dict-file affix-file hash-file
buildhash -s count affix-file
tryaffix [-p|-s] [-c] expanded-file affix [+addition]
icombine [-T type] [-w chars] [aff-file]
Ispell is fashioned after the spell program from ITS (called ispell on Twenex systems.) The most common usage is "ispell filename". In this case, ispell will display each word which does not appear in the dictionary at the top of the screen and allow you to change it. If there are "near misses" in the dictionary (words which differ by only a single letter, a missing or extra letter, a pair of transposed letters, or a missing space or hyphen), then they are also displayed on following lines. As well as "near misses", ispell may display other guesses at ways to make the word from a known root, with each guess preceded by question marks. Finally, the line containing the word and the previous line are printed at the bottom of the screen. If your terminal can display in reverse video, the word itself is highlighted. You have the option of replacing the word completely, or choosing one of the suggested words. Commands are single characters as follows (case is ignored):
If the -M switch is specified, a one-line mini-menu at the bottom of the screen will summarize these options. Conversely, the -N switch may be used to suppress the mini-menu. (The minimenu is displayed by default if ispell was compiled with the MINIMENU option, but these two switches will always override the default).
If the -L flag is given, the specified number is used as the number of lines of context to be shown at the bottom of the screen (The default is to calculate the amount of context as a certain percentage of the screen size). The amount of context is subject to a system-imposed limit.
If the -V flag is given, characters that are not in the 7-bit ANSI printable character set will always be displayed in the style of "cat -v", even if ispell thinks that these characters are legal ISO Latin-1 on your system. This is useful when working with older terminals. Without this switch, ispell will display 8-bit characters "as is" if they have been defined as string characters for the chosen file type.
"Normal" mode, as well as the -l, -a, and -A options and interactive mode (see below) also accepts the following "common" flags on the command line:
The -H, -n, -t, and -o options select whether ispell runs in HTML (-H), nroff/troff (-n), TeX/LaTeX (-t), or ordinary text (-o) input mode. mode. (The default mode is controlled by the DEFTEXFLAG installation option, but is normally nroff/troff mode for historical reasons.) Unless overridden by one of the mode-selection switches, TeX/LaTeX mode is automatically selected if an input file has the extension ".tex", and HTML mode is automatically selected if an input file has the extension ".html" or ".htm".
In HTML mode, HTML tags delimited by <> signs are skipped, except that the "ALT=" construct is recognized if it appears with no spaces around the equals sign, and the text inside is spell-checked.
In TeX/LaTeX mode, whenever a backslash ("\") is found, ispell will skip to the next whitespace or TeX/LaTeX delimiter. Certain commands contain arguments which should not be checked, such as labels and reference keys as are found in the \cite command, since they contain arbitrary, non-word arguments. Spell checking is also suppressed when in math mode. Thus, for example, given
ispell will find "Ckapter" but not "SCH". The -t option does not recognize the TeX comment character "%", so comments are also spell-checked. It also assumes correct LaTeX syntax. Arguments to infrequently used commands and some optional arguments are sometimes checked unnecessarily. The bibliography will not be checked if ispell was compiled with IGNOREBIB defined. Otherwise, the bibliography will be checked but the reference key will not.
References for the tib (if available on your system), bibliography system, that is, text between a ``[.'' or ``<.'' and ``.]'' or ``.>'' will always be ignored in TeX/LaTeX mode.
The -b and -x options control whether ispell leaves a backup (.bak) file for each input file. The .bak file contains the pre-corrected text. If there are file opening / writing errors, the .bak file may be left for recovery purposes even with the -x option. The default for this option is controlled by the DEFNOBACKUPFLAG installation option.
The -B and -C options control how ispell handles run-together words, such as "notthe" for "not the". If -B is specified, such words will be considered as errors, and ispell will list variations with an inserted blank or hyphen as possible replacements. If -C is specified, run-together words will be considered to be legal compounds, so long as both components are in the dictionary, and each component is at least as long as a language-dependent minimum (3 characters, by default). This is useful for languages such as German and Norwegian, where many compound words are formed by concatenation. (Note that compounds formed from three or more root words will still be considered errors). The default for this option is language-dependent; in a multi-lingual installation the default may vary depending on which dictionary you choose.
The -P and -m options control when ispell automatically generates suggested root/affix combinations for possible addition to your personal dictionary. (These are the entries in the "guess" list which are preceded by question marks.) If -P is specified, such guesses are displayed only if ispell cannot generate any possibilities that match the current dictionary. If -m is specified, such guesses are always displayed. This can be useful if the dictionary has a limited word list, or a word list with few suffixes. However, you should be careful when using this option, as it can generate guesses that produce illegal words. The default for this option is controlled by the dictionary file used.
The -S option suppresses ispell's normal behavior of sorting the list of possible replacement words. Some people may prefer this, since it somewhat enhances the probability that the correct word will be low-numbered.
The -d option is used to specify an alternate hashed dictionary file, other than the default. If the filename does not contain a "/", the library directory for the default dictionary file is prefixed; thus, to use a dictionary in the local directory "-d ./xxx.hash" must be used. This is useful to allow dictionaries for alternate languages. Unlike previous versions of ispell, a dictionary of /dev/null is illegal, because the dictionary contains the affix table. If you need an effectively empty dictionary, create a one-entry list with an unlikely string (e.g., "qqqqq").
The -p option is used to specify an alternate personal dictionary file. If the file name does not begin with "/", $HOME is prefixed. Also, the shell variable WORDLIST may be set, which renames the personal dictionary in the same manner. The command line overrides any WORDLIST setting. If neither the -p switch nor the WORDLIST environment variable is given, ispell will search for a personal dictionary in both the current directory and $HOME, creating one in $HOME if none is found. The preferred name is constructed by appending ".ispell_" to the base name of the hash file. For example, if you use the English dictionary, your personal dictionary would be named ".ispell_english". However, if the file ".ispell_words" exists, it will be used as the personal dictionary regardless of the language hash file chosen. This feature is included primarily for backwards compatibility.
If the -p option is not specified, ispell will look for personal dictionaries in both the current directory and the home directory. If dictionaries exist in both places, they will be merged. If any words are added to the personal dictionary, they will be written to the current directory if a dictionary already existed in that place; otherwise they will be written to the dictionary in the home directory.
The -w option may be used to specify characters other than alphabetics which may also appear in words. For instance, -w "&" will allow "AT&T" to be picked up. Underscores are useful in many technical documents. There is an admittedly crude provision in this option for 8-bit international characters. Non-printing characters may be specified in the usual way by inserting a backslash followed by the octal character code; e.g., "\014" for a form feed. Alternatively, if "n" appears in the character string, the (up to) three characters following are a DECIMAL code 0 - 255, for the character. For example, to include bells and form feeds in your words (an admittedly silly thing to do, but aren't most pedagogical examples):
Numeric digits other than the three following "n" are simply numeric characters. Use of "n" does not conflict with anything because actual alphabetics have no meaning - alphabetics are already accepted. Ispell will typically be used with input from a file, meaning that preserving parity for possible 8 bit characters from the input text is OK. If you specify the -l option, and actually type text from the terminal, this may create problems if your stty settings preserve parity.
It is not possible to use -w with certain characters. In particular, the flag-marker character for the language (defined in the affix file, but usually "/") can never be made into a word character.
The -W option may be used to change the length of words that ispell always accepts as legal. Normally, ispell will accept all 1-character words as legal, which is equivalent to specifying "-W 1." (The default for this switch is actually controlled by the MINWORD installation option, so it may vary at your installation.) If you want all words to be checked against the dictionary, regardless of length, you might want to specify "-W 0." On the other hand, if your document specifies a lot of three-letter acronyms, you would specify "-W 3" to accept all words of three letters or less. Regardless of the setting of this option, ispell will only generate words that are in the dictionary as suggested replacements for words; this prevents the list from becoming too long. Obviously, this option can be very dangerous, since short misspellings may be missed. If you use this option a lot, you should probably make a last pass without it before you publish your document, to protect yourself against errors.
The -T option is used to specify a default formatter type for use in generating string characters. This switch overrides the default type determined from the file name. The type argument may be either one of the unique names defined in the language affix file (e.g., nroff) or a file suffix including the dot (e.g., .tex). If no -T option appears and no type can be determined from the file name, the default string character type declared in the language affix file will be used.
The -k option is used to enhance the behavior of certain deformatters. The name parameter gives the name of a deformatter keyword set (see below), and the list parameter gives a list of one or more keywords that are to be treated specially. If list begins with a plus (+) sign, it is added to the existing keywords; otherwise it replaces the existing keyword list. For example, -ktexskip1 +bibliographystyle adds "bibliographystyle" to the TeX skip-1 list, while -khtmlignore pre,strong replaces the HTML ignore list with "pre" and "strong". The lists available are:
All of the above keyword lists can also be modified by environment variables whose names are the same as above, except in uppercase, e.g., TEXSKIP1. The -k switch overrides (or adds to) the environment variables, and the environment variables override or add to the built-in defaults.
The -F switch specifies an external deformatter program. This program should read data from its standard input and write to its standard output. The program must produce exactly one character of output for each character of input, or ispell will lose synchronization and corrupt the output file. Whitespace characters (especially blanks, tabs, and newlines) and characters that should be spell-checked should be passed through unchanged. Characters that should not be spell-checked should be converted into blanks or other non-word characters. For example, an HTML deformatter might turn all HTML tags into blanks, and also blank out all text delimited by tags such as "code" or "kbd".
The -F switch is the preferred way to deformat files for ispell, and eventually will become the only way.
If ispell is invoked without any filenames or mode switches, it enters an interactive mode designed to let the user check the spelling of individual words. The program repeatedly prompts on standard output with "word:" and responds with either "ok" (possibly with commentary), "not found", or "how about" followed by a list of suggestions.
The -l or "list" option to ispell is used to produce a list of misspelled words from the standard input.
The -a option is intended to be used from other programs through a pipe. In this mode, ispell prints a one-line version identification message, and then begins reading lines of input. For each input line, a single line is written to the standard output for each word checked for spelling on the line. If the word was found in the main dictionary, or your personal dictionary, then the line contains only a '*'. If the word was found through affix removal, then the line contains a '+', a space, and the root word. If the word was found through compound formation (concatenation of two words, controlled by the -C option), then the line contains only a '-'.
If the word is not in the dictionary, but there are near misses, then the line contains an '&', a space, the misspelled word, a space, the number of near misses, the number of characters between the beginning of the line and the beginning of the misspelled word, a colon, another space, and a list of the near misses separated by commas and spaces. Following the near misses (and identified only by the count of near misses), if the word could be formed by adding (illegal) affixes to a known root, is a list of suggested derivations, again separated by commas and spaces. If there are no near misses at all, the line format is the same, except that the '&' is replaced by '?' (and the near-miss count is always zero). The suggested derivations following the near misses are in the form:
(e.g., "re+fry-y+ies" to get "refries") where each optional pfx and sfx is a string. Also, each near miss or guess is capitalized the same as the input word unless such capitalization is illegal; in the latter case each near miss is capitalized correctly according to the dictionary.
Finally, if the word does not appear in the dictionary, and there are no near misses, then the line contains a '#', a space, the misspelled word, a space, and the character offset from the beginning of the line. Each sentence of text input is terminated with an additional blank line, indicating that ispell has completed processing the input line.
These output lines can be summarized as follows:
For example, a dummy dictionary containing the words "fray", "Frey", "fry", and "refried" might produce the following response to the command "echo 'frqy refries | ispell -a -m -d ./test.hash":
(#) International Ispell Version 3.0.05 (beta), 08/10/91 & frqy 3 0: fray, Frey, fry & refries 1 5: refried, re+fry-y+ies
This mode is also suitable for interactive use when you want to figure out the spelling of a single word.
The -A option works just like -a, except that if a line begins with the string "&Include_File&", the rest of the line is taken as the name of a file to read for further words. Input returns to the original file when the include file is exhausted. Inclusion may be nested up to five deep. The key string may be changed with the environment variable INCLUDE_STRING (the ampersands, if any, must be included).
When in the -a mode, ispell will also accept lines of single words prefixed with any of '*', '&', '@', '+', '-', '~', '#', '!', '%', '`', or '^'. A line starting with '*' tells ispell to insert the word into the user's dictionary (similar to the I command). A line starting with '&' tells ispell to insert an all-lowercase version of the word into the user's dictionary (similar to the U command). A line starting with '@' causes ispell to accept this word in the future (similar to the A command). A line starting with '+', followed immediately by tex or nroff will cause ispell to parse future input according the syntax of that formatter. A line consisting solely of a '+' will place ispell in TeX/LaTeX mode (similar to the -t option) and '-' returns ispell to nroff/troff mode (but these commands are obsolete). However, the string character type is not changed; the '~' command must be used to do this. A line starting with '~' causes ispell to set internal parameters (in particular, the default string character type) based on the filename given in the rest of the line. (A file suffix is sufficient, but the period must be included. Instead of a file name or suffix, a unique name, as listed in the language affix file, may be specified.) However, the formatter parsing is not changed; the '+' command must be used to change the formatter. A line prefixed with '#' will cause the personal dictionary to be saved. A line prefixed with '!' will turn on terse mode (see below), and a line prefixed with '%' will return ispell to normal (non-terse) mode. A line prefixed with '`' will turn on verbose-correction mode (see below); this mode can only be disabled by turning on terse mode with '%'.
Any input following the prefix characters '+', '-', '#', '!', '%', or '`' is ignored, as is any input following the filename on a '~' line. To allow spell-checking of lines beginning with these characters, a line starting with '^' has that character removed before it is passed to the spell-checking code. It is recommended that programmatic interfaces prefix every data line with an uparrow to protect themselves against future changes in ispell.
To summarize these:
In terse mode, ispell will not print lines beginning with '*', '+', or '-', all of which indicate correct words. This significantly improves running speed when the driving program is going to ignore correct words anyway.
In verbose-correction mode, ispell includes the original word immediately after the indicator character in output lines beginning with '*', '+', and '-', which simplifies interaction for some programs.
The -s option is only valid in conjunction with the -a or -A options, and only on BSD-derived systems. If specified, ispell will stop itself with a SIGTSTP signal after each line of input. It will not read more input until it receives a SIGCONT signal. This may be useful for handshaking with certain text editors.
The -f option is only valid in conjunction with the -a or -A options. If -f is specified, ispell will write its results to the given file, rather than to standard output.
The -v option causes ispell to print its current version identification on the standard output and exit. If the switch is doubled, ispell will also print the options that it was compiled with.
The -c, -e[1-5], and -D options of ispell, are primarily intended for use by the munchlist shell script. The -c switch causes a list of words to be read from the standard input. For each word, a list of possible root words and affixes will be written to the standard output. Some of the root words will be illegal and must be filtered from the output by other means; the munchlist script does this. As an example, the command:
BOTHER BOTHE/R BOTH/R
The -e switch is the reverse of -c; it expands affix flags to produce a list of words. For example, the command:
An optional expansion level can also be specified. A level of 1 (-e1) is the same as -e alone. A level of 2 causes the original root/affix combination to be prepended to the line:
BOTH/R BOTH BOTHER
A level of 3 causes multiple lines to be output, one for each generated word, with the original root/affix combination followed by the word it creates:
BOTH/R BOTH BOTH/R BOTHER
A level of 4 causes a floating-point number to be appended to each of the level-3 lines, giving the ratio between the length of the root and the total length of all generated words including the root:
BOTH/R BOTH 2.500000 BOTH/R BOTHER 2.500000
A level of 5 causes multiple lines to be output, one for each generated word. If the generated word did not use any affixes, the line is just that word. If one or more affixes were used, the original root and the affixes actually used are printed, joined by a plus sign; then the generated word is printed:
BOTH BOTH+R BOTHER
Finally, the -D flag causes the affix tables from the dictionary file to be dumped to standard output.
Ispell is aware of the correct capitalizations of words in the dictionary and in your personal dictionary. As well as recognizing words that must be capitalized (e.g., George) and words that must be all-capitals (e.g., NASA), it can also handle words with "unusual" capitalization (e.g., "ITCorp" or "TeX"). If a word is capitalized incorrectly, the list of possibilities will include all acceptable capitalizations. (More than one capitalization may be acceptable; for example, my dictionary lists both "ITCorp" and "ITcorp".)
Normally, this feature will not cause you surprises, but there is one circumstance you need to be aware of. If you use "I" to add a word to your dictionary that is at the beginning of a sentence (e.g., the first word of this paragraph if "normally" were not in the dictionary), it will be marked as "capitalization required". A subsequent usage of this word without capitalization (e.g., the quoted word in the previous sentence) will be considered a misspelling by ispell, and it will suggest the capitalized version. You must then compare the actual spellings by eye, and then type "I" to add the uncapitalized variant to your personal dictionary. You can avoid this problem by using "U" to add the original word, rather than "I".
The rules for capitalization are as follows:
The buildhash program builds hashed dictionary files for later use by ispell. The raw word list (with affix flags) is given in dict-file, and the the affix flags are defined by affix-file. The hashed output is written to hash-file. The formats of the two input files are described in ispell(5). The -s (silent) option suppresses the usual status messages that are written to the standard error device.
The munchlist shell script is used to reduce the size of dictionary files, primarily personal dictionary files. It is also capable of combining dictionaries from various sources. The given files are read (standard input if no arguments are given), reduced to a minimal set of roots and affixes that will match the same list of words, and written to standard output.
Input for munchlist contains of raw words (e.g from your personal dictionary files) or root and affix combinations (probably generated in earlier munchlist runs). Each word or root/affix combination must be on a separate line.
The -D (debug) option leaves temporary files around under standard names instead of deleting them, so that the script can be debugged. Warning: on a multiuser system, this can be a security hole. To avoid possible destruction of important files, don't run the script as root, and set MUNCHDEBUGDIR to the name of a directory that only you can access.
The -v (verbose) option causes progress messages to be reported to stderr so you won't get nervous that munchlist has hung.
If the -s (strip) option is specified, words that are in the specified hash-file are removed from the word list. This can be useful with personal dictionaries.
The -l option can be used to specify an alternate affix-file for munching dictionaries in languages other than English.
The -c option can be used to convert dictionaries that were built with an older affix file, without risk of accidentally introducing unintended affix combinations into the dictionary.
The -T option allows dictionaries to be converted to a canonical string-character format. The suffix specified is looked up in the affix file (-l switch) to determine the string-character format used for the input file; the output always uses the canonical string-character format. For example, a dictionary collected from TeX source files might be converted to canonical format by specifying -T tex.
The findaffix shell script is an aid to writers of new language descriptions in choosing affixes. The given dictionary files (standard input if none are given) are examined for possible prefixes (-p switch) or suffixes (-s switch, the default). Each commonly-occurring affix is presented along with a count of the number of times it appears and an estimate of the number of bytes that would be saved in a dictionary hash file if it were added to the language table. Only affixes that generate legal roots (found in the original input) are listed.
If the "-c" option is not given, the output lines are in the following format:
where strip is the string that should be stripped from a root word before adding the affix, add is the affix to be added, count is a count of the number of times that this strip/add combination appears, and bytes is an estimate of the number of bytes that might be saved in the raw dictionary file if this combination is added to the affix file. The field separator in the output will be the tab character specified by the -t switch; the default is a slash ("/").
If the -c ("clean output") option is given, the appearance of the output is made visually cleaner (but harder to post-process) by changing it to:
where strip, add, count, and bytes are as before, and <tab> represents the ASCII tab character.
The method used to generate possible affixes will also generate longer affixes which have common headers or trailers. For example, the two words "moth" and "mother" will generate not only the obvious substitution "+er" but also "-h+her" and "-th+ther" (and possibly even longer ones, depending on the value of min). To prevent cluttering the output with such affixes, any affix pair that shares a common header (or, for prefixes, trailer) string longer than elim characters (default 1) will be suppressed. You may want to set "elim" to a value greater than 1 if your language has string characters; usually the need for this parameter will become obvious when you examine the output of your findaffix run.
Normally, the affixes are sorted according to the estimate of bytes saved. The -f switch may be used to cause the affixes to be sorted by frequency of appearance.
To save output file space, affixes which occur fewer than 10 times are eliminated; this limit may be changed with the -l switch. The -M switch specifies a maximum affix length (default 8). Affixes longer than this will not be reported. (This saves on temporary disk space and makes the script run faster.)
Affixes which generate stems shorter than 3 characters are suppressed. (A stem is the word after the strip string has been removed, and before the add string has been added.) This reduces both the running time and the size of the output file. This limit may be changed with the -m switch. The minimum stem length should only be set to 1 if you have a lot of free time and disk space (in the range of many days and hundreds of megabytes).
The findaffix script requires a non-blank field-separator character for internal use. Normally, this character is a slash ("/"), but if the slash appears as a character in the input word list, a different character can be specified with the -t switch.
The tryaffix shell script is used to estimate the effectiveness of a proposed prefix (-p switch) or suffix (-s switch, the default) with a given expanded-file. Only one affix can be tried with each execution of tryaffix, although multiple arguments can be used to describe varying forms of the same affix flag (e.g., the D flag for English can add either D or ED depending on whether a trailing E is already present). Each word in the expanded dictionary that ends (or begins) with the chosen suffix (or prefix) has that suffix (prefix) removed; the dictionary is then searched for root words that match the stripped word. Normally, all matching roots are written to standard output, but if the -c (count) flag is given, only a statistical summary of the results is written. The statistics given are a count of words the affix potentially applies to and an estimate of the number of dictionary bytes that a flag using the affix would save. The estimate will be high if the flag generates words that are currently generated by other affix flags (e.g., in English, bathers can be generated by either bath/X or bather/S).
The dictionary file, expanded-file, must already be expanded (using the -e switch of ispell) and sorted, and things will usually work best if uppercase has been folded to lower with 'tr'.
The affix arguments are things to be stripped from the dictionary file to produce trial roots: for English, con (prefix) and ing (suffix) are examples. The addition parts of the argument are letters that would have been stripped off the root before adding the affix. For example, in English the affix ing normally strips e for words ending in that letter (e.g., like becomes liking) so we might run:
tryaffix ing ing+e
to cover both cases.
All of the shell scripts contain documentation as commentary at the beginning; sometimes these comments contain useful information beyond the scope of this manual page.
When all options are enabled, ispell may take several seconds to generate all the guesses at corrections for a misspelled word; on slower machines this time is long enough to be annoying.
The hash table is stored as a quarter-megabyte (or larger) array, so a PDP-11 or 286 version does not seem likely.
Ispell should understand more troff syntax, and deal more intelligently with contractions.
Although small personal dictionaries are sorted before they are written out, the order of capitalizations of the same word is somewhat random.
When the -x flag is specified, ispell will unlink any existing .bak file.
There are too many flags, and many of them have non-mnemonic names.
The -e flag should accept mnemonic arguments instead of numeric ones.
Munchlist does not deal very gracefully with dictionaries which contain "non-word" characters. Such characters ought to be deleted from the dictionary with a warning message.
Findaffix and munchlist require tremendous amounts of temporary file space for large dictionaries. They do respect the TMPDIR environment variable, so this space can be redirected. However, a lot of the temporary space needed is for sorting, so TMPDIR is only a partial help on systems with an uncooperative sort(1). ("Cooperative" is defined as accepting the undocumented -T switch). At its peak usage, munchlist takes 10 to 40 times the original dictionary's size in Kb. (The larger ratio is for dictionaries that already have heavy affix use, such as the one distributed with ispell). Munchlist is also very slow; munching a normal-sized dictionary (15K roots, 45K expanded words) takes around an hour on a small workstation. (Most of this time is spent in sort(1), and munchlist can run much faster on machines that have a more modern sort that makes better use of the memory available to it.) Findaffix is even worse; the smallest English dictionary cannot be processed with this script in a mere 50Kb of free space, and even after specifying switches to reduce the temporary space required, the script will run for over 24 hours on a small workstation.
Collected, revised, and enhanced for the Usenet by Walt Buehring, 1987.
Table-driven multi-lingual version by Geoff Kuenning, 1987-88.
Large dictionaries provided by Bob Devine (vianet!devine).