At what level of schematicity should we investigate argument structure variation?

editorial reviewedA recurring discussion in formal-theoretical approaches to language is at which level of schematicity to describe argument structure (Müller 2006; Goldberg 2013). On the one hand, so-called ‘lexical’ approaches describe argument structure at the level of the individual verbs, viz. in the verb’s lexical entry (e.g. Sag et al. 2003; Müller 2018). On the other hand, so-called ‘phrasal’ approaches posit highly schematic argument structure constructions (e.g. Goldberg 1995; Tomasello 2003). These constructions may in principle combine freely with verbs, although they can have preferences as to with which verbs they combine in actual practice (Goldberg 2006: 22; Goldberg 2013: 439–440). This discussion is highly relevant when conducting quantitative analyses of variation in argument structure. Consider the Dutch examples in (1)-(3), where the preposition naar ‘to’ is optional. Do these examples instantiate a single alternation between a schematic transitive construction and a schematic naar-construction? Or are we actually dealing with three different alternations, viz. one between a transitive and prepositional zoek-variant, another one between a transitive and prepositional verlang-variant, and one between a transitive and prepositional peil-variant? (1) Voor iedereen dien je (naar) een individuele oplossing te zoeken. For everyone ought you (to) an individual solution to search ‘You ought to look for an individual solution for everyone’. (Sonar-corpus, id: WS-U-T-B-0000000143.p.7.s.3) (2) Dirk De Wilde verlangt (naar) wat meer creativiteit. Dirk De Wilde desires (to) some more creativity ‘Dirk De Wilde desires some more creativity.’ (Sonar-corpus, id: WS-U-E-A-0000202851.p.1.s.23) (3) 'What's in a name' dachten we en peilden (naar) jullie mening. What’s in a name thought we and gauged (to) your opinion ‘”What’s in a name”, we thought, and gauged your opinion.’ (Sonar-corpus, id: WR-P-E-C-0000009624.p.1.s.4) Hybrid approaches have also been proposed that allow argument structure to be described either at a lower or higher level of schematicity or that combine elements from both proposals (e.g. van Trijp 2011; Boas 2014; Diessel 2019: 115–141). Still, the question remains: how can we determine which level of schematicity is most appropriate? I argue that this essentially constitutes an empirical question, and propose a methodological procedure to deal with it. The procedure consists of three steps: (i) define a number of possible levels of schematicity, based on theoretical grounds; (ii) formulate hypotheses at several of these levels; (iii) systematically put these hypotheses to the test. The alternation(s) in (1)-(3) are put under scrutiny to illustrate this procedure. By applying the Lexical Origin Hypothesis at various levels of schematicity (Goldberg 1999; Perek & Lemmens 2010), the following more concrete hypotheses are derived. These hypotheses will be tested on data of the Sonar-corpus of written Dutch, using distributional vectors and mixed regression modelling (Oostdijk et al. 2013). • The naar-construction expresses directionality, the transitive construction does not. • The meaning of ‘search’ has specialized to ‘seek to make or to acquire’ for the transitive variant, and to ‘look for’ for the prepositional variant. • The meaning of ‘desire’ has specialized to ‘demand’ for the transitive variant, and to ‘long for’ for the prepositional variant. • The meaning of ‘gauge’ has specialized to ‘directly assess’ for the transitive variant, and to ‘gauge by asking’ for the prepositional variant