Much effort is taking place today to understand the best type of interface between a driver and his/her intelligent vehicle, e.g. the best combination of input / output modalities [1, 2, 3, 5, 6, 13]. In a scenario in which an in-vehicle system answers the driver?s queries perhaps with a mixture of speech and graphical displays such as maps, there is an orthogonal issue that needs to be addressed: of all the information collected to answer the query, how much should be communicated and at what level of detail. In this paper, we discuss a system that provides driving directions (in written form) to a user, and its evaluation. Traditional literature provides an abundance of optimal algorithms that can be applied to a representation of a territory to generate text output for navigating between two arbitrary termini within that territory. In addition to the termini, the output generally consists of the minimal chain of intermediate landmarks and interconnecting roadways. By using simple templates, the algorithm can then realize this "essential" information into text. A prime example is MapQuest, the well-known web-based application (www.mapquest.com) that uses the NavTech geographical database for generating route directions within the United States. The final output, however, lacks the type of cueing that occurs in discourse between two people exchanging route information [4, 16]. In this paper, we show that providing those additional cues improves wayfiding performance: simulated drivers had fewer incorrect turns when following enhanced directions than those following essential directions.

We will describe the software system we developed, that can generate both essential and enhanced directios. It consists mainly of 1) a route generator that uses factual map data to generate routes, and 2) a natural language front-end. Using this software system and two sample groups, a single-factor experiment was conducted comparing the effectiveness of essential vs. enhanced form directions in wayfinding. Most notable in the experiment was the difference in incorrect turns, a measure of wayfinding; it approached significance in favor of the experimental group, namely, of the group that read enhanced directions. There were no other significant or marginally significant differences in the other measures we collected. More details on the software system and the experiment described in this paper can be found in [14].