"Good vs. Bad" in research article abstracts - a corpus-based analysis