Childrens poor emotional self-regulation is associated withpoor mental health outcomes. This study presents methods thatimprove prediction rates of polite and frustrated speech usinglinguistic cues. These improvements can be used to help auto-matically identify characteristics of poor self-regulation in fu-ture studies. This work adds to previous research by consider-ing existing computer science, psychology, and psycholinguis-tics methodologies and findings. More specifically, featuresassociated with childrens cognitive control capacities acrossage groups are considered to investigate acoustic, semantic,and syntactic features in speech. The current analyses indi-cate that the features most predictive for polite and frustratedspeech differ, a combination of features work best for predict-ing both speech types, and the predictive quality of featuresdo not vary substantially by age. Further work should be con-ducted to clarify how well these findings transfer to general andclinical populations as well as to consider the developmentalnorms of different age groups.