12.07.2015 Views

file - ChaSen - 奈良先端科学技術大学院大学

file - ChaSen - 奈良先端科学技術大学院大学

file - ChaSen - 奈良先端科学技術大学院大学

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

30, 77, 154] and those asking reputations and opinions [58, 71]. The number ofpapers on reasons [55, 57] and methods [6, 40, 79, 135–137, 184] is increasing, butthere are still not many.Since descriptive answers often consist of several sentences, it is possible toclassify the answers by their discursive features and explanatory strategy, to whichthe conventional general discourse analysis method can be applied.4.2.2 Discourse analysisDiscourse analysis has a long history, is very extensive, and encompasses manystudy cases. The scope extends from the analysis of natural interaction [49, 180,181] to that with literal “reading” [189]. Here, I introduce cases that deal with explanatorywritten texts. Textual discourse analysis identifies text segment typessuch as clauses, sentences, and paragraphs, and the logical and rhetorical relationsamong them [16, 45, 46, 55, 57, 65, 177, 183, 190]. The Rhetorical StructureTheory(RST) [16] is one of the most often used discourse analysis methods in naturallanguage processing. Mann et al. built a bottom-up dependency tree calleda rhetorical structure by defining logical and rhetorical relations between clausesand fixing the dependency among the clauses. Based on their idea, Marcu et al.proposed a method for automatically generating a rhetorical structure tree fromthe corpus [85]. Rhetorical structure tags based on RST have been appended tosome large corpora [110].Some previous work studied Japanese corpus annotation based on descriptiontype. For instance, in annotation by human, there are categorization of definitionof word in dictionary [150], annotation of causal relation between sentences[57], and in annotation by computer, automatic tagging to definition statementsof words in web pages [31]. Those work mentioned problems of this kind ofannotation as follows;• Huge amount of corpus are required to prove statistically any hypotheses,because the number of annotated tags for description types per an articleare relatively a few comparing other linguistic annotation.• Low efficiency of annotation due to read the long context of expression whenassigning a tag to the expression.49

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!