Network Working Group                               Ken Whistler, Sybase
Internet Draft                                     Glenn Adams, Spyglass
                                         <draft-whistler-plane14-01.txt>


                 Language Tagging in Unicode Plain Text

                            February 15, 1998



                          Status of this Memo

This document is an Internet-Draft.  Internet-Drafts are working
documents of the Internet Engineering Task Force (IETF), its areas, and
its working groups. Note that other groups may also distribute working
documents as Internet- Drafts.

Internet-Drafts are draft documents valid for a maximum of six months.
Internet-Drafts may be updated, replaced, or obsoleted by other
documents at any time.  It is not appropriate to use Internet-Drafts as
reference material or to cite them other than as a "working draft" or
"work in progress".

To learn the current status of any Internet-Draft, please check the
1id-abstracts.txt listing contained in the Internet-Drafts Shadow
Directories on ds.internic.net (US East Coast), nic.nordu.net (Europe),
ftp.isi.edu (US West Coast), or munnari.oz.au (Pacific Rim).

1.    Abstract

This document proposed a mechanism for language tagging in [UNICODE]
plain text. A set of special-use tag characters on Plane 14 of
[ISO10646] (accessible through UTF-8, UTF-16, and UCS-4 encoding forms)
are proposed for encoding to enable the spelling out of ASCII-based
string tags using characters which can be strictly separated from
ordinary text content characters in ISO10646 (or UNICODE).

One tag identification character and one cancel tag character are also
proposed. In particular, a language tag identification character is
proposed to identify a language tag string specifically; the language
tag itself makes use of [RFC1766] language tag strings spelled out
using the Plane 14 tag characters. Provision of a specific,
low-overhead mechanism for embedding language tags in plain text is
aimed at meeting the need of Internet Protocols such as ACAP, which
require a standard mechanism for marking language in UTF-8 strings.

The tagging mechanism as well the characters proposed in this document
have been approved by the Unicode Consortium for inclusion in The
Unicode Standard.  However, implementation of this decision awaits
formal acceptance by ISO JTC1/SC2/WG2, the working group responsible
for ISO10646. Potential implementers should be aware that until this
formal acceptance occurs, any usage of the characters proposed herein
is strictly experimental and not sanctioned for standardized character
data interchange.

2.    Definitions and Notation

No attempt is made to define all terms used in this document. In
particular, the terminology pertaining to the subject of coded
character systems is not explicitly specified. See [UNICODE],
[ISO10646], and [RFC2130] for additional definitions in this area.

2.1   Requirements Notation

This document occasionally uses terms that appear in capital letters.
When the terms "MUST", "SHOULD", "MUST NOT", "SHOULD NOT", and "MAY"
appear capitalized, they are being used to indicate particular
requirements of this specification. A discussion of the meanings of
these terms appears in [RFC2119].

2.2   Definitions

The terms defined below are used in special senses and thus warrant
some clarification.

2.2.1 Tagging

The association of attributes of text with a point or range of the
primary text. (The value of a particular tag is not generally
considered to be a part of the "content" of the text. Typical examples
of tagging is to mark language or font of a portion of text.)

2.2.2 Annotation

The association of secondary textual content with a point or range of
the primary text. (The value of a particular annotation *is* considered
to be a part of the "content" of the text. Typical examples include
glossing, citations, exemplication, Japanese yomi, etc.)

2.2.3 Out-of-band

An out-of-band channel conveys a tag in such a way that the textual
content, as encoded, is completely untouched and unmodified. This is
typically done by metadata or hyperstructure of some sort.

2.2.4 In-band

An in-band channel conveys a tag along with the textual content, using
the same basic encoding mechanism as the text itself. This is done by
various means, but an obvious example is SGML markup, where the tags
are encoded in the same character set as the text and are interspersed
with and carried along with the text data.

3.0   Background

There has been much discussion over the last 8 years of
language tagging and of other kinds of tagging of Unicode plain
text. It is fair to say that there is more-or-less universal
agreement that language tagging of Unicode plain text is
required for certain textual processes. For example, language
"hinting" of multilingual text is necessary for multilingual
spell-checking based on multiple dictionaries to work well.
Language tagging provides a minimum level of required
information for text-to-speech processes to work correctly.
Language tagging is regularly done on web pages, to enable
selection of alternate content, for example.

However, there has been a great deal of controversy regarding
the appropriate placement of language tags. Some have
held that the only appropriate placement of language tags
(or other kinds of tags) is out-of-band, making use of
attributed text structures or metadata. Others have argued
that there are requirements for lower-complexity in-band
mechanisms for language tags (or other tags) in plain text.

The controversy has been muddied by the existence and widespread
use of a number of in-band text markup mechanisms (HTML,
text/enriched, etc.) which enable language tagging, but
which imply the use of general parsing mechanisms which
are deemed too "heavyweight" for protocol developers and
a number of other applications. The difficulty of using
general in-band text markup for simple protocols derives
from the fact that some characters are used both for textual
content and for the text markup; this makes it more difficult
to write simple, fast algorithms to find only the textual
content and ignore the tags, or vice versa. (Think of this
as the algorithmic equivalent of the difficulty the human
reader has attempting to read just the content of raw
HTML source text without a browser interpreting all the
markup tags.)

The Plane 14 proposal addresses the recurrent and persistent
call for a lighter-weight mechanism for text tagging than
typical text markup mechanisms in Unicode. It proposes a special set
of characters used *only* for tagging. These tag characters
can be embedded into plain text and can be identified and/or
ignored with trivial algorithms, since there is no overloading
of usage for these tag characters--they can only express
tag values and never textual content itself.

The Plane 14 proposal is not intended for general annotation
of text, such as textual citations, phonetic readings (e.g.
Japanese Yomi), etc. In its present form, its use is intended
to be restriced solely to specifying in-line language tags.
Future extensions may widen this scope of intended usage.

4.0   Proposal

This proposal suggests the use of 97 dedicated tag characters
encoded at the start of Plane 14 of ISO/IEC 10646 consisting of
a clone of the 94 printable 7-bit ASCII graphic characters and
ASCII SPACE, as well as a tag identification character and a tag
cancel character.

These tag characters are to be used to spell out any ASCII-
based tagging scheme which needs to be embedded in Unicode
plain text. In particular, they can be used to spell out
language tags in order to meet the expressed requirements
of the ACAP protocol and the likely requirements of other
new protocols following the guidelines of the IAB character
workshop (RFC 2130).

The suggested range in Plane 14 for the block reserved for
tag characters is as follows, expressed in each of the
three most generally used encoding schemes for ISO/IEC
10646:

UCS-4

U-000E0000 .. U-000E007F

UTF-16

U+DB40 U+DC00 .. U+DB40 U+DC7F

UTF-8

0xF3 0xA0 0x80 0x80 .. 0xF3 0xA0 0x81 0xBF

Of this range, U-000E0020 .. U-000E007E is the
suggested range for the ASCII clone tag characters themselves.


4.1   Names for the Tag Characters

The names for the ASCII clone tag characters should be exactly
the ISO 10646 names for 7-bit ASCII, prefixed with the word
"TAG".

In addition, there is one tag identification character
and a CANCEL TAG character. The use and syntax of these characters
is described in detail below.

The entire encoding for the proposed Plane 14 tag characters and
names of those characters can be derived from the following list.
(The encoded values here and throughout this proposal are listed
in UCS-4 form, which is easiest to interpret. It is assumed that
most Unicode applications will, however, be making use either
of UTF-16 or UTF-8 encoding forms for actual implementation.)

U-000E0000  <reserved>
U-000E0001  LANGUAGE TAG
U-000E0002  <reserved>
....
U-000E001F  <reserved>
U-000E0020  TAG SPACE
U-000E0021  TAG EXCLAMATION MARK
....
U-000E0041  TAG LATIN CAPITAL LETTER A
....
U-000E007A  TAG LATIN SMALL LETTER Z
....
U-000E007E  TAG TILDE
U-000E007F  CANCEL TAG

4.2   Range Checking for Tag Characters

The range checks required for code testing for tag characters
would be as follows. The same range check is expressed here
in C for each of the three significant encoding forms for 10646.

Range check expressed in UCS-4:

     if ( ( *s >= 0xE0000 ) || ( *s <= 0xE007F ) )

Range check expressed in UTF-16 (Unicode):

    if ( ( *s == 0xDB40 ) && ( *(s+1) >= 0xDC00 ) && ( *(s+1) <= 0xDC7F ) )

Expressed in UTF-8:

    if ( ( *s == 0xF3 ) && ( *(s+1) == 0xA0 ) && ( *(s+2) & 0xE0 == 0x80 )

Because of the choice of the range for the tag characters, it would also
be possible to express the range check for UCS-4 or UTF-16 in terms of
bitmask operations, as well.

4.3   Syntax for Embedding Tags

The use of the Plane 14 tag characters is very simple. In order
to embed any ASCII-derived tag in Unicode plain text, the tag
is simply spelled out with the tag characters instead, prefixed
with the relevant tag identification character. The
resultant string is embedded directly in the text.

The tag identification character is used as a mechanism for
identifying tags of different types. This enables multiple
types of tags to coexist amicably embedded in plain text and
solves the problem of delimitation if a tag is concatenated
directly onto another tag. Although only one type of tag is
currently specified, namely the language tag, the encoding
of other tag identification characters in the future would
allow for distinct tag types to be used.

No termination character is required for a tag. A tag terminates
either when the first non Plane 14 Tag Character (i.e. any
other normal Unicode value) is encountered, or when the next
tag identification character is encountered.

All tag arguments must be encoded only with the tag characters
U-000E0020 .. U-000E007E. No other characters are valid for
expressing the tag argument.

A detailed BNF syntax for tags is listed below.

4.4   Tag Scope and Nesting

The value of an established tag continues from the point the
tag is embedded in text until either:

   A. The text itself goes out of scope, as defined by the
      application. (E.g. for line-oriented protocols, when
      reaching the end-of-line or end-of-string; for text
      streams, when reaching the end-of-stream; etc.)

or

   B. The tag is explicitly cancelled by the CANCEL TAG
      character.

Tags of the same type cannot be nested in any way. The appearance
of a new embedded language tag, for example, after text which
was already language tagged, simply changes the tagged value for
subsequent text to that specified in the new tag.

Tags of different type can have interdigitating scope, but
not hierarchical scope. In effect,
tags of different type completely ignore each other, so that
the use of language tags can be completely asynchronous with the
use of character set source tags (or any other tag type) in the
same text in the future.

4.5   Cancelling Tag Values

U-000E007F CANCEL TAG is provided to allow the specific cancelling
of a tag value. The use of CANCEL TAG has the following syntax.
To cancel a tag value of a particular type, prefix the CANCEL
TAG character with the tag identification character of the
appropriate type. For example, the complete string to cancel
a language tag is:

U-000E0001 U-000E007F

The value of the relevant tag type returns to the default state
for that tag type, namely: no tag value specified, the same as
untagged text.

The use of CANCEL TAG without a prefixed tag identification
character cancels *any* Plane 14 tag values which may be
defined. Since only language tags are currently provided with
an explicit tag identification character, only language tags
are currently affected.

The main function of CANCEL TAG is to make possible such
operations as blind concatenation of strings in a tagged context
without the propagation of inappropriate tag values across the
string boundaries. For example, a string tagged with a Japanese
language tag can have its tag value "sealed off" with a terminating
CANCEL TAG before another string of unknown language value is
concatenated to it. This would prevent the string of unknown
language from being erroneously marked as being Japanese simply
because of a concatenation to a Japanese string.

4.6   Tag Syntax Description

An extended BNF (Backus-Naur Form) description of the tags specified
in this proposal is found below.  Note the following BNF extensions
used in this formalism:

1. Semantic constraints are specified by rules in the form of an
   assertion specified between double braces; the variable $$ denotes
   the string consisting of all terminal symbols matched by the
   this non-terminal.

   Example:   {{ Assert ( $$[0] == '?' ); }}

   Meaning:   The first character of the string matched by this
              non-terminal must be '?'

2. A number of predicate functions are employed in semantic constraint
   rules which are not otherwise defined; their name is sufficient for
   determining their predication.

   Example:   IsRFC1766LanguageIdentifier ( tag-argument )

   Meaning:   tag-argument is a valid RFC1766 language identifier

3. A lexical expander function, TAG, is employed to denote the tag
   form of an ASCII character; the argument to this function is either
   a character or a character set specified by a range or enumeration
   expression.

   Example:   TAG('-')

   Meaning:   TAG HYPHEN-MINUS

   Example:   TAG([A-Z])

   Meaning:   TAG LATIN CAPITAL LETTER A ...
              TAG LATIN CAPITAL LETTER Z

4. A macro is employed to denote terminal symbols that are character
   literals which can't be directly represented in ASCII. The argument
   to the macro is the UNICODE (ISO/IEC 10646) character name.

   Example:   '${TAG CANCEL}'

   Meaning:   character literal whose code value is U-000E007F

5. Occurrence indicators used are '+' (one or more) and '*' (zero
   or more); optional occurrence is indicated by enclosure in '['
   and ']'.

4.6.1 Formal Tag Syntax

tag                     :   language-tag
                        |   cancel-all-tag
                        ;

language-tag            :   language-tag-introducer language-tag-argument
                        ;

language-tag-argument   :   tag-argument
              {{ Assert ( IsRFC1766LanguageIdentifier ( $$ ); }}
                        |   tag-cancel
                        ;

cancel-all-tag          :   tag-cancel
                        ;

tag-argument            :   tag-character+
                        ;

tag-character           :   { c : c in
              TAG( { a : a in printable ASCII characters or SPACE } ) }
                        ;

language-tag-introducer :   '${TAG LANGUAGE}'
                        ;

tag-cancel              :   '${TAG CANCEL}'
                        ;


5.0   Tag Types

5.1   Language Tags

Language tags are of general interest and should have a high
degree of interoperability for protocol usage. To this end, a
specific LANGUAGE TAG tag identification character is provided.
A Plane 14 tag string prefixed by U-000E0001 LANGUAGE TAG is
specified to constitute a language tag. Furthermore, the tag values
for the language tag are to be spelled out as specified in RFC
1766, making use only of registered tag values or of user-defined
language tags starting with the characters "x-".

For example, to embed a language tag for Japanese, the Plane 14
characters would be used as follows. The Japanese tag from RFC 1766
is "ja" (composed of ISO 639 language id) or, alternatively,
"ja-JP" (composed of ISO 639 language id plus ISO 3166 country id).
Since RFC 1766 specifies that language tags are not case significant,
it is recommended that for language tags, the entire tag be
lowercased before conversion to Plane 14 tag characters. (This
would not be required for Unicode conformance, but should be followed
as general practice by protocols making use of RFC 1766 language tags,
to simplify and speed up the processing for operations which need to
identify or ignore language tags embedded in text.) Lowercasing,
rather than uppercasing, is recommended because it follows the majority
practice of expressing language tag values in lowercase letters.

Thus the entire language tag (in its longer form) would be converted
to Plane 14 tag characters as follows:

U-000E0001 U-000E006A U-000E0061 U-000E002D U-000E006A U-000E0070

The language tag (in its shorter, "ja" form) could be expressed
as follows:

U-000E0001 U-000E006A U-000E0061

The value of this string is then expressed in whichever encoding
form (UCS-4, UTF-16, UTF-8) is required and embedded in text at
the relevant point.

5.2   Additional Tags

Additional tag identification characters might be defined in the
future. An example would be a CHARACTER SET SOURCE TAG, or a
GENERIC TAG for private definition of tags.

In each case, when a specific tag identification character is encoded,
a corresponding reference standard for the values of the tags associated
with the identifier should be designated, so that interoperating
parties which make use of the tags will know how to interpret the
values the tags may take.

6.0   Display Issues

All characters in the tag character block are considered to have
no visible rendering in normal text. A process which interprets
tags may choose to modify the rendering of text based on the tag
values (as for example, changing font to preferred style for
rendering Chinese versus Japanese). The tag characters
themselves have no display; they may be considered similar to
a U+200B ZERO WIDTH SPACE in that regard. The tag characters also
do not affect breaking, joining, or any other format or layout
properties, except insofar as the process interpreting the
tag chooses to impose such behavior based on the tag value.

For debugging or other operations which must render the tags
themselves visible, it is advisable that the tag characters be
rendered using the corresponding ASCII character glyphs (perhaps
modified systematically to differentiate them from normal ASCII
characters). But, as noted below, the tag character values are
chosen so that even without display support, the tag characters
will be interpretable in most debuggers.

8.0   Unicode Conformance Issues

The basic rules for Unicode conformance for the tag characters are
exactly the same as for any other Unicode characters. A conformant
process is not required to interpret the tag characters. If it does
not interpret tag characters, it should leave their values undisturbed
and do whatever it does with any other uninterpreted characters. If
it does interpret them, it should interpret them according to the
standard, i.e. as spelled-out tags.

So for a non-TagAware Unicode application, any language tag characters
(or any other kind of tag expressed with Plane 14 tag characters)
encountered would be handled exactly as for uninterpreted Tibetan
from the BMP, uninterpreted Linear B from Plane 1, or uninterpreted
Egyptian hieroglyphics from private use space in Plane 15.

A TagAware but TagPhobic Unicode application can recognize the tag
character range in Plane 14 and choose to deliberately strip them
out completely to produce plain text with no tags.

The presence of a correctly formed tag cannot be taken as a
guarantee that the data so tagged is correctly tagged. For example,
nothing prevents an application from erroneously labelling French
data as Spanish, or from labelling JIS-derived data as Japanese, even
if it contains Greek or Cyrillic characters.

8.1   Note on Encoding Language Tags

The fact that this proposal for encoding tag characters in
Unicode includes a mechanism for specifying language tag values
does not mean that Unicode is departing from one of its
basic encoding principles:

    Unicode encodes scripts, not languages.

This is still true of the Unicode encoding (and ISO/IEC 10646), even
in the presence of a mechanism for specifying language tags
in plain text.  There is nothing obligatory about the use of Plane 14
tags, whether for language tags or any other kind of tags.

Language tagging in no way impacts current encoded characters
or the encoding of future scripts.

It is fully anticipated that implementations of Unicode which
already make use of out-of-band mechanisms for language tagging
or "heavy-weight" in-band mechanisms such as HTML will continue
to do exactly what they are doing and will ignore Plane 14
tag characters completely.

9.0   Security Considerations

Security issues are not discussed in this memo.

************************************************************************

References

[ISO10646]

    ISO/IEC 10646-1:1993 International Organization for Standardization.
    "Information Technology -- Universal Multiple-Octet Coded Character
    Set (UCS) -- Part 1: Architecture and Basic Multilingual Plane",
    Geneva, 1993.

[RFC1766]

    Alvestrand, H., "Tags for the Identification of Languages", RFC 1766.

[RFC2070]

    F. Yergeau, G. Nicol, G. Adams, and M. Duerst, "Internationalization
    of the Hypertext Markup Language", RFC 2070, January 1997.

[RFC2119]

    S. Bradner, "Key words for use in RFCs to Indicate Requirement Levels",
    RFC 2119, March 1997.

[RFC2130]

    C. Weider, C. Preston, K. Simonsen, H. Alvestrand, R. Atkinson,
    M. Crispin, and P. Svanberg, "The Report of the IAB Character Set
    Workshop held 29 February - 1 March, 1996", RFC 2130, April 1997.

[UNICODE]

    The Unicode Standard, Version 2.0, The Unicode Consortium,
    Addison-Wesley, July 1996.

Acknowledgements

The following people also contributed to this document, directly or
indirectly: Chris Newman, Mark Crispin, Rick McGowan, Joe Becker,
John Jenkins, and Asmus Freytag. This document also was reviewed by
the Unicode Technical Committee, and the authors wish to thank all
of the UTC representatives for their input. The authors are, of course,
responsible for any errors or omissions which may remain in the text.

Authors' Addresses

Ken Whistler
Sybase, Inc.
6475 Christie Ave.
Emeryville, CA 94608-1050
Phone: +1 510 922 3611
Email: kenw@sybase.com

Glenn Adams
Spyglass, Inc.
One Cambridge Center
Cambridge, MA 02142
Phone: +1 617 679 4652
Email: glenn@spyglass.com