4.3 Selection of classes and data acquisition methodology

In this thesis it is obviously necessary to select a subset of Auslan to be considered for testing of GRASP, since it would take vast amounts of time to sample most or all of the signs in Auslan. Several issues then arise in the selection:

• How many signs do you choose?

• On what basis do you select the signs?

• How do you uniquely identify a sign?

The number of signs should at least be sufficient to incorporate as many of the common aspects of signs as is possible.

This includes things such as:

• The 31 handshapes.

• The various types of movements possible, such as shaking, rotatory, to and fro, pointing, slow and fast etc.

• The various areas of the body.

• The various sign types as far as hand involvement is concerned, such as one-handed, symmetrical double-handed (both hands do exactly the same thing), true double-handed (both hands have the same handshape but are not doing the same thing), two-handed with simple subordinate handshape (the hands doing different things, but the subordinate handshape being something like a flat palm or a fist) and true two-handed signs (each hand does completely independent things that are roughly equal in complexity).

At the same time, it is also important to focus attention on distinguishing signs that you would expect it to have difficulty differentiating. Thus, when signs were selected, one of the possible reasons for selection was its similarity to other signs. This can be seen, for example, in the selection of a large number of gestures around the right side of the face, such as when, think, know, forget, head, and in families of signs with similar gestures, such as shop, spend, buy, money, pen, draw, write and right, is-true, juice. This was in line with the goal of expanding the system to a full-lexicon system, and seeing if it would have trouble telling them apart. As will be shown, this did not appear to be a major obstacle.

While it is impossible to capture all aspects of sign language with anything less than, say, a thousand signs, almost all features can probably be encapsulated in one hundred or so signs.

The above aspect also cross in to the second area which is how we select which signs we are going to include in the project.

At a global level, we wish to meet the above criteria. But there are many sets that can fulfil the above criteria, so in addition, the following method was used for the selection of the signs to be included.

• A family of commonly used signs were included. To establish ``common use'' a number of methods were used. First, a brief dictionary of signs ([JRC88]) was consulted to see what one group of Deaf people believed were important signs. Second, the Auslan Dictionary [Joh89] was scanned, and the importance of each sign in the dictionary was personally assessed on the basis of the author's day-to-day experience.

• It was then seen how many of the signs covered interesting aspects. The Auslan Dictionary was rescanned for signs that were ``interesting'' as far as having aspects that were not found in the common signs. This included many shapes that made use of the rarer handshapes, as well as finding signs that were similar to see how well the system coped with them.

The complete list of signs can be found in Appendix B, and with the explanation of the basis for its selection. Some statistics about the signs are shown in table 4.2.

  
Table 4.2: Some important statistics on the signs used.

In table 4.2, there are a large proportion of signs that are one-handed. This is because most common signs are one-handed, even though one-handed signs only form approximately 40 per cent of the signs in Auslan. A sign was selected on the basis of ``Sign complexity'' because of some interesting aspect, like the large number of critical points or a changing handshape. An ``Uncommon handshape'' indicates the sign contained a handshape that was not found in the other signs selected. ``Similarity'' indicates that the sign was chosen because it was similar to some of the other signs selected. ``Other'' indicates that was chosen because the author felt like it, or some other obscure reason, like its relevance to the investigation.

Another problem was how to uniquely identify each sign. Since there is often more than one sign for a given word or a word does not translate well to a sign, it is not sufficient for a single word to be given. To solve this, a single name was given to each sign, but to uniquely identify it, the Auslan-dictionary reference number is given.

• 4.3.1 Data acquisition protocol