Previous work has shown that young children can be prompted to revise an utterance by an unspecific prompt like "hm?" or "what?". In a game setting this technique is re-introduced with two-year-old children who name pictures on cards that they collect to put on a game-board. The pictures elicit words that start with an onset cluster, like /tr/, /kl/, /st/, which are often reduced to singleton consonants by two-year-old speakers. The prompt is assumed to trigger their self-monitoring system, which could highlight a mismatch between the uttered word - with a reduced cluster - and the intended target - with a full cluster - if this full cluster is indeed part of the young speaker's segmental representation of the word. We compared the initial production and the prompted repeated productions of 20 children and studied the changes that were triggered by the prompt. Results show that (1) the technique works very well with these young speakers and triggers repeats in most cases, (2) changes involve improvements in most cases, ranging from segmentally more accurate productions and signs of knowledge of an initially absent additional consonant, to full consonant cluster productions, (3) the technique can help to disentangle word-form encoding errors from errors resulting from the stored segmental representation.