Interpreting strange results

#1
So effectively I have two variables, language problems, and emotional problems, both of which independently cause an increase in a third variable, which is behaviour problems. I hypothesized that high emotional problems would significantly moderate the relationship between language problems and behaviour problems. However, my interaction term is negative. Therefore high emotional problems are apparently moderating relationship between language problems and behaviour problems by weakening the relationship. How is this possible? Surely this is entirely counter-intuitive. The results are basically saying that emotional problems are good for behaviour outcomes?! Or am I interpreting this interaction wrong.