Re: [agtk-devel] Re: Question about events and format.

[Steven B. <sb...@cs...>]

Dear Fredrik,

> The first question addresses the need for labeling of events. 
> In acoustical phonetic analysis, there are quite a few 
> situations where the object in question is not a time interval, 
> but a point in time. Such example in phonetics is the onset of 
> voicing, turing points in the formant track or the articulatory 
> release of a plosive. How do you propose that this kind of 
> annotation should be handles in the AG library, given the 
> definition of the graph.

Haejoong has described two good approaches to this.  There are
two others to consider.

a) Put the label of the durationless event on an arc which ends (or begins)
at that point in time.  This is what our ToBI example does - the break
indices are specified as properties of the preceding word, even though they
pertain to the strength of the prosodic break between words.

Annotation graphs as a framework for multidimensional linguistic data
analysis http://www.ldc.upenn.edu/sb/home/publications.html#9907003

b) Saying that phonetic events are durationless is an abstraction.  It may
be more faithful to represent them using small intervals, and to use
annotations that span a small amount of time.

> The second issue regards format extentions. In my project, I 
> use the Praat software since it's annotation format facilitates 
> the distinction between intervalls and points.
> Therefore, in order for me to use the AGTK, I would have to 
> have an object that parses the Praat files and inherits from 
> agfio. The thing is that I have already constructed a similar 
> object in Python (using regexp), which would mean that 
> writing such a module should be close to trivial.
> However, since I guess that I am not the only one interrested 
> in such an extension, it would probably be better if that 
> module would be developed within the development team of 
> AGTK. The quality and consistency would probably increase, 
> and changes in the API would probably be easier to transfer 
> to all modules that way.
> Therefore, if you think that it is a good idea to include 
> support of the Praat program to AGTK, I suggest that I send 
> you Perl/Python code (from my project) including regexp in 
> order for you to be able to include support for Praat 
> annotations in AGTK. That way, I and many others would be 
> able to use the AG library for the Praat format as well.

We should definitely support Praat format, and will add it to our long list
of features to add.  Note that there's nothing inherently difficult about
loading from a format that represents instants, and representing them in
any of the four ways we've described.  Steve Cassidy has already had to
deal with this in representing EMU data using annotation graphs
[http://www.shlrc.mq.edu.au/emu/].

The quickest way to go (as in all open source projects) would of course be
for you to contribute code to the file I/O library.  Would you consider
doing that?  Haejoong could then look it over and it would show up in our
next release?

Thanks for getting in touch, and we hope you decide to use the library and
contribute to its further development.

Steven Bird

--
Steven Bird        Email: <sb...@cs...>  Web: http://www.cs.mu.oz.au/~sb/
A/Prof, Dept of Computer Science, University of Melbourne, Vic 3010, AUSTRALIA
Senior Research Assoc, Linguistic Data Consortium, University of Pennsylvania