"... In PAGE 13: ...change variable for the six conditions can be found in Table4 . The results for the proposed model were similar to the results found for the change score cor- relations; namely, the relationship between the third variables and change was unaffected by the relationship between the third variables and baseline.... ..."

"... In PAGE 3: ... By discourse relations, we mean those bindings that relate two units of text by means of coherence and rhetorical effects [16]. In Table1 , the discourse relations that are paired with the dash interpolations are shown. Note that although there is a distribution pattern, these relations are due to the semantics and context of the sentences; thus their distribution could be slightly different for another interpreter who tries to classify them into the same set of relations.... In PAGE 3: ...f texts (i.e., news articles ). The row denoting other (nonrelational) usages in Table1 consists of corpus- specific reference mechanisms, title introduction, list introduction using dashes, or one-off usages such as introducing quoted sentences. As can generally be seen, the distribution of discourse relations in dash interpolated sentences is not completely ad hoc and is worthy of special consideration.... ..."

"... In PAGE 3: ... By discourse relations, we mean those bindings that relate two units of text by means of coherence and rhetorical effects [16]. In Table1 , the discourse relations that are paired with the dash interpolations are shown. Note that although there is a distribution pattern, these relations are due to the semantics and context of the sentences; thus their distribution could be slightly different for another interpreter who tries to classify them into the same set of relations.... In PAGE 3: ...f texts (i.e., news articles ). The row denoting other (nonrelational) usages in Table1 consists of corpus- specific reference mechanisms, title introduction, list introduction using dashes, or one-off usages such as introducing quoted sentences. As can generally be seen, the distribution of discourse relations in dash interpolated sentences is not completely ad hoc and is worthy of special consideration.... ..."

"... In PAGE 16: ... The errors which fall into this class range in severity from making the text seem \choppy quot; to making the text incomprehensible (due to the reader apos;s inability to resolve pronouns, for example). Table2 contains a subset of the classes in our error taxonomy (derived from sample analy- sis) which fall into the category of discourse-level errors. Although, some of these errors appear to manifest themselves at the sentence level (e.... ..."

"... In PAGE 5: ... It returns values between 0 (identical segmentations) and 1 (maximally different segmen- tations) and differs from accuracy in that predicted boundaries which are only slightly off are penalised less than those which are completely wrong. Human agreement is relatively high4 on both seg- mentation and span labelling (see Table1 ), which can be explained by the fact that (i) the RST-DT annotators were given very detailed and precise in- structions and (ii) assigning boundaries and labels is an easier task than creating full-scale discourse trees. 4.... In PAGE 5: ... It is nevertheless interesting to see how far one can go with a modest feature space and considerably less structural information. Table1 shows the results. A set of diacritics is used to indicate significance (on accuracy) through- out this paper, see Table 2.... In PAGE 6: ... As with one-step chunking, we also im- plemented a stacked variant, stacking both the seg- mentation and the labelling models. It can be seen in Table1 that the two-step mod- els outperform the one-step models. This difference is significant except for the stacked model on the la- belling task (labelled).... ..."

"... In PAGE 18: ...RESULTS Table1 reports the posterior mean and 90% credible interval for the rela- tive risk of severe wheeze for boys compared to girls, given by the poste- rior distribution of exp( 0). Table 2 reports the posterior mean and 90% credible interval for the number of cases per 100 children aged seven to nine years that are attributed to each of the excess risk factors baseline (j=1), dampness (j=2), passive smoking (j=3), NO2 exposure (j=4) and the latent spatial factor (j=5).... ..."

"... In PAGE 3: ... A set of discourse markers that can be used by a gen- eration system to indicate the nucleus and satellite of each relation. Table1 presents part of the statistical data that we derived for the rhetorical relations that we use in the examples given in this paper.... ..."