The Unicode Bidirectional Algorithm (UBA), formally defined in Unicode Standard Annex #9 (UAX #9), is a specification developed by the Unicode Consortium that determines how text containing a mixture of left-to-right and right-to-left scripts is displayed. It is a normative part of the Unicode Standard and is required for conformance wherever characters from right-to-left scripts such as Arabic or Hebrew are rendered.
Background
Most writing systems display text from left to right, but several scripts—including Arabic, Hebrew, Thaana, and Syriac—are written from right to left. When text from both directions appears in the same document, the result is known as bidirectional text (or bidi text). Without a clear specification, ambiguities arise in determining the correct display order of characters.
The Unicode Standard prescribes a logical order for storing characters in memory, regardless of their visual direction. The UBA translates this logical order into a correct visual display order.
Directional Formatting Characters
The UBA defines several categories of special control characters used to influence text direction:
Implicit Directional Marks
Lightweight, zero-width characters that act as directional anchors without affecting display:
Abbreviation
Code Point
Name
LRM
U+200E
LEFT-TO-RIGHT MARK
RLM
U+200F
RIGHT-TO-LEFT MARK
ALM
U+061C
ARABIC LETTER MARK
Explicit Directional Embeddings
Signal that a piece of text is to be treated as embedded in a given direction:
Abbreviation
Code Point
Name
LRE
U+202A
LEFT-TO-RIGHT EMBEDDING
RLE
U+202B
RIGHT-TO-LEFT EMBEDDING
Explicit Directional Overrides
Force characters to be treated as strongly directional, overriding their implicit types:
Abbreviation
Code Point
Name
LRO
U+202D
LEFT-TO-RIGHT OVERRIDE
RLO
U+202E
RIGHT-TO-LEFT OVERRIDE
Explicit Directional Isolates
Introduced in Unicode 6.3, isolates prevent the enclosed text from affecting the surrounding text's ordering:
Abbreviation
Code Point
Name
LRI
U+2066
LEFT-TO-RIGHT ISOLATE
RLI
U+2067
RIGHT-TO-LEFT ISOLATE
FSI
U+2068
FIRST STRONG ISOLATE
PDI
U+2069
POP DIRECTIONAL ISOLATE
Terminating Characters
Abbreviation
Code Point
Name
Terminates
PDF
U+202C
POP DIRECTIONAL FORMATTING
LRE, RLE, LRO, RLO
PDI
U+2069
POP DIRECTIONAL ISOLATE
LRI, RLI, FSI
The Algorithm
The UBA processes text in four main phases:
1. Paragraph Separation
Text is split into paragraphs at paragraph separator characters (type B). Each paragraph is processed independently.
2. Initialization
Each character is assigned a bidirectional character type (e.g., L, R, AL, EN, AN) from the Unicode Character Database. An embedding level list is also initialized.
3. Resolving Embedding Levels
A series of rules resolves the embedding level of each character:
P1–P3: Determine the paragraph embedding level (0 for LTR, 1 for RTL).
X1–X10: Assign explicit embedding levels based on directional formatting characters.
W1–W7: Resolve weak types (e.g., European numbers, separators).
N0–N2: Resolve neutral and isolate formatting types, including bracket pairs.
I1–I2: Resolve implicit embedding levels.
The maximum embedding depth is 125 levels, a value guaranteed not to change in future versions of the standard.
4. Reordering
Rules L1–L4 reorder characters on each line for display:
L1: Resets trailing whitespace and separators to the paragraph embedding level.
L2: Reverses contiguous sequences of characters at the highest embedding levels, progressively down to the lowest odd level.
L3: Reorders combining marks relative to their base characters.
L4: Applies glyph mirroring to characters with the Bidi_Mirrored property when their resolved direction is right-to-left (e.g., "(" becomes ")").
Bidirectional Character Types
Characters are classified into the following categories:
Category
Type
Description
Strong
L
Left-to-Right (e.g., Latin, Han)
R
Right-to-Left (e.g., Hebrew)
AL
Right-to-Left Arabic (e.g., Arabic, Syriac)
Weak
EN
European Number
ES
European Number Separator
ET
European Number Terminator
AN
Arabic Number
CS
Common Number Separator
NSM
Nonspacing Mark
Neutral
B
Paragraph Separator
S
Segment Separator
WS
Whitespace
ON
Other Neutrals
Conformance
A conforming implementation must:
Display all visible characters in the order described by the UBA (UAX9-C1).
Only apply higher-level protocol overrides as defined in Section 4.3 of the specification (UAX9-C2).
Higher-Level Protocols
The UBA permits six higher-level protocol overrides (HL1–HL6), including:
HL4: Apply the UBA independently to segments of structured text (e.g., XML, source code).
HL6: Apply additional glyph mirroring beyond the standard Bidi_Mirrored property.
HTML and CSS Equivalents
On web pages, Unicode directional formatting characters can be replaced by HTML5 and CSS3 markup:
Unicode
HTML
CSS
RLI...PDI
dir="rtl"
direction:rtl; unicode-bidi:isolate
LRI...PDI
dir="ltr"
direction:ltr; unicode-bidi:isolate
FSI...PDI
<bdi>, dir="auto"
unicode-bidi:plaintext
Security Considerations
The misuse of bidirectional formatting characters poses significant security risks, as they can be used to make malicious code or text appear benign. This is documented in Unicode Technical Report #36 (UTR36). Directional overrides (LRO, RLO) are particularly dangerous and should be avoided where possible.
History
Unicode 1.0 (1991): Basic bidirectional support introduced.
Unicode 6.3 (2013): Major revision introducing directional isolates (LRI, RLI, FSI, PDI) and bracket pair resolution (rule N0). These additions were made to address the overly strong effect of directional embeddings on surrounding text.
Unicode 17.0 (2025): Current version (Revision 51).
