Ar Tonelico III

Lots and lots of text and lots of obscure kanji! First I'll finish the game, then I can re-watch the cosmosphere events from the extra menu. I will probably do the cosmospheres (they're ridiculously funny), perhaps some talk events and if there should be interested, let's see.

Translations

• Japanese to English

(or should I translate to German??)

Tips

I used the Japanese IME "canna" under ubuntu, compiled from source, and changed kana-kanji dictionaries, so that kanji+furigana is written upon entering and converting Japanese te

Script

Now I managed to dump the script. Much better than having to write by hand:) And I also ripped the voice clips and bgm, character poses, the textures, I can view a few models (not the character models though)... The dump can be found here: [1] or [2] or [3] Now updated with speaker information! For anyone who's interested, what I found out...

AT3 ebd script files

• consists of EVENT_MESSAGE_SW[2digit-NUMBER]_[3digit-NUMBER].ebm (called DIAG from now on) and EVENT_SW[2digit-NUMBER]_[3digit-NUMBER].ebm (called CTRL from now on) 
• each DIAG corresponds do a CTRL file with the same NUMBER's
•  DIAG to contains the main dialogue lines, while CTRL is probably system-related
• DIAG files are also usually only a few hundred bytes long
• DIAG has a header of 4 bytes, then comes the main part

• the first 26 bytes of CTRL are as follows (decimal):
[#1] 000 000 000 000 000 000 000 005 000 000 000 110 097 109 101 000 005 000 000 000 144 224 150 190 000 [#2] [#3] 000 000 [#4] 000 [#5] [#6] [#7] [#8] [#8] 000 [#9] 000
whereas #[n] are 
-- #1takes many different values, 001 is very frequent (~50%)
-- #2 takes many different values
-- #3 mostly 000, a few times 001, 002,  4 times 003, 3 times 004
-- #4 mostly small bytes <=021, 021 and 00x frequently occur in adjacent files together, takes 044 in two instances
-- #5 either 000, 016, 049, or 064
-- #6 always either 113, 116, 117, 119, or 127
-- #7 bytes <= 025, either 00x or 021x with x<=5 except a handful of times
-- #8 almost always 000, except 10 and 7 files respectively
-- #9 either 000, 001, 002, 003, 004, 005, 017, 019, 021. 025 with the lower bytes much more common
• the byte of CTRL always seems to be <bh:7f>, the last 26 bytes only being somewhat similiar
• in general, CTRL displays a high ration of <bh:00>
• CTRL contains no UTF8 chars
• the main part of CTRL, apart from the man 0's, contains only ASCII chars, most of which are LATIN characters and punctuation, with a few special chars such as <bh:f4>, <bh:dc> (Ü)

• the main part of DIAG is in the following format, after the 3-byte header comes:
[SEPARATOR]  [UTF8-sequence][SEPARATOR][UTF8-sequence] ... [UTF8-sequence][SEPARATOR]
• as the text is Japanese, [UTF8-sequence] is usually a multiple of 3-byte blocks, each block representing a multi-byte for one Japanese character; it terminates on a zero-byte
• the main text may contain a ※削除※ line, [LEADING] is then <bh:ff>
• [SEPARATOR] always consists of 36 bytes, each byte smaller than <bd:192>,  with the only exception it may also contain <bh:ff>. Not counting the <bh:00> byte UTF8 terminating byte.
• [SEPARATOR]: most bytes are constant, except the following meaningful bytes
• the 25th byte: it is a [LEADING] number, counting the dialogue lines
• a [LEADING] byte <bh:ff> this line is outside the "normal" dialogue flow, ie a system message ("You got item..") or "Party member xyz joined." or "……。"  or "…！？" &c.
• the 13th byte: this indicates the [SPEAKER]. [SPEAKER] is <bh:ff> when there is no speaker
• the first byte indicates the [MODE]
<bh:00> - talk with speech bubbles at character's 3D models
<bh:01> - talk with 2D character portraits
<bh:02> - item get
TO SUMMARIZE
• dialogue in EVENT-MESSAGE file: [3 byte header][36-byte separator][UTF8 byte sequence, terminating on <bh:00>], repeat
• 13th byte [SEPARATOR] is speaker, 26th [SEPARATOR] marks "normal" spoken text

And here is an improved lua script I wrote that looks for valid UTF8 sequences in a file, works much better and doesn't need specific information on separators &c.:

--PARSE BYTES FOR VALID UTF8 string terminated sequences (except ASCII, ie the first bit non-zero)
--specify how what's between UTF8 should be interpreted, don't forget the newline!
function process_separator(sep)
 -- return "" --just use this if you only the UTF8 data
 if #sep>25 then
  return "(" .. sep[1] .. "・" .. sep[13] .. "・" .. sep[25] .. ")"
 else
  local tag = ""
  local to_number = {A=10, B=11, C=12, D=13, E=14, F=15, a=10, b=11, c=12, d=13, e=14, f=15}
  to_number["0"] = 0
  to_number["1"] = 1
  to_number["2"] = 2
  to_number["3"] = 3
  to_number["4"] = 4
  to_number["5"] = 5
  to_number["6"] = 6
  to_number["7"] = 7
  to_number["8"] = 8
  to_number["9"] = 9
  for _,v in ipairs(sep) do
   local h1=to_number[string.sub(v,1,1)]
   local h2=to_number[string.sub(v,2,2)]
   if h2 then
    tag = tag .. string.char(16*h1+h2)
   else
    tag = tag .. string.char(h1)
   end
  end
  return tag
 end
end

--true is interpreted as 1, nil or false as 0
function dec_to_8bit(dec,byte) --byte must point to an initialized table of wrong (false or nil) values
 local exp = 128
 for i=1,8 do
  if dec >= exp then
   byte[i] = true
   dec = dec-exp
  end
  exp = exp*.5
 end
end

function get_utf8(filename, outname)
 local infile = io.open(filename,"rb")
 if infile then
  print("Searching for valid UTF8 in file: " .. filename .. "...")
  out = io.open(outname,"a+") --change "w+" to "a+" to append to end of file, not deleting previous data
  out:write("#FILE: " .. filename .. "\n")
  local occ    = 0 --just count how many valid chars we found
  local cur_pos
  local len    = 0
  local len2
  local utf8   = ""
  local betw   = {} --what's between the utf8 sequences
  local betw2
  local insert_line_break = false
  local insert_sep        = false
  local file_len = infile:seek("end")
  infile:seek("set")
  repeat
   local dec  = string.byte(infile:read(1))
   cur_pos = infile:seek("cur")
   local byte = {}
   dec_to_8bit(dec,byte)
   if len >= 1 then
    if not byte[1] or byte[2] then
     --UTF8 multibyte chars MUST start with 10 except the first byte!
     utf8 = ""
     insert_sep = true
     table.insert(betw,string.format("%x",betw2))
     --return to where we wrongly assumed UTF8 started...
     cur_pos = cur_pos+len2-len-1
     infile:seek("set",cur_pos)
     len  = 0
    else
     --valid utf8 found, dumping...
     utf8 = utf8 .. string.char(dec)
     len = len - 1
     occ = occ + 1
     if len==1 then
      --utf8 sequence end
      len = 0
      if insert_sep and #betw>0 then
       out:write(process_separator(betw) .. utf8)
       betw = {}
       insert_sep = false
      else
       out:write(utf8)
      end
      insert_line_break = true
      utf8 = ""
     end
    end
   else
    if byte[1] and byte[2] then --we are not interested in ASCII chars... otherwise allow b2=="0"
     -- now determine byte length of glyph
     len = 2
     repeat
      len = len+1
     until not byte[len]
     len = len-1
     if len > 6 then
      --UTF8 only allows for 6byte chars at most
      table.insert(betw,string.format("%x",dec))
      len = 0
      insert_sep = true
     else
      utf8 = utf8 .. string.char(dec)
      len2 = len
      betw2= dec
     end
    else
     if (dec == 0) and insert_line_break then
      --zero terminated :)
      out:write("\n")
     else
      table.insert(betw,string.format("%x",dec))
     end
     insert_sep = true
    end
    insert_line_break = false
   end
  until cur_pos >= file_len
  out:write("\n")
  infile:close()
  out:close()
  print("Found " .. occ .. " valid UTF8 chars, except ASCII.\nWritten to "  .. outname .. ".\nDone.")
 end
 return occ
end

get_utf8(arg[1],arg[2])

Oh, and btw, the AT3 script contains 1026643 characters : )