P. Dybala, M. Ptaszynski, R. Rzepka, and K.Araki (Japan)
artificial intelligence, human-computer interaction,evaluation methods
Human-likeness of dialogue systems is an important,
albeit neglected issue. In this paper, basing on evaluation
experiments of humor-equipped chatterbot, we propose a
method of measuring the distance between humans and
systems and relation between human-likeness and humor.
The results show that the presence of humor can enhance
the performance of dialogue systems. A humor-equipped
chatterbot was evaluated as more human like and
generally better than one without humor, by both first and
third person evaluators. The implications of this fact and
novelty of evaluation method are discussed, and some
ideas for the future are given.