"... In PAGE 7: ... These results are contained in the DIKU data, and are shown in Tables 3-4. The first row of Table3 shows the results on deadline extrac- tion. In each case we selected an operating point with a target rate of C8CU of around 5%.... In PAGE 7: ... By evaluating the errors, we found that errors usually result when dates are given in tables; we aim to develop new parsers to improve processing in this domain. The second row of Table3 summarizes the performance on the program committee extraction task. Of the 1455 program commit- tee members, we found 1252 with our system.... ..."

"... In PAGE 7: ... In order to evaluate the proposed DC slice system, we have compared it to the system developed using the cus- tomized JVM approach [3] from the viewpoint of cost and module size necessary for the dynamic analysis. In the eval- uation, we have used the programs shown in Table3 as the input of the systems. P1 is a simple database program which contains few el- ements of object-oriented language.... ..."

"... In PAGE 5: ...1 Metrics Using our slicing system, we have evaluated the proposed method. Table3 shows features of sample programs we have used; C8BD loads some data from text files and generates HTML files, C8BE is a paint application using a mouse. We have used the following metrics: Slice Size: Comparison with static slice and dynamic slice [Table 4] Since we have implemented OODC slice only, we compute static slice and dynamic slice by hand.... ..."

"... In PAGE 8: ... We have made an experiment, and we have evaluated the slice size and analysis cost. Table3 lists the target programs for the eval- uation. Program P1(which consists of 4 classes, 262 state- ments) is a database management program, and the program P2(which consists of 5 classes, 231 statements) is a sorting program.... In PAGE 8: ...2 Analysis Cost We have compared the extended JVM with the original JVM with respect to the execution time and the memory usage. The target programs are listed ones in Table3 . Table 5 shows the results.... ..."

"... In PAGE 10: ... A detail discussion on each of these measures is presented in Annex 2. Table3 shows that the four housing programs have an average rating between poor and fair (an aggregate score of less than 10). In contrast, both the sanitation programs examined are rated higher with scores of 12.... In PAGE 31: ... In Karnataka, a partial survey was undertaken to get a sense of these unaccounted for cost components. Computations of the real value of the subsidies, based on the results of the survey, are presented in Table3 . They show that, in relation to housing costs, the value of land accounts for a very high proportion of the subsidy, often around 75 percent of total costs.... In PAGE 31: ... Thus, if Table 3 were representative of all programs, and, if even only half of the land provided as an unbudgeted subsidy could instead be converted into cash, the current direct subsidy program could be increased to more than two and one half times its current size. Table3 : Hidden Subsidy on Land in Bangalore Slum Assisted by VAMBAY, NSDP and Urban Shelter Programs Second, when the underlying value of the land in the programs is high, beneficiaries cash out because they prefer to consume less land and housing, and more of other things. This is not to argue that the slum dwellers should not be entitled to land and secure tenure, but rather that building low cost housing on high-value sites is likely to be an inefficient means of providing a housing subsidy to the poor.... ..."

"... In PAGE 10: ...2 Testing the Hypotheses Since the three hypotheses are obviously related in their outcome, namely to reduce cost of non-quality and improve field performance, we will try to first summarize the achievements in one picture. Table1 provides average values of the different process metrics in the discussed three-year timeframe. They look promising but can not imme- diately show detailed relations.... ..."

"... In PAGE 18: ... If, for instance, a third of the top 75 stations were gearing their broadcasts for a New York audience by delivering news, weather and traffic reports specific to New York, this would suggest that diversity in programming is being limited by the location of the top broadcasters, even though their content is available internationally. As seen in Table5 , 47% of the 75 Internet radio stations measured by Arbitron are created for a national market, while the balance are targeted to listeners in specific locations. By comparison, when the target market locations are compared according to listening hours, 56% of the actual tuning hours are being spent listening to the national stations.... ..."

"... In PAGE 28: ... The best results are obtained with the -csn option. In Table2 , we compare FGSyn with Chortle-crf [10], ASYL [1] and mis-pga (new) [18]. We do not compare our results with those of FlowMap [7] as that program focuses on generating a set of mapping solutions with area and depth trade-off while our program (as well as Chortle-crf, ASYL and mis-pga (new)) targets minimum area with no depth constraints.... In PAGE 28: ...5 MIPS) with 64 MB of memory. As seen in Table2 , FGSyn does 20:6% better than Chortle-crf, 16:8% better than ASYL and 13:0% better than mis-pga (new). The memory requirement of FGSyn is only 30% more than that of the mis-pga (new).... ..."