Mathematical Markup Language (MathML) Version 2.0
4 Content Markup
5 Combining Presentation and Content Markup
5.1 Why Two Different Kinds of Markup?
5.2 Mixed Markup
5.2.1 Reasons to Mix Markup
5.2.2 Combinations that are prohibited
5.2.3 Presentation Markup Contained in Content Markup
5.2.4 Content Markup Contained in Presentation Markup
5.3 Parallel Markup
5.3.1 Top-level Parallel Markup
5.3.2 Fine-grained Parallel Markup
5.3.3 Parallel Markup via Cross-References:
5.4 Tools, Style Sheets and Macros for Combined Markup
5.4.1 Notational Style Sheets
5.4.2 Content-Faithful Transformations
5.4.3 Style Sheets for Extensions
6 Entities, Characters and Fonts
Presentation markup and content markup can be combined in one of two ways. The first manner is to intersperse content and presentation elements in what is essentially a single tree. This is called mixed markup. The second manner is to provide both an explicit presentation and an explicit content in a pair of trees. This is called parallel markup. This chapter describes both mixed and parallel markup, and how they may used in conjunction with style sheets and other tools.
Chapters 3 and 4 describe two kinds of markup for encoding mathematical material in documents.
Presentation markup captures notational structure. It encodes the notational structure of an expression in a sufficiently abstract way to facilitate rendering to various media. Thus, the same presentation markup can be rendered with relative ease on screen in either wide and narrow windows, in ASCII or graphics, in print, or it can be enunciated in a sensible way when spoken. It does this by providing information such as structured grouping of expression parts, classification of symbols, etc.
Presentation markup does not directly concern itself with the mathematical structure or meaning of an expression. In many situations, notational structure and mathematical structure are closely related, so a sophisticated processing application may be able to heuristically infer mathematical meaning from notational structure, provided sufficient context is known. However, in practice, the inference of mathematical meaning from mathematical notation must often be left to the reader.
Employing presentation tags alone may limit the ability to re-use a MathML object in another context, especially evaluation by external applications.
Content markup captures mathematical structure. It encodes mathematical structure in a sufficiently regular way in order to facilitate the assignment of mathematical meaning to an expression by application programs. Though the details of mapping from mathematical expression structure to mathematical meaning can be extremely complex, in practice, there is wide agreement about the conventional meaning of many basic mathematical constructions. Consequently, much of the meaning of a content expression is easily accessible to a processing application, independently of where or how it is displayed to the reader. In many cases, content markup could be cut from a Web browser and pasted into a mathematical software tool (such as future versions of Axiom, Maple or Mathematica) with confidence that sensible values will be computed.
Since content markup is not directly concerned with how an expression is displayed, a renderer must infer how an expression should be presented to a reader. While a sufficiently sophisticated renderer and style-sheet mechanism could in principle allow a user to read mathematical documents using personalized notational preferences, in practice, rendering content expressions with notational nuances may still require human intervention of some sort.
Employing content tags alone may limit the ability of the author to precisely control how an expression is rendered.
It is important to emphasize that both content and presentation tags are necessary in order to provide the full expressive capability one would expect in a mathematical markup language. Often the same mathematical notation is used to represent several completely different concepts. For example, the notation xi may be intended (in polynomial algebra) as the i-th power of the variable x, or (in vector analysis) as the i-th component of a vector x, for example in tensor calculus in general relativity. In other cases, the same mathematical concept may be displayed in one of various notations. For instance, the factorial of a number might be expressed with an exclamation mark, a Gamma function, or a Pochhammer symbol.
Thus the same notation may represent several mathematical ideas, and, conversely, the same mathematical idea often has several notations. In order to provide authors with the ability to precisely control notational nuances while at the same time encoding meanings in a machine-readable way, both content and presentation markup are needed.
In general, if it is important to control exactly how an expression is rendered, presentation markup will generally be more satisfactory. If it is important that the meaning of an expression can be interpreted dependably and automatically, then content markup will generally be more satisfactory.
MathML offers authors elements for both content and presentation markup. Whether to use one or the other, or a combination of both, depends on what aspects of rendering and interpretation an author wishes to control, and what kinds of re-use he or she wishes to facilitate.
In many common situations, an author or authoring tool may choose to generate either presentation or content markup exclusively. For example, a program for translating legacy documents would most likely generate pure presentation markup. Similarly, an educational software package might very well generate only content markup for evaluation in a computer algebra system. However, in many other situations, there are advantages to mixing both presentation and content markup within a single expression.
If an author is primarily presentation-oriented, interspersing some content markup will often produce more accessible, more re-usable results. For example, an author writing about linear algebra might write:
<mrow> <apply> <power/> <ci>x</ci><cn>2</cn> </apply> <mo>+</mo> <msup> <mi>v</mi><mn>2</mn> </msup> </mrow>
where v is a vector and the superscript denotes a vector component, and x is a real variable. On account of the linear algebra context, a visually impaired reader may have directed his or her voice synthesis software to render superscripts as vector components. By explicitly encoding the power, the content markup yields a much better voice rendering than would likely happen by default.
If an author is primarily content-oriented, there are two reasons to intersperse presentation markup. First, using presentation markup provides a way of modifying or refining how a content expression is rendered. For example, one might write:
<apply> <in/> <ci><mi fontweight="bold">v</mi></ci> <ci>S</ci> </apply>
In this case, the use of embedded presentation markup allows the author to specify that v should be rendered in boldface.
A second reason to intersperse presentation in content markup is that there is a continually growing list of areas of discourse that do not have pre-defined content elements for encoding their objects and operators. As a consequence, any system of content markup inevitably requires an extension mechanism that combines notation with semantics in some way. MathML content markup specifies several ways of attaching an external semantic definitions to content objects. However, it is necessary to use MathML presentation markup to specify how such user-defined semantic extensions should be rendered.
For example, the `rank' operator from linear algebra is not
included as a pre-defined MathML content element. Thus, to express
we use the
mo presentation element inside a
ci element to achieve the proper presentation, along
semantics element to bind a semantic
definition to the symbol. The
mo element indicates
to a renderer that it should use standard operator spacing around the
content identifier `rank', just as it would for
`sin' or `log':
<apply> <eq/> <apply> <fn> <semantics> <ci><mo>rank</mo></ci> <annotation-xml encoding="OpenMath"> <OMS cd="linalg1" name="rank"/> </annotation-xml> </semantics> </fn> <apply> <times/> <apply> <transpose/> <ci>u</ci> </apply> <ci>v</ci> </apply> </apply> <cn>1</cn> </apply>
Here, the semantics of the presentation sub-expressions have been given using symbols from OpenMath content dictionaries (CD).
The main consideration when presentation markup and content markup are mixed together in a single expression is that the result should still make sense. When both kinds of markup are contained in a presentation expression, this means it should be possible to render the resulting mixed expressions simply and sensibly. Conversely, when mixed markup appears in a content expression, it should be possible to simply and sensibly assign a semantic interpretation to the expression as whole. These requirements place a few natural constraints on how presentation and content markup can be mixed in a single expression, in order to avoid ambiguous or otherwise problematic expressions.
Two motivating examples illustrate the kinds of problems that must be avoided in mixed markup. Consider:
<mrow> <plus/> <mi> x </mi> <mi> y </mi> </mrow>
In this example, the content element
plus has been
indiscriminately embedded in a presentation expression. Should the plus
sign appear in its usual infix position, as it would in content markup, or
should it render as the first thing in the row? Neither choice is very
satisfactory, and consequently, this kind of mixing is not
allowed. Similarly, consider:
<apply> <ci> x </ci> <mo> + </mo> <ci> y </ci> </apply>
As before, the
mo element is problematic. Should a
renderer infer that the usual arithmetic operator is intended, and act as
if the prefix content element
plus had been used?
Again, there is no compelling answer, and thus this kind of mixing of
content and presentation markup is also prohibited.
The use of presentation markup within content markup is limited to situations that do not effect the ability of content markup to unambiguously encode mathematical meaning. More specifically, presentation markup may only appear in content markup in three ways:
Any other presentation markup occurring within a content markup is a MathML error. More detailed discussion of these three cases follows:
cnare permitted to contain any sequence of
PCDATA, presentation elements, and
sepempty elements. Contiguous blocks of
cnelements are rendered as if they were wrapped in
mielements. A contiguous block of
cnshould be rendered as if wrapped in
mn. If a token element contains both
PCDATAand presentation elements, contiguous blocks of
PCDATA(if any) are treated as if wrapped in
mnelements as appropriate, and the resulting collection of presentation elements are rendered as if wrapped in an
sepelement is only meaningful in identifiers and numbers defined to be of complex type, where it separates
PCDATAinto real and imaginary parts. When a token elements contains both
sepelements and presentation elements, the
sepelements are ignored.
csymbolelement may contain either
PCDATAinterspersed with presentation markup, or content elements of the container type. It is a MathML error for a
csymbolelement to contain both presentation and content elements. When the
csymbolelement contains both raw data and presentation markup, the same rendering rules that apply to content elements of the token type should be used.
semanticselement is to provide a mechanism for incorporating arbitrary MathML expressions into content markup in a semantically meaningful way. In particular, any valid presentation expression can be embedded in a content expression by placing it as the first child of a
semanticselement. The meaning of this wrapped expression should be indicated by one or more annotation elements also contained in the
semanticselement. Suggested rendering for a
semanticselement is discussed in section 4.2.6 [Syntax and Semantics].
Stéphane Dalmas has commented that the paragraph below is not clear enough.
The guiding principle for embedding content markup within presentation expressions is that the resulting expression should still have an unambiguous rendering. In general, this means that embedded content expressions must be semantically meaningful, since rendering of content markup depends on its meaning. This translates into the condition that elements of the content, operator and relation type are permitted, while elements of the qualifier and condition type are not.
As a rule, content elements other than containers derive part of their
semantic meaning from the surrounding context, such as
bvar element is qualifying an integral,
logical quantifier or lambda expression. Another example
would be whether a
degree element occurs in
Thus, in a presentation context, elements such as these do not have a clearly
defined meaning, and hence there is no obvious choice for a rendering.
Consequently, they are not allowed.
Stéphane Dalmas has suggested to list only those elements that cannot appear as a child.
The complete list of content elements that may appear as a child
in a presentation element is:
Note that within presentation markup, content expressions may only appear in locations where it is valid for any MathML expression to appear. In particular, content expressions may not appear within presentation token elements. In this regard mixing presentation and content are asymmetrical.
For rendering purposes, when a permitted content element appears within
a presentation context, a processing application should treat it as if it
were replaced with an
mrow containing a
presentation encoding of the rendering the application would ordinarily
generate for that content markup. For example, consider:
<mfrac> <mi>x</mi> <interval closure="open-closed"> <cn>1</cn> <cn>3</cn> </interval> </mfrac>
In this case, a visual renderer would typically render the
interval construct as (1,3]. Using presentation markup,
this might be encoded as:
<mfenced close="]"> <mn>1</mn> <mn>3</mn> </mfenced>
Consequently, the original mixed markup should be visually rendered as
<mfrac> <mi>x</mi> <mfenced close="]"> <mn>1</mn> <mn>3</mn> </mfenced> </mfrac>
Some applications are able to make use of both presentation and content information. For these applications it is desirable to provide both forms of markup for the same mathematical expression. This is called parallel markup.
Parallel markup is achieved with the
semantics element. Parallel markup for an expression
can be used on its own, or can be incorporated as part of a larger content
or presentation tree.
In many cases what is desired is to provide
presentation markup and content markup for a mathematical
expression as a whole.
To achieve this, a single
semantics element is used
pairing two markup trees, with the first branch being the MathML presentation
markup, and the second branch being the MathML content markup.
The following example encodes the boolean arithmetic expression (a+b)(c+d) in this way.
<semantics> <mrow> <mrow><mo>(</mo><mi>a</mi> <mo>+</mo> <mi>b</mi><mo>)</mo></mrow> <mo>⁢</mo> <mrow><mo>(</mo><mi>c</mi> <mo>+</mo> <mi>d</mi><mo>)</mo></mrow> </mrow> <annotation-xml encoding="MathML-Content"> <apply><and/> <apply><xor/><ci>a</ci> <ci>b</ci></apply> <apply><xor/><ci>c</ci> <ci>d</ci></apply> </apply> </annotation-xml> </semantics>
This example is non-trivial in the sense that the content markup could not be easily derived from the presentation markup alone.
Top-level pairing of independent presentation and content markup is
sufficient for many, but not all, situations. Applications that allow
treatment of sub-expressions of mathematical objects require
the ability to associate presentation, content or information with the
parts of an object with mathematical markup. Top-level
pairing with a
semantics element is insufficient in
this type of situation; identification of a sub-expression in one branch of
semantics element gives no indication of the
corresponding parts in other branches.
The ability to identify corresponding sub-expressions is required in
applications such as mathematical expression editors. In this situation,
selecting a sub-expression on a visual display can identify a particular
portion of a presentation markup tree. The application then needs to
determine the corresponding annotations of the sub-expressions; in
particular, the application requires the sub-expressions of the
annotation-xml tree in MathML content notation.
It is, in principle, possible to provide annotations for
each presentation node by incorporating
semantics elements recursively.
<semantics> <mrow> <semantics> <mrow><mo>(</mo><mi>a</mi> <mo>+</mo> <mi>b</mi><mo>)</mo></mrow> <annotation-xml encoding="MathML-Content"> <apply><plus/><ci>a</ci> <ci>b</ci></apply> </annotation-xml> </semantics> <mo>⁢</mo> <semantics> <mrow><mo>(</mo><mi>c</mi> <mo>+</mo> <mi>d</mi><mo>)</mo></mrow> <annotation-xml encoding="MathML-Content"> <apply><plus/><ci>c</ci> <ci>d</ci></apply> </annotation-xml> </semantics> </mrow> <annotation-xml encoding="MathML-Content"> <apply><times/> <apply><plus/><ci>a</ci> <ci>b</ci></apply> <apply><plus/><ci>c</ci> <ci>d</ci></apply> </apply> </annotation-xml> </semantics>
To be complete this example would be much more verbose, wrapping each of
the individual leaves
mn in a further seven
This approach is very general and works for all kinds of annotations (including non-MathML annotations and multiple annotations). It leads, however, to O(n2) increase in size of the document. This is therefore not a suitable approach for fine-grained parallel markup of large objects.
To better accomodate applications that must deal with sub-expressions
of large objects, MathML uses cross-references between the branches of a
semantics element to identify corresponding
Cross-referencing is achieved using
xref attributes within the branches of a containing
semantics element. These attributes may optionally be
placed on MathML elements of any type.
The following example shows this cross-referencing for the boolean arithmetic expression (a+b)(c+d).
<semantics> <mrow id="E"> <mrow id="E.1"> <mo id="E.1.1">(</mo> <mi id="E.1.2">a</mi> <mo id="E.1.3">+</mo> <mi id="E.1.4">b</mi> <mo id="E.1.5">)</mo> </mrow> <mo id="E.2">⁢</mo> <mrow id="E.3"> <mo id="E.3.1">(</mo> <mi id="E.3.2">c</mi> <mo id="E.3.3">+</mo> <mi id="E.3.4">d</mi> <mo id="E.3.5">)</mo> </mrow> </mrow> <annotation-xml encoding="MathML-Content"> <apply xref="E"> <and xref="E.2"/> <apply xref="E.1"> <xor xref="E.1.3"/><ci xref="E.1.2">a</ci><ci xref="E.1.4">b</ci> </apply> <apply xref="E.3"> <xor xref="E.3.3"/><ci xref="E.3.2">c</ci><ci xref="E.3.4">d</ci> </apply> </apply> </annotation-xml> <annotation-xml encoding="OpenMath"> <OMA xref="E"> <OMS cd="logic" name="and" xref="E"/> <OMA xref="E.1"> <OMS cd="logic" name="xor" xref="E.1.3"/> <OMV name="a" xref="E.1.2"/> <OMV name="b" xref="E.1.4"/> </OMA> <OMA xref="E.3"> <OMS cd="logic" name="xor" xref="E.3.3"/> <OMV name="c" xref="E.3.2"/> <OMV name="d" xref="E.3.4"/> </OMA> </OMA> </annotation-xml> </semantics>
OMV are elements defined in the OpenMath
standard for representing application, symbol and variable, respectively.
id attribute and a corresponding
xref appearing within the same
semantics element create a correspondence between
In creating these correspondences by cross-reference, all
id attributes referenced by any
xref must be in the same branch of an
semantics element. This constraint
guarantees that these correspondences do not create unintentional cycles.
(Note that this restriction does not exclude the use of
id attributes within the other branches of the enclosing
semantics element. It does, however, exclude
references to these other
originating in the same
There is no restriction on which branch of the
semantics element may contain the destination
id attributes. It is up to the application to determine
which branch to use.
In general, there will not be a one-to-one correspondence between nodes
in parallel branches. For example, a presentation tree may contain
elements, such as parentheses, that have no correspondents in the content
tree. It is therefore often useful to put the
attributes on the branch with the finest-grained node structure. Then all
of the other branches will have
to some subset of the
In absence of other criteria, the first branch of the
semantics element is a sensible choice to contain the
id attributes. Applications that add or remove
annotations will then not have to re-assign attributes to the
In general, the use of
xref attributes allows a full correspondence between
sub-expressions to be given in text that is at most a constant factor
larger than the original. The direction of the references should not be
taken to imply that sub-expression selection is intended to be permitted
only on one child of the
semantics element. It is
equally feasible to select a subtree in any branch and to recover the
corresponding subtrees of the other branches.
The interaction of presentation and content markup can be greatly enhanced through the use of various tools. While the set of tools and standards for working with XML applications is rapidly evolving at the present, we can already outline some specific techniques.
In general, the interaction of content and presentation is handled via transformation rules on MathML trees. These transformation rules are sometimes called `macros'. In principle, these rules can be expressed using any one of a number of mechanisms, including DSSSL, Java programs operating on a DOM, etc. We anticipate, however, that the principal mechanism for these transformations in most applications shall be XSLT.
In this section we discuss transformation rules for two specific purposes: for notational style sheets, and to simplify parallel markup.
Authors who make use of content markup may be required to deploy their
documents in locales with notational conventions different than the default
content rendering. It is therefore expected that transformation tools will
be used to determine notations for content elements in different
settings. Certain elements, e.g.
widely varying common notations and will often require notational
selection. Some examples of notational variations are given below.
Other elements, for example
sin, are less likely to require these features.
We observe that selection of notational style is sometimes necessary for correct understanding of documents by locale. For instance, the binomial coefficient in French notation is equivalent to in Russian notation.
A natural way for a MathML application to bind a particular notation to the set of content tags is with an XSLT style sheet [XSLT]. The examples of this section shall assume this is the mechanism to express style choices. (Other choices are equally possible, for example an application program may provide menus offering a number of rendering choices for all content tags.)
When writing XSLT style sheets for mathematical notation, some transformation rules can be purely local, while others will require multi-node context to determine the correct output notation. The following example gives a local transformation rule that could be included in a notational style sheet displaying open intervals as ]a,b[ rather than as (a,b).
<xsl:template match="interval"> <mrow> <xsl:choose> <xsl:when test="@closure='closed'"> <mfenced open="[" close="]" separators=","> <xsl:apply-templates/> </mfenced> </xsl:when> <xsl:when test="@closure='open'"> <mfenced open="]" close="[" separators=","> <xsl:apply-templates/> </mfenced> </xsl:when> <xsl:when test="@closure='open-closed'"> <mfenced open="]" close="]" separators=","> <xsl:apply-templates/> </mfenced> </xsl:when> <xsl:when test="@closure='closed-open'"> <mfenced open="[" close="[" separators=","> <xsl:apply-templates/> </mfenced> </xsl:when> <xsl:otherwise> <mfenced open="[" close="]" separators=","> <xsl:apply-templates/> </mfenced> </xsl:otherwise> </xsl:choose> </mrow> </xsl:template>
An example of a rule requiring context information would be:
<xsl:template match="apply[factorial]"> <mrow> <xsl:choose> <xsl:when test="not(*=ci) and not(*=cn)"> <mrow> <mo>(</mo> <xsl:apply-templates select="*" /> <mo>)</mo> </mrow> </xsl:when> <xsl:otherwise> <xsl:apply-templates select="*" /> </xsl:otherwise> </xsl:choose> <mo>!</mo> </mrow> </xsl:template>
Other examples of context-dependent transformations would be, e.g.
apply of a
to render a-b+c, rather than
a+ -b+c, or for the
apply of a
to render sin2x, rather than
Notational variation will occur both for built-in content elements as well as extensions. Notational style for extensions can be handled as described above, with rules matching the names of any extension tags, and with the content handling (in a content-faithful style sheet) proceeding as described in section 5.4.3 [Style Sheets for Extensions].
There may be a temptation to view notational style sheets as a transformation from content markup to equivalent presentation markup. This viewpoint is explicitly discouraged, since information will be lost and content-oriented applications will not function properly.
We define a `content-faithful' transformation to be a transformation that retains the original content in parallel markup (section 5.3 [Parallel Markup]).
Tools that support MathML should be `content-faithful', and not gratuitously convert content elements to presentation elements in their processing. Notational style sheets should be content-faithful whenever they may be used in interactive applications.
It is possible to write content-faithful style sheets in a number of ways. Top-level parallel markup can be achieved by incorporating the following rules in an XSLT style sheet:
<xsl:template match="/"> <semantics> <xsl:apply-templates/> <annotation-xml encoding="MathML-Content"> <xsl:copy-of select="."/> </annotation-xml> </semantics> </xsl:template> <xsl:template match="*"> <xsl:copy> <xsl:apply-templates/> </xsl:copy> </xsl:template>
The notation would be generated by additional rules for producing presentation from content, such as those in section 5.4.1 [Notational Style Sheets].
The presentation tags of MathML form a closed vocabulary of notational structures, but are quite rich and can be used to express a rendering of most mathematical notations. Although the basic layout schemata are rich enough, they may be composed to layout notations that are quite complex, or that might not have yet been considered. In this sense, the presentation ability of MathML is open-ended. It is often useful, however, to give a name to a new notational schema if it is going to be used often.
The content tags form a fixed vocabulary of concepts covering the types
of mathematics seen in most common applications. It is not reasonable to
expect users to compose existing MathML content tags to construct new
content concepts. (This approach is frought with technical difficulties
even for professional mathematicians.) Instead, it is anticipated that
applications whose mathematical content concepts extend beyond what is
offered by MathML, will use annotations within
semantics elements, and that these annotations will use
content description languages outside of MathML.
Often the naming of a notation and the identification of a new semantic concept are related. This allows a single transformation rule to capture both a presentation and a content markup for an expression. This is one of the areas of MathML that benefits most strongly from the use of macro processing. With any of the current document transformation standards, for instance XSLT or DSSSL, it is trivial to define rules that take, for example
and respectively transform them to
<semantics> <ci><mo>rank</mo></ci> <annotation-xml encoding="OpenMath"> <OMS cd="linalg1" name="rank"/> </annotation-xml> </semantics>
<apply> <transpose/> <ci>X</ci> </apply>
The lengthy sample encoding of rank(uTv)=1, from section 5.2.1 [Reasons to Mix Markup] could then be condensed to
<apply> <eq/> <apply> <rank/> <apply> <times/> <tr>u</tr> <ci>v</ci> </apply> </apply> <cn>1</cn> </apply>
From this example we see how the combination of presentation and content markup could become much simpler and effective to generate as standard style-sheet libraries become available.