.\"
.\" $Id: ispell.4,v 1.30 1995/08/05 23:19:39 geoff Exp $
.\"
.\" Copyright 1992, 1993, Geoff Kuenning, Granada Hills, CA
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\"
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\" 3. All modifications to the source code must be clearly marked as
.\"    such.  Binary redistributions based on modified source code
.\"    must be clearly marked as modified versions in the documentation
.\"    and/or other materials provided with the distribution.
.\" 4. All advertising materials mentioning features or use of this software
.\"    must display the following acknowledgment:
.\"      This product includes software developed by Geoff Kuenning and
.\"      other unpaid contributors.
.\" 5. The name of Geoff Kuenning may not be used to endorse or promote
.\"    products derived from this software without specific prior
.\"    written permission.
.\"
.\" THIS SOFTWARE IS PROVIDED BY GEOFF KUENNING AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED.  IN NO EVENT SHALL GEOFF KUENNING OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $Log: ispell.4,v $
.\" Revision 1.30  1995/08/05  23:19:39  geoff
.\" Fix a place where a line was eaten because it was seen as an nroff
.\" command.
.\"
.\" Revision 1.29  1995/01/08  23:23:45  geoff
.\" Fix a tiny typo.
.\"
.\" Revision 1.28  1994/11/02  06:56:07  geoff
.\" Remove the anyword feature, which I've decided is a bad idea.
.\"
.\" Revision 1.27  1994/10/26  05:12:31  geoff
.\" Document the new compound-word options for German and Scandinavian
.\" languages, and the always-OK flag for French.
.\"
.\" Revision 1.26  1994/05/25  04:29:19  geoff
.\" Document the new restriction that boundary characters must appear
.\" singly.
.\"
.\" Revision 1.25  1994/01/25  07:11:42  geoff
.\" Get rid of all old RCS log lines in preparation for the 3.1 release.
.\"
.\"
.TH ISPELL 4 local
.SH NAME
ispell \- format of ispell dictionaries and affix files
.SH DESCRIPTION
.PP
.IR Ispell (1)
requires two files to define the language that it is spell-checking.
The first file is a dictionary containing words for the language,
and the second is an "affix" file that defines the meaning of special
flags in the dictionary.
The two files are combined by
.I buildhash
(see
.IR ispell "(1))"
and written to a hash file which is not described here.
.PP
A raw
.I ispell
dictionary (either the main dictionary or your own personal
dictionary) contains a list of words, one per line.
Each word may optionally be followed by a slash ("/") and one or more
flags, which modify the root word as explained below.
Depending on the options with which
.I ispell
was built, case may or
may not be significant in either the root word or the flags, independently.
Specifically, if the compile-time option CAPITALIZATION is defined, case
is significant in the root word;
if not, case is ignored in the root word.
If the compile-time option MASKBITS is set to a value of 32, case is ignored
in the flags;
otherwise case is significant in the flags.
Contact your system administrator or
.I ispell
maintainer for more information (or use the
.B \-vv
flag to find out).
The dictionary should be sorted with the
.B \-f
flag of
.IR sort (1)
before the hash file is built;
this is done automatically by
.IR munchlist (1),
which is the normal way of building dictionaries.
.PP
If the dictionary contains words that have string characters (see the
affix-file documentation below), they must be written in the format
given by the
.B defstringtype
statement in the affix file.
This will be the case for most non-English languages.
Be careful to use this format, rather than that of your favorite
formatter, when adding words to a dictionary.  (If you add words to
your personal dictionary during an
.I ispell
session, they will automatically be converted to the correct format.
This feature can be used to convert an entire dictionary if necessary:)
.PP
.RS
.nf
	echo qqqqq > dummy.dict
	buildhash dummy.dict \fIaffix-file\fP dummy.hash
	awk '{print "*"}END{print "#"}' \fIold-dict-file\fP \e
	| ispell -a -T \fIold-dict-string-type\fP \e
	  -d ./dummy.hash -p ./\fInew-dict-file\fP \e
	  > /dev/null
	rm dummy.*
.fi
.RE
.PP
The case of the root word controls the case of words accepted by
.IR ispell ,
as follows:
.IP (1)
If the root word appears only in lower case (e.g., 
.IR bob "),"
it will be accepted in lower case, capitalized, or all capitals.
.IP (2)
If the root word appears capitalized (e.g.,
.IR Robert "),"
it will not
be accepted in
all-lower case, but will be accepted capitalized or all in capitals.
.IP (3)
If the root word appears all in capitals (e.g.,
.IR UNIX "),"
it will only be accepted all in capitals.
.IP (4)
If the root word appears with a "funny" capitalization (e.g.,
.IR ITCorp "),"
a word will be accepted only if it follows that capitalization, or if
it appears all in capitals.
.IP (5)
More than one capitalization of a root word may appear in the dictionary.
Flags from different capitalizations are combined by OR-ing them together.
.PP
Redundant capitalizations (e.g.,
.I bob
and
.IR Bob ")"
will be combined
by
.I buildhash
and by
.I ispell
(for personal dictionaries),
and can be removed from a raw dictionary by
.IR munchlist .
.PP
For example, the dictionary:
.PP
.RS
.nf
bob
Robert
UNIX
ITcorp
ITCorp
.fi
.RE
.PP
will accept
.IR bob ,
.IR Bob ,
.IR BOB ,
.IR Robert ,
.IR ROBERT ,
.IR UNIX ,
.IR ITcorp ,
.IR ITCorp ,
and
.IR ITCORP ,
and will reject all others.
Some of the unacceptable forms are
.IR bOb ,
.IR robert ,
.IR Unix ,
and
.IR ItCorp .
.PP
As mentioned above, root words in any dictionary may be extended by flags.
Each flag is a single alphabetic character, which represents a prefix or
suffix
that may be added to the root to form a new word.
For example, in an English dictionary
the
.B D
flag can be added to
.I bathe
to make
.IR bathed .
Since flags are represented as a single bit in the hashed dictionary, this
results in significant space savings.
The
.I munchlist
script will reduce an existing raw dictionary by adding flags when possible.
.PP
When a word is extended with an affix, the affix will be accepted only
if it appears in the same case
as the initial (prefix) or final (suffix) letter of the word.
Thus, for example, the entry
.I UNIX/M
in the main dictionary
.RB "(" M
means
add an apostrophe and an "s" to make a possessive) would accept
.I "UNIX'S"
but would reject
.IR "UNIX's" .
If
.I "UNIX's"
is legal, it must appear as a separate dictionary entry,
and it will not be combined by
.IR munchlist .
(In general, you don't need to worry about these things;
.I munchlist
guarantees that its output dictionary will accept the same set of
words as its input, so all you have to do is add words to the dictionary
and occasionally run munchlist to reduce its size).
.PP
As mentioned, the affix definition file describes the affixes associated
with particular flags.
It also describes the character set used by the language.
.PP
Although the affix-definition
grammar is designed for a line-oriented layout, it is actually
a free-format yacc grammar and can be laid out weirdly if you want.
Comments are started by a pound (sharp) sign (#),
and continue to the end of the line.
Backslashes are supported in the usual fashion (\fB\e\fInnn\fR, plus
specials
.BR \en ,
.BR \er ,
.BR \et ,
.BR \ev ,
.BR \ef ,
.BR \eb ,
and the new hex format \fB\ex\fInn\fR).
Any character
with special meaning to the parser can be changed to an uninterpreted
token by backslashing it;
for example, you can declare a flag named 'asterisk' or 'colon' with
.I "flag \e*:"
or
.IR "flag \e::" .
.PP
The grammar will be presented in a top-down fashion, with discussion
of each element.
An affix-definition file must contain exactly one table:
.PP
.RS
.nf
\fItable\fR	:	[\fIheaders\fR] [\fIprefixes\fR] [\fIsuffixes\fR]
.fi
.RE
.PP
At least one of
.I prefixes
and
.I suffixes
is required.
They can appear in either order.
.PP
.RS
.nf
\fIheaders\fR	:	[ \fIoptions\fR ] \fIchar-sets\fR
.fi
.RE
.PP
The headers describe options global to this dictionary and language.
These include the character sets to be used and the formatter, and
the defaults for certain
.I ispell
flags.
.PP
.RS
.nf
\fIoptions\fR :	{ \fIfmtr-stmt\fR | \fIopt-stmt\fR | \fIflag-stmt\fR | \fInum-stmt\fR }
.fi
.RE
.PP
The options statements define the defaults for certain ispell flags
and for the character sets used by the formatters.
.PP
.RS
.nf
\fIfmtr-stmt\fR	:	{ \fInroff-stmt\fR | \fItex-stmt\fR }
.fi
.RE
.PP
A
.I fmtr-stmt
describes characters that have special meaning to a formatter.
Normally, this statement is not necessary, but some languages may have
preempted the usual defaults for use as language-specific characters.
In this case, these statements may be used to redefine the special characters
expected by the formatter.
.PP
.RS
.nf
\fInroff-stmt\fR	:	{ \fBnroffchars\fR | \fBtroffchars\fR } \fIstring\fR
.fi
.RE
.PP
The
.B nroffchars
statement allows redefinition of certain
.I nroff
control characters.
The string given must be exactly five characters long, and must list
substitutions for the left and right parentheses ("()") , the period ("."),
the backslash ("\e"), and the asterisk ("*").
(The right parenthesis is not currently used, but is included for
completeness.)
For example, the statement:
.PP
.RS
.nf
\fBnroffchars\fR {}.\e\e*
.fi
.RE
.PP
would replace the left and right parentheses with left and right curly
braces for purposes of parsing
.IR nroff / troff
strings, with no effect on the others (admittedly a contrived example).
Note that the backslash is escaped with a backslash.
.PP
.RS
.nf
\fItex-stmt\fR	:	{ \fBTeXchars\fR | \fBtexchars\fR } \fIstring\fR
.fi
.RE
.PP
The
.B TeXchars
statement allows redefinition of certain TeX/LaTeX control characters.
The string given must be exactly thirteen characters long, and must list
substitutions for the left and right parentheses ("()") , the left
and right square brackets ("[]"), the left and right curly braces ("{}"),
the left and right angle brackets ("<>"),
the backslash ("\e"), the dollar sign ("$"), the asterisk ("*"),
the period or dot ("."), and the percent sign ("%").
For example, the statement:
.PP
.RS
.nf
\fBtexchars\fR ()\e[\|]<\e><\e>\e\e$*.%
.fi
.RE
.PP
would replace the functions of the left and right curly braces with the
left and right angle brackets for purposes of parsing TeX/LaTeX constructs,
while retaining their functions for the
.I tib
bibliographic preprocessor.
Note that the backslash, the left square bracket, and the right angle bracket
must be escaped with a backslash.
.PP
.RS
.nf
\fIopt-stmt\fR	:	{ \fIcmpnd-stmt\fR | \fIaff-stmt\fR }
.sp
\fIcmpnd-stmt\fR	:	\fBcompoundwords\fR \fIcompound-opt\fR
.sp
\fIaff-stmt\fR		:	\fBallaffixes\fR \fIon-or-off\fR
.sp
\fIon-or-off\fR	:	{ \fBon\fR | \fBoff\fR }
.sp
\fIcompound-opt\fR :	{ \fIon-or-off\fR | \fBcontrolled\fR \fIcharacter\fR }
.fi
.RE
.PP
An
.I opt-stmt
controls certain ispell defaults that are best made language-specific.
The
.B allaffixes
statement controls the default for the
.B \-P
and
.B \-m
options to
.I ispell.
If
.B allaffixes
is turned
.B off
(the default),
.I ispell
will default to the behavior of the
.I \-P
flag:
root/affix suggestions will only be made if there are no "near misses".
If
.B allaffixes
is turned
.BR on ,
.I ispell
will default to the behavior of the
.I \-m
flag:
root/affix suggestions will always be made.
The
.B compoundwords
statement controls the default for the
.B \-B
and
.B \-C
options to
.I ispell.
If
.B compoundwords
is turned
.B off
(the default),
.I ispell
will default to the behavior of the
.I \-B
flag:
run-together words will be reported as errors.
If
.B compoundwords
is turned
.BR on ,
.I ispell
will default to the behavior of the
.I \-C
flag:
run-together words will be considered as compounds if both are in
the dictionary.
This is useful for languages such as German and Norwegian, which
form large numbers of compound words.
Finally, if
.B compoundwords
is set to
.IR controlled ,
only words marked with the flag indicated by
.I character
(which should not be otherwise used)
will be allowed to participate in compound formation.
Because this option requires the flags to be specified in the dictionary,
it is not available from the command line.
.PP
.RS
.nf
\fIflag-stmt\fR	:	\fBflagmarker\fR \fIcharacter\fR
.fi
.RE
.PP
The
.B flagmarker
statement describes the character which is used to separate affix
flags from the root word in a raw dictionary file.
This must be a
character which is not found in any word (including in string characters;
see below).
The default is "/" because this character is not normally
used to represent special characters in any language.
.PP
.RS
.nf
\fInum-stmt\fR	:	\fBcompoundmin\fR \fIdigit\fR
.fi
.RE
.PP
The
.B compoundmin
statement controls the length of the two components of a compound
word.
This only has an effect if
.B compoundwords
is turned
.B on
or if the
.B \-C
flag is given to
.IR ispell .
In that case, only words at least as long as the given minimum will be
accepted as components of a compound.
The default is 3 characters.
.PP
.RS
.nf
\fIchar-sets\fR	:	\fInorm-sets\fR [ \fIalt-sets\fR ]
.fi
.RE
.PP
The character-set section describes the characters that can be part of
a word, and defines their collating order.
There must always be a definition of "normal" character sets;  in
addition, there may be one or more partial definitions of "alternate"
sets which are used with various text formatters.
.PP
.RS
.nf
\fInorm-sets\fR :	[ \fIdeftype\fR ] charset-group
.fi
.RE
.PP
A "normal" character set may optionally begin with a
definition of the file suffixes that make use of this set.
Following this are one or more character-set declarations.
.PP
.RS
.nf
\fIdeftype\fR :	\fBdefstringtype\fR \fIname\fR \fIdeformatter\fR \fIsuffix\fR*
.fi
.RE
.PP
The
.B defstringtype
declaration gives a list of file suffixes which should make use of the
default string characters defined as part of the base character set;
it is only necessary if string characters are being defined.
The
.I name
parameter
is a string giving the unique name associated with these suffixes;
often it is a formatter name.
If the formatter is a member of the troff family, "nroff" should be
used for the name associated with the most popular macro package;
members of the TeX family should use "tex".
Other names may be chosen freely, but they should be kept simple,
as they are used in
.I ispell 's
.B \-T
switch to specify a formatter type.
The
.I deformatter
parameter
specifies the deformatting style to use when processing files with the
given suffixes.
Currently, this must be either
.B tex
or
.BR nroff .
The
.I suffix
parameters are a whitespace-separated list of strings which, if
present at the end of a filename, indicate that the associated set of
string characters should be used by default for this file.  For
example, the suffix list for the troff family typically includes
suffixes such as ".ms", ".me", ".mm", etc.
.PP
.RS
.nf
\fIcharset-group\fR :	{ \fIchar-stmt\fR | \fIstring-stmt\fR | \fIdup-stmt\fR}*
.fi
.RE
.PP
A
.I char-stmt
describes single characters;
a
.I string-stmt
describes characters that must appear together as a string, and which
usually represent a single character in the target language.
Either may
also describe conversion between upper and lower case.
A
.I dup-stmt
is used to describe alternate forms of string characters, so that a
single dictionary may be used with several formatting
programs that use different conventions for representing non-ASCII
characters.
.PP
.RS
.nf
\fIchar-stmt\fR	:	\fBwordchars\fR \fIcharacter-range\fR
		|	\fBwordchars\fR \fIlowercase-range\fR \fIuppercase-range\fR
		|	\fBboundarychars\fR \fIcharacter-range\fR
		|	\fBboundarychars\fR \fIlowercase-range\fR \fIuppercase-range\fR
\fIstring-stmt\fR	:	\fBstringchar\fR \fIstring\fR
		|	\fBstringchar\fR \fIlowercase-string\fR \fIuppercase-string\fR
.fi
.RE
.PP
Characters described with the
.B boundarychars
statement are considered
part of a word only if they appear singly,
embedded between characters declared with the
.B wordchars
or
.B stringchar
statements.
For example, if the hyphen is a boundary character (useful in French),
the string "foo-bar" would be a single word, but "-foo" would be the
same as "foo", and "foo--bar" would be two words separated by non-word
characters.
.PP
If two ranges or strings are given in a
.I char-stmt
or
.IR string-stmt ,
the first describes
characters that are interpreted as lowercase and the second describes
uppercase.
In the case of a
.B stringchar
statement, the two strings must be of the same length.
Also, in a
.B stringchar
statement, the actual strings may contain
both uppercase and characters themselves without difficulty;
for instance, the statement
.PP
.RS
.nf
stringchar	"\e\e*(sS"	"\e\e*(Ss"
.fi
.RE
.PP
is legal and will not interfere with (or be interfered with by) other
declarations of of "s" and "S" as lower and upper case, respectively.
.PP
A final note on string characters:
some languages collate certain special characters as if they were strings.
For example, the German "a-umlaut"
is traditionally sorted as if it were "ae".
Ispell is not capable of this;
each character must be treated as an individual entity.
So in certain cases,
ispell will sort a list of words into a different order than the standard
"dictionary" order for the target language.
.PP
.RS
.nf
\fIalt-sets\fR	:	\fIalttype\fR [ \fIalt-stmt\fR* ]
.fi
.RE
.PP
Because different formatters use different notations to represent
non-ASCII characters,
.I ispell
must be aware of the representations used by these formatters.
These are declared as alternate sets of string characters.
.PP
.RS
.nf
\fIalttype\fR	:	\fBaltstringtype\fR \fIname\fR \fIsuffix\fR*
.fi
.RE
.PP
The
.B altstringtype
statement introduces each set by declaring the associated formatter
name and filename suffix list.
This name and list are interpreted exactly as in the
.B defstringtype
statement above.
Following this header are one or more \fIalt-stmt\fRs which declare
the alternate string characters used by this formatter.
.PP
.RS
.nf
\fIalt-stmt\fR		:	\fBaltstringchar\fR \fIalt-string\fR \fIstd-string\fR
.fi
.RE
.PP
The
.I altstringchar
statement describes alternate representations for string
characters.
For example, the \-mm macro package of
.I troff
represents the German "a-umlaut" as 
.IR a\e*: ,
while
.I TeX
uses the sequence \fI\e"a\fR.
If the
.I troff
versions are declared as the standard versions using
.BR stringchar ,
the
.I TeX
versions may be declared as alternates by using the statement
.PP
.RS
.nf
altstringchar	\e\e\e"a	a\e\e*\:
.fi
.RE
.PP
When the
.B altstringchar
statement is used to specify alternate forms,
all forms for a particular formatter must be declared together as a group.
Also, each formatter or macro package
must provide a complete set of characters, both
upper- and lower-case, and the character sequences used for each formatter
must be completely distinct.
Character sequences which describe upper- and lower-case versions of
the same printable character must also be the same length.
It may be necessary to define some new macros for a given formatter to
satisfy these restrictions.
(The current version of
.I buildhash
does not enforce these restrictions, but failure to obey them may
result in errors being introduced into files that are processed with
.IR ispell .)
.PP
An important minor point is that
.I ispell
assumes that all characters declared as
.B wordchars
or
.B boundarychars
will occupy exactly
one position on the terminal screen.
.PP
A single character-set statement can declare either a single character
or a contiguous range of characters.
A range is given as in egrep and the shell:
[a-z] means lowercase alphabetics;
[^a-z] means all but lowercase, etc.
All character-set statements are combined (unioned) to produce
the final list of characters that may be part of a word.
The collating order of the characters is defined by the order of their
declaration;
if a range is used, the characters are considered to have been declared
in ASCII order.
Characters that have case are collated next to each other, with the
uppercase character first.
.PP
The
character-declaration statements have a rather strange behavior caused by its
need to match each lowercase character with its uppercase equivalent.
In any given
.B wordchars
or
.B boundarychars
statement, the characters in each range are
first sorted into ASCII collating sequence, then matched one-for-one
with the other range.
(The two ranges must have the same number of characters).
Thus, for example, the two statements:
.PP
.RS
.nf
\fBwordchars\fP [aeiou] [AEIOU]
\fBwordchars\fP [aeiou] [UOIEA]
.fi
.RE
.PP
would produce exactly the same effect.
To get the vowels to match
up "wrong", you would have to use separate statements:
.PP
.RS
.nf
\fBwordchars\fP a U
\fBwordchars\fP e O
\fBwordchars\fP i I
\fBwordchars\fP o E
\fBwordchars\fP u A
.fi
.RE
.PP
which would cause uppercase 'e' to be 'O', and lowercase 'O' to be 'e'.
This should normally be a problem only with languages which have been
forced to use a strange ASCII collating sequence.
If your uppercase and lowercase letters both collate in the same order,
you shouldn't have to worry about this "feature".
.PP
The prefixes and suffixes sections have exactly the same syntax, except
for the introductory keyword.
.PP
.RS
.nf
\fIprefixes\fR	:	\fBprefixes\fI flagdef\fR*
\fIsuffixes\fR	:	\fBsuffixes\fI flagdef\fR*
\fIflagdef\fR	:	\fBflag\fR [\fB*\fR|\fB~\fR] \fIchar\fB : \fIrepl\fR*
.fi
.RE
.PP
A prefix or suffix table consists of an introductory keyword and a list
of flag definitions.
Flags can be defined more than once, in which case
the definitions are combined.
Each flag controls one or more
.IR repl s
(replacements)
which are conditionally applied to the beginnings or endings of various
words.
.PP
Flags are named by a single character
.IR char .
Depending on a configuration option,
this character can be either any uppercase letter (the default
configuration) or any 7-bit ASCII character.
Most languages should be
able to get along with just 26 flags.
.PP
A flag character may be prefixed with one or more option characters.
(If you wish to use one of the option characters as a flag character,
simply enclose it in double quotes.)
.PP
The asterisk (\fB*\fP) option
means that this
flag participates in
.I cross-product
formation.
This only matters if the
file contains both prefix and suffix tables.
If so, all prefixes and
suffixes marked with an asterisk will be applied in all cross-combinations
to the root word.
For example, consider the root
.I fix
with prefixes
.I pre
and
.IR in ,
and suffixes
.I es
and
.IR ed .
If all flags controlling these prefixes and suffixes are marked with an
asterisk, then the single root
.I fix
would also generate
.IR prefix ,
.IR prefixes ,
.IR prefixed ,
.IR infix ,
.IR infixes ,
.IR infixed ,
.IR fix ,
.IR fixes ,
and
.IR fixed .
Cross-product formation can produce a large number of words quickly, some
of which may be illegal, so watch out.
If cross-products produce illegal
words,
.I munchlist
will not produce those flag combinations, and the flag will not be useful.
.PP
.RS
.nf
\fIrepl\fR	:	\fIcondition\fR* \fB>\fR [ \fB- \fIstrip-string \fB,\fR ] \fIappend-string\fR
.fi
.RE
.PP
The \fB~\fR option specifies that the associated flag is only active
when a compound word is being formed.
This is useful in a language like German, where the form of a word
sometimes changes inside a compound.
.PP
A
.I repl
is a conditional rule for modifying a root word.
Up to 8
.I conditions
may be specified.
If the
.I conditions
are satisfied, the
rules on the right-hand side of the
.I repl
are applied, as follows:
.IP (1)
If a strip-string is given, it is first stripped from
the beginning or ending (as appropriate) of the root word.
.IP (2)
Then the append-string is added at that point.
.PP
For example, the
.I condition
.B .
means "any word", and the
.I condition
.B Y
means "any word ending in Y".
The following (suffix) replacements:
.PP
.RS
.nf
\&.	>	MENT
Y	>	-Y,IES
.fi
.RE
.PP
would change
.I induce
to
.I inducement
and
.I fly
to
.IR flies .
(If they were controlled by the same flag, they would also change
.I fly
to
.IR flyment ,
which might not be what was wanted.
.I Munchlist
can be used to protect against this sort of problem;
see the command sequence given below.)
.PP
No matter how much you might wish it, the strings on the right must be
strings of specific characters, not ranges.
The reasons are rooted deeply in the way
.I ispell
works, and it would be difficult or impossible to provide
for more flexibility.
For example, you might wish to write:
.PP
.RS
.nf
[EY]	>	-[EY],IES
.fi
.RE
.PP
This will not work.
Instead, you must use two separate rules:
.PP
.RS
.nf
E	>	-E,IES
Y	>	-Y,IES
.fi
.RE
.PP
The application of
.IR repl s
can be restricted to certain words with
.IR conditions :
.PP
.RS
.nf
\fIcondition\fR	:	{ \fB.\fR | \fIcharacter\fR | \fIrange\fR }
.fi
.RE
.PP
A
.I condition
is a restriction on the characters that adjoin, and/or are
replaced by, the right-hand side of the
.IR repl .
Up to 8
.I conditions
may be given, which should be enough context for anyone.
The right-hand side will be applied only if the
.I conditions
in the
.I repl
are satisfied.
The
.I conditions
also implicitly define a length;
roots shorter than the number of
.I conditions
will not pass the test.
(As a special case, a
.I condition
of a single dot "." defines a length of zero,
so that the rule applies to all words indiscriminately).
This length is independent of the separate test that insists that
all flags produce an output word length of at least four.
.PP
.I
Conditions
that are single characters should be separated by white space.
For example, to specify words ending in "ED", write:
.PP
.RS
.nf
E D	>	-ED,ING		# As in covered > covering
.fi
.RE
.PP
If you write:
.PP
.RS
.nf
ED	>	-ED,ING
.fi
.RE
.PP
the effect will be the same as:
.PP
.RS
.nf
[ED]	>	-ED,ING
.fi
.RE
.PP
As a final minor, but important point, it is sometimes useful to rebuild
a dictionary file using an incompatible suffix file.
For example,
suppose you expanded the "R" flag to generate "er" and "ers" (thus
making the Z flag somewhat obsolete).
To build a new dictionary
.I newdict
that, using
.IR newaffixes ,
will accept exactly the same list of
words as the old list
.I olddict
did using
.IR oldaffixes ,
the
.B \-c
switch of
.I munchlist
is useful, as in the following example:
.PP
.RS
.nf
$ munchlist -c oldaffixes -l newaffixes olddict > newdict
.fi
.RE
.PP
If you use this procedure, your new dictionary will always accept the
same list the original did, even if you badly screwed up the affix
file.
This is because
.I munchlist
compares the words generated by a flag with the original word list, and
refuses to use any flags that generate illegal words.
(But don't forget that the
.I munchlist
step takes a long time and eats up temporary file space).
.SH EXAMPLES
.PP
As an example of conditional suffixes, here is the specification of the
.B S
flag from the English affix file:
.PP
.RS
.nf
flag *S:
    [^AEIOU]Y	>	-Y,IES	# As in imply > implies
    [AEIOU]Y	>	S		# As in convey > conveys
    [SXZH]	>	ES		# As in fix > fixes
    [^SXZHY]	>	S		# As in bat > bats
.fi
.RE
.PP
The first line applies to words ending in Y, but not in vowel-Y.
The second takes care of the vowel-Y words.
The third then handles those words that end in a sibilant
or near-sibilant, and the last picks up everything else.
.PP
Note that the
.I conditions
are written very carefully so that they apply
to disjoint sets of words.
In particular, note that the fourth line
excludes words ending in Y as well as the obvious SXZH.
Otherwise, it would convert "imply" into "implys".
.PP
Although the English affix file does not do so, you can also have a flag
generate more than one variation on a root word.
For example, we could extend the English "R" flag as follows:
.PP
.RS
.nf
flag *R:
   E			>	R		# As in skate > skater
   E			>	RS		# As in skate > skaters
   [^AEIOU]Y	>	-Y,IER	# As in multiply > multiplier
   [^AEIOU]Y	>	-Y,IERS	# As in multiply > multipliers
   [AEIOU]Y	>	ER		# As in convey > conveyer
   [AEIOU]Y	>	ERS		# As in convey > conveyers
   [^EY]		>	ER		# As in build > builder
   [^EY]		>	ERS		# As in build > builders
.fi
.RE
.PP
This flag would generate both "skater" and "skaters" from "skate".
This capability can be very useful in languages that make use of noun, verb,
and adjective endings.
For instance, one could define a single flag
that generated all of the German "weak" verb endings.
.SH "SEE ALSO"
ispell(1)