We extend the n-gram-based data-driven prediction approach (Elghafari, Meurers and Wunsch, 2010) to identify function word errors in non-native academic texts as part of the Helping Our Own (HOO) Shared Task. We focus on substitution errors for four categories: prepositions, determiners, conjunctions, and quantifiers. These error types make up 12 % of the errors annotated in the HOO training data. In our best submission in terms of the error detection score, we detected 67 % of preposition and determiner substitution errors, 40% of conjunction substitution errors, and 33% of quantifier substitution errors. For approximately half of the errors detected, we were also able to provide an appropriate correction. 1