We conducted a preliminary study to evaluate the contribution of confidence measure for improving the comprehension of an automatic transcription. The future framework is deaf children studying in a standard school classroom. We defined local confidence measures that can be estimated as soon as possible without waiting for the recognition process to be completed. We then defined different modalities to highlight words of low confidence in an automatic transcription and we presented the different modified transcriptions to several subjects who had to answer to some questions of comprehension and to restore the sentences originally uttered. We showed that highlighting words of low confidence can improve the comprehension of the automatic transcription.