Alignment of (dis)preferred properties during the production of referring expressions
Martijn B. Goudbeek, Emiel Krahmer, Marc G. J. Swerts · 2009
It is generally assumed that speakers prefer cer-tain properties over others when referring to tar-gets, and various algorithms for the generation of referring expressions model this assumption. These preferences are often claimed to be fixed for a given domain. We argue that this is not the case, and that speakers ’ preferences are at least in part determined by the preferences of other speak-ers. In an experiment in which participants could describe people or furniture pieces based on pre-ferred and dispreferred properties, we found that speakers frequently aligned with respect to the previously heard (preferred or dispreferred) prop-erty, even though they could always refer with the preferred property. Speakers aligned more with the dispreferred property in the more clearly struc-tured furniture domain.