Given this first configuration, an evaluation was carried out which resulted in a precision of 59.0% for pronominal anaphora resolution and 23.7% for adjectival anaphora resolution. Needless to say, these results are very low for pronominal anaphora and extremely poor for adjectival anaphora. In evaluating the errors, we concluded that the defined anaphoric accessibility space was too constrained and too arbitrarily defined. It simply disregarded the relationship between anaphora and dialogue structure. Consequently, we proposed the following changes for the second experiment.