Abstract

Early literacy and language skills are a significant precursor to children’s later educational success. Using social robots as a learning companion is one method to augment efforts to support this aim. SPARC has been shown as a useful way to combine reinforcement learning with human experts to reduce the state search space and for human experts to
intervene before any negative reward actions take place, however human expertise is not consistent. In this paper we investigate if SPARC can be counter-useful as expertise levels decline. Results from the study suggest that once a skill level below 90% of an expert is breached,
alternative LfD methods maybe more suitable.