The role of ambiguity in child errors

This chapter compares two contrasting approaches to accounting for the verb placement errors in child Swiss German that are described in Schönenberger (2001). The first is a learning model that captures the errors because it both learns from ambiguous input and has a rich hypothesis space of interacting parameters (Gould 2017). The second captures the errors instead by means of a cognitive bias early in development, namely a heuristic for Dependency Length Minimization (DLM) (cf. Futrell et al. 2015). The latter approach is notable in that it (a) does not rely on learning from ambiguous input to capture child errors (cf. Sakas and Fodor 2001), (b) offers a prima facie simpler way of capturing the errors, and (c) is novel in applying DLM to account for child errors. Nevertheless, closer investigation shows that a DLM-based model does not provide a principled account of the children’s developmental trajectory and is clearly not any simpler than the alternative. Further, there is some reason to think more generally that DLM does not play a role in the development of the Swiss German children during the course of Schönenberger’s study. In contrast, an approach based on parameter interaction does provide the desired principled account. This comparison provides support for a non-biased learning model that has parameter interaction and learns from ambiguous input.

Failure Based Design Process

A modified design specification process was used to efficiently resolve field failures. The new process is focused on rapid experimentation to mitigate parameter interaction and facilitate quick field implementation. The change focuses on effectively using customer use data, material information, and manufacturing experience to create replication tests. The normal development process and the new specification processes were used on the same two field failure issues and are presented as case studies. The change yields design improvements in half the time and with increased effectiveness.