Department of English Language Teaching, Farhangian University, P.O. Box 14665-889, Tehran, Iran
Abstract: (72 Views)
This study examines how scaffolded retrieval practice mediated by a large language model (LLM) influences the receptive knowledge of verb–noun collocations among intermediate learners of English as a foreign language, in comparison with a closely matched corpus-based approach. Drawing on usage-based perspectives on lexical entrenchment and cognitive accounts of retrieval and contextual variation, it was hypothesized that stronger short-term learning and more durable retention would be produced when adaptive LLM scaffolding was combined with safeguards ensuring linguistic authenticity. Sixty-five intermediate learners were randomly assigned to either an LLM-mediated condition or a corpus-based condition and took part in six retrieval-oriented sessions over a two-week period. Receptive collocational knowledge was measured through parallel pretests, immediate posttests, and delayed posttests administered four weeks later, while process data documented retrieval attempts, contextual diversity, and time-on-task. In addition, a purposive subsample of 12 participants engaged in semi-structured interviews, which were analyzed using reflexive thematic analysis supported by MAXQDA 2022 software. The findings indicated that learners in the LLM condition achieved significantly larger gains on both posttests. These advantages were primarily associated with higher retrieval frequency and greater contextual variability, whereas time-on-task played only a minor role. Interview data suggested increased noticing, growing confidence in validated LLM examples, and sustained engagement throughout repeated retrieval cycles. The results indicate that carefully scaffolded and monitored LLM use, embedded within structured practice schedules, can support collocational learning and complement corpus-based resources in second-language instruction.
Type of Study:
Research |
Subject:
Special