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.

element of sentence, called anaphora resolution [52, 53, 64, 156], is generally difficult,then have not been achieved enough high accuracy to be able to used inpractical tasks. As an alternative to avoid anaphora resolution, it is considerableto chunk additional sentences possibly including elided elements. In this point ofview, I will enhance question segmentation and question type identification as infollowing paragraphs.In question segment extraction, the portion and structure of question segmentin a query have not been identified before the processing, thus bag-of-wordsapproach only using words in the query and hypothesizing no question type isplausible. However if question segment comprises many ellipses, the approachonly using bag-of-words is not enough to extract features of question segments.As a enhancement to solve this problem, it is considerable to perform only accurateellipsis analysis over the entire query as preprocessing of chunking.In experimental results of question type identification, the performances incondition using only features of a chunked segment present better evaluationvalues than using features of contextual sentences before and after the chunkedsentence together. Thus it is considered that it is difficult to improve the accuracyof question type identification by simply adding contexts of chunked sentence. Onthe other hand, because existence of ellipsis in chunked sentences is problem inquestion type identification as well as in question segment extraction, any solutionof this problem is required. As already shown in previous paragraphs, anaphoraresolution conducts not enough accurately in the current technology. In this kindof condition, the solution has to select approaches that acquire any informationabout elided elements even if anaphora resolution fails to identify those elements.As an expectable way, instead of completely identifying each ellipsis in questionsegment, selecting chunks involved elided elements and merging features in thechunks to that of target sentence can be considered. Moreover, by using chunkingresult, it can be possible to remove redundant sentences in a query from searchspace to identify elided elements.44

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

Saved successfully!

Ooh no, something went wrong!