Looking at the collates.txt file, I see a lot of replacements that don't seem to make sense.
Code:
「」。、.,-_ '`!?:;*´"()[]<>{}/\*«»‘’‚‛“”„‟…=
aÀÁÂÃÄÅÆĀĂĄǠǢǍǞǺǼȀȂȦàáâãäåæāăąǡǣǎǟǻǽȁȃȧ=A
bƀƁƂƃƄƅ=B
cÇĆĈĊČƆƇçćĉċčƈ=C
dĎĐƉƊƋďđƌƍ=D
eÈÉÊËĒĔĖĘĚƎƏƐȄȆȨèéêëēĕėęěȅȇȩ=E
fƑƒ=F
gĜĞĠĢƓǤǦǴĝğġģǥǧǵ=G
hĤĦǶȞĥħƕǷȟ=H
iÌÍÎÏĨĪĬĮİIJƖȈȊìíîïĩīĭįıijƗȉȋ=I
jĴĵ=J
kĶƘǨķĸƙǩ=K
lĹĻĽĿŁĺļľŀłƛƚ=L
mƜµ=M
nÑŃŅŇŊƝñńņňʼnŋƞ=N
oÒÓÔÕÖØŌŎŐŒƟƠƢǑǪǬǾȌȎȪȬȮȰòóôõöøōŏőœơƣǒǫǭǿȍȏȫȭȯȱ=O
pƤƥ=P
q=Q
rŔŖŘƦȐȒŕŗřȑȓ=R
sߌŜŞŠƧƩȘśŝşšƨƪș=S
tŢŤŦƬƮȚţťŧƫƭț=T
uÙÚÛÜŨŪŬŮŰŲƯƱǓǕǗǙǛȔȖùúûüũūŭůűųưƲǔǖǘǚǜȕȗ=U
v=V
wŴŵ=W
x=X
yÝŶŸƳȲýÿŷƴȳ=Y
zŹŻŽƵƷƹƻǮȤźżžƶƸƺǯȥ=Z
アァ=あ
イィ=い
ウゥ=う
エェ=え
オォ=お
カ=か
キ=き
ク=く
ケ=け
コ=こ
サ=さ
シ=し
ス=す
セ=せ
ソ=そ
タ=た
チ=ち
ツ=つ
テ=て
ト=と
ナ=な
ニ=に
ヌ=ぬ
ネ=ね
ノ=の
ハ=は
ヒ=ひ
フ=ふ
ヘ=へ
ホ=ほ
マ=ま
ミ=み
ム=む
メ=め
モ=も
ラ=ら
リ=り
ル=る
レ=れ
ロ=ろ
ワヮ=わ
ヤ=や
ユ=ゆ
ヨ=よ
ヲ=を
ャ=ゃ
ュ=ゅ
ョ=ょ
ガ=が
ギ=ぎ
グ=ぐ
ゲ=げ
ゴ=ご
ザ=ざ
ジ=じ
ズ=ず
ゼ=ぜ
ゾ=ぞ
ダ=だ
ヂ=ぢ
ヅ=づ
デ=で
ド=ど
バ=ば
ビ=び
ブ=ぶ
ベ=べ
ボ=ぼ
パ=ぱ
ピ=ぴ
プ=ぷ
ペ=ぺ
ポ=ぽ
ン=ん
ッ=っ
For instance there are `ヘ=へ` are 'ペ=ぺ' which replace a symbol with the same symbol?
And 'ジ=じ' and 'ヅ=づ' replace the same symbol with 2 different ones. (Probably only one is used.)
As they are just nice characters to me without any reference, I can't really say whether they make sense or improve your dictionary search. If your understanding of Japanese is good enough, would you be willing to look through it?
Looking at the keyboard.txt file for jaK, I see 4 pages with keyboard layouts. Lower-case roman, Upper-case roman and then two nearly identical pages with characters --small upper right strokes versus upper right circles-- separated with a new separator. The first is separated with a ' ~ - ' and 2 empty lines. The second also with a ' ~ - ' and no empty lines. Then a wholly new separator '--' between two pages with characters.
Code:
JP: Japan
qwertyuio
asdfghjkl
zxcvbnmp'
~ -
QWERTYUIO
ASDFGHJKL
ZXCVBNMP'
~ -
わ ら ま は な た さ か あ bs
や り み ひ に ち し き い cl
ゆ る む ふ ぬ つ す く う ん
よ れ め へ ね て せ け え sh
を ろ も ほ の と そ こ お ok
--
わ ら ぱ ば な だ ざ が あ bs
ゃ り ぴ び に ぢ じ ぎ い cl
ゅ る ぷ ぶ っ づ ず ぐ う ん
ょ れ ぺ べ ね で ぜ げ え sh
を ろ ぽ ぼ の ど ぞ ご お ok
If you can't see any Japanese characters I would start with changing all the separators to the ones used between the lower and upper case roman character keyboard layouts, recompile and see whether it has an impact.