"... In PAGE 3: ... Most of these features are treated as defects in lumber grading and need to be removed in manufacturing processes. Table1 shows common features that are important in determining lumber grade and the common sensing modality used to automatically detect those features. Recognizing that not all grading features can be detected with one single sensing mechanism, current R amp;D efforts are focussing on developing lumber grading systems that combine 2 or more of these sensing modalities.... ..."

"... In PAGE 8: ... This compari- son can be made for all the sentences of the test set for which some of the patterns assigned by the evaluators match the automatic or reference phrasing of the sen- tence. Table5 shows the percentage of the evaluators that assigned the same phrasing pattern, the number of sentences for which it occurred and the average clas- sification of those patterns. For example, there were 2 sentences for which all the evaluators assigned the same phrasing pattern that matched the automatic or reference phrasing.... In PAGE 9: ... Number of phrasing patterns assigned by the evaluators for each sentence length. The sum of the second column of Table5 gives the total number of patterns assigned by the evaluators that were also evaluated. Four of those 77 patterns received the worst grading, for which 5 of the 10 evaluators found the phrasing unacceptable.... ..."

"... In PAGE 4: ..., forest, etc.) 39% Business services and marketing 19% Banking, nance and insurance 12% Medicine, health, pharmaceutic 3 % Transportation 3 % Utilities and energy 3% Wholesale and retail trade 1% Other 20% Table2 : Industry sectors of neural networks in Europe. has signi cant variation both within and between species, making it a di cult material for automatic grading.... ..."

"... In PAGE 5: ... The quizzes also need to contain more detailed, team-specific project questions. In Table1 , each of the above grading schemes has been mapped to the grading criteria discussed earlier. Note that no single scheme meets all the grading criteria.... ..."

"... In PAGE 22: ... Because of increased enrollment and class attendance rates, classrooms of FFE schools are more crowded than non-FFE school classrooms. Data in Table4 indicate that, on the average, FFE school classrooms have about 22 percent more students than non-FFE school classrooms. Table 4 Average class size: School survey results Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 All grades ... ..."

"... In PAGE 4: ... An interesting result of the test is to compare the phrasing patterns assigned by the test subjects with the automatic and reference phrasing. Table3 shows the percentage of agreement in the phrasing performed by the test subjects and the evalua- tion of that phrasing. Even when 90% of the evaluators agreed on a phrasing, 10% found it unacceptable.... ..."

"... In PAGE 8: ... Having achieved 95% agreement, we concluded that our grading system was reliable, and we proceeded to have a the third author grade the remaining 80 percent of the code solutions. Table4 presents our semantic accuracy results by treatment group. As can be seen, the Automatic group generally outperformed the On Request condition, which, in turn, outperformed the No Feedback condition.... In PAGE 8: ... According to a Shapiro-Wilk test for normality, these data are not normally distributed, thus requiring us to use non-parametric Kruskall Wallis ANOVAs to test for significant differences. Consult column 5 of Table4 for the p-values corresponding to each measure. As can be seen from Table 4, although there exists no statistically-significant difference with respect to overall accuracy (df = 2, H = 4.... In PAGE 8: ... Consult column 5 of Table 4 for the p-values corresponding to each measure. As can be seen from Table4 , although there exists no statistically-significant difference with respect to overall accuracy (df = 2, H = 4.46, p = 0.... ..."