Mojibake

Mojibake (文字化ã') (IPA: [modÍ¡Ê'ibake]; lit. "character transformation"), from the Japanese 文字 (moji) "character" + 化ã' (bake) "transform", is the garbled text that is the result of text being decoded using an unintended character encoding. The result is a systematic replacement of symbols with completely unrelated ones, often from a different writing system. This display may include the generic replacement character � in places where the binary representation is considered invalid. A replacement can also involve multiple consecutive symbols, as viewed in one encoding, when the same binary code constitutes one symbol in the other encoding. This is either because of differing constant length encoding (as in Asian 16-bit encodings vs European 8-bit encodings), or the use of variable length encodings (notably UTF-8 and UTF-16).

The rendering of glyphs, which often fails because of missing fonts, is a different issue that is not to be confused with mojibake. Symptoms include blocks with the codepoint displayed in hexadecimal, or even the generic replacement character �. Importantly, these replacements are valid, and the result of correct error handling by the software.

Causes

To correctly reproduce the original text that was encoded, the correspondence between two things must be preserved: the encoded data, and the notion of its encoding. As mojibake is the instance of incompliance between these, it can be achieved by manipulating the data itself, or just relabeling it.

Mojibake is often seen with text data that have been tagged with a wrong encoding; or not tagged at all, but moved between computers with different default encodings. A major source of trouble are communication protocols that rely on settings on each computer rather than sending or storing metadata together with the data.

The differing default settings between computers are in part due to differing deployments of Unicode among operating system families, and partly the legacy encodings' specializations for different writing systems of human languages. Whereas Linux distributions mostly switched to UTF-8 (around 2004) for all uses of text, Microsoft Windows still uses codepages for text files, that differ between languages. For some languages, an example being Japanese, several encodings have historically been employed, causing users to see mojibake relatively often. As a Japanese example, the word mojibake "文字化ã'", when encoded in UTF-8, is incorrectly displayed as "æâ€"‡åâ€"Ã¥Å'â€"け" in software that assumes text to be in the Windows-1252 or ISO-8859-1 encodings, usually labelled Western.

Underspecification

If the encoding is not specified, it is up to the software to decide it by other means. Depending on type of software, the typical solution is either configuration or charset detection heuristics. Both are prone to mispredict in not-so-uncommon scenarios.

The encoding of text files are usually governed by the OS level setting, which depends on brand of operating system and possibly the user's language. Therefore, the assumed encoding is systematically wrong for files that come from a computer with a different setting, for example when transferring files between Windows and Linux. One solution is to use a byte order mark, but for source code and other machine readable text, many parsers don't tolerate this. Another is storing the encoding as metadata in the filesystem. Filesystems that support extended file attributes can store this as user.charset. This also requires support in software that wants to take advantage of it, but does not disturb other software.

While a few encodings are easy to detect, in particular UTF-8, there are many that are hard to distinguish (see charset detection). A web browser may not be able to distinguish a page coded in EUC-JP and another in Shift-JIS if the coding scheme is not assigned explicitly using HTTP headers sent along with the documents, or using the HTML document's meta tags that are used to substitute for missing HTTP headers if the server cannot be configured to send the proper HTTP headers; see character encodings in HTML.

Misspecification

Mojibake also occurs when the encoding is wrongly specified. This often happens between encodings that are similar. For example, the Eudora email client for Windows was known to send emails labelled as ISO-8859-1 that were in reality Windows-1252. The Mac OS version of Eudora did not exhibit this behaviour. Windows-1252 contains extra printable characters in the C1 range, that were not displayed properly in software complying with the ISO standard; this especially affected software running under other operating systems (e.g. Unix).

Human ignorance

Of the encodings still in use, many are partially compatible with each other, with ASCII as the predominant common subset. This sets the stage for human ignorance:

• Compatibility can be a deceptive property, as the common subset of characters are unaffected by a mixup of two encodings (see Problems in different writing systems).
• People think they are using ASCII, and tend to label whatever superset of ASCII they actually use as "ASCII". Maybe for simplification, but even in academic literature, the word "ASCII" can be found used as an example of something not compatible with Unicode, where evidently "ASCII" is Windows-1252 and "Unicode" is UTF-8. Note that UTF-8 is backwards compatible with ASCII.

Overspecification

When there are layers of protocols, each trying to specify the encoding based on different information, the least certain information may be misleading to the recipient. For example, consider a web server serving a static HTML file over HTTP. The character set may be communicated to the client in any number of 3 ways:

• in the HTTP header. This information is based on server configuration, since the server cannot otherwise know the encoding of files on disk. Furthermore, the configuration may be more or less global, i.e. does not necessarily reflect anything about the particular file being served. For the Apache web server, adding an encoding to the HTTP header is achieved with the AddDefaultCharset option.
• in the file, as an HTML meta tag (http-equiv or charset). This is the encoding that the author meant to save the particular file in.
• in the file, as a byte order mark. This is the encoding that the author's editor actually saved it in. Unless an accidental encoding conversion has happened (by opening it in one encoding and saving it in another), this will be correct.

Resolutions

Applications using UTF-8 as a default encoding may achieve a greater degree of interoperability because of its widespread use and backward compatibility with US-ASCII.

The difficulty of resolving an instance of mojibake varies depending on the application within which it occurs and the causes of it. Two of the most common applications in which mojibake may occur are web browsers and word processors. Modern browsers and word processors often support a wide array of character encodings. Browsers often allow a user to change their rendering engine's encoding setting on the fly, while word processors allow the user to select the appropriate encoding when opening a file. It may take some trial and error for users to find the correct encoding.

The problem gets more complicated when it occurs in an application that normally does not support a wide range of character encoding, such as in a non-Unicode computer game. In this case, the user must change the operating system's encoding settings to match that of the game. However, changing the system-wide encoding settings can also cause Mojibake in pre-existing applications. In Windows XP or later, a user also has the option to use Microsoft AppLocale, an application that allows the changing of per-application locale settings. Even so, changing the operating system encoding settings is not possible on earlier operating systems such as Windows 98; to resolve this issue on earlier operating systems, a user would have to use third party font rendering applications.

Problems in different writing systems

English

Mojibake in English texts generally occurs in punctuation, such as em dashes (â€"), en dashes (â€"), and curly quotes (“, ”), but rarely in character text, since most encodings agree with ASCII on the encoding of the English alphabet. For example, the pound sign "£" will appear as "£" if it was encoded by the sender as UTF-8 but interpreted by the recipient as CP1252 or ISO 8859-1. If iterated, this can lead to "£", "ÃÆ'‚£", "ÃÆ'ƒâ€šÃÆ'‚£", etc.

Some computers did in older eras have vendor-specific encodings which caused mismatch also for English text. Commodore brand 8-bit computers used PETSCII encoding, particularly notable for inverting the upper and lower case compared to standard ASCII. PETSCII printers worked fine on other computers of the era, but flipped the case of all letters. IBM mainframes use the EBCDIC encoding which does not match ASCII at all.

Central European

Users of Central and Eastern European languages can also be affected. Because most computers were not connected to any network during the mid- to late-1980s, there were different character encodings for every language with diacritical characters.

Russian and other Cyrillic alphabets

Mojibake may be colloquially called krakozyabry (кракозя́бры, IPA:krɐkɐˈzʲæbrɪ̈) in Russian, which was and remains complicated by several systems for encoding Cyrillic. The Soviet Union and early Russian Federation developed KOI encodings (Kod Obmena Informatsiey, Код Обмена Информацией, which translates to “Code for Information Exchange”). This began with Cyrillic-only 7-bit KOI7, based on ASCII but with Latin and some other characters replaced with Cyrillic letters. Then came 8-bit KOI8 encoding that is an ASCII extension which encodes Cyrillic letters only with high-bit set octets corresponding to 7-bit codes from KOI7. That's why KOI8 text, even Russian, remains partially readable after stripping the eighth bit, which was considered as a major advantage in the age of 8BITMIME-unaware email systems. Eventually KOI8 gained different flavors for Russian/Bulgarian (KOI8-R), Ukrainian (KOI8-U), Belarusian (KOI8-RU) and even Tajik (KOI8-T).

Meanwhile in the West, Code page 866 supported Ukrainian and Belarusian as well as Russian/Bulgarian in MS-DOS. For Microsoft Windows, Code Page 1251 added support for Serbian and other Slavic variants of Cyrillic.

Most recently, the Unicode encoding includes code points for practically all the characters of all the world's languages, including all Cyrillic characters.

Before Unicode, it was necessary to match text encoding with a font using the same encoding system. Failure to do this produced unreadable gibberish whose specific appearance varied depending on the exact combination of text encoding and font encoding. Accented vowels were symptomatic of trying to view any Cyrillic encoding with a font limited to the Latin alphabet. In general, Cyrillic gibberish was symptomatic of using the wrong Cyrillic font.

In Bulgarian, mojibake is often called maymunitsa (маймуница), meaning “monkey's alphabet”. In Serbian, it is called Ñ'убре (Ä'ubre), meaning “trash”. Unlike the former USSR, South Slavs never used something like KOI8, and Code Page 1251 was the dominant Cyrillic encoding there before Unicode. Therefore these languages experienced fewer encoding incompatibility troubles than Russian. In the 1980s, Bulgarian computers used their own MIK encoding, which is superficially similar to (although incompatible with) CP866.

Polish

Prior to the creation of ISO 8859-2 in 1987, users of various computing platforms used their own character encodings such as AmigaPL on Amiga, Atari Club on Atari ST and Masovia, IBM CP852, Mazovia and Windows CP1250 on IBM PCs. Polish companies selling early DOS computers created their own mutually-incompatible ways to encode Polish characters and simply reprogrammed the EPROMs of the video cards (typically CGA, EGA, or Hercules) with the needed glyphs for Polish â€" arbitrarily located without reference to where other computer sellers had placed them.

The situation began to improve when, after pressure from academic and user groups, ISO 8859-2 succeeded as the "Internet standard" with limited support of the dominant vendors' software (today largely replaced by Unicode). With the numerous problems caused by the variety of encodings, even today some users tend to refer to Polish diacritical characters as krzaczki ([kshach-kih], lit. "little shrubs").

Western European

The alphabets of the North Germanic languages, Finnish, German, French and Spanish are all extensions of the Latin alphabet. The additional characters are typically the ones that become corrupted, making texts only mildly unreadable with mojibake:

• Ã¥, ä and ö in Finnish and Swedish
• Ã¥, æ and ø in Norwegian and Danish
• ä, ö, ü and ß in German
• á, ð, í, ó, ú, ý, æ and ø in Faroese
• á, ð, é, í, ó, ú, ý, þ, æ and ö in Icelandic
• à, â, ç, è, é, ë, ê, ï, î, ö, ô, ù, û, ÿ, æ, Å" in French
• à, è, é, ì, ò, ù in Italian
• á, é, í, ñ, ó, ú, ï, ü, ¡, ¿ in Spanish
• £ in British

... and their uppercase counterparts, if applicable.

These are languages for which the iso-8859-1 character set (also known as Latin 1 or Western) have been in use. However, iso-8859-1 have been obsoleted, by two competing standards, the backward compatible windows-1252, and the slightly altered iso-8859-15. Both adds the euro sign € and the French Å", most notably, but otherwise, any mixups of these 3 character sets do not create mojibake in these languages. Furthermore, it is always safe to interpret iso-8859-1 as windows-1252, and fairly safe to interpret it as iso-8859-15, in particular with respect to the euro sign, which replaces a never-used poor man's currency sign. However, with the advent of utf-8, mojibake has become the norm in certain scenarios, e.g. exchange of text files between Unix and Windows computers. Utf-8 reverts compatibility back to ASCII â€" all characters in iso-8859-1 that are not in ASCII take two bytes in utf-8.

In Swedish, Norwegian, Danish and German, vowels are rarely repeated, and it is usually obvious when one character gets corrupted, such as the second letter in "kÃ⁠¤rlek" (kärlek, "love"). This way, even though the reader has to guess between å, ä and ö, almost all texts remain legible. Finnish text, on the other hand, does feature repeating vowels in words like hääyö ("wedding night") which can sometimes render text very hard to read (e.g. hääyö appears as "hÃ⁠¤Ã⁠¤yÃ⁠¶"). Icelandic and Faroese have ten and eight possibly confounding characters, respectively, which thus can make it more difficult to guess corrupted characters; Icelandic words like þjóðlöð ("outstanding hospitality") become almost entirely unintelligible when rendered as "þjóðlöð".

In German, Buchstabensalat ("letter salad") is a common term for this phenomenon, and in Spanish, deformación (literally deformation).

Some users transliterate their writing when using a computer, either by omitting the problematic diacritics, or by using digraph replacements (å -> aa, ä/æ -> ae, ö/ø -> oe, etc.). Thus, an author might write "ueber" instead of "über", which is standard practice in German when umlauts are not available. The latter practice seems to be better tolerated in the German language sphere than in the Nordic countries. For example, in Norwegian, digraphs are associated with archaic Danish, and may be used jokingly. However, digraphs are useful in communication with other parts of the world. As an example, the Norwegian football player Ole Gunnar Solskjær had his name spelled "SOLSKJAER" on his back when he played for Manchester United.

An artifact of UTF-8 misinterpreted as ISO-8859-1, "Ring meg nÃ¥" ("Ring meg nå"), was seen in an SMS scam raging in Norway in June 2014.

Asian encodings

Another type of mojibake occurs when text is erroneously parsed in a multi-byte encoding, such as one of the east Asian encodings. With this kind of mojibake more than one (typically two) characters are corrupted at once, e.g. "k舐lek" (kärlek) in Swedish, where "är" is parsed as "舐". Compared to the above mojibake, this is harder to read, since letters unrelated to the problematic å, ä or ö are missing, and is especially problematic for short words starting with å, ä or ö such as "än" (which becomes "舅"). Since two letters are combined, the mojibake also seems more random (over 50 variants compared to the normal three, not counting the rarer capitals). In some rare cases, an entire text string which happens to include a pattern of particular word lengths, such as the sentence "Bush hid the facts", may be misinterpreted.

Japanese

In Japanese, the phenomenon is, as mentioned, called mojibake 文字化ã'. It is often encountered by non-Japanese when attempting to run software written for the Japanese market.

Chinese

In Chinese, the same phenomenon is called 乱码 (Simplified Chinese) / 亂碼 (Traditional Chinese) (pinyin: Luàn mǎ, meaning chaotic code), and can occur when computerised text is encoded in one Chinese character encoding but is displayed using the wrong encoding. When this occurs, it is often possible to fix the issue by switching the character encoding without loss of data. The situation is complicated because of the existence of several Chinese character encoding systems in use, the most common ones being: Unicode, Big5, and Guobiao (with several backward compatible versions), and the possibility of Chinese characters being encoded using Japanese encoding.

An additional problem is caused when encodings are missing characters, which is common with rare or antiquated characters that are still used in personal or place names. Examples of this are Taiwanese politicians Wang Chien-shien (Chinese: 王建煊; pinyin: Wáng Jiànxuān) and Yu Shyi-kun (simplified Chinese: 游é"¡å ƒ; traditional Chinese: 游錫堃; pinyin: Yóu XíkÅ«n), ex-PRC Premier Zhu Rongji (Chinese: 朱镕基; pinyin: ZhÅ« RóngjÄ«) and Taiwanese singer David Tao (Chinese: 陶喆; pinyin: Táo Zhé).

Newspapers have dealt with this problem in various ways, including using software to combine two existing, similar characters; using a picture of the personality; or simply substituting a homophone for the rare character in the hope that the reader would be able to make the correct inference.

Yugoslav languages

Croatian, Bosnian, Serbian, the variants of the Yugoslav Serbo-Croatian language, add to the basic Latin alphabet the letters Å¡, Ä', č, ć, ž, and their capital counterparts Å , Đ, ÄŒ, Ć, Ž. All of these letters are defined in Latin-2 and Windows-1250, while only some (Å¡, Å , ž, Ž, Đ) exist in the usual OS-default Windows-1252, and are there because of some other languages.

Although Mojibake can occur with any of these characters, the letters that are not included in Windows-1252 are much more prone to errors. Thus, even nowadays, "Å¡Ä'čćž ŠĐČĆŽ" is often displayed as "šðèæž ŠÐÈÆŽ", despite the fact that ð, è, æ, È, Æ are never used in Slavic languages.

When confined to basic ASCII (most user names, for example), common replacements are: Å¡â†'s, Ä'â†'dj, čâ†'c, ćâ†'cj, žâ†'z (capital forms analogously, with Đâ†'Dj or Đâ†'DJ depending on word case). All of these replacements introduce ambiguities, so reconstructing the original from such a form is usually done manually if required.

The Windows-1252 encoding is important because the English versions of the Windows operating system are most widespread, not localized ones. The reasons for this include a relatively small and fragmented market, increasing the price of high quality localization, a high degree of software piracy (in turn caused by high price of software compared to income), which discourages localization efforts, and people preferring English versions of Windows and other software .

The drive of differentiating Croatian from Serbian, Bosnian from Croatian and Serbian, and now even Montenegrin from the other three creates many problems. There are many different localizations, using different standards, and of different quality. There are no common translations for the vast amount of computer terminology originating in English. In the end, people use adopted English words ("kompjuter" for "computer", "kompajlirati" for "compile," etc.), and if they are unaccustomed to the translated terms may not understand what some option in a menu is supposed to do based on the translated phrase. Therefore, the people who understand English, as well as those who are accustomed to English terminology (which are most, because English terminology is also mostly taught in schools because of these problems) regularly choose the original English versions of non-specialist software.

When Cyrillic script is used (for Macedonian and partially Serbian language), the problem is similar to other Cyrillic-based scripts.

Newer versions of English Windows allow the ANSI codepage to be changed (older versions require special English versions with this support), but this setting can be and often was incorrectly set. For example, Windows 98/Me can be set to most non-RTL SBCS codepages including 1250, but only at install time.

Hungarian

Hungarian is another affected language, which uses the 26 basic English characters, plus the accented forms á, é, í, ó, ú, ö, ü (all present in the Latin-1 character set), plus the 2 characters Å' and ű, which are not in Latin-1. These 2 characters can be correctly encoded in Latin-2, Windows-1250 and Unicode. Before Unicode became common in e-mail clients, e-mails containing Hungarian text often had the letters Å' and ű corrupted, sometimes to the point of unrecognizability. It is common to respond to an e-mail rendered unreadable (see examples below) by character mangling (referred to as "betűszemét", meaning "garbage lettering") with the phrase "ÁrvíztűrÅ' tükörfúrógép", a nonsense phrase (literally "Flood-resistant mirror-drilling machine") containing all accented characters used in Hungarian.

Examples

Indic text

A similar effect can occur in Indic text, even if the character set used is properly recognized by the application. This is because, in many Indic scripts, the rules by which individual letter symbols combine to create symbols for syllables may not be properly understood by a computer missing the appropriate software, even if the glyphs for the individual letter forms are available.

A particularly notable example of this is the old Wikipedia logo, which attempts to show the character analogous to "wi" (the first syllable of "Wikipedia") on each of many puzzle pieces. The puzzle piece meant to bear the Devanagari character for "wi" instead used to display the "wa" character followed by an unpaired "i" modifier vowel, easily recognizable as mojibake generated by a computer not configured to display Indic text. The logo as redesigned as of May 2010 has fixed these errors.

The idea of Plain Text requires the operating system to provide a font to display Unicode codes. This font is different from OS to OS for Singhala and it makes orthographically incorrect glyphs for some letters (syllables) across all operating systems. For instance, the 'reph', the short form for 'r' is a diacritic that normally goes on top of a plain letter. However, it is wrong to go on top of some letters like 'ya' or 'la' but it happens in all operating systems. This appears to be a fault of internal programming of the fonts. In Macintosh / iPhone, the muurdhaja l (dark l) and 'u' combination and its long form both yield wrong shapes.

Some Indic and Indic-derived scripts, most notably Lao, were not officially supported by Windows XP until the release of Vista. However, various sites have made free-to-download fonts.

Arabic

Another affected language is Arabic (see below). The text in general gets totally unreadable when the encodings do not match.

Example

See also

• Newline â€" To the theme of unfortunate text encoding diversity comes the notable special case of representing the line break. It too has had its historical conventions, and to date differs between Windows and Unix systems. Though most software supports both conventions (which is trivial), software that must preserve or display the difference (e.g. version control systems and data comparison tools) can get substantially more difficult to use if not adhering to one convention.
• Byte order mark â€" The most in-band way to store the encoding together with the data â€" prepend it. This is by intention invisible to humans using compliant software, but will by design be perceived as "garbage characters" to incompliant software (including many interpreters).

References

External links

• Avoiding Mojibake - Reprinted article from the Japan Times
• Coldfusion Developers Guide article
• Online decoder of Hebrew text â€" Online decoder converts input into readable Hebrew text
• Translators guide
• Recovery tool â€" Recovery of Japanese text (Japanese)
• Chinese E-mail Fixer â€" Recovery of Chinese text
• Chinese Encoding Converter - Convert file encoding of text files
• Decodr.ru - Fast Cyrillic decoder of e-mail and charsets
• Universal Cyrillic decoder â€" Recovery of Cyrillic (and other) text
• Multilingual online text decoder â€" Online decoder converts scrambled input into readable text. Supported languages are Russian, Bulgarian, Greek, Hebrew and Thai.
• Noto pri ĉapeloj: Resuma tabelo â€" Covers most cases of Esperanto mojibake.
• Encoding Repair Kit â€" Fixes mojibake, originally intended for Japanese. Windows freeware, 1998. Not compatible with Windows Vista or 7.
• Krzaki krzaki krzaki - Online decoder for Polish mojibake cases.