5 Notational system

Two important criteria for transcription are machine readability and rendering for human readability. We first propose a feature-based vector notation for machine readability. Abbreviations for human readability are under development.

A gesture respresents a set of values of the gesture attributes in the following order:

  1. source:

  2. trajectory:

  3. target:

  4. symmetry

A gesture can consist of a source, a trajectory and a target. Whereas for dynamic gestures all three features need to be specified, for static gestures only the source needs to be described, since there is no trajectory and thus the source and the target are identical.

5.0.0.1 Source and Target.

When making a complete annotation without gaps (e.g. empty or unsystematic notations), the target location and target hand shape of a gesture and the source location and source hand shape of the following gesture are identical. Thus, the target labels of the gestures can be omitted for reasons of efficiency.

The categories source and target both consist of the features location and hand shape. The category trajectory contains the features directionality, shape, hand shape, and modifiers.
The attributes size, repetitions, and speed are subsumed in the category of modifiers.
If a gesture is produced with a pair of body parts, e.g. hands, it must be added whether it is produced either in a parallel or symmetrical fashion, or whether the limbs act independently.

The transcription system is tentative and is still under development. We currently propose that transcription symbols are concatenated in two ways: in order to connect values on different attribute levels, a ``,'' is used. The following ASCII-based labels for values apply:

5.0.0.2 Location.

Location is described by using a modified and extended matrix-like version of the HamNoSys 4.0 charts on Hand and Body Position and on Distance from the Body (see Table 2).


Table 2: Possible Hand and Body Positions according to HamNoSys 4.0
left of left side left side of middle of right side of right of right side
above the head 1ll 1l 1m 1r 1rr
head 2ll 2l 2m 2r 2rr
forehead 3ll 3l 3m 3r 5rr
nose 4ll 4l 4m 4r 4rr
under the nose 5ll 5l 5m 5r 5rr
mouth 6ll 6l 6m 6r 6rr
tongue 7ll 7l 7m 7r 7rr
teeth 8ll 8l 8m 8r 8rr
chin 9ll 9l 9m 9r 9rr
under the chin 10ll 10l 10m 10r 10rr
neck 11ll 11l 11m 11r 11rr
shoulder region 12ll 12l 12m 12r 12rr
breast region 13ll 13l 13m 13r 13rr
stomach region 14ll 14l 14m 14r 14rr
lap 15ll 15l 15m 15r 15rr
thighs 16ll 16l 16m 16r 16rr
knees 17ll 17l 17m 17r 17rr
lower legs 18ll 18l 18m 18r 18rr
feet 19ll 19l 19m 19r 19rr

5.0.0.3 Hand Shape.

Hand shape symbols are taken from the FORM hand shape inventory (see Martell, 2002). The label A5, for instance, stands for a relaxed hand. The label A7 for folded hands has been added to the categories.

Figure 4: FORM inventory of hand shapes (Martell, 2002).
\includegraphics[scale=0.6]{handshapes.eps}

5.0.0.4 Directionality.

The labels for directionality are:

Combinations of these directions are joined with a slash (``$/$''), i.e. $up/fo$ for an up-forward direction.

5.0.0.5 Shape.

The labels for trajectory shape are:

Further categories may, of course, be added if necessary.

5.0.0.6 Modifiers.

The basic functional transcription as just outlined can be supplemented by additional symbols, some of which only apply to particular parts of the body (e.g. hands):

5.0.0.7 Symmetry.

When describing a gesture which is executed with a pair of body parts or limbs, e.g. the hands, this category becomes relevant because it is necessary to define which part of the pair is used or how both parts are related to each other when moving about. If a gesture is executed with only one body part, the category can be assigned the value $rp$ (right part, relative to the speaker) or $lp$ (left part). If a gesture is executed with both parts, the general case would be first to describe the gesture produced with the right part, then the gesture produced with the left part, concatenated by ``$;$''. For gestures where the body parts move independently within the same time frame, this annotation does not have to be changed. However, a great number of gestures tend to be of a symmetrical or parallel nature, i.e. the body parts move either symmetrically or parallelly. In these cases, the two notational strings are merged and are assigned the values $sy$ (symmetrical) or $pa$ (parallel) under this category. This can, however, also be added on to. It might, for example, be necessary to define the type of symmetry as symmetrical to an axis (radial or bilateral symmetry) vs. symmetrical to a center.

If it is deemed necessary by the individual researcher to make an inter-rater comparison, all attributes need to be assigned a value. So even the optional ones would get their default value. In order to avoid notational ambiguity, it is necessary to keep in mind that the notational order is fixed since some values can occur more than once in a single string.

Thus, a gesture can be described in form of a vector with a maximum of values as listed below:

  1. location of source: see Table 2
  2. hand shape of source: see Figure 4
  3. directionality of trajectory: $ri, le, fo, ba, up, do$ and combinations of those
  4. shape of trajectory: $ci, li, wl, ar, zl, el, sq$
  5. hand shape of trajectory: see Figure 4
  6. size of trajectory: $xs, s, m, l, xl$
  7. repetitions of trajectory: $r(0), r(1), r(2),$ etc.
  8. speed of trajectory: $sl, me, fa$

  9. (location of target: see Table 2)
  10. (hand shape of target: see Figure 4)

  11. symmetry: $rp, lp, sy, pa$

To give an example, the transcription

$15m,5A,ri,ci,1B,l,r(0),me,15m,5A,rp$
for the hands describes an unrepeated gesture carried out with medium speed with the right hand tracing a large circle with a pointed index finger, which starts and ends with the hands in a relaxed position on the lap.



Subsections
Thorsten Trippel 2003-08-12