Representing Myanmar in Unicode

Details and Examples

Martin Hosken1

Introduction
The first edition of this technical note addressed the issue of how Myanmar text was encoded using the
Unicode standard as it stood until version 5.1. With Unicode 5.1 various new characters were added to the
Myanmar block which had the effect of simplifying the encoding model considerably. Such a change could
only come about with agreement from all implementors and those with existing data because they will need
to update and change to the new model. This is nearly impossible to achieve if existing implementations are
already in widespread use, which was not the case at the time for the Myanmar block. In addition, such a
change was necessary to facilitate the encoding of minority scripts. So with a necessity and a unique
opportunity for change, the characters were added and the encoding model simplified.
This technical note describes the simplified model and keeps the older model description as a later section
for comparison. The information is structured to follow closely the previous edition of this technical note.
The author wishes to thank the Myanmar Language Commission, the Myanmar NLP Lab and the Myanmar
Computer Federation for reviewing and providing input to this version of the document.

1
SIL International and Payap University, Chiang Mai, THAILAND
Representing Myanmar in Unicode Page 1 of 15
Unicode 5.1 Model

Basic Myanmar
The basic consonants and vowels are relatively obvious in how they are encoded. Thus:

pm 1005 102C letter

Here we show the Myanmar word, the underlying Unicode codes that would be stored to represent this and
an English gloss of the word. As this example shows, characters are stored in the order in which they are
read.

cg 1001 102B to shake

odumQ 101E 102D 1000 1039 1001 102C dignity

o°Ýg 101E 1012 1039 1013 102B faith

In this example, we highlight the code of interest. Notice how the g (U+102B MYANMAR VOWEL SIGN TALL
AA) has a different code to the m (U+102C MYANMAR VOWEL SIGN AA). The Myanmar character underlying the
two codes is the same, and there are rendering rules that can give the correct form, so why has the tall -aa
been given its own code? The primary reason is that Sgaw Karen, among other minority scripts, only has the
tall form, and so a rendering system that works for the Myanmar language is not going to work for Sgaw
Karen and vice versa. A Myanmar language specific keyboarding implementation could choose to enforce a
particular variant of the -aa vowel in the context of certain consonants (in Burmese following c, }, i, °, y,
or &), medial combinations and syllable chainings, but this is not required.

nKd 100A 102D 102F brown

xk:H 1011 102F 1036 1038 to tie

Notice how the two forms of ¬k (U+102F MYANMAR VOWEL SIGN U) have the same code. It is up to the
rendering system to choose which form should be shown and different fonts can have different rules
depending on the designer’s preference.
U+1031 –e vowel
We will see later why the vowels are stored in this relative order. But for now it is important to note that the
Unicode standard states that vowels are stored after the consonant, according to how they are read,
regardless of where they are rendered. This introduces one of the complexities of implementing Myanmar
script:

ae 1014 1031 the sun

ayg 1015 1031 102B plentiful

The a vowel is rendered in front of the consonant that it is read (and so stored) following. Notice that this
says nothing about the relative order for typing, but it does mean that anyone implementing the Myanmar
script needs to take special care of this character. In general people are used to and want to type the a vowel
in front of the consonant, and so implementors need to address issues of keyboarding as well as rendering.
Medials
The medial characters have their own codes and are always stored after the base consonant and before any
vowels. Although the character j has traditionally been typed in non-Unicode fonts before the consonant, it
is consistent with normal spelling to store U+103C MYANMAR CONSONANT SIGN MEDIAL RA after the
consonant.

zsm: 1016 103B 102C 1038 fever

Representing Myanmar in Unicode Page 2 of 15

 aju: 1000 103C 1031 1038 grime

arG: 1019 103D 1031 1038 give birth

rë 1019 103E 102F regard important

Syllable Chaining
In the case of syllable chaining, subjoined characters are not given their own codes. Instead a virama
character is used to indicate that the following character is subjoined and should take a subjoined form.

ywWm 1015 1010 1039 1010 102C hinge

Devoweliser
There are two ways of representing the devowelising process. The first is by creating a medial or syllable
chained form, using U+1039 to mark the devowelising (as shown above). The second is to use the visible
virama character ¬f (U+103A MYANMAR SIGN ASAT) in conjunction with a base consonant.

xif 1011 1004 103A think

juOf 1000 103C 1009 103A avoid

aumf 1000 1031 102C 103A glue

The second example also illustrates that Of is encoded with U+1009 followed by U+103A even though the
glyph shape closely resembles the independent vowel O U+1025 MYANMAR LETTER U. Keyboard
implementors may wish to enforce this.
The third example is not a true devowelising, but it shows that U+103A can also be used as a tone mark in
combination with U+102B and U+102C.
Kinzi
The remaining issue regarding representation needed for the modern Myanmar language is how kinzi is
represented in Unicode. Glyph based encodings give the kinzi its own code. But linguistically, the kinzi is
merely a special form of a devowelised nga (U+1004 MYANMAR LETTER NGA). We encode kinzi as a
devowelised nga with the following letter underneath, subjoined. But the difference is that when rendered,
the devowelised nga changes shape and the subjoined base character remains a full character. Thus we use
U+1004 U+103A U+1039.

pîuà 1005 1004 103A 1039 1000 103C 1036 path

oabFm 101E 1004 103A 1039 1018 1031 102C ship

Like the –e vowel, kinzi is particularly problematic to implement since people want to type it following the
base consonant and it also needs careful handling during rendering.
Diacritic storage order
It is possible for a Myanmar syllable to have a number of diacritics surrounding a base consonant. Since all
these diacritics are not spacing, how do we know in which order they should be stored? For example, nKd can
be stored as U+100A U+102D U+102F or as U+100A U+102F U+102D. But what happens if one person stores it
one way and then someone searches for that word spelled the other way? It is important that there is a
consistent way of storing strings so that applications can work consistently.

Representing Myanmar in Unicode Page 3 of 15

The following list gives the relative order that each diacritic should be stored, if it occurs, following a base
consonant.
Name Specification Example
Consonant [U+1000 .. U+102A, U+103F, U+104E]2 u
Asat 3
U+103A ¬f
Stacked U+1039 [U+1000 .. U+1019, U+101C, U+101E, U+1020, U+1021] ¬N
Medial Y U+103B ¬s
Medial R U+103C ]¬
Medial W U+103D ¬G
Medial H U+103E ¬S
E vowel U+1031 a¬
Upper Vowel [U+102D, U+102E, U+1032] ¬D
Lower Vowel [U+102F, U+1030] ¬k
A Vowel [U+102B, U+102C] ¬m
Anusvara U+1036 ¬H
Visible virama U+103A ¬f
Lower Dot U+1037 ¬h
Visarga U+1038 ¬:

Notice the general order of: initial consonant cluster, vowels, tones.
For example:

yoûL: 1015 101E 103B 103E 1030 1038 Malay

]rÏm 1019 103C 103D 103E 102C segmentalize

101E 103B 103E 1031 102C 1004
aoûmif 103A top knot
Notice also that the diacritic storage order does not define a phonetic syllable. If Asat or Stacked are
present, then a syllable break occurs in the middle of the order, following them. In addition, A syllable
containing a devowelised consonant will follow the order twice, once for the main consonant and once for
the devowelised one.

Advanced Issues
So far we have covered what is explained in the Unicode Standard4. In this section we examine some of the
more difficult areas of the Myanmar language including some implementation details regarding line
breaking and sorting; further examination of the kinzi question; contractions and some issues with respect to
Old Myanmar.
Line breaking
Myanmar does not have interword spaces like English. Instead spaces are used to mark phrases. Some
phrases are relatively short (two or three syllables, 1.5em, or 2.3 times the width of U+1000 u) while others
can be quite long (8.5em or 13 times the width of U+1000 u). A common approach to addressing line
breaking issues is to adjust the phrase spacing so that a line breaks at a phrase break. If this approach fails
and a phrase must be continued onto a second line, U+200B ZERO WIDTH SPACE may be used to indicate a
possible line break point in the text.
2
Notice the extension of the list here to include independent vowels. The Unicode Standard V4.0 only lists values up to
U+102A. U+104E has changed glyph and can function in consonant position as in ¨if: (104E 1004 103A 1038)
3
Only for use with Kinzi and contractions
4
Version 5.1
Representing Myanmar in Unicode Page 4 of 15
The problem with this approach is that when phrases are quite long or a lot of text is to be typeset, the
manual adjustment of phrasing or the introduction of zero width spaces can be onerous. A further option is to
break lines automatically within phrases when needed. The clearest solution is to have a line break occurring
at a word boundary, but since there are no word breaks in Myanmar this is not immediately possible. Most
words, though, are mono-syllabic and so a mechanism of breaking lines at syllable boundaries is usually
sufficient. From this we can say that a syllable break may occur before a Myanmar digit, an independent
vowel, one of the various signs or a base consonant so long as the consonant:
• is not devowelised with an asat and
• has no stacked consonant below it and
• is not a kinzi.
These same syllable breaking rules are used for sorting purposes, with the addition of non-line breaking
syllable breaks, such as those occurring between the two characters in a syllable chain. For example these
phrases show possible inter-syllable line breaks.
1000 1031 102C 1004 103A | 101C 1031 1038 |
aumifav:awGausmif: 1010 103D 1031 | 1000 103B 1031 102C 1004 103A the kids are
uko
d mG :juw<f` 1038
1000
| 1000 102D 102F | 101E 103D 102C 1038 |
103C | 1010 101A 103A 104B
going to school

1021 102D 1015 103A | 1001 1014 103A 1038 | to the bedroom
f ef:wHcg:ukd
tdyc 1010 1036 | 1001 102B 1038 | 1000 102D 102F door
Notice how in the second example the word 1010 1036 | 1001 102B 1038 is a single word with multiple
syllables. Is there some way, without a dictionary, that we can ensure that the word is not line broken? There
is a Unicode character that was added for version 4.0: U+2060 WORD JOINER. Previous to this the character
U+FEFF ZERO WIDTH NON-BREAKING SPACE was used. Since U+FEFF is most commonly used at the start of a
Unicode text file to both identify it as being Unicode data and to indicate the encoding form of the data,
U+2060 was added to the standard to take over the function of zero width non-breaking space. The role of
this character is to indicate a non-breaking point in a text. Lines should not be broken at that point.
Therefore, if we want to ensure that no line-break occurs at the syllable boundary within our poly-syllabic
word, we can insert a U+2060 into our data stream between the two syllables and a rendering engine should
not break a line at that point. Thus:
1021 102D 1015 103A | 1001 1014 103A 1038 | to the bedroom
f ef:wHcg:ukd
tdyc 1010 1036 2060 1001 102B 1038 | 1000 102D 102F door
In summary, therefore, we propose three levels of line breaking support: breaking at phrase spaces; breaking
at syllable breaks and support for polysyllabic words. A rendering engine may choose the sophistication of
line breaking support it provides.
Sorting
Sorting Myanmar strings is a complex process involving significant string transformation and four levels of
comparison. The string transformation is a syllable based operation for which the identification of syllable
boundaries (but not word boundaries) are required. The same techniques that are used for line-breaking,
therefore, may be used for sorting.
Kinzi revisited
One of the significant improvements brought about by the addition of the asat character is that kinzi is now
unambiguously encoded. Thus:

ta&F 1021 1004 103A 1039 101D 1031

taiG 1021 1004 103D 1031

Contractions
The Myanmar language has a system of double acting consonants, where a consonant acts as both the final
of a syllable and the initial of a following syllable. These are significant for sorting purposes. Double acting
consonants are rare, but occur in two common words.

Representing Myanmar in Unicode Page 5 of 15

 101A 1031 102C 1000 103A 103B
a<musfm: 102C 1038 man, husband
1000 103B 103D 1014 103A 102F
uïóky
f f 1015 103A I (1st person singular)
Notice how the visible virama (103A) occurs immediately after the double acting consonant. This position is
not listed in the standard diacritic order, but is most appropriate for a double acting consonant. In order to
identify such contractions, we propose that the visible virama be stored immediately following the
consonant. This storage approach will also affect the syllable definition since a devowelised consonant with
a vowel acts like a normal base consonant with its preceding syllable break.
There are also words with double acting consonants which are unmarked. Since these are unmarked, it has
been decided that despite their etymology, these words should be sorted as if there were no double acting
consonant.

&gus 101D 102B 1000 103B sentence

}drmS ef 1002 102D 1019 103E 102C 1014 103A momentum

Old Myanmar
There are a few issues that storing old Myanmar text introduce, although again, most of these are resolved
due to the simplified encoding model.
Stacked Ya
There are occasions where a medial ya (U+103B) representation is used for a stacking ya. What is needed is a
syllable break between the base consonant and the ya. Thus we propose:

O<sme 1025 101A 200C 103B 102C 1014 O<smOf garden/orchard

The extra column gives the modern spelling of the word. The use of U+200C ZERO WIDTH NON-JOINER
indicates the break in the syllable. It makes no difference to rendering and is only used in Pali sorting.
LaSwe (Medial la)
In some words, a subscript la (U+101C) acts as a medial rather than as the start of a new syllable. This is
simply encoded using a virama:

uì 1000 1039 101C us drop

usìyf 1000 103B 1039 101C 1015 103A usyf tight

AMyint (Archaic tone mark)

Old Myanmar includes a medial form of U+1021 which causes no particular problems since it is just treated
as any other medial, and occurs very rarely.

ed<†f 1014 102D 101A 1039 1021 103A aeð day

Minority Language Extensions

With the addition of support for a number of minority languages which are based on the Myanmar script, it
is possible that some characters that look like presentation forms of sequences found in Myanmar will be
encoded. For example in Unicode 5.1 there is the character: \S (U+1061 MYANMAR LETTER SGAW KAREN SHA).
This looks like the sequence \ (U+101B MYANMAR LETTER RA) followed by ¬S (U+103E MYANMAR CONSONANT
SIGN MEDIAL HA). But the two are completely unrelated and the sequence must always be used in Myanmar
language and the unit in the Sgaw Karen language.

Searching and Comparison

The approach we have taken here to add markers to handle syllable breaking can introduce problems when
searching and during string comparison. Since the codes used do not make any change to rendering, it is

Representing Myanmar in Unicode Page 6 of 15

likely that in many cases the codes will be left out of a text. Therefore there is a need for searching and
comparison code to take into account and ignore the extra codes inserted.
The following codes should be ignored when searching and comparing: U+200B ZERO WIDTH SPACE and
U+2060 WORD JOINER. In addition, with the simplified encoding, U+200C ZERO WIDTH NON-JOINER may
also be ignored.

Representing Myanmar in Unicode Page 7 of 15

Pre Unicode 5.1 Encoding Model

Introduction
This section contains most of the text of the original edition of UTN#11 and was written in conjunction with
Maung Tuntunlwin5.

Basic Myanmar
The basic consonants and vowels are relatively obvious in how they are encoded. Thus:

pm 1005 102C letter

Here we show the Myanmar word, the underlying Unicode codes that would be stored to represent this and
an English gloss of the word. As this example shows, characters are stored in the order in which they are
read.

cg 1001 102C to shake

In this example, we highlight the code of interest. Notice how the g has the same code as the m (U+102C
MYANMAR VOWEL SIGN AA) and that it is up to the rendering system to decide which form of the character is
to be displayed. The same goes for diacritics. There is only one code for a particular character and it is up to
the rendering system to ensure that the diacritic is appropriately placed.

nKd 100A 102F 102D brown

xk:H 1011 102F 1036 1038 to tie

Here the two forms of ¬k (U+102F MYANMAR VOWEL SIGN U) are decided by the rendering system.
U+1031 –e vowel
We will see later why the vowels are stored in this relative order. But for now it is important to note that the
Unicode standard states that vowels are stored after the consonant, according to how they are read,
regardless of where they are rendered. This introduces one of the complexities of implementing Myanmar
script:

ae 1014 1031 the sun

ayg 1014 1031 102C plentiful

The a vowel is rendered in front of the consonant that it is read (and so stored) following. Notice that this
says nothing about the relative order for typing, but it does mean that anyone implementing the Myanmar
script needs to take special care of this character. In general people are used to and want to type the a vowel
in front of the consonant, and so implementors need to address issues of keyboarding as well as rendering.
Medials and Syllable Chaining
So much for what is clearly visible on the Unicode chart. What about all those glyphs that are not there?
How are words including medials or involve syllable chaining stored?

ywWm 1015 1010 1039 1010 102C hinge

zsm: 1016 1039 101A 102C 1038 fever

aju: 1000 1039 101B 1031 1038 grime

arG: 1019 1039 101D 1031 1038 give birth

rë 1019 1039 101F 102F regard important

In the linguistic model, a medial is formed by devowelising the inherent vowel of the preceding consonant.
Likewise, for a syllable chained letter, the inherent vowel at the end of the previous syllable is devowelised.

5
Myanmar World Distribution
Representing Myanmar in Unicode Page 8 of 15
In Unicode this devowelising process is marked using the virama code (U+1039 MYANMAR SIGN VIRAMA).
Thus we store a consonant followed by the virama and then follow it with the consonant of interest.
Devoweliser
There are two ways of representing the devowelising process. The first is by creating a medial or syllable
chained form, using U+1039 to mark the devowelising. The second is to use the visible virama character (¬f)
in conjunction with a base consonant. But if U+1039 is being used to mark medials and syllable chaining,
how is the visible character to be represented? The Unicode standard gives the answer. The sequence U+1039
MYANMAR SIGN VIRAMA followed by U+200C ZERO WIDTH NON-JOINER is used to represent a visual virama
(¬f).6

xif 1011 1004 1039 200C I think

Kinzi
The remaining issue regarding representation needed for the modern Myanmar language is how kinzi is
represented in Unicode. Glyph based encodings give the kinzi its own code. But linguistically, the kinzi is
merely a special form of a devowelised nga (U+1004 MYANMAR LETTER NGA). Thus we encode kinzi as
U+1004 U+1039.

pîuà 1005 1004 1039 1000 1039 101B 1036 path

Like the –e vowel, kinzi is particularly problematic to implement since people want to type it following the
base consonant and it also needs careful handling during rendering.
Diacritic storage order
It is possible for a Myanmar syllable to have a number of diacritics surrounding a base consonant. Since all
these diacritics are not spacing, how do we know in which order they should be stored? For example, nKd can
be stored as U+100A U+102D U+102F or as U+100A U+102F U+102D. But what happens if one person stores it
one way and then someone searches for that word spelled the other way? It is important that there is a
consistent way of storing strings so that applications can work consistently.
The following list gives the relative order that each diacritic should be stored, if it occurs, following a base
consonant.
Name Specification Example
kinzi U+1004 U+1039 ¬F
Consonant [U+1000 .. U+102A] 7
u
Stacked U+1039 [U+1000 .. U+1019, U+101C, U+101E, U+1020, U+1021] ¬N
Medial Y U+1039 U+101A ¬s
Medial R U+1039 U+101B ]¬
Medial W U+1039 U+101D ¬G
Medial H U+1039 U+101F ¬S
E vowel U+1031 a¬
Lower Vowel [U+102F, U+1030] ¬k
Upper Vowel [U+102D, U+102E, U+1032] ¬D
A Vowel U+102C ¬m
Anusvara U+1036 ¬H
Visible virama U+1039 U+200C ¬f
Lower Dot U+1037 ¬h
Visarga U+1038 ¬:

6
For fallback purposes, U+1039 also displays a ¬f if not followed by a consonant. This is an implementation detail and is not
used in spelling words. I.e. all such occurrences should be considered wrong spellings.
7
Notice the addition to the list of independent vowels. The Unicode Standard v4.0 only lists values up to U+1021.
Representing Myanmar in Unicode Page 9 of 15
Notice the general order of: initial consonant cluster, vowels, tones.
For example:
1015 101E 1039 101A 1039 101F
yoûL: 1030 1038 Malay
1019 1039 101B 1039 101D 1039
]rÏm 101F 102C segmentalize
101E 1039 101A 1039 101F 1031
aoûmif 102C 1004 1039 200C top knot

Advanced Issues
So far we have covered what is explained in the Unicode Standard8. In this section we examine some of the
more difficult areas of the Myanmar language including some implementation details regarding line
breaking and sorting; further examination of the kinzi question; contractions and some issues with respect to
Old Myanmar.
Line breaking
Myanmar does not have interword spaces like English. Instead spaces are used to mark phrases. Some
phrases are relatively short (two or three syllables, 1.5em, or 2.3 times the width of U+1000 u) while others
can be quite long (8.5em or 13 times the width of U+1000 u). A common approach to addressing line
breaking issues is to adjust the phrase spacing so that a line breaks at a phrase break. If this approach fails
and a phrase must be continued onto a second line, U+200B ZERO WIDTH SPACE may be used to indicate a
possible line break point in the text.
The problem with this approach is that when phrases are quite long or a lot of text is to be typeset, the
manual adjustment of phrasing or the introduction of zero width spaces can be onerous. A further option is to
break lines automatically within phrases when needed. The clearest solution is to have a line break occurring
at a word boundary, but since there are no word breaks in Myanmar this is not immediately possible. Most
words, though, are mono-syllabic and so a mechanism of breaking lines at syllable boundaries is usually
sufficient. From this we can say that a syllable break may occur before a base consonant so long as the
consonant:
• is not devowelised with a visible virama and
• has no stacked consonant below it (ignoring true medials: –y –r –w –h) and
• is not a kinzi.
These same syllable breaking rules are used for sorting purposes, with the addition of non-line breaking
syllable breaks, such as those occuring between the two characters in a syllable chain. For example these
phrases show possible inter-syllable line breaks.
1000 1031 102C 1004 1039 200C | 101C 1031 1038
aumifav:awGausmif: | 1010 1039 101D 1031 | 1000 1039 101A 1031
the kids are
102C 1004 1039 200C 1038 | 101E 102F 102D 1037
oko
h d mG :juonf` | 101E 1039 101D 102C 1038 | 1000 1039 101B | going to school
101E 100A 1039 200C 104B
1021 102D 1015 1039 200C | 1001 1014 1039 200C
to the bedroom
f ef:wHcg:ukd
tdyc 1038 | 1010 1036 | 1001 102C 1038 | 1000 102F
door
102D
Notice how in the second example the word 1010 1036 | 1001 102C 1038 is a single word with multiple
syllables. Is there some way without a dictionary, that the we can ensure that the word is not line broken?
There is a Unicode character that was added for version 4.0: U+2060 WORD JOINER. Previous to this the
character U+FEFF ZERO WIDTH NON-BREAKING SPACE was used. Since U+FEFF is most commonly used at the
start of a Unicode text file to both identify it as being Unicode data and to indicate the encoding form of the
data, U+2060 was added to the standard to take over the function of zero width non-breaking space. The role
of this character is to indicate a non-breaking point in a text. Lines should not be broken at that point.

8
Version 4.0, 2003
Representing Myanmar in Unicode Page 10 of 15
Therefore, if we want to ensure that no line-break occurs at the syllable boundary within our poly-syllabic
word, we can insert a U+2060 into our data stream between the two syllables and a rendering engine should
not break a line at that point. Thus:
1021 102D 1015 1039 200C | 1001 1014 1039 200C
to the bedroom
f ef:wHcg:ukd
tdyc 1038 | 1010 1036 2060 1001 102C 1038 | 1000
door
102F 102D
In summary, therefore, we propose three levels of line breaking support: breaking at phrase spaces; breaking
at syllable breaks and support for polysyllabic words. A rendering engine may choose the sophistication of
line breaking support it provides.
Sorting
Sorting Myanmar strings is a complex process involving, significant string transformation and four levels of
comparison. The string transformation is a syllable based operation for which the identification of syllable
boundaries (but not word boundaries) are required. The same techniques that are used for line-breaking,
therefore, may be used for sorting.
Kinzi revisited
Consider the word taiG. How can it be represented? The normal encoding we would expect would be 1021
1004 1039 101D 1031. But there are two ways of interpreting this string:

ta&F 1021 (1004 1039) 101D 1031

taiG 1021 1004 (1039 101D) 1031

The question is how different systems will interpret the string. One approach is to say that kinzi above a
consonant is rare and that kinzi above one of the for medial consonants (U+101A U+101B U+101D U+101F) is
very rare, and so we can say that for the sequence U+1004 U+1039 U+10xx we interpret as (U+1004
U+1039) U+10xx if U+10xx is a normal non medial consonant and that we interpret the sequence as U+1004
(U+1039 U+10xx) if U+10xx is one of the medial consonants.

But 1004 1039 represents a kinzi:

ta&F 1021 1004 1039 101D 1031

taif& 1021 1004 1039 200C 101D 1031

But neither of these representations are the word we want, so how can we represent this word? The main
issue is whether we interpret 1021 1004 1039 101D 1031 as 1021 (1004 1039) 101D 1031 or as 1021
1004 (1039 101D) 1031?

There are different approaches we can take, the approach we propose here is to introduce either a U+200C
ZERO WIDTH NON-JOINER or a U+200D ZERO WIDTH JOINER between the U+1004 and the following U+1039.
The effect of this is to break the kinzi string and to make the U+101D look more like a medial than a main
consonant. The sub-sequence 1004 200C 1039 101D results in the rendering we want. But it also marks that
there is a syllable break between the U+1004 and the U+101D, while we want the syllable break to occur
before the U+1004. So a better solution is: 1004 200D 1039 101D. Here the ZERO WIDTH JOINER indicates
that the syllable should be held together and therefore that there is a break before the syllable. Notice that all
this talk of syllable breaks makes no difference for rendering (even line breaking). It is only needed for
sorting purposes.
If we consider all the sequences that can occur, we can see what the rendering will be and where the syllable
break will occur when sorting.

ta&Fh 1021 1004 1039 101D 1031 1037 ✓ (good)

taiðG 1021 1004 200C | 1039 101D 1031 1037 ✗ (bad)

taiGð 1021 | 1004 200D 1039 101D 1031 1037 ✓ (good)

tifa&Ð 1021 1004 1039 200C | 101D 1031 1037 ✓ (good)

Representing Myanmar in Unicode Page 11 of 15

 ta&Fh 1021 | 1004 1039 200D 101D 1031 1037 ✗ (bad)
The final column lists whether the syllable breaking is appropriate. Notice that there is no sequence that
explicitly marks a kinzi (only the final sequence does this), and that also is appropriate for indicating that the
kinzi is associated with the previous syllable. Therefore we have to make the unmarked initial sequence
always represent a kinzi (with the syllable break following the kinzi).
The principles presented here form the basis for the solutions for other encoding problems in Myanmar.
Contractions
The Myanmar language has a system of double acting consonants, where a consonant acts as both the final
of a syllable and the initial of a following syllable. These are significant for sorting purposes. Double acting
consonants are rare, but occur in two common words.
101A 1031 102C 1000 1039 200C
a<musfm: 1039 101A 102C 1038 man, husband
1000 1039 101A 1039 101D 1014
uïóky
f f 1039 200C 102F 1015 1039 200C I (1st person singular)
Notice how the visible virama (1039 200C) occurs immediately after the double acting consonant. This
position is not listed in the standard diacritic order, but is most appropriate for a double acting consonant. In
order to identify such contractions, we propose that the visible virama be stored immediately following the
consonant. This storage approach will also affect the syllable definition since a devowelised consonant with
a vowel acts like a normal base consonant with its preceding syllable break.
There are also words with double acting consonants which are unmarked. Since these are unmarked, it has
been decided that despite their etymology, these words should be sorted as if there were no double acting
consonant.

uvsmE 1000 101C 1039 101A 100F womanly virtues

1021 101D 1039 101F 1014 1039
t&Sef 200C momentum

Old Myanmar
There are a few issues that storing old Myanmar text introduce.
Stacked Ya
There are occasions where a medial ya (U+101A) representation is used for a stacking ya. What is needed is a
syllable break between the base consonant and the ya. Thus we propose:

O<sme 1025 101A 200C 1039 101A 102C

1014
O<smOf garden/orchard
The extra column gives the modern spelling of the word. The use of U+200C ZERO WIDTH NON-JOINER
indicates the break in the syllable. It makes no difference to rendering and is only used in Pali sorting.
LaSwe (Medial la)
In some words, a subscript la (U+101C) acts as a medial rather than as the start of a new syllable. We propose
this change of sorting be marked using U+200D ZERO WIDTH JOINER between the base consonant and the
medial:

uì 1000 200D 1039 101C us drop

usìyf 1000 1039 101A 200D 1039 101C

1015 1039 200C
usyf tight
In the second example, while the U+200D is not necessary (since from context it is possible to identify that
the U+101C is a medial), it is included for consistency, making implementation easier.
AMyint (Medial A)
Old Myanmar includes a medial form of U+1021 which causes no particular problems since it is just treated
as any other medial, and occurs very rarely.

Representing Myanmar in Unicode Page 12 of 15

 ed<†f 1014 102D 101A 1039 1021 1039
200C
aeð day

Searching and Comparison

The approach we have taken here to add markers to handle syllable breaking can introduce problems when
searching and during string comparison. Since the codes used do not make any change to rendering, it is
likely that in many cases the codes will be left out of a text. Therefore there is a need for searching and
comparison code to take into account and ignore the extra codes inserted.
The following codes should be ignored when searching and comparing: U+200B ZERO WIDTH SPACE and
U+2060 WORD JOINER.

U+200C ZERO WIDTH NON-JOINER and U+200D ZERO WIDTH JOINER are more problematic. At the strictest
level of comparison, they should be included but since in many cases they are only used to control syllable
breaking for sorting a helpful approach to searching and comparison would only consider these character
when they affect rendering. U+200C affects rendering only when it follows U+1039 MYANMAR SIGN VIRAMA.
U+200D only affects rendering when it occurs in the sequence U+1004 U+200D U+1039.Conclusion

With the change to the Myanmar encoding model comes a much greater simplicity while not changing the
original character of the model which is both linguistic and practical. The model may come as a surprise to
those who are used to a glyph based encoding in which each glyph shape and position receives its own code,
or more radically each cluster receives its own code.

Representing Myanmar in Unicode Page 13 of 15

References
Bechert, et al 1979, Burmese Manuscripts, Part 1 Wiesbaden.
Department of the Myanmar Language Commission 1993, Myanmar – English Dictionary Ministry of
Education, Union of Myanmar.
Okell, John 1994, Burmese: An Introduction to the Script SOAS, London.
The Unicode Consortium 2003, The Unicode Standard, Version 4.0 Addison-Wesley, Massachusetts.

Representing Myanmar in Unicode Page 14 of 15

Appendix
This is a subset of the complete chart of the Myanmar block in the Unicode standard. The other letters are
not used.

100 101 102 103 104

0 u w V ¬l 0
1 c x t a¬ 1
2 } ' ¬J 2
3 C " Ã 3
4 i e T 4
5 p y O 5
6 q z OD ¬H 6
7 Z A { ¬Ð 7
8 ç b ¬: 8
9 Ù r jo ¬ 9
A n < ajomf ¬f ?
B õ \ ¬g ¬s `
C X v ¬m j¬ Y
D ø & ¬d ¬G I
E Í o ¬D ¬S ၎
F E [ ¬k ù >

Representing Myanmar in Unicode Page 15 of 15