This study explored the effect on vocabulary retention of vocalizations involving three cognitive processing depths ("repetition," "manipulation," and "generation"). Eight participants in an English for academic purposes (EAP) context encountered five unknown words when working alone and five different words when working in pairs. In each condition, they studied the words from a text and dictionary, completed puzzle and question tasks, and did a stimulated recall. Tests one week and one month later assessed short- and long-term retention. The data were examined to investigate whether vocalization of deeper processing was associated with better retention, how deeper processing facilitated matches between task utterances and test responses, and which condition (solitary or collaborative) produced higher test scores. Quantitative results show that delayed test scores correlated inversely with repetition during collaboration. Qualitative analysis suggests that vocalizations were better remembered when participants deployed deeper processing to create mnemonics, connect input with L1/L2 knowledge, and express opinions. Solitary and collaborative conditions were equally effective in promoting retention.