In reference coding, one of the categories is the "reference group". The intercept will be for that group, and effects for other groups are relative to that one. (This is appropriate coding if you want to compare all other groups as deviations from some baseline. For example, you might consider "non-drinkers" as a baseline, and then categories "1-6 drinks/week", "7-13 drinks/week", etc, as deviations from the baseline.)
In effects coding, there is no reference group. The intercept will correspond to the mean of all groups, and effects for all groups are relative to mean. (This is appropriate when no group is appropriate as a baseline. For example, "Republicans", "Democrats", "Green Party", etc., might have no particular reference group.)
I'm not clear on the use of "full". My guess is that this is most relevant when one is testing hypotheses between particular groups, but I really don't know for sure. (A fair amount googling did not clarify this for me.)
