"... In PAGE 2: ... We can now capture the necessary details in our trip planning example. Table1 shows an interaction with the key communicative acts and relationships identi ed. Notice the a19a21a20 and a19a23a22 are two hotels; a19a23a22 is contacted when a19a24a20 cancels.... In PAGE 3: ... For any utterance a90a18a95 that a90 a92 responds to, the corresponding commitments or acts of a90a18a95 constructs the antecedent. By executing the algorithm on Table1 , we obtain the follow- ing list of commitments. Here TBD indicates to be decided.... In PAGE 8: ... 6. CONCLUSIONS Given an agent interaction marked up with the various columns of a conversation table, as in Table1 , our approach rst extracts commitments and commitment relations from it. It then gener- ates the agent models by ne-tuning the relation diagrams.... ..."

"... In PAGE 9: ... In this case, combinations of foods with larger number of calories have a better chance of satisfying this constraint. We only consider four plans (plans I, II, V, and VI in Table1 ) for this query because it is obviously not an appropriate choice to consider foods with lower number of calories ... ..."

"... In PAGE 45: ... Table5 : Introduction of a cache Operator (Third Rule) The path between two cache operators is reordered by promoting the underlined nested-loop-join to lie above the two caches, as depicted to the right in Table 5. This transformation requires that Xy contains no cache operator.... In PAGE 45: ...Table 5: Introduction of a cache Operator (Third Rule) The path between two cache operators is reordered by promoting the underlined nested-loop-join to lie above the two caches, as depicted to the right in Table5 . This transformation requires that Xy contains no cache operator.... ..."

"... In PAGE 38: ... This will lead to sub-optimal solutions in nearly all the cases. Table11 gives examples of some of the realistic and theoretical approaches studied in this article. The first example [134], has the shortest planning period of all.... In PAGE 38: ...eriod longer than 4 days in that model. The short planning period leads to a rather simple rostering problem. However, it is not a very good idea not to fix the coverage constraints long beforehand because that does not correspond to the needs in real hospitals. The next rows in Table11 present other approaches with fixed and semi-fixed planning periods. More complex models, with longer planning periods, are listed lower down in the table.... ..."

"... In PAGE 8: ...5 Rather, as Table 4 shows, we also consider more substantial discrepancies involving a deliberate (though possibly not planned-in-advance) behavior of the crew. A traditional planning approach typically does not take into consideration some of the items highlighted in Table4 or the concatenated circumstantial effects caused by the highlighted discrepancies. In contrast to typical planning approaches, by virtue of representing behaviors and not just abstracted tasks, the Brahms simulation is capable of showing how the practice of onboard activities often diverges, both in timing and execution, from the originally scheduled activities and procedures.... ..."

"... In PAGE 4: ...that performs a best-first search across belief states. Our algorithm, presented in Table1 , is supplied with a domain model a0 , and a problem statement consisting of: an initial belief state a1 , a user-defined utility-function a0 , a utility suc- cess threshold a7a33a8a11a10a6a12a15a14a17a16a17a10 , and a value specifying a confidence threshold for termination a18 a8a11a10a6a12a15a14a17a16a17a10 . GenerateOrExecutePlan(domain a2 , belief state a3 , utility function a4 , goal threshold a5a7a6a9a8a11a10a13a12a15a14a16a8 , satisfiability threshold a17 a6a9a8a11a10a18a12a19a14a16a8 ) 1.... ..."

"... In PAGE 2: ... Figure 1. Methodology for building and testing HAPRC The Planner Our planning system, called HAP, is highly adjustable and can be customized by the user through a number of pa- rameters, which are illustrated in Table1 along with their value sets. These parameters concern the type of search, the quality of the heuristic and several other features that affect the planning process.... In PAGE 3: ... If equal_estimation is set to 1, then the search strategy will prefer the state, which is closer to the starting state. Statistical Analysis In order to find out which is the best configuration for our planner we have run a large number of planning problems (97) from 8 domains (Problem Set A, see Table 2), using 432 different configurations ( Table1 ) on our HAP planner. Then we tried to statistically analyze the results (plan length Lij and planning time Tij, for the i-th problem and the j-th configuration) obtained from those runs in order to find out their potential relationship with the planner set- tings.... In PAGE 4: ... This was done in order to decide which the best default setting is for each planner parame- ter, when the rule-based configuration cannot set one or more parameters. The results are shown in the third column of Table1 . We notice here that there was no clear winner value for each parameter since some values had better avg j L while others had better avg j T .... In PAGE 5: ... Selected problem characteristics The data about the selected attributes for the 97 prob- lems of set A were subsequently joined with the data about the parameters and performance from the runs of HAP with all possible 432 configurations on the same problems. This led to a dataset of 41904 instances with 27 attributes: 19 problem characteristics (B01-B19), 6 parameters of HAP (first column of Table1 ) and its performance (number of steps in plan Lij and execution time Tij). The next step was to select the type of learning task that should be applied to discover a model of the dependencies between problem characteristics, planner parameters and good planning performance.... ..."

"... In PAGE 29: ...21s P4 11 5 13 0.81s P5 11 7 15 327s Table2 : Parallel Blocks World - cheapest plans: Minimal number of moves at fixed plan length (AB) column on the plan length. Column six shows the execution time for finding the shortest plan in an incremental plan length search starting from 0, similar to the method used for CCALC.... In PAGE 30: ... This becomes drastically apparent when execution times seem to explode from one instance to the next, in a highly non-linear manner as in Table 1 where a solution for P3 can be found in reasonable time whereas P4 and P5 could not be solved within the time limit of 4000 seconds. This observation is also confirmed in the other tables (instance P5 in Table2 , etc.) and is partly explained by the behavior of the underlying DLV system, which is not geared towards plan search, and as a... ..."

"... In PAGE 6: ... This paper details results from the first experi- ment. Table1 summarizes the experiment as it applied to the planner. The five frames of Figure 5 depict snapshots of the path plan and projected lighting on the terrain at various times over the 24-hour Experiment 1 traverse.... ..."