Horizontal links in the constructional network and their role in constructional contamination : connections between Dutch krijgen-constructions

In this dissertation, we investigated the intricacies of the phenomenon of constructional contamination, which is the effect whereby one grammatical construction influences a subset of instances of another construction in its realization due to (coincidental) formal similarity between both constructions. This was done through an extensive empirical study of two cases of grammatical alternation in Dutch: (i) the alternation between the [Aux+Part] and [Part+Aux] word order in receptive and resultative verb clusters with krijgen ‘to get’ and (ii) the variation between the prepositions van ‘from/by’ and door ‘through/by’ in the prepositional phrase encoding the agent of the receptive construction. We studied potential contamination effects in these alternations in two national varieties of Dutch, viz. Belgian Dutch and Netherlandic Dutch, and with a range of methodological approaches, viz. corpus studies, an acceptability survey, sentence embeddings generated by a large language model and a psycholinguistic priming experiment. (i) The mechanism underlying contamination. In earlier work on contamination, the phenomenon was explained as the consequence of language users storing unanalyzed strings of a contaminating construction as exemplar chunks, which they later recycle in a target construction. Construction grammar assumes that formally similar constructions are connected in the minds of language users, so the question rises if these so-called “horizontal links” contribute to contamination, as a complementary mechanism to chunking-related processes. We indeed found several indications for this. For example, contamination also occurs in cases where constructions do not exhibit exact string similarity, but are only similar at a more schematic level. In another case, a reverse contamination effect was found, meaning that the alternant for which chunks are supposedly stored is actually avoided in the target construction. (ii) Lectal variation in contamination. We have shown that contamination effects can differ between national varieties. More specifically, it was found that contamination occurs with a more restricted set of verbs in Netherlandic Dutch compared to Belgian Dutch, and that in the latter variety, contamination occurs more easily based on abstract similarity. We explained this as a consequence of weaker horizontal links between certain constructions in Netherlandic Dutch, and, tentatively, as a consequence of the relatively more exemplar-based nature of Netherlandic Dutch, which might result in stronger horizontal links at lower levels (i.e. less abstract) in the constructional taxonomy. (iii) Bidirectionality of contamination. We confirmed the hypothesis that contamination is bidirectional in nature. However, we have also shown that it is not necessarily symmetric: contamination effects can, in some pairs of constructions, occur more easily in one direction than the other. (iv) Contamination affects acceptability judgments. We have shown that the effects of constructional contamination extend beyond language production: if language users evaluate the acceptability of instances of a construction, they might be influenced by the connections that construction has with formally similar constructions in their minds. (v) Contamination shapes grammar. We have demonstrated that contamination is very likely blocking the [Aux+Part]-order in resultative clusters. This indicates that, in some cases, connections between constructions can better explain tendencies in grammar than postulated grammatical properties of constructions.