What is the level of precision of phonological representations in working memory?

The precision with which verbal information is represented in working memory (WM) is a debated question. Some studies suggest that verbal WM precision is limited to abstract phonological levels of representation while other studies, using specifically designed paradigms, indicate that information can reach phonetic-level precision. The present study investigated at which linguistic level verbal WM operates by default, by probing memory for phonological versus phonetic information in a non-word WM paradigm. In three experiments, we presented non-word lists followed by a non-word probe, with negative probes differing from targets by a single phoneme. This phoneme was either a phonetic variant of the target (e.g., /t/ - /t*/), a phonologically close phoneme (e.g., /t/ - /d/) or a phonologically distant phoneme (e.g., /t/ - /v/). In the three experiments, we observed reliable rejection of negative probes differing by a phonologically distant phoneme, while rejection of negative probes differing by either a phonologically close phoneme or a phonetic variant was much less robust. This study shows that verbal WM preferentially involves phonological levels of representation, and with limited precision at this level.