commit 11ca79bcb3442fcdd6d804499b1a79e3d06c501c
author Noriyuki Takahashi <noriyukit@google.com> Wed Dec 17 07:17:51 2014 +0000
committer Yohei Yukawa <yukawa@google.com> Wed Dec 17 07:17:51 2014 +0000
tree 71082d6b2037603946f0e08d68a8516214b979ad
parent ec6babed0322adbce77efadc3ade78ebed704997

Learn content word for real time conversion in the mobile mode

To propagate the segmentation information in mobile use, this CL extends Segment::Candidate::inner_segment_boundary to include content key and value lengths.  This extra information is utilized in UserHistoryPredictor to reconstruct content word.

Motivation:
In mobile mode, it would be useful if content words can be learned.  Here are some examples.

Case 1:
1. Type がっこうに
2. Select 学校に
学校に (which is one segment) is learned but 学校 (content word) is not learned.  Here learning 学校 makes sense for mobile since particles are suggested by zero query suggestion in the mobile mode.

Case 2:
1. Type とうきょうかなごやにいきたい
2. Select 東京か名古屋に行きたい
This is similar to the case 1 but we can learn both 東京 and 名古屋.

Note that this CL should not change anything in the desktop mode.

BUG=none
TEST=unittest

git-svn-id: https://mozc.googlecode.com/svn/trunk@449 a6090854-d499-a067-5803-1114d4e51264