In plain English, whereas n-grams are consecutive word patterns, such as bigrams, trigrams, four-grams, etc., skip-grams are not entirely consecutive but contain gaps or spaces, thus skipping over words. Skip-gram models arestatistical language models, which may be used for purposes such as prediction. Skip-gram models would be more fuzzy than regular n-gram models, but also more data intensive.
See also my quick and dirty webpage: