See also: IRC log
We have nearly completed the cycle of initial discussions. Today, global metadata and the sampling mechanism were discussed. For global metadata, we have identified two major options: simple key-value lists or a meta element below which arbitrary content can be placed. This will need further discussion. For the sampling mechanism, there was preliminary agreement on a format which would avoid introducing textual content by placing the trace values in an attribute, and which allows for representing confidence that changes over time.
Discussion on representing global metadata, based on Felix' document:
http://lists.w3.org/Archives/Public/public-xg-emotion/2008Aug/0015.html
Felix refers to voiceXML spec:
<felix> http://www.w3.org/TR/2004/REC-voicexml20-20040316/#dml2.3.5
Marc mentions that the examples given by Felix aren't global but rather specific to a particular annotation
Christian agrees. Thinks that example 2 could be global as well, if e.g. this one classifier is used throughout the experiment
To make clear what is meant with "global metadata": they should refer to e.g. technical settings (sensors used, version of used software, local circumstances (lab or outside study, temperatures...)
Marc: could we use approaches of other markup languages? RDF, EMMA?
<marc> http://www.w3.org/TR/emma/#s4.1.4
Basically, two options: simple list of key-value items, as suggested by Felix, or a parent element with arbitrary subelements, as in EMMA.
Felix: suggests to discuss this in France
On to sampling mechanism, based on Marc's prepared document
http://lists.w3.org/Archives/Public/public-xg-emotion/2008Aug/0028.html
To clarify, this is to represent episodes with changing scale-values (e.g. valence, intensity...):
<trace freq="10ms">
0.1 0.1 0.2 0.2 0.2 0.5 0.7 0.7 0.8 0.7 0.7
</trace>
Christian likes it
Felix: wonders if it is general enough
Marc: it is similar to the inkML approach
<Nestor> I like it, but I have a question: frequency or time-period?
Nestor: points out that frequency is measured in Hz, not ms
Marc: True, we have to think of a new attribute name, or value here
Christian thinks it might be usable for storing episodes in automatic analysis as well
Marc: asks if confidence should be in here as well and if can change as well
Christian: agrees
<Nestor> I agree
Marc: is it a problem that we use plain text here within the tags? To avoid that, we could write:
<marc> <trace freq="100Hz" samples="0.3 0.3 0.4 0.3 0.3 0.3 0.2 0.3 0.3 0.3 0.2" />
Marc: we could do the same with confidence now, so we have a trace tag with a confidence associated with that dimension
Christian: we could simply add confidence as an attribute to the trace tag, like confidence ="0.3 0.3 0.4 0.3 0.3 0.3 0.2 0.3 0.3 0.3 0.2"
<emotion>
<dimensions>
<arousal>
<trace freq="100Hz" samples="0.3 0.3 0.4 0.3 0.3 0.3 0.2 0.3 0.3 0.3 0.2"
confidence="0.3 0.3 0.4 0.3 0.3 0.3 0.2 0.3 0.3 0.3 0.2" />
</arousal>
</dimensions>
</emotion>
Felix: what about the confidence being equal for all the trace
Marc: it would be on the same level as arousal then
Christian: it should be okay to provide for both, confidence at the arousal level for the whole emotion, and a confidence for the specific trace values, in case confidence changes for particular trace values.
Marc: we can discuss this detail lateron
Marc: suggests to have the option to use traces for any scale value in the emotion core section of the requirements
Christian: likes this
Christian: it can be used for dimensions, appraisal, action tendency and intensity
<felix> wants to shortly discuss the links to the rest of the world again, the sample idea:http://lists.w3.org/Archives/Public/public-xg-emotion/2008Aug/0024.html
(Example5)
Marc: what is the time base here? Maybe expressing duration in "samples" is too system-centric?
should discuss this in France as well