|Title:||Testing knowledge of whole English collocations available for use in written production: Developing tests for use with intermediate and advanced Danish learners||Add Dissertation|
|Author:||Robert Revier||Update Dissertation|
|Email:||click here to access email|
|Institution:||Aarhus University, Denmark, Department of English|
|Linguistic Subfield(s):||Applied Linguistics; Semantics; Language Acquisition;|
|Abstract:||This foreign language testing research aimed to explore two assumptions: (a) testing of collocation knowledge should target knowledge of whole collocations and (b) the learning burden of a collocation is in part a function of its semantic composition. The CONTRIX, developed to serve as the main testing instrument, is equipped with a constituent matrix capable of eliciting responses which can illuminate knowledge of whole English collocations. The CONTRIX features three semantic categories of collocation: transparent, semi-transparent, and non-transparent. This internal structure serves to explore whether collocation knowledge is supported by semantic transparency. Five empirical studies were conducted to generate validation information.
Studies 1 and 2 served to (a) generate test reliability and validity data, (b) investigate the extent to which collocation knowledge is supported by semantic transparency, and (c) chart the development of collocation knowledge among Danish EFL learners representing three levels of formal education. Test scores were found to be reliable and valid. Overall collocation knowledge appeared to have developed in step with general English proficiency, and was partially supported by semantic transparency. The dominant knowledge component for all three education levels was the knowledge of transparent collocations. The gap between knowledge of semi-transparent and transparent collocations narrowed over time. Although knowledge of non-transparent collocations also increased over time, the relative gap separating knowledge of non-transparent collocations and transparent collocations remained largely constant.
Study 3 investigated the productive quality of the CONTRIX test. CONTRIX-3 was administered together with a receptive and a productive version of COLTRANS-2, which was designed to test knowledge of whole English collocations on the basis of translation. As expected, the CONTRIX-3 total test scores aligned more closely with scores obtained on the productive version of the COLTRANS than they did with the receptive scores. The section scores obtained with the productive version of the COLTRANS did not, however, show the expected stair-step pattern in relation to the three categories of transparency.
Study 4 investigated the CONTRIX’s capacity to track the development of collocation knowledge over real time. To generate true longitudinal data, CONTRIX-1 was administered a second time to the same groups of learners who, one year earlier, had participated in Study 1. The results showed that the sensitivity of the CONTRIX was sufficient to capture significant growth in collocation knowledge over the period of a year. These findings give support to the validity of using cross-sectional CONTRIX data to chart the development of collocation knowledge.
Study 5 served to validate CONTRIX-4 against other language tests. CONTRIX-4 featured semi-transparent collocations. These were selected to represent three successive frequency bands, which served to investigate the extent to which collocation knowledge was supported by frequency of occurrence. To generate correlational data, CONTRIX-4 was administered together with four language tests. CONTRIX-4 total test scores indicated desirable discrimination, and section scores showed the expected stair-step frequency profile. The observed inter-correlations showed signs of both convergence validity between CONTRIX-4 and other measures of multiword lexical knowledge and divergence validity between CONTRIX-4 and a measure of general language use.