W3C home > Mailing lists > Public > www-multimodal@w3.org > September 2012

Fwd: AW: Call for Implementations of EmotionML published

From: abe kazemzadeh <kazemzad@usc.edu>
Date: Wed, 29 Aug 2012 05:45:56 -0700
Message-ID: <CAFaJ0ryQdj-esbu6vr9t9=2Sbg=TJdr4RPumg-dtBc8WKhwsmQ@mail.gmail.com>
To: www-multimodal@w3.org
Hi Kazuyuki, and other multimodal group list members,

Thanks for adding me to the list. I'm forwarding the original,
off-list thread, but I'll summarize for those reading this for the
first time, and also explain more about the spacing issue.

I'm providing an implementation report for EmotionML based on an
application I've worked on for my dissertation, a dialog system called
Emotion Twenty Questions (EMO20Q), where a dialog agent tries to guess
the emotion that the user is thinking about (as opposed to actually
feeling the emotion, which may or not be the case).  The demo of the
application is at http://ark.usc.edu/~abe/wsgi_questioner .

The implementation report is attached (and a human readable version as
well).  The issue that I noticed was that spaces in the name attribute
of a vocabulary item cause the xml file to fail to validate against
the xsd file.  Since the application uses colloquial names of
emotions, they may have spaces, like "let down". I noticed this when
using the `xml` commandline tool (xmlstarlet,

xml   val  -e --xsd emotionml.xsd emotionml.xml
emotionml.xml:65.15: Element
'{http://www.w3.org/2009/10/emotionml}item', attribute 'name': 'let
down' is not a valid value of the atomic type 'xs:NMTOKEN'.
(these files can be found at

In the implementation report I'm attaching, I ignored this error by
changing the emotion "let down" to "letDown".  It turns out this might
not be desirable behavior (in which case I should change it to fail
for test case 100).  I look forward to getting feedback about this
possible issue, but I should say that I'm on vacation right now, so I
might not be able to reply immediately.

---------- Forwarded message ----------
From: Kazuyuki Ashimura <ashimura@w3.org>
Date: Tue, Aug 28, 2012 at 2:13 PM
Subject: Re: AW: Call for Implementations of EmotionML published
To: kazemzad@usc.edu
Cc: Felix.Burkhardt@telekom.de, Deborah Dahl

Hi Abe,
Just CCing Felix and Debbie (the WG chair)

I've just added you to the www-multimodal@w3.org list, so
you don't have to send a "subscribe" message to the W3C
email management system.

However, your original message which included the XML version
and PDF version of your implementation report is not archived
in the W3C email archive though Felix's response [1] is archived.

So could you please resend your report to the list for archival

[1] http://lists.w3.org/Archives/Public/www-multimodal/2012Aug/0009.html



On 08/28/2012 09:55 PM, Felix.Burkhardt@telekom.de wrote:
> Thanks Abe
> Yes please, submit to the list
> Cite Kaz:
>> Could you please send your implementation report to the MMI public list (www-multimodal@w3.org) as the EmotionML Candidate Recommendation announcement [1] says?
> [1] http://lists.w3.org/Archives/Public/www-multimodal/2012May/0010.html
> And interesting point you raise about the "no space" requirement, I myself was not aware of this and there is an example ("being hurt") in the official WD by Marc and Catherine [2] that includes a space.
> Also it says in the spec [3]:
> name: a name for the item, used to refer to this item. An <item> MUST NOT have the same name as any other <item> within the same <vocabulary>.
> So, I don't see a problem with spaces in names for the vocabulary.
> Cheers,
> Felix
> [2]http://www.w3.org/TR/2011/WD-emotion-voc-20110407/
> [3] http://www.w3.org/TR/2012/CR-emotionml-20120510/#s3.1.2
> Von: abe.kazemzadeh@gmail.com [mailto:abe.kazemzadeh@gmail.com] Im Auftrag von abe kazemzadeh
> Gesendet: Dienstag, 28. August 2012 09:07
> An: Burkhardt, Felix
> Cc: christian@becker-asano.de; patrick.gebhard@dfki.de; tim.llewellynn@nviso.ch; r.cowie@qub.ac.uk; begolie@ornl.gov; schuller@tum.de; marc.schroeder@dfki.de
> Betreff: Re: Call for Implementations of EmotionML published
> Hi Felix and all,
> Here's an implementation report for the EMO20Q agent that you demoed.
> I'm not sure if this is the right format for the report... Let me know if it need any
> fixing. I also have a human readable version (pdf) as well as the xml
> format given in http://www.w3.org/2002/mmi/2012/emotionml-irp/ .
> The only issue is that there were a few words in our vocabulary with
> spaces (eg, "let down"). As a computer readable format, it is
> possible to record this as "letDown", but if the no space requirement
> is not strongly motivated, I think it might make sense to accept
> spaces (e.g., "pissed off", "culture shock", or maybe "deja vu". It
> seems like multiword emotions eventually get lexicalized, like
> "homesick" or "carefree", but one could make the case for containing
> space in order to make the format more general).
> I wasn't sure if I should submit this report to the list mentioned on the specification site, www-multimodal@w3.org.  I haven't been following this list, so please let me know if should join and submit via the list or if the organizers here are collecting them off several threads, like this one.
> I'll be on vacation and traveling for the next 3 weeks but I should be
> able to get to email mostly within a day or so.
> Thanks,
> Abe
> On Thu, Aug 23, 2012 at 9:23 AM, abe kazemzadeh
> <abe.kazemzadeh@gmail.com> wrote:
>> Hi Felix,
>> On Thu, Aug 23, 2012 at 1:18 AM, <Felix.Burkhardt@telekom.de> wrote:
>>> Congratulations, I just played it and it only took 12 questions to guess my emotion (jealousy). There was only one strange situation, when I first answered "no" on the question "is it like sadness?" and the next question was "is it sadness?".
>> Thanks for playing the emo20q demo! I'm glad it guessed correctly,
>> but you're right, there are some non-sequitur responses. I'm still
>> trying to decide whether more data or an improved algorithm will be
>> the best way to fix these...
>>> So when will you send the report? Are you clear on the format?
>> I hope to send it soon. I've reviewed the report requirements
>> (http://www.w3.org/2002/mmi/2012/emotionml-irp/ ). It seems clear, so
>> no questions at the moment, but if there are any example reports
>> available that might help.
>>> You're all aware we extended the deadline to mid September?
>> I wasn't aware of the extension, but that's great.
>>> I'll be on the Eusipco conference in Bukarest next week in case anyone is also there and we could meet.
>> Have a good trip. I just checked with Shri, unfortunately no one from
>> SAIL is going to be at Eusipco this year.
>> Take care,
>> Abe
>>> -----Ursprüngliche Nachricht-----
>>> Von: abe kazemzadeh [mailto:abe.kazemzadeh@gmail.com]
>>> Gesendet: Mittwoch, 22. August 2012 22:08
>>> An: Burkhardt, Felix
>>> Cc: christian@becker-asano.de; patrick.gebhard@dfki.de; tim.llewellynn@nviso.ch; r.cowie@qub.ac.uk; begolie@ornl.gov; schuller@tum.de; marc.schroeder@dfki.de
>>> Betreff: Re: Call for Implementations of EmotionML published
>>> Hi Felix,
>>> I'm sorry that I was delayed with the implementation report for my use of EmotionML. I just recently made a usable demo and if I could still submit a report, I would be very glad if I could help with the EmotionML effort. The demo is at http://ark.usc.edu/~abe/wsgi_questioner . It basically uses the EmotionML vocabulary idiom with a list of 110 emotion words for implementing emotion twenty questions (EMO20Q). After each question, the agent updates the probabilities/potentials associated with each word and hopefully the belief update will narrow down the candidate words (lower the entropy of the categorical distribution over the
>>> vocabulary) so that the agent can guess the emotion word in less than
>>> 20 questions.
>>> I would have submitted the report earlier, but it just wasn't ready.
>>> Actually the EmotionML helped make the emo20q demo practically usable because earlier I had been serializing a big object in between the http requests, but now I only serialize an EmotionML vocabulary (with associated weights) and a dialog turn history.
>>> Thanks,
>>> Abe
>>> On Mon, Jun 18, 2012 at 2:37 AM, <Felix.Burkhardt@telekom.de> wrote:
>>>> Hi prospective implementers of EmotionML This is a reminder to deliver
>>>> Implementation Reports until 10th August, I attach my own implementation report as a sample.
>>>> Marc has sadly left us and I'm the new editor of EmotionML, so if you have any questions I'd be happy to assist you.
>>>> It would be great to get some feedback on who actually works on implementation reports and when you think you can deliver.
>>>> Regards,
>>>> Felix
>>>> I include Marc's last mail(s)
>>>> -----Original mail-----
>>>> Von: Marc Schroeder [mailto:marc.schroeder@dfki.de]
>>>> Gesendet: Freitag, 11. Mai 2012 09:16
>>>> An: Burkhardt, Felix; abe.kazemzadeh@gmail.com;
>>>> christian@becker-asano.de; patrick.gebhard@dfki.de;
>>>> tim.llewellynn@nviso.ch; r.cowie@qub.ac.uk; begolie@ornl.gov
>>>> Betreff: Call for Implementations of EmotionML published
>>>> Dear prospective implementors of EmotionML 1.0,
>>>> the W3C has published the Candidate Recommendation and the Call for Implementations of EmotionML yesterday:
>>>> http://www.w3.org/News/2012#entry-9449
>>>> The specification as such has not changed much since the previous version, just some clarifications here and there:
>>>> http://www.w3.org/TR/2012/CR-emotionml-20120510/
>>>> The most relevant bit for you guys will be the Implementation Report Plan, in which we have basically listed as verifiable assertions the various properties that an implementation of different aspects of EmotionML should guarantee:
>>>> http://www.w3.org/2002/mmi/2012/emotionml-irp/
>>>> A key issue here might be to clarify whether you are implementing a "producer" and/or a "consumer" of EmotionML. In the Introduction of the Implementation Report Plan, e have tried to give clear descriptions what it means for a producer and a consumer to "pass", "fail" or "not-impl" a given assertion.
>>>> I'll be happy to work with you in the next few weeks to clarify what needs to be done so that your implementation reports can help move EmotionML forward. Simply get back to me with any questions you have.
>>>> I'd say if you think the question is of relevance to other prospective implementors, it should be OK to "reply all" to this email.
>>>>> I have just completed an implementation of an EmotionML checker in java, which performs a full validation of input documents with respect to all assertions in the IRP. Aspects of the specification that cannot be >verified through schema validation are verified through java code.
>>>>> This means that if the tool accepts any given document (or document fragment), I am reasonably confident it can be treated as valid EmotionML.
>>>>> I have placed the code in the public domain:
>>>>> https://github.com/marc1s/emotionml-checker-java
>>>> Mit freundlichen Grüßen / Viele Grüße / Best Regards
>>>> Felix Burkhardt
>>>> Deutsche Telekom AG
>>>> T-Labs (Research & Innovation)
>>>> Dr. Felix Burkhardt
>>>> Winterfeldtstr. 21, 10781 Berlin
>>>> +4930835358136 (Tel.)
>>>> +4952192100512 (Fax)
>>>> E-Mail: felix.burkhardt@telekom.de
>>>> www.telekom.com
>>>> Erleben, was verbindet.
>>>> Deutsche Telekom AG
>>>> Aufsichtsrat: Prof. Dr. Ulrich Lehner (Vorsitzender)
>>>> Vorstand: René Obermann (Vorsitzender), Dr. Manfred Balz, Reinhard
>>>> Clemens, Niek Jan van Damme, Timotheus Höttges, Claudia Nemat, Prof.
>>>> Dr. Marion Schick
>>>> Handelsregister: Amtsgericht Bonn HRB 6794 Sitz der Gesellschaft: Bonn
>>>> WEEE-Reg.-Nr. DE50478376
>>>> Große Veränderungen fangen klein an - Ressourcen schonen und nicht jede E-Mail drucken.

Kaz Ashimura, W3C Staff Contact for Web&TV, MMI and Voice
Tel: +81 466 49 1170

Received on Saturday, 1 September 2012 11:43:20 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 22:42:44 UTC