Author:

Abstract:

Corpus-based approaches to constructional semantics are a rapidly developing area of contemporary linguistics. This dissertation, which presents a multivariate quantitative analysis of the Dutch causative constructions with 'doen' and 'laten', contributes to this research area by introducing several theoretical and methodological innovations. First, this study integrates the inter- and intraconstructional perspectives on semantics (cf. onomasiology and semasiology). To describe the internal structure of each constructional category, different exemplars of the constructions and their features are evaluated with regard to their intracategorial family resemblance. Several quantitative techniques, such as multidimensional scaling and hierarchical cluster analysis, are used to establish the relevant dimensions of semantic variation and the senses, which are operationalized as groups of similar exemplars. The similarity is based on thirty-five manually coded semantic and formal variables. The analyses show that the main conceptual dimensions along which the exemplars are distributed correspond to the distinction between mental and non-mental caused events, and the one between more or less autonomous Causees, or ‘direct’ vs. ‘indirect’ causation. The exemplars form a continuum along these dimensions, rather than a set of discrete senses, although there are several densely populated clusters, mostly associated with highly frequent low-level schemata, such as doen denken aan “remind of” and laten weten “let know, inform”. From the interconstructional perspective, the study explores the most distinctive exemplars and features, which are associated with the two constructions. These features and exemplars are related to the dimension ‘direct vs. indirect causation’, in accordance with the previous studies. This conclusion is made on the basis of a series of quantitative analyses, which involve cutting-edge statistical techniques: logistic regression with mixed effects, conditional inference trees and random forests. The analyses reveal that the intracategorial prototypicality and intercategorial distinctiveness of the exemplars and their semantic features do not correlate in the case of laten, which is more semantically heterogeneous than doen. This has important implications for Empirical Cognitive Semantics and especially for the ongoing debate about the nature of salience phenomena and their corpus-based operationalization. The study also explores variation in the use of the constructions in different registers and national varieties of Dutch, and proposes an innovative account of language variation and change. The analyses demonstrate that the difference between the lectal variants of the constructions is due to the different prominence of specific low-level schemata and referential situations. The results also suggest that the causative doen may be an obsolescent form. The main methodological novelty of this study is that constructional semantics is modelled in a fully bottom-up fashion, with the individual constructional exemplars as a starting point, and the generalizations (the senses and semantic dimensions), which emerge as a result of a fully automatic procedure. The approach thus integrates objectivity and measurability with intuitive interpretability of the results, which can be seen as one of the main challenges for contemporary corpus-based models of semantics.