4 Conclusion and Outlook

The present report presents a systematisation and formalisation of the syntax of the CoGesT 1.0 gesture transcription notation, resulting in the simplified and hierarchically structured human-readable and machine-readable CoGesT 1.1 notation.

The syntax formulations are intended to be used as follows:

Regular expressions:

corpus pre-processing (to be implemented in Perl).

BNF:

semantic interpretation, lexical induction (to be implemented in Prolog) and lexical inference (to be implemented in DATR).

XML:

archiving and interchange (to be implemented with XML tools, and imported into the TAMINO XML database).

An attribute-value representation for CoGesT 1.1 syntax, based on this preliminary formalisation of syntax and implementation in XML, is being developed in order to provide a platform for integrating the description of gesture into a general grammatical description and for the description of CoGesT semantics will be developed later.

The requirements defined at the outset of the report have been fulfilled: provision of a formal basis for a lexical representation of gestures, contribution towards optimising the CoGesT transcription system, checking of the consistency of the CoGesT system, both in definition and in practical use, mapping of CoGesT 1.0 on to a formal language CoGesT 1.1 which is a human-readable, machine-readable and automatically processable formal language for gesture modelling.

In the present report, a formal reconstruction of the syntax of the CoGesT 1.0 gesture transcription notation system was performed, and a proposal for simplifying and introducing a well-motivated grouping of vector elements was proposed. Syntaxes in regular expression, BNF and XML DTD notations were developed.

The ``semantics'' of the CoGesT notation are complex (cf. Gibbon et al. (2003); Thies (2003); Gut et al. (2002)), and ongoing work is concerned with further formalisation in this area.

1. a ``meaning'' in the conventional sense, e.g. of an emblem gesture meaning ``farewell'', ``the bill please'', etc.;
2. a mapping to the motor and visual properties of the gestue;
3. a mapping to the concurrent linguistic features vectors, amounting to an autosegmental subvector of segmental phonetic features, as well as vectors defining categories at morphological, phrasal, textual and discourse levels.

Each of these cases implies the provision of a specific range of `possible worlds' for its interpretation. However, first case will be handled conservatively in terms of intuitively interpretable attribute-value structures, at least initially. The second and third cases are being fully formalised within an event logic framework and Time Map framework, following Carson-Berndsen (1998).