1Introduction to MEI

1.1.1MEI Design Principles

This section of the Guidelines defines principles and criteria for designing, developing, and maintaining an XML-based encoding scheme for music notation documents.

1.1.2Acknowledgments

Many institutions and individuals assisted in the preparation of these Guidelines and in the overall development of the Music Encoding Initiative framework and community.

Grateful acknowledgment is given to the following institutions for their generous contributions: the Akademie der Wissenschaften und der Literatur (AdW) in Mainz for serving as hosting institution for the MEI Community, and the National Endowment for the Humanities (NEH) and the Deutsche Forschungsgemeinschaft (DFG) for their joint financial support of the MEI project in its early stages. We thank several institutions that hosted Music Encoding Conferences or other MEI-related meetings in the past: The AdW Mainz, the University of Virginia Library, the Biblioteca Umanistica of the Università degli Studi Firenze, McGill University Montréal, the Centre d’études supérieures de la Renaissance Tours, the Maryland Institute for Technology in the Humanities (MITH) in College Park, the Oxford e-Research Centre, the Universität Paderborn and the Österreichische Akademie der Wissenschaften Wien in conjunction with the Universität Wien and the Mozarteum Salzburg. We also thank all other institutions that allow their researchers to invest time into both the community and the encoding framework. It is their interest that makes MEI an incredible platform for interchange and scholarly progress.

The Text Encoding Initiative is also owed a special debt of gratitude. In addition to providing much of the inspiration for MEI, the TEI organization supplied funding for the MEI Technical Group in its efforts to adopt ODD. The editors of these Guidelines are grateful for those of the TEI, which provided a stellar exemplar and from which we have borrowed shamelessly.

MEI has been a community-driven effort for more than a decade, and many individuals have provided significant and much-appreciated commitments of time and energy to the development of MEI: Nikolaos Beer; Vincent Besson; Benjamin W. Bohl; Margrethe Bue; Donald Byrd; Irmlind Capelle; Tim Crawford; David A. Day; Giuliano Di Bacco; Norbert Dubowy; Richard Freedman; Ichiro Fujinaga; Andrew Hankinson; Maja Hartwig; Kristin Herold; Franz Kelnreiter; Johannes Kepper; Robert Klugseder; Zoltán Kömíves; David Lewis; Urs Liska; Elsa De Luca; Erin Mayhood; Stefan Morent; Stefan Münnich; Markus Neuwirth; Kevin Page; Daniel Pitti; Laurent Pugin; Klaus Rettinghaus; Kristina Richts; Daniel Röwenstrunk; Perry Roland; Craig Sapp; Agnes Seipelt; Eleanor Selfridge-Field; Christine Siegert; Peter Stadler; Axel Teich Geertinger; Martha Thomae; Joachim Veit; Raffaele Viglianti; Thomas Weber; and Sonia Wronkowska.

Thanks to Bernhard R. Appel; Richard Chesser; Morgan Cundiff; J. Stephen Downie; Oliver Huck; Fotis Jannidis; John Rink; Federica Riva; Frans Wiering and Barbara Wiermann for providing expertise on a wide range of topics related to music notation modelling.

Also thanks to Syd Bauman, Terry Catapano, and Sebastian Rahtz for their invaluable problem-solving assistance during the development of the 2010 RNG schema. Thanks to Sebastian Rahtz and James Cummings of the Text Encoding Initiative (TEI) for their help with making ODD work with MEI, their assistance in more closely aligning MEI and TEI, and their quick responses to questions and Roma bug reports.

Finally, the members of the Music Encoding Initiative would like to thank Perry Roland for his foresight, engagement and dedication in laying the foundations of this initiative.

1.2.1Musical Domains

The term "music" has many different notions, ranging from audible sounds over written performance instructions or transcriptions of such events to conceptual rulesets that establish different theories of what music is, and what is allowed in music. In 1965, Milton Babbitt distinguished between graphemic, acoustic and auditory aspects of music (Babbitt, Milton: The Use of Computers in Musicological Research, in: Perspectives of New Music 3/2 (1965), p. 76).

Various music encoding formats took up this distinction, most notably SMDL, the Standard Music Description Language (ISO/IEC DIS 10743). While the format itself was hardly ever used for its impractical implementation details, parts of its design certainly influenced the development of other formats, including MEI. In a documentation draft (http://xml.coverpages.org/smdl10743-pdf.gz, p.5), SMDL identifies four different musical domains:

logical domain: The logical domain is the basic musical content – the essence from which all performances and editions of the work are derived, including virtual time values, nominal pitches, etc. The logical domain is describable as “the composer’s intentions with respect to pitches, rhythms, harmonies, dynamics, tempi, articulations, accents, etc.,” and it is the primary focus of SMDL. It can also be described as “the abstract information common to both the gestural and visual domains.” […]

gestural domain: The gestural domain is comprised of any number of performances, each of which may specify how and when components of the logical domain is rendered in a specific performance, including all the means whereby the performer actually “expresses” (acoustically instantiates) the music (intonation, agogic and dynamic stress, etc.). The gestural domain is perhaps most succinctly described as “the information added by performers,” or “how the music actually sounds during particular performances.” […]

visual domain: The visual domain is comprised of any number of scores, each of which somehow specifies exactly how components of the logical domain is rendered visually in some particular printable (and/or displayable) edition, including such graphical details as symbology, symbol sets, fonts, page layout, beaming conventions and exceptions, etc. The visual domain is perhaps most succinctly described as “the information added by human editors, engravers, and typesetters,” or “how the music actually looks in some particular edition.” […]

analytical domain: The analytical domain is comprised of any number of theoretical analyses and/or commentaries, each of which somehow specifies opinions, exegeses, etc. about any or all of the information in the other three domains. […]

On a generic level, MEI follows the same definition, and it definitely shares the same terminology. However, not all four domains are available throughout the MEI schema, and quite frequently, two domains fall together in MEI. Very often, MEI prioritizes the visual domain over the gestural domain by (partly) conflating the logical and the visual domains. For example, MEI utilizes the pname (pitch name) attribute on notes to capture the written pitch of a note, whereas the sounding pitch may be described with the pname.ges attribute. Here, the logical and visual domains go without a special indication, whereas the gestural domain is identified by a special suffix. However, in case of transposing instruments, additional markup (namely the attributes trans.diat and trans.semi from MEI’s attribute class att.staffDef.log) will create a distinction between the logical and visual domain (see chapter 4.2.2 Defining Score Parameters for CMN). In that case, pname will be restricted to the visual domain, while the logical aforementioned attributes provide additional information for the logical domain.

Even though the technical implementation of MEI prioritizes the visual domain to some degree, this does not mean that any given encoding has to provide visual information. MEI takes no assumption on what data is required: While an OMR project (optical music recognition) may generate strictly visually oriented data only, another project focussed on audio transcriptions may generate gestural data only. A third project could integrate both approaches.

In order to avoid ambiguous encodings, MEI is very strict and specific on the scope of its individual markup elements. For an encoder, the suffixes mentioned above provide clear hints on which domain is addressed by specific markup: Many attributes carry a suffixed .log (logical), .ges (gestural), .vis (visual), or .anl (analytical) in their name. In addition, the internal structure of MEI heavily relies on those different domains. When customizing MEI (see chapter 1.2.5 Customizing MEI), it is possible to turn off either visual or gestural domain encoding completely. That way, MEI allows to address the four most eminent musical domains specifically and independent of each other.

1.2.2Events and Controlevents

MEI differentiates between two essential aspects of music notation: Events and ControlEvents. There are other examples for such a separation of concerns with regard to music. In Greg’s Copy-Text Theory (W.W. Greg: The Rationale of Copy-Text, 1950), a distinction between primary and secondary text is made; similar attempts have been made for music specifically.

In MEI, elements describing the basic musical text are referred to as Events. They are the building blocks for the stream of music – mostly those are notes, rests, and chords. In contrast, ControlEvents make no independent contribution to that flow of music. Instead, they provide additional information about the encoded Events, they control their performance. Examples for such ControlEvents are dynamic markings, tempos indications, or performance directives. Depending on the encoding strategy used, slurs and ties often also fall into this category (they may be encoded as attributes instead, in which case they become a property of the basic events). Simply put, Events describe what needs to be performed, and ControlEvents indicate how it needs to be performed. In (4 Repertoire: Common Music Notation-based) MEI, Events are nested inside a layer element, while ControlEvents are direct children of the first measure they apply to, following all staff elements there. These structural differences result in different markup concepts. As Events are encoded inside layers, their semantic position inside the encoded work can be derived from their structural position – the measure, staff and layer they're nested in, and within that layer by their position inside the sequence of all layer children. As mentioned above, it is highly recommended to encode ControlEvents inside the first measure they apply to, but they still require references to the actual events they apply to. There are two common concepts to provide such a connection, both of which offering specific benefits and drawbacks. A technically very stable connection between ControlEvents and Events can be established by using pointers. In this case, all events that need to be referenced need an xml:id attribute, which holds a globally unique identifier for this very element. The referencing controlevent then uses a startid and, if necessary, endid attribute to create a link to where in the stream of music it is supposed to start or end.

<measure n="10">

<staff n="1">

<layer>

<note pname="f" oct="4" dur="4"/>

<note pname="g" oct="4" dur="4" xml:id="c4ded06ff"/>

<note pname="a" oct="4" dur="4"/>

<note pname="c" oct="5" dur="4"/>

</layer>

</staff>

<dynam startid="#c4ded06ff">f</dynam>

</measure>

In the example above, the dynam element references the second quarter in the given measure. Additional attributes like place may be used to describe the position of the forte indication within the score. A hairpin element may use the endid attribute to indicate the duration of the hairpin using the same mechanism as above.

A ControlEvent encoded like above will be strictly tied to the referenced Events – if their position inside the XML document changes for whatever reason, they will keep that connection. This means that the semantic position to which they are bound may change without affecting the binding. An example could be an inserted additional note in front – the dynamic marking would not start on the second quarter, but perhaps on the third instead.

As this behavior may not be desired in all cases, an alternative binding between ControlEvents and Events is possible, relying on timestamps instead. This mechanism is illustrated in the following example:

<measure n="10">

<staff n="1">

<layer n="1">

<note pname="f" oct="4" dur="4"/>

<note pname="g" oct="4" dur="4"/>

<note pname="a" oct="4" dur="4"/>

<note pname="c" oct="5" dur="4"/>

</layer>

</staff>

<dynam staff="1" layer="1" tstamp="2">f</dynam>

</measure>

Here, no xml:id is required on notes. Instead, the dynam element uses the staff and layer attributes to indicate to which set of events the following tstamp attribute refers to.

This mechanism actually depends on what has been only recommended above: placing the controlevent inside the measure where it starts. The startid reference mechanism would work equally well if all controlevents where positioned in the very first or last measure, or actually even inside a separate file. The tstamp references however would not, they depend on correct placement of the controlevents inside the XML tree. For consistency, it is therefore recommended to always use this placement.

The benefit of this concept is that controlevents are tied to a semantic position, but not necessarily to a given XML element. The forte may still be placed on the second quarter, even though the composer may have replaced that quarter G4 with a different pitch and / or duration. Actually, it is not required that an Event can be found at the position indicated by a timestamp. This may be useful to encode a slur ending at an arbitrary position between two events, or dynam markings spread across otherwise empty measures.

If the ending of a ControlEvent shall be given by timestamp, the tstamp2 attribute is used.

Because of potential inconsistencies, an encoding should not offer both startid and tstamp or endid and tstamp2. Though not being recommendable, it is possible to mix startid with tstamp2 and tstamp with endid. In general, it is easier for software to process startid and endid. When no other arguments apply, using xml:id-based pointers is therefore the most common way to connect ControlEvents with Events.

The details on how timestamps are calculated and used in MEI are given in 1.2.3 Timestamps in MEI.

1.2.3Timestamps in MEI

In MEI, timestamps are treated in a slightly simplified way: they have no notion of beat. Instead, timestamps rely solely on the numbers given in the meter signature. In a measure of 4/4, timestamps will range from 1 to 4. The second eighth note will be 1.5 in this case. If the same measure would be given in 2/2, it would be 1.25 instead.

At this point, MEI uses real numbers only to express timestamps. In case of (nested or complex) tuplets, this solution is inferior to fractions because of rounding errors. It is envisioned to introduce a fraction-based value for timestamps in a future revision of MEI. For now, it is recommended to round the fractional part of the number to no more than five digits to avoid such problems.

Durations may also be expressed based on timestamps. In this case, the values are a combination of the count of measures that need to be moved forward to reach the measure in which an encoded feature ends, and the timestamp within that measure.

The following example contains a number of slur examples illustrating durations expressed by timestamps.

<!-- slur starting on timestamp 1, ending on timestamp 4 of the same measure -->

<slur tstamp="1" tstamp2="0m+4"/>

<!-- slur ending on timestamp 1 of the following measure -->

<slur tstamp="1" tstamp2="1m+1"/>

<!-- slur ending on timestamp 2.5 in the second next measure -->

<slur tstamp="1" tstamp2="2m+2.5"/>

Sometimes, timestamps are used to indicate positions where no music Events are located (see 1.2.2 Events and Controlevents). Therefore, the allowed range of timestamps stretches from 0 to the current meter count + 1. By definition, a timestamp of 1 indicates the position of the left bar line, while a timestamp of 5 (in case of a 4/4 meter) indicates the right bar line. This makes it possible to encode open-ended slurs in a graphical way. However, it should be kept in mind that such timestamps may not be converted to startid and endid, and not every application may be able to render them correctly, even though they are perfectly valid MEI, and sometimes are necessary to faithfully transcribe a source.

1.2.4MEI Profiles

MEI is an encoding framework, not a data format. This means that MEI provides recommendations for encoding music documents, but it depends on the encoder's needs and requirements to which features and solutions are appropriate to the task and should be used. MEI offers specific models for different notation types and music repertoires, but it is rarely advisable to use them all side by side in one encoding.

In order to use MEI, it is advised to use a restricted version of the schema, which will make it easier both for an encoder and a reader of the encoded files. MEI provides a number of pre-defined profiles, which focus on specific uses of MEI while still maintaining a great level of flexibility. For projects that need even better control over their data, it is highly recommended to create a more specific customized version of MEI (see chapter 1.2.5 Customizing MEI). The following customizations are provided with every release of MEI:

mei-CMN: For most users, this will be the best starting point into music encoding with MEI. The mei-CMN customization targets at documents that use Common Western Music Notation. The specific rules for that notation are specified in chapter 4 Repertoire: Common Music Notation, even though other chapters of these Guidelines apply as well.

mei-Mensural: For documents written in Mensural Notation (both black and white), MEI offers the mei-Mensural customization. The specific rules for that notation are specified in chapter 5 Repertoire: Mensural Notation, even though other chapters of these Guidelines apply as well.

mei-Neumes: This profile allows to encode medieval Neume Notation with MEI. The specific rules for that notation are specified in chapter 6 Repertoire: Neume Notation, even though other chapters of these Guidelines apply as well. Please note that the mei-Neumes profile has undergone significant changes from MEI version 3 to version 4.

mei-Basic: As an encoding framework, MEI offers multiple approaches to encode certain features at various levels of detail. While this flexibility is at the core of MEI and often required for research projects, it is an obstacle when developing software and converters for MEI. The mei-Basic profile is a subset of MEI which restricts it to one way of encoding for every feature of music notation. It covers Common Western Music Notation only, and excludes all editorial markup. In essence, it has the same functionality as most other music encoding formats like MusicXML or MNX. The purpose of mei-Basic is to serve as common ground for data interchange, both between projects using different profiles of MEI, and other encoding schemes.

mei-all: This is the full definition of MEI. It includes all different repertoires, which has certain side effects and enables encoding options that are neither intended nor advocable. For example, in mensural notation music is organized by staves. In contrast, Common Music Notation utilizes measures, which in turn contain staves. These staves have a different meaning here, and are modeled differently in MEI. mei-all mixes those models and thus invites encoding errors. In general, you should almost never use mei-all except for testing purposes.

mei-all_anyStart : This profile includes all of mei-all, but extends it even further so that it allows any MEI element as root of conforming MEI instances. In regular MEI, the only allowed starting elements are mei, meiHead, music and meiCorpus. The sole purpose of this customization is to simplify validation at tutorial sessions and other educational purposes. It should not be used in production.

The first three profiles provide good starting points to encode music from the respective repertoires. They may also serve as template for further, project-specific customizations. The latter two profiles target very specific use cases and should not be used by default.

1.2.5Customizing MEI

In production, it is best to use a customized version of MEI, restricted to the very needs of a project. Such a custom schema will guide the encoders and will help to ensure consistency and data quality throughout a project’s files. A customization typically provides a subset of MEI’s encoding models (typically starting from one of the official profiles mentioned in chapter 1.2.4 MEI Profiles), with only one solution for any given situation being allowed. The customization will help to reflect the scope of a project into its data: Only those aspects of music notation a project is interested in will be allowed, so that the absence of a specific information can not be misunderstood as an oversight of the encoders. Larger editorial projects like Complete Works editions typically use Editorial Guidelines (german: Editionsrichtlinien) for the same purposes: (internal) quality control and (external) documentation. In that sense, MEI customizations may serve as Editorial Guidelines in digital form.

MEI is implemented in ODD. ODD, or One Document Does-it-all, is another XML-based markup language developed and maintained by the TEI. TEI's documentation for ODD can be found in the TEI Guidelines chapter 22: Documentation Elements, chapter 23: Using the TEI, and the "Getting Started with P5 ODDs" document.

At this point, there is no specific documentation on how to customize MEI with ODD beyond the generic TEI documentation. However, the provided 1.2.4 MEI Profiles are based on ODD customizations, and may serve as starting point for further project-specific restrictions. They can be found at https://github.com/music-encoding/music-encoding/tree/develop/customizations. In addition, several projects have shared their customizations on GitHub, such as Freischütz Digital or Beethovens Werkstatt.

MEI provides a web service at http://custom.music-encoding.org which allows to compile such customizations against the MEI sources in order to generate RelaxNG schemata, which can be used for validation. More documentation on customizing MEI will be provided as time permits; until then, it is recommended to reach out to the MEI Community for additional assistance.