Listeners need to track continuous speech and phoneme perception to communicate effectively. However, children’s speech listening is usually assessed using sentence repetition tasks, primarily using one short sentence at a time (e.g. the BKB). These tasks do not measure the effort of mentally repairing and reconstructing the speech that was initially misheard.

To investigate how children perceive mispronunciations in continuous speech, we have created an event-related potential (ERP) paradigm using 10-minute narrated short stories. Here we present the results from 1) a behavioural study validating the perceptual distinctiveness of our mispronunciation manipulations; and 2) the pilot ERP study for one of the short stories.

In our stimuli validation study, we are currently testing 20 typically developing hearing children aged 6-11 years on a three-alternative forced-choice response paradigm (odd-one-out). Our stimuli consist of one syllable nouns (e.g., 'bag'). The odd-ones-out are mispronunciations of the trials’ noun altered by systematically changing the initial consonants’ place of articulation ('dag') or voicing ('pag'), while maintaining the manner, to yield non-words.

We predict that the children will successfully detect the odd-ones-out, with accuracy significantly higher than chance. We also predict that detection (as measured by RT and accuracy) will significantly differ for place compared to voicing articulation manipulations. Finally, we predict that detection will significantly improve with the children’s age, receptive vocabulary and processing speed.

In our pilot ERP study, nouns in the middle and at the end of sentences have been randomly assigned as target words every few sentences throughout an Enid Blyton story. A third of these target words remain in their original form, a third are mispronounced as per the stimuli validation study (but with up to four syllables), and the final third are nonsense words. To maintain attention the children are asked comprehension questions at the end.

We predict that the hearing children will show a graded N400 response, with the largest amplitudes for nonsense words followed by mispronunciations and the original words. This would indicate that the nonsense and manipulated words were unpredictable, and that the children were listening to them. We also predict that the children will show reduced processing efficiency, due to fatigue, over the course of the story. As evidenced by smaller N400 amplitudes for the mispronounced and original words. However, we expect N400 amplitude to remain large for the nonsense words as they are perceptually distinct.