File novel-pinyin-fixes-train-factor.patch of Package novel-pinyin

According to Funda Wang's description of ibus-pinyin,
I found that I make a mistake on self-learning:
When novel-pinyin guess a wrong result, I should not
correct it by one mistake. Instead only three
 same consecutive wrong guess will correct the guess
result.
(Note: for example, you input two different 
Chinese names with the same pronunciation.)

commit 66d10c5d843dc3bfbd04d6e5cd0aa158fcb383e3
Author: Peng Wu <epico@linux-zyun.(none)>
Date:   Sun Mar 8 11:25:46 2009 +0800

    Modify train_factore from 70 to 23.

diff --git a/src/lookup/pinyin_lookup.cpp b/src/lookup/pinyin_lookup.cpp
index 603ab7a..c335453 100644
--- a/src/lookup/pinyin_lookup.cpp
+++ b/src/lookup/pinyin_lookup.cpp
@@ -427,7 +427,7 @@ bool PinyinLookup::train_result(PinyinKeyVector keys, CandidateConstraints const
     //TODO: verify the new training method.
     phrase_token_t last_token = sentence_start;
     // constraints->len + 1 == results->len
-    guint32 train_factor = 70;
+    guint32 train_factor = 23;
     for ( size_t i = 0; i < constraints->len; ++i){
 	phrase_token_t * token = &g_array_index(results, phrase_token_t, i);
 	if ( *token == NULL )