"... In PAGE 13: ... Depending on the gesture commands are sent to the devices and feedback is applied. The correlation of gestures and commands is shown in Table3 . If the dialogue nishes by time out or by the pointing gesture or a certain termination gesture the control ow enters the direction determination (Figure 12).... In PAGE 30: ...evice at a time. First a device is selected by the unique pointer click. Depending on the selected device, the gesture will execute a certain command. Table3 shows a possible mapping. The stars indicate that the device in that column supports the function of the row.... ..."

"... In PAGE 6: ...24 18 21 24 27 30 Number of terms MAP Rocchio KLD CHI Figure 4. Feedback methods on various feedback terms Table4 gives the performance comparisons between RF 1 and Okapi using Rocchio blind feedback method as shown in Equation (1) on the 150 old queries. All results are reported in terms of MAP.... In PAGE 6: ... All results are reported in terms of MAP. Only description queries and short queries results are reported here in Table4 . The performance improvement results on long queries using blind feedback are not reported because of lack of statistical significance.... In PAGE 7: ... The performance gain using both techniques is almost the sum of the gains by applying the individual techniques alone. Our results on the 150 old queries in Table4 show that combining ranking function tuning and blind feedback can improve the search performance by almost 30% over the baseline Okapi system. This is very encouraging considering the fact that most ad hoc search queries are very short and many of these queries are repetitive.... ..."

"... In PAGE 8: ... into a gate that performs the same function as the short, as in Figure 74 It turns out that 34% of the nonpropagatable non-feedback shorts can be found this way. Table2 shows the complete division of non-feedback shorts into nonpropagatable and nonstimulatable non-feedback shorts as well as the subdivision of each of these categories into LD0 and Non-LD0 categories. Approximately 50% of the 264 undetectable non-feedback shorts were LD0.... ..."

"... In PAGE 6: ....2.1 Usefulness of haptic feedback: Experiment Iteration 1: ANOVA results show that when considered across all cognitive load tasks, the addition of haptic feedback to our Progress Bar task significantly improved response times; however, significant performance benefits were not found when haptic feedback was added in the Button task. Table1 shows response times associated with the Progress Bar task that are significantly lower with haptic feedback than without. No other statistically significant effect of haptic augmentation was found on performance time or accuracy for these two tasks.... In PAGE 6: ... In addition, almost half of participants felt that they performed better in the Progress Bar task with haptic feedback (consistently with performance data). Experiment Iteration 2: The ANOVA results show that the addition of haptic feedback to our Scroll Bar task significantly improved task completion times ( Table1 ). No statistically significant effect of haptic feedback on accuracy was found.... ..."

"... In PAGE 6: ... Some topics had no true relevant documents among the top-10 retrieved documents; for these topics, we did not use any feed- back. We compared UIUCrelfb (relevance feedback) with both Baseline1 (no feedback) and Baseline2 (pseudo rel- evance feedback) in Table5 . Both relevance feedback and pseudo relevance feedback improve the performance in document and passage measures, but the aspect-level performance drops.... ..."

"... In PAGE 20: ... It is assumed that an error message about a constraint can influence the outcome of the next attempt at the constraint, resulting in a satisfaction (desired) or a violation (not desired). Table11 characterises this as a utility function Table 11. The utility function for feedback selection.... ..."