This paper analyzes results on light-verb construction identification, distinguishing between known cases that could be directly learned from training data from unknown cases that require an extra level of semantic processing. We propose a simple baseline that beats the best results of the PARSEME 1.1 shared task (Savary et al., 2018) for the known cases, and couple it with another simple baseline to handle the unknown cases. We additionally present two other classifiers based on a richer set of features, with results surpassing these best results by 7 percentage points.