpus of Spoken Dutch (Oostdijk et al. 2002), we have investigated the factors driving this alternation for the verbs ergeren (‘to annoy’), interesseren (‘to interest’), storen (‘to disturb’) and verbazen (‘to amaze’). The statistical work horse technique was logistic regression. Many theoretical frameworks deal with the argument variation of psych verbs through some form of the agentivity hypothesis, which by and large claims that more agentive participants are more likely to take up subject position (Hopper and Thompson 1980; Grimshaw 1990; Dowty 1991; Langacker 1991; Croft 1993; Pesetsky 1995; Vanhoe 2002). However, we will claim that it is important to distinguish between the type and token level agentivity hypotheses, which are often conflated, yet have fundamentally different theoretical implications (Dowty 1991: 579–581; Goldberg 1995: 220–221; Levin and Grafmiller 2012: 220–221). The type level compares the preference for argument constructions between verbs (Van de Velde 2004). Among the verbs under scrutiny, it was found that those verbs whose lexical meaning implies a more agentive experiencer, did not more often realize this experiencer in subject position. Meanwhile, the token level concerns differences within the occurrences of a single verb. Using the operationalization of Levin and Grafmiller (2012), we did observe that the participant that causes the mental state was more likely to appear in subject position if it was animate, and hence more capable of agentive action. Other hypotheses were tested as well. The pronominality of the participants turned out to be an even better predictor of the employed argument construction than animacy, while the etymology of the verbs was not shown to provide critical information to predict the dominant construction (cf. Klein and Kutscher 2005: 41–45). To conclude, we believe that our failure to confirm the agentivity hypothesis at the type level indicates that caution may be in order when predicting the dominant argument construction of a verb based on its lexical meaning. However, the confirmation of the agentivity hypothesis at the token level can be taken to show that argument constructions do add meaning to the utterance, on top of the lexical meaning of the verb (Goldberg 1995; Colleman and De Clerck 2009). Perhaps most importantly, by not confirming the agentivity hypothesis at the type level while confirming it at the token level, this study may serve as an example of the importance to distinguish between both.status: publishe