W3C home > Mailing lists > Public > public-xg-emotion@w3.org > May 2008

Re: [EMOXG] Deliverable report published as first draft: Emotion Markup Language: Requirements with Priorities

From: Bill Jarrold <jarrold@AI.SRI.COM>
Date: Wed, 7 May 2008 09:10:26 -0700
Message-Id: <D1715097-2457-46BD-92B5-F9490D8E5DBB@AI.SRI.COM>
Cc: Marc Schroeder <schroed@dfki.de>, EMOXG-public <public-xg-emotion@w3.org>
To: Christian Peter <Christian.Peter@igd-r.fraunhofer.de>

Hi Christian,

Thanks for the comments.

A few responses below....

On May 7, 2008, at 4:26 AM, Christian Peter wrote:

> Hi Bill (and all),
> thanks for making these points. The [0,1] approach is the  
> technically most practical and most obvious, which probably is the  
> reason nobody else stumbled over the applicability issues raised by  
> you.

> I think I do agree with you that [0,1] is not sufficient to  
> properly specify an emotion's intensity. At least it might not be  
> feasable for some use cases for reasons provided by you.
> Your class-based approach sounds very attractive to me. Actually  
> {0, ..., 1} are possible classes which would allow easy mapping to  
> the [0,1] standard representations used for other attributes. But  
> other classes (low, medium, high ...) are also imaginable which  
> would need to be specified, probably along with information on how  
> to map them to [0,1] scales.

A given annotator may or may not want to map low, medium, high to [0,1].

Here are two examples of a context in which an annotator would not  
want to map high, low medium to [0,1]:

a) the annotator was doing a quick first pass over a data set.  All  
she/he needs is a quick coarse judgement about emotion intensity and  
does not want to commit to something over precise.  Maybe later, in a  
second or third pass a specific [0, 1] value or range of values could  
be assigned to each of low, medium, high.

b) the annotator is trying to represent natural language of research  
participants.  So, if there is some transcribed audio with e.g.  
Research Participant 42 saying "Oh, yeah, like, my mother was super  
angry at me when I ...." The annotator would merely want to map the  
English "super angry" to an emotion annotation of High.  The  
annotator would not want to convert "super angry" something like 0.75  
because this would be arbitrary putting words into the Research  
Participant's mouth.

> Specification of the classes used (and mapping rules) would need to  
> be done as meta information then, which means that we would need to  
> add it to section 2.2: Meta-information about emotion annotation I  
> guess?

Perhaps.  However, I am thinking that annotators should provide their  
own classes or import these classes from some other ontology.  After  
all there is limitless variety of such classes imaginable (see next  
two paragraphs).  However, what we should require of our markup  
language is that it have vocabulary for associating generic  
qualitative values such as High, Medium Low with specific classes.

E.g. An annotator may wish to import someone's ontology of US Persons  
and use e.g. NativeNewYorker and CaliforniaHippie because each of  
those groups have e.g. different anger thresholds.  A researcher  
interested in animal affect might want to use an ontology of animals  
to allow for different meanings of high , medium and low depending on  
whether one is talking about Sloths, Tasmanian Devils, etc.

I would claim that anticipating the variety of taxonomonies that  
annotators would wish to anchor their qualitative values to is well  
beyond our scope.



> What do others think?
> Christian
> --
> Bill Jarrold schrieb:
>> Hi Marc,
>> Thanks for putting this together.  Looks great.  I have just a few  
>> comments.  Although there have been other discussion of your draft  
>> in recent days, that discussion seems unrelated to my comments so,  
>> rather than replying to them, I am replying directly to your message.
>> On Apr 29, 2008, at 2:33 AM, Marc Schroeder wrote:
>>> Dear all,
>>> the last few weeks have shown little visible activity in this  
>>> group, but Felix Burkhardt and I have been working on the  
>>> requirements document that is intended to reflect the outcomes of  
>>> our discussions during the last few months, regarding mandatory  
>>> vs. optional requirements for an emotion markup language.
>>> So here it is, as a first draft, for you to comment upon:
>>> http://www.w3.org/2005/Incubator/emotion/XGR-requirements-20080429/
>>> I kindly request that any comments to this document should be  
>>> sent to this list. I suggest that we allow for
>>>     two weeks, i.e. until 13 May 2008
>>> time for commenting on this draft. Comments expressed during this  
>>> time can be discussed and, if agreement can be reached, will be  
>>> taken on board. If no objections from group members are raised,  
>>> the report will be considered final by the above date.
>> My comments have to do with requirement for expressing amounts.   
>> These comments apply to Core 6 and Core 8.
>> Core 6 says "The emotion markup must provide an emotion attribute  
>> to represent the intensity of an emotion.  The intensity is a  
>> dimension with values in [0,1]"
>> Core 8 says "The emotion markup shuld provide emotion attributes  
>> to represent the various kinds of regulation.  The value of these  
>> attributes should be in [0,1]."
>> I consider a value range of [0,1] to be a unitless value range.
>> This is good.  But I think we want to give annotators more options  
>> for these values than just [0,1].
>> One reason that merely [0,1] may not be enough is this: Is it  
>> really true that there is a maximum amount of a particular  
>> emotion?  A maximal amount of experiencable joy?  How depressing !  
>> (-:  Having a fixed max or min seems to be taking a theoretical  
>> stand that might be quite valid for some theories but not all  
>> theories.
>> Okay, so, the basic idea is that we should allow additional means  
>> to express intensity. In particular we should allow [a]  
>> qualitative values [b] qualitative values relative to a particular  
>> class [c] a partial ordering of values [d] unit based values.
>> Do we need [a] AND [b] AND [c] AND [d]?  I don't know.  Lets  
>> consider each one as a separate amendment.  Maybe we'll want just  
>> [a] but not [b], [c], or [d].  Let's discuss which if any of these  
>> additions we want.
>> I will now describe each of [a], [b], [c] and [d] and try to  
>> provide an evocative use case (expressed in English)
>> [a] Qualitative Values
>> In a mushy area sometimes all we can assert are values like high,  
>> medium or low.  E.g.  "Fred felt a lot of anger."  How much is a  
>> lot?  .7, .8, .9?  The value can be quite arbitrary.  Thus we want  
>> a set of qualitative values such as "high", "medium", "low", "very  
>> high", "very low" etc......[Sure, a given researcher should be  
>> able map high to .9 or 0.77 if he decides to later on down the  
>> road in his/her markup activity.  So, there would also need to be  
>> vocab to optionally map such vocab to actual numbers.]
>> [b] qualitative values relative to a particular class
>> This is basically [a] except you can relativize the qualitative value
>> to a particular class.  E.g.
>> "Fred felt a low amount of anger for a New Yorker"
>> "Joe felt a low amount of love for a California Hippie During the
>> Summer of Love"
>> "Subject24601 expressed a medium amount of surprise for Sample Set 35
>> As Rated by Graduate Student Joe"
>> We would do this kind of thing (i.e. [a] or [b]) all the time with  
>> a knowledge rep /
>> ontology system such as Cyc or the Knowledge Machine (and the
>> Component Library associated with it) [references happily  
>> furnished if
>> you ask].  For example the express the concept that "Fred is short  
>> for
>> a basketball player and tall for a pigmy." (We'll really we'd say  
>> that
>> the value of the height property for Fred was Low with respect to the
>> class Basketball Player but High for the class Pigmy".  It stands  
>> to reason that if this kind of scheme has been developed for these  
>> projects for quantities like height we should also do it for emotion.
>> [Of course, the user of Cyc or KM can specify stuff like "Short  
>> for a Basketball Player is 5 feet 9 inches." as well]
>> [c] a partial ordering of values
>> The idea here is we want to handle cases where we do not wish to nail
>> down any particular value for an emotion intensity but do wish to say
>> that one value is greater or less than another.
>> E.g. "The level of Fred's happiness today is higher than it was
>> yesterday."
>> [d] unit based values.
>> I don't know of any scheme for quantifying emotion intensity on some
>> sort of scale.  However, we should allow (and encourage) researchers
>> who have devised such a scale to use use it.  This scheme would allow
>> the annotator to associate a value with a particular set of units  
>> of his/her
>> chosing."
>> E.g. "Using his new instrument, Joe the Scientist measured Fred's
>> happiness as 75 felicitons."
>>  "Using her new instrument, Sally the Scientist measured  
>> Fred'shappiness
>> as 3.2 kilofelicitons."
>>  "Using her new instrument, Jing the Scientist measured Fran's  
>> pride as -1.98 shameitrons."
>> Well, the examples might sound a little silly, but hopefully they  
>> illustrate the need for what
>> I am making the case for.  We can make the examples more sober and  
>> scientifically plausible in
>> due course as necessary.
>> Hope this helps,
>> Bill
>>> This document gives us a pretty concrete starting point for  
>>> thinking about syntax now. I believe we should come to a  
>>> conclusion about which format to use (XML vs. RDF vs. OWL) in  
>>> May, and start drafting a syntax specification from June onwards.
>>> Does this sound reasonable? Let's get back to work! :-)
>>> Best regards,
>>> Marc
>>> -- 
>>> Dr. Marc Schröder, Senior Researcher at DFKI GmbH
>>> Coordinator EU FP7 Project SEMAINE http://www.semaine-project.eu
>>> Chair W3C Emotion ML Incubator http://www.w3.org/2005/Incubator/ 
>>> emotion
>>> Portal Editor http://emotion-research.net
>>> Team Leader DFKI Speech Group http://mary.dfki.de
>>> Project Leader DFG project PAVOQUE http://mary.dfki.de/pavoque
>>> Homepage: http://www.dfki.de/~schroed
>>> Email: schroed@dfki.de
>>> Phone: +49-681-302-5303
>>> Postal address: DFKI GmbH, Campus D3_2, Stuhlsatzenhausweg 3,  
>>> D-66123 Saarbrücken, Germany
>>> -- 
>>> Official DFKI coordinates:
>>> Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
>>> Trippstadter Strasse 122, D-67663 Kaiserslautern, Germany
>>> Geschaeftsfuehrung:
>>> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
>>> Dr. Walter Olthoff
>>> Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
>>> Amtsgericht Kaiserslautern, HRB 2313
> -- 
> ---------------------------------------------------------------------- 
> --
> Christian Peter
> Fraunhofer Institute for Computer Graphics Rostock
> Human-Centered Interaction Technologies
> Joachim-Jungius-Str. 11, 18059 Rostock, Germany
> Phone: +49 381 4024-122, Fax: +49 381 4024-199
> email: christian.peter@igd-r.fraunhofer.de
> ---------------------------------------------------------------------- 
> --
> Problems with the electronic signature? Please load the current root
> certificate of the Fraunhofer-Gesellschaft into your browser!
> (http://pki.fraunhofer.de/EN/)
> ---------------------------------------------------------------------- 
> --
Received on Wednesday, 7 May 2008 16:36:41 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 19:52:15 UTC