Word Overuse and Alignment in Large Language Models: The Influence of Learning from Human Feedback