Do the right thing . . . but expect the unexpected

Abstract

This paper has itself obviously been selective; but it is not intended to be perniciously so. Computational Linguistics Volume 24, Number ? even the best emulator will always generate unexpected utterances, which violate Spike Lee's maxim, and cause unwanted implicatures. Some would argue that the semblance of paradox is simply evidence that there are limits to the experimental paradigm. That there are such limits is not in doubt. But there remains something odd about that fact that not only would I myself generate the unexpected utterance, but everyone else would too. Thus, over time, I really should come to expect this unexpected utterance---and it should therefore stop being unexpected. Perhaps I am exposed to too few instances of the paradoxical behaviour in my lifetime to become attuned to it. But the language community as a whole has already had plenty of time to adjust its expectations. It has not done so: the asymmetry in behaviour appears to be stable. Thus, individual language users will continue to be surprised by the behaviour of their language community. By contrast, researchers in natural language generation should not be surprised at such asymmetries in behaviour. Once we understand better both the behaviour, and the interplay of expectations which underlies it, analysts and engineers alike will know exactly when they should expect the unexpected. Acknowledgments