Design requirements

Huron (1992: 15-37) specifies a list of twelve properties of good representations.

These properties may be used to characterize representations and to point out differences between representations. Huron lists the properties as follows:

1. Unique 2. Mnemonic 3. Consistent 4. Reversible 5. Terse 6. Non-cryptic

7. Structurally isomorphic 8. Context-free

9. Idiomatic

10. Explicit (interpreted) 11. Optional

12. Extendable

By unique, Huron means that “no two signifieds may share the same signifier”

(ibid.: 16). This property guarantees error-free (or ambiguity-free) translation or interpretation of the representation. Mnemonic refers to associational relation-ships or mappings between signifiers and signifieds, said relationrelation-ships meaning to help the user to learn or remember the representation. Huron mentions several types of mnemonic relationships. An example of literal mapping is the relation-ship between a graphic music notation symbol and its written name (for exam-ple, a fermata symbol and the word “fermata”). An “initial” mnemonic relationship occurs when a signifier is a letter that equals the written name of the signified; for example, when the letters W, H, and E represent durational values of a whole note, half note, and eighth note, respectively. “Pictorial” relationship refers to (approximate) visual resemblance between signifiers and signifieds; for example, the characters | and & as signifiers for a barline and a treble clef, respectively. Northern Indian tabla notation is an example of “onomatopoeic”

mapping (Huron 1992: 20; Bel & Kippen 1992: 208).

Other types of mnemonic mappings listed by Huron are operational, seman-tic convention, isotonic convention, and topological correspondence. Isotonic convention describes the mapping of two quantitatively ordered parameters. For example, in MIDI, numbers 0 to 127 are mapped to the keys of a piano or

syn-thesizer keyboard. Isotonic mapping of dynamic markings would be achieved by assigning consecutive integer values for each dynamic level from ppp (or pppp, if desired) through fff (or ffff).

A representation is consistent when general rules and conventions of the rep-resentation are followed and applied without exceptions. Consistency is harder to achieve in terse representations than in verbose ones. This is due to limits placed upon the size and thereby by the number of possible signifiers for a given parameter. For example, the DARMS representation mixes initialism and iso-tonic convention in the encoding of durations (Huron 1992: 23).

In a completely reversible representation, the signifiers may be derived from the signifieds with the same effort required to derive the respective signifieds from the representation’s signifiers. According to Huron, however, reverse map-ping from signifieds to signifiers may be less mnemonic than is the opposite mapping (Huron 1992: 24).

A terse representation is often more efficient to use (for example, to write or to read) than a “verbose” representation. Huron uses a programming language to exemplify the advantages of terse representation (1992: 25). According to this example, it is more convenient to keep (local) variable names short rather than unnecessarily verbose. Here, “verbose” means that variable names are full, descriptive names rather than single letters or abbreviations as used in terse rep-resentations. The same principle is discussed by Kernighan and Pike (1999: 3).

A terse representation is more economical in terms of space and often faster to comprehend, especially by an expert user or programmer. However, Kernighan and Pike point out that short names should be used for local variables whereas descriptive names should be used for global ones (ibid.).

Terseness is beneficial especially for a text-based representation intended for data input. The less keystrokes, the faster the encoding process. However, terse representations become easily – but not necessarily – more cryptic than verbose ones do. Obviously, non-cryptic representations are easier to understand and to process manually than are cryptic ones.

Structural isomorphism means that the signifiers are structurally organized similarly to the signifieds in the system to be represented. For example, a struc-turally isomorphic computer representation of music notation would somehow maintain the two-dimensional graphical structure of music notation. In context-free representations signifiers are self-contained and independent of other signi-fiers.

A representation is idiomatic if it takes advantage of practical features, “idi-oms”, provided by computing environments. One such feature is the ASCII

character set. Another is the principle in which computer memory is physically organized as 8 bit bytes and other power-of-two quantities.

According to Huron, a representation is a result of interpretation. A represen-tation is explicit when signifieds are mapped to signifiers so that they form a cor-rect and sufficient interpretation of the target system for a particular application or purpose. For example, a different kind of explicit mapping may be needed in a representation intended for music printing purposes than for music analysis.

According to Huron, it is important that the signifiers are well-suited (i.e., pro-vide a correct interpretation) to the application for which they are used (Huron 1992: 31-34).

A representation is optional, if it allows the user to omit those attributes that are unnecessary for a particular task. For example, a user might want to omit pitch when encoding only a rhythmic structure. An extendable representation is not restricted to the set of signifiers defined by the designer of the representa-tion.

It may be difficult for a music representation to meet all of the above require-ments. For example, to be both terse and non-cryptic, may be impossible unless it is for a fairly simple or restricted target system. Further, if a representation is too idiomatic, then expressiveness, extensibility or even storage capability may suffer. Many binary formats, for instance, use a fixed number of bits for storing data lengths or other information about amounts. For example, the RIFF file for-mat, which is used for many music and video applications, uses 32-bit data-length fields, which limits the storage capacity of RIFF to 4 gigabytes. This makes a single RIFF incapable of storing lengthy, high-resolution video or multi-track audio recordings.

Huron´s requirements provide a usable set of features for characterizing and assessing music representations. Still, some additional criteria and complemen-tary remarks can be brought forward.

First of all, a representation should be complete (i.e., provide a signifier for every signified). This completeness may be difficult, if not impossible, to achieve in a dynamic target system. This situation emphasizes the importance of extensibility. Extensibility addresses the issue of completeness indirectly. To be usable, however, a representation should be complete enough to serve the pur-pose for which it is intended. An overly restricted initial set of signifiers may result in the use of terse or idiomatic signifiers that will later lead to inconsisten-cies when more signifiers must be added.

As mentioned by Huron, a central issue in representation is mapping between signifiers and signifieds. In general, four different types of mapping can be defined: (1) one-to-one, (2) one-to-many, (3) one, and (4) many-to-many. In one-to-one mapping, there is always exactly one signifier for each

sig-nified and vice versa. In one-to-many type mapping, a signifier is mapped to a group of signifieds. In many-to-one mapping, in turn, a group of signifiers is mapped to a single signifier. In many-to-many mapping the mapping is done between a group of signifiers and a group of signifieds. Of these, one-to-one mapping is the most explicit and unambiguous. One-to-one mapping also makes reverse mapping easier to achieve.