"... In PAGE 15: ... According to subject C #5Cit is almost less the coding style than the design style that is important quot;. Additionally, subjects C and D agreed that not too deep an inheritance tree was good style #28although from Table2 it can be seen their de#0Cnitions of deep di#0Ber#29. Subject H added #5Cthere seems to be few points of reference for object-oriented code quality quot;.... In PAGE 30: ...ode must be put through code review, guidelines are subjective, e.g., don apos;t use gotos, global variables, but bad design produces bad code B One #0Cle for one class is not good style; inconsistency C Long methods are a sign that something wants thinking about; excessive inheritance depth; #5Cit apos;s almost less the coding style than the design style that apos;s important quot; D Short methods, protected instead of public inheritance where possible; not too deep an inheritance tree E I apos;ve not enough programming experience, it apos;s vital that common programming guidelines are accepted though F Good design usually means good code: wehave code reviews earlier now G I don apos;t likeoverloading, it complicates things too much H Should have code reviews to sort things out, seems to be few points of reference for OO code quality I Identi#0Ccation of correct objects; inappropriate use of inheritance: all at the design level J Small methods; mixing object-oriented code with procedural code is sloppy K #5CMechanism not policy quot; #28based on good design#29 L Question not asked M Good design then good code, bad design then bad code Table 19: Answers and remarks to Q. 16 Subject Remark A Di#0Ecult to say,changes should be localized: only when encapsulation is broken is there a worry, code reviews help B Di#0Ecult to know, code reviews help C Its no more impossible than under the procedural type of method; breaking encapsulation is a bad idea, code reviews help D No, but this is true of any system E It depends on the utility of the method changed F Use metrics and code reviews to checksoyes it is possible G No, but a code review should spot it H It apos;s subjective, but code reviews spot these things I Question not asked J Yes, the changes should be tested adequately and encapsulated K No, di#0Ecult to tell; code reviews help L Encapsulation so it should not have degraded M Very di#0Ecult to tell Table2 0: Answers and remarks to Q. 17... In PAGE 31: ...Remark A Trouble with availability and quality of design documentation Subjects B, C, D, F, G, H, J, K, amp; M made similar statements E Documentation is usually available G #5CDesign documentation is important otherwise you are back to where you were before quot; L Documentation for OO programs more important aid to understanding than documentation for non OO systems M #5CThe documentation is essential quot; Table2 1: Answers and remarks to Q. 18 Subject Remark A No, without dynamic binding no object-oriented programming B No, really useful; Objective-C far better than C++ for it C No, it allows very elegant code, but there can be problems of #0Cnding methods bound at run-time D No, it apos;s a big help: simpli#0Ces the code; it apos;s more intuitive E No, if managed well it maps reality F No, good#2Fclear abstractions are aided byit G If you understand it, then it causes few problems H No, it makes your design simpler I Question not asked J No, it allows generic code, no switches etc K No, it provides lots of advantages L No, only problem is calling a non existent method M Yes if you are careless, no if you are not Table 22: Answers and remarks to Q.... In PAGE 31: ...Remark A Trouble with availability and quality of design documentation Subjects B, C, D, F, G, H, J, K, amp; M made similar statements E Documentation is usually available G #5CDesign documentation is important otherwise you are back to where you were before quot; L Documentation for OO programs more important aid to understanding than documentation for non OO systems M #5CThe documentation is essential quot; Table 21: Answers and remarks to Q. 18 Subject Remark A No, without dynamic binding no object-oriented programming B No, really useful; Objective-C far better than C++ for it C No, it allows very elegant code, but there can be problems of #0Cnding methods bound at run-time D No, it apos;s a big help: simpli#0Ces the code; it apos;s more intuitive E No, if managed well it maps reality F No, good#2Fclear abstractions are aided byit G If you understand it, then it causes few problems H No, it makes your design simpler I Question not asked J No, it allows generic code, no switches etc K No, it provides lots of advantages L No, only problem is calling a non existent method M Yes if you are careless, no if you are not Table2 2: Answers and remarks to Q. 19... In PAGE 32: ...Remark A No, using a message should have the same e#0Bect on any object that responds to it; abstraction B Yes, it apos;s a great bene#0Ct, but it can make things di#0Ecult to understand. If used properly and sparingly it apos;s advantageous C Yes, it produces generic code, but easy to lose track of what apos;s going on D No, it simpli#0Ces things rather than making them more di#0Ecult: you are looking at the behaviour of each object but not the details E Maintain semantic consistency and no problem, otherwise problems F No, methods with same name should achieve the same aim G Yes but related to learning the concept H It simpli#0Ces things; I think it apos;s wonderful I It apos;s the most important property in OO systems, but must maintain high level semantics J It can cause confusion: you just have to be careful K No problem, it improves understanding L It apos;s an aid to human understanding, it apos;s more intuitive M No problem with the size of systems I apos;velooked at Table2 3: Answers and remarks to Q. 20 Subject Remark A No problems, perhaps with operator overloading, but tools help this B No problems, but I avoid operator overloading as it is not predictable C No problems, no operator overloading experience D Yes, naming is big issue in object-oriented programming, no great problems with operator overloading, but no great experience either E No real experience of any such problems F Not with polymorphism, but experience of operator overloading not pleasant G No, but little use; operator overloading used sensibly then useful H Yes, it apos;s hard to know the best time to use the same names I This is one of the major hidden problems in object-oriented systems, programmers must understand about consistent naming.... In PAGE 32: ... 20 Subject Remark A No problems, perhaps with operator overloading, but tools help this B No problems, but I avoid operator overloading as it is not predictable C No problems, no operator overloading experience D Yes, naming is big issue in object-oriented programming, no great problems with operator overloading, but no great experience either E No real experience of any such problems F Not with polymorphism, but experience of operator overloading not pleasant G No, but little use; operator overloading used sensibly then useful H Yes, it apos;s hard to know the best time to use the same names I This is one of the major hidden problems in object-oriented systems, programmers must understand about consistent naming. Operator overloading for thoroughly de#0Cned operators only J None experienced; no use of operator overloading K Yes, easily make a mistake renaming a member function, and this can lead to subtle errors that are hard to #0Cnd because everything looks ok L Not personally M No Table2 4: Answers and remarks to Q. 21... In PAGE 33: ...g., protected inheritance#29, doesn apos;t look object-oriented, doesn apos;t encourage object-oriented way of thinking L Doesn apos;t encourage object-oriented programming: you may write C instead of C++; it apos;s strongly typed M Strong typing; sold as a better C: allows you to feel as if you are an object-oriented programmer, but you are not, allows you to write C; not forced to write object-oriented code Table2 5: Answers and remarks to Q. 22 Subject Remark A Its e#0Eciency and popularity B Its fast C Its performance D Familiarity with C and speed E Di#0Ecult to say if it has any F Control over everything G Its performance H Its performance I Allows use existing C libraries and OO is taking o#0B because of C++ J Its e#0Eciency; existing C libraries K More e#0Ecient L E#0Eciency and it apos;s street credibility M E#0Ecient; backed byAT amp;T Table 26: Answers and remarks to Q.... In PAGE 33: ...nde#0Cned areas #28e.g., protected inheritance#29, doesn apos;t look object-oriented, doesn apos;t encourage object-oriented way of thinking L Doesn apos;t encourage object-oriented programming: you may write C instead of C++; it apos;s strongly typed M Strong typing; sold as a better C: allows you to feel as if you are an object-oriented programmer, but you are not, allows you to write C; not forced to write object-oriented code Table 25: Answers and remarks to Q. 22 Subject Remark A Its e#0Eciency and popularity B Its fast C Its performance D Familiarity with C and speed E Di#0Ecult to say if it has any F Control over everything G Its performance H Its performance I Allows use existing C libraries and OO is taking o#0B because of C++ J Its e#0Eciency; existing C libraries K More e#0Ecient L E#0Eciency and it apos;s street credibility M E#0Ecient; backed byAT amp;T Table2 6: Answers and remarks to Q. 22... In PAGE 34: ... It apos;s common practice quot; #5CTime pressure does exist quot; B It depends on the maintainer. It apos;s #5Ceasier in C++ quot; C #5CEntirely possible quot; #5COnly thing you can do in the time quot; D #5CAbsolutely, someone pushed in a hurry will do it the easiest way quot; E #5CIt apos;s too easy in C++ quot; F Yes because of time pressures G #5CYes, but that apos;s bad maintenance rather than design quot; H #5CThere apos;s a good chance it might happen quot; I apos;ve only done it once I #5CYes particularly in C++ quot; J In C++ a lot easier to do quick #0Cx; depends on nature of the project as well as individual K #5CYes absolutely quot; L #5COne of the dangers in these hybrid languages quot; M #5CIt apos;s all too easy, it apos;s really easy, and the temptation is always there Table2 7: Answers and remarks to Q. 23 Subject Remark A time constraints have caused quick #0Cxes to OO code in the past Subjects C, D, F, and K made similar statements D Must try to remain competitive and that can cause quick #0Cxes K Documentation is the #0Crst thing that su#0Bers because of time constraints Table 28: Remarks about time constraints when performing quick #0Cxes... In PAGE 34: ... It apos;s #5Ceasier in C++ quot; C #5CEntirely possible quot; #5COnly thing you can do in the time quot; D #5CAbsolutely, someone pushed in a hurry will do it the easiest way quot; E #5CIt apos;s too easy in C++ quot; F Yes because of time pressures G #5CYes, but that apos;s bad maintenance rather than design quot; H #5CThere apos;s a good chance it might happen quot; I apos;ve only done it once I #5CYes particularly in C++ quot; J In C++ a lot easier to do quick #0Cx; depends on nature of the project as well as individual K #5CYes absolutely quot; L #5COne of the dangers in these hybrid languages quot; M #5CIt apos;s all too easy, it apos;s really easy, and the temptation is always there Table 27: Answers and remarks to Q. 23 Subject Remark A time constraints have caused quick #0Cxes to OO code in the past Subjects C, D, F, and K made similar statements D Must try to remain competitive and that can cause quick #0Cxes K Documentation is the #0Crst thing that su#0Bers because of time constraints Table2 8: Remarks about time constraints when performing quick #0Cxes... ..."

"... In PAGE 3: ... The inter-class transformation parameters and were trained from 9 speakers except the test speaker in the 10 evalua- tion speakers. Table1 summarizes the word error rates (WER) after supervised adaptation with 1 or 3 adaptation sentences for each speaker with correct transcriptions. The adaptation sentences and test sen- tences were different.... In PAGE 3: ... Table 2 shows corresponding results for unsupervised adaptation on sentences from the test set. As in Table1 , the columns in the table indicate the number of sentences used in performing the adaptation for each speaker (1,3, or 20). The MLLR parameters are estimated from blind transcriptions of the sentences by the baseline recognition system, and the test sentences are subse- quently recognized again using the adapted models.... In PAGE 4: ... Table 3 describes the WER observed when the inter-class transformations were trained from different data. The first row in Table 3 repeats the results from the case of Inter-class MLLR without weights in Table1 . The WER using the inter-class transformations obtained from the native speakers [5] was worse than the results from the case of the non-native speakers.... ..."

"... In PAGE 3: ... The inter-class transformation parameters and were trained from 9 speakers except the test speaker in the 10 evalua- tion speakers. Table1 summarizes the word error rates (WER) after supervised adaptation with 1 or 3 adaptation sentences for each speaker with correct transcriptions. The adaptation sentences and test sen- tences were different.... In PAGE 3: ... Table 2 shows corresponding results for unsupervised adaptation on sentences from the test set. As in Table1 , the columns in the table indicate the number of sentences used in performing the adaptation for each speaker (1,3, or 20). The MLLR parameters are estimated from blind transcriptions of the sentences by the baseline recognition system, and the test sentences are subse- quently recognized again using the adapted models.... In PAGE 4: ... Table 3 describes the WER observed when the inter-class transformations were trained from different data. The first row in Table 3 repeats the results from the case of Inter-class MLLR without weights in Table1 . The WER using the inter-class transformations obtained from the native speakers [5] was worse than the results from the case of the non-native speakers.... ..."

"... In PAGE 39: ... To begin a geometry problem is usually expressed by descriptions such as, Two transversals intersect two parallel lines and intersect with each other at a point x between the two parallel lines. In the geometry theorem proving program, such statements, as well as the geometry definitions and theorems, would be encoded in some descriptive notation ( Table1 ) such as a language composed of terms and relations. Problem Statement (line l1) (line l2) (line t1) (line t2) (parallel l1 l2) (transversal t1 l1 l2) (transversal t2 l1 l2) (intersect t1 t2 x) (between x l1 l2) Geometry Definitions (production rules of the program) (production-rule (alternate-interior lt;a1 gt; lt;a2 gt;) (angle lt;a1 gt; lt;a2 gt;) - gt; (congruent lt;a1 gt; lt;a2 gt;)) Table 1.... In PAGE 39: ... In the geometry theorem proving program, such statements, as well as the geometry definitions and theorems, would be encoded in some descriptive notation (Table 1) such as a language composed of terms and relations. Problem Statement (line l1) (line l2) (line t1) (line t2) (parallel l1 l2) (transversal t1 l1 l2) (transversal t2 l1 l2) (intersect t1 t2 x) (between x l1 l2) Geometry Definitions (production rules of the program) (production-rule (alternate-interior lt;a1 gt; lt;a2 gt;) (angle lt;a1 gt; lt;a2 gt;) - gt; (congruent lt;a1 gt; lt;a2 gt;)) Table1 . Given data structure and program in a semi-formal notation (extracted from Larkin and Simon, Table 4, p.... In PAGE 47: ...47 Stockholm: Cognition, Education, and Communication Technology Symposium A comparison of the points of view ( Table1 ) shows that the alternative Newell and Simon reject also includes a linguistic deep structure, again the symbol structures of their IPS models. Although they raise the possibility of nonlinguistic deep structure, they cannot conceive of this standing on its own because they fundamentally view understanding and thinking as mapping to and manipulating linguistic (symbol) strings.... ..."

"... In PAGE 8: ...superv = trained in supervisedmode, unsuperv = trained in unsupervisedmode Same#training instances and parameters as in Table1 , judgment threshold = 0.1 Table 2: Results using both Supervised and Unsupervised Learning... In PAGE 9: ...particular noun are done on the same set of testing sentences and as the size of the training set increases, all training instances from the previous training set are incorporated into the new training set. Table1 presents the results when the algorithm is trained using only supervised mode. Incorrect classi cations are divided into three groups: misclassi cations not caughtbythecheck against the di erence threshold, correct classi cations not strong enough to pass the threshold, and misclassi cations caughtby the threshold check.... ..."

"... In PAGE 10: ...http://genomebiology.com/2007/8/6/R101 exercise ( Table3 ; Binomial test P lt; 1.9e-07).... In PAGE 10: ... Like with the FTT and YMF methods, we also evaluated tis- sue- and stage-specific subsets of CRMs using this learning algorithm and a leave-one-out-cross-validation strategy. The apos;blastoderm apos;, and apos;embryo apos; CRMs gave significantly high clas- sification accuracy in similar cross-validation experiments ( Table3 ). As we saw with the other methods, the blastoderm CRMs have the most pronounced differences compared to the other CRM subsets and to the entire REDfly analysis set.... ..."

"... In PAGE 10: ... Surprising, however, is the fact that prices received also vary even when controlling for the market, the season of the sale, and the transactions volume. There is for example a 24 to 32 percent variation in the prices received within the same market ( Table3 ).12 The high variation is consistent with the hypothesis that the price markup is indeed important and as such the ability to bargain may be a crucial factor in the determination of this market price.... In PAGE 10: ... Even though some farmers reported that they managed to resolve these types of problems, the majority did not, and may have had to settle for a lower price than expected. Characteristics of information and search costs are also reported in Table3 . Ex-ante information on the price in the market in which farmers are selling varies from 24 percent for farmgate sales to 53 percent and 85 percent for local and distant market sales.... ..."

"... In PAGE 2: ... The next step is then to compare the general data from the source and target texts and see if we can conclude something about the translations. We do this by comparing the relative proportions of number of sentences, number of word tokens and recurrent sentence rates, by using the following simple measures (the figures summarized in Table4 below): Gb7 ST-Sentence = the number of source sentences/number of target sentences Gb7 ST-Word ratio = number of source words/number of target words, Gb7 ST-Recurrent sentence ratio = Recurrent sentence rate(Source text)/Recurrent sentence rate(Target text). The figures tells us that only two of the texts have more source sentences than target sentences (namely OS2 and InfoWin).... ..."

"... In PAGE 10: ... The mitochondrial dataset includes a total of 38 taxa and 13 mitochondrion-encoded proteins (2,619 amino acid positions), including homologs from the complete mitochondrial genome of Capsaspora that we have determined. Phylogenetic trees were inferred using a variety of methods (see Table1 and Supplementary Material). Because phylogenetic analyses regularly suffer from systematic error such as long-branch attraction (LBA) (Felsenstein 1978), we used methods and evolutionary models known to minimize these artifacts (see Table 1 and Supplementary Material).... In PAGE 11: ...11 the nuclear dataset (see Table1 and Figure 1) (Ruiz-Trillo et al. 2002; Lavrov and Lang 2005; Philippe et al.... In PAGE 11: ...56.60% missing data; see Table S1). Consistent with this hypothesis, a tree excluding those taxa with more than 50% missing data shows a ML bootstrap support of 100% for Metazoa (Figure S1). An important point is that the position of Capsaspora, ichthyosporeans and choanoflagellates remained identical regardless of the method ( Table1 ). Curiously, the nuclear tree shows Capsaspora as the sister-group to ichthyosporeans (Figure 1), whereas the mitochondrial tree shows Capsaspora in an intermediate position between ichthyosporeans and choanoflagellates (Figure 2).... ..."

"... In PAGE 9: ... Table 2 shows the resulting IPC loss from Baseline scheduling to the two feedback-adjusted prediction schemes. The relative depth in Table2 is the constant by which the pipeline depths in Table 1 were multiplied. Table 2.... In PAGE 9: ... This means that the predictors can be expected to perform about as well on deep pipelines as they do on shorter ones. However, the results in Table2 assume that the scheduler remains unpipelined and capable of scheduling dependent instructions back-to-back, which is somewhat unrealistic. Table 3 shows the same data for processors where the wakeup logic (predicted or otherwise) is pipelined.... ..."