zend japan did a modification to the file lib/ezi18n/classes/ezcodemapper.php to ensure that the search for japanese / chinese / korean words was working correctly.

attached is the original 3.9.2 file, the modified file, and a unix diff file.

basically, the code reacts on certain characters, and appends a space, and the character at the end of the $normalizedTextArray array, where the regular code would only append the space.

this new code uses n-grams where n=2 (bigram) to better match search words to the search index values.