BerkeleyParser 1.7 modified by NICT Masao Utiyama mutiyama@nict.go.jp Fri Jan 15 11:54:45 JST 2016 * Introduction BerkeleyParser is one of the most accurate parsers. Please refer to README.txt for the original README.txt. Machine Translation Group at NICT has extended BerkeleyParser 1.7 so that it is better suited to our purposes. Please refer to changes.txt for the modifications we made. There are two important modifications. 1. We added '-robust' option. When you specify '-robust' as java -jar BerkeleyParser-1.7g1.jar -robust -gr Then, the parser try to output a fragmental parsing result when it fails to obtain a full parsing tree. As a result, there are few cases in which the parser outputs '(( ))' 2. We made BerkeleyParser run faster. Please refer to README_NICT_fast.txt for details. This work was done by Jun Kawai. In addition, we have also added options '-constraint', '-weightIn', '-weightOut', and '-wall', which were incorporated by Isao Goto into BerkeleyParser 1.5, while he was in NICT. These options were used in Isao Goto, Masao Utiyama, Eiichiro Sumita, and Sadao Kurohasi. (2015) Preordering using a Target-Language Parser via Cross-Language Syntactic Projection for Statistical Machine Translation. ACM TALLIP, vol. 14, Issue 3, Article No.13. However, we are no longer sure if these options in version 1.7 work properly. * LICENSE Because the original BerkeleyParser is licensed as GPL-2, the license of this software is also GPL-2. If you use this code in your research and would like to acknowledge it, please write in your paper as We used the BerkeleyParser 1.7 modified by NICT\footnote{http://www2.nict.go.jp/univ-com/multi_trans/member/mutiyama/software.html}