Abstract
Corpora are well-known for the affordance to make linguistic regularities salient. Since the coinage of the term ‘data-driven learning’ (DDL) in the 1990s, much has been done to investigate the effects of DDL on learning vocabulary, most notably lexico-grammatical patterns. However, less researched is how learners construct vocabulary knowledge from scratch with DDL approaches. While generalizing recurring lexico-grammatical patterns is relatively straightforward, deducing meanings of unknown words from truncated concordances can be challenging. In this exploratory study, learners constructed knowledge of new words with the inductive approach or deductive one; and their retention of word knowledge was measured holistically using a word knowledge framework. The ways in which learners availed themselves of corpus data in the construction of word knowledge were also examined. The results provide evidence that the inductive approach is more conducive to developing collocational knowledge; and the deductive approach provides the opportunity to consolidate definitional knowledge of the newly acquired words.
Disclosure statement
No potential conflict of interest was reported by the author.
Notes
1 Available at https://corpus.byu.edu/coca/
2 Available at https://www.macmillandictionary.com/
3 Available at https://www.apowersoft.tw/free-online-screen-recorder
4 Available at https://www.lextutor.ca/list_learn/bnc_coca/
5 Following Nation’s (Citation2001) word knowledge framework, the word form recall test measures the production of written forms; the POS recognition test measures the recognition of word parts; the definition recall test measures the production of word meanings; the definition recognition test measures the recognition of word meanings; the collocation recall test measures the production of collocations and/or grammatical functions; the collocation recognition test measures the recognition of collocations and/or grammatical functions. Note that class time constraints precluded the possibility of measuring all aspects on the framework.
6 Plonsky and Oswald’s (Citation2014) benchmark for interpreting effect sizes in L2 research: r = 0.25 (small effect); r = 0.40 (medium effect); and r = 0.60 (large effect).