The role of semantic transparency in the acquisition of English collocations by Saudi learners

The importance of collocations in L2 language development is well established, but it is also widely acknowledged that collocations are difficult for L2 learners to acquire. Several factors have been investigated that affect their learnability, such as their frequency and the influence of the learner’s first language. Nevertheless, the possible effect of the degree of semantic transparency of the collocation being learnt has not yet been sufficiently addressed. Furthermore, the few existing studies have failed to account for the effect of prior familiarity with the constituent words when assessing the learning of collocations. This thesis addresses these gaps by investigating whether the degree of semantic transparency of L2 collocations affects their learnability when prior familiarity is controlled for. Forty-four English verb-noun collocations were selected on the basis of rigorous criteria. Two types of semantic transparency measure were obtained for each collocation. First, human ratings of transparency were elicited from 46 native speakers of American English. Second, distributional semantic measures of transparency were calculated on the basis of co-occurrence frequencies in a corpus. Both the collocations and their constituent words were then explicitly taught to 94 pre-intermediate Saudi students who did not recognize either the collocations or their constituents in a pretest. After five weeks of daily teaching, based on the results of two tests, the extent to which a collocation was learnt was found to be significantly positively correlated with its degree of semantic transparency, whether estimated from human ratings or computational measures. Collocations with lower semantic transparency were less well learnt compared to more transparent ones. However, overall, large learning gains were demonstrated by all the learners in the study, indicating the effectiveness of explicit instruction in promoting the learning of L2 collocations. These findings have important pedagogical implications, since they suggest that collocations can be successfully taught even to low level learners but that most effort needs to go into teaching less transparent combinations. The strong correlation between human ratings of transparency and the cosine similarity measure of the constituent words suggests that this measure can successfully be used to estimate transparency. This is a much less costly and laborintensive method than obtaining human ratings, which therefore has the potential to be scaled up for application in materials writing and curriculum design more generally.