Attraction through formal resemblance. Five case studies on constructional contamination

peer reviewedPhonological resemblance can exert an influence on two constructions leading them to converge (see e.g. Van de Velde & Van der Horst 2013). In this talk we will look into one particular case of formal attraction, which we call ‘constructional contamination’ (see Pijpops & Van de Velde 2016; Pijpops, De Smet & Van de Velde 2018; Van de Velde & Pijpops 2018). More concretely, constructional contamination arises when a construction has two slightly different formal variants, like He liked to please/meet other people (…) vs. He liked pleasing/meeting other people, and the distribution of the variants is skewed as a result of formal resemblance with another construction. In the case at hand, we have an alternance between to-infinitives and -ing forms as complements of like, and constructional contamination with object-referring nominalizations or gerunds would boost the use of the -ing alternant. This effect would measurably show up if, controlling for (known) semantic-syntactic factors the skew was found to be larger for those verbs that are more likely to occur as -ing forms in other contexts: the fact that meeting has a lexicalized meaning, as opposed to pleasing, would lead us to expect that the distribution of like to INF and like V-ING would be more skewed towards the latter variant with meet than with please (provided pleasing is not more frequent than (to) please for other reasons). In essence, constructional contamination is the paradigmatic version of what in syntagmatic contexts has been referred to as β-persistence by Szmrecsanyi (2005), and is related to structural priming (Bock 1986). In this talk, we will argue that constructional contamination is ubiquitous. The reason is that language users sometimes employ shallow parsing (Ferreira, Bailey and Ferraro 2002; Ferreira and Patson 2007; Dąbrowska 2014). What is distinct for linguists need not be distinct for language users, who are more likely to be influenced by superficial phonetic resemblance. We will present five cases of constructional contamination, and show that the effect is statistically robust