After including information about dialogue structure and removing the linguistic preferences, precision rates rose to 62.3% for pronominal anaphora resolution and 65.8% for adjectival anaphora resolution. A considerable increase is gained in the resolution of adjectival anaphora by simply changing the definition of the accessibility space. That is due to the fact that adjectival anaphora need a larger space than that used in Experiment 0.
But these results are still low and demonstrate that dialogue structure information alone is not sufficient. Thus, a third experiment was carried out using both dialogue structure and linguistic information. Several variations in the preference system were investigated independently.