<div dir="ltr">
<p class="MsoNormal" style="line-height:normal"><b><span style="font-size:13.5pt;font-family:"Times New Roman","serif""><a href="http://www.speech.kth.se/maptask/">KTH Human-Computer Map Task Corpora</a></span></b></p>
<p class=""><span style="font-family:"Times New Roman","serif"" lang="EN-GB">A common procedure in
modelling human-like dialogue systems is to collect data on human–human
dialogue and then train models that predict the behaviour of the interlocutors.
However, we think that it might be problematic to use a corpus of human–human
dialogue as a basis for implementing dialogue system components. One problem is
the interactive nature of the task. If the system produces a slightly different
behaviour than what was found in the original data, this would likely result in
a different behaviour in the interlocutor. Another problem is that humans are
likely to behave differently towards a system as compared to another human
(even if a more human-like behaviour is being modelled). Yet another problem is
that much dialogue behaviour is optional and therefore makes the actual
behaviour hard to use as a gold standard.</span></p>
<p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">The KTH Human-Computer Map Task Corpora has been collected as part of our
efforts towards building data-driven models for Response Location Detection:
detecting when in the user's speech is it appropriate for a system to provide a
response (</span>Skantze,
2012<span style="font-family:"Times New Roman","serif"" lang="EN-GB">; </span>Meena
et al., 2013a<span style="font-family:"Times New Roman","serif"" lang="EN-GB">; </span>Meena
et al., 2014<span style="font-family:"Times New Roman","serif"" lang="EN-GB">). Map Task is a common experimental paradigm for studying human-human
dialogue, where one subject (the information giver) is given the task of
describing a route on a map to another subject (the information follower) (</span>Anderson
et al., 1991<span style="font-family:"Times New Roman","serif"" lang="EN-GB">). For example, </span>Cathcart
et al. (2003)<span style="font-family:"Times New Roman","serif"" lang="EN-GB"> used the HCRC Map Task data to train a shallow model for prediction of
backchannel continuers in an interaction. Similarly, </span>Koiso
et al. (1998)<span style="font-family:"Times New Roman","serif"" lang="EN-GB"> conducted the Japanese Map Task dialogue and presented their findings on
turn-transition and backchannel relevant places. In the KTH Human-Computer Map
Task the user acts as the giver and the system (a dialogue system) as the
follower. Our main objective behind this was to be able to collect corpus and build
data-driven models for detecting when in the user's speech is it appropriate
for a system to provide a response. The nature of the response could be
anything: a back-channel, a clarification request or a question. However, it
was not the objective of the study to identify the nature of the response. We
only wanted to predict the appropriateness in terms of timing. <br></span></p><p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB"><br></span></p>
<p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">We believe the data could be useful for researchers in the dialogue system
community and have now made it public. The corpora can be downloaded from <a href="http://www.speech.kth.se/maptask/">http://www.speech.kth.se/maptask/</a>
. It comprises of two data-sets, the first is the Training-Set, which was
collected to train various data-driven models of RLD (</span>Skantze,
2012<span style="font-family:"Times New Roman","serif"" lang="EN-GB">). The trained model was then integrated into the same system (used for
data collection) and evaluated through new users in the same Map Task interaction
(</span>Meena
et al., 2013a<span style="font-family:"Times New Roman","serif"" lang="EN-GB">; </span>Meena
et al., 2013b<span style="font-family:"Times New Roman","serif"" lang="EN-GB">; </span>Meena
et al., 2014<span style="font-family:"Times New Roman","serif"" lang="EN-GB">). The interaction data collected from the user evaluation comprises the
second data-set.</span><span style="font-size:11pt;font-weight:normal" lang="EN-GB"><br></span></p><p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-size:11pt;font-weight:normal" lang="EN-GB"><br></span></p><p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-family:times new roman,serif"><font><span style="font-weight:normal" lang="EN-GB">The corpora is released only for research purpose and
presentation at scientific conferences. If you use this corpus in your
research, please cite the following article: </span></font></span>
</p><ul type="disc"><li class="MsoNormal" style="line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">Meena, R., Skantze, G., & Gustafson, J. (2014). Data-driven Models
for timing feedback responses in a Map Task dialogue system. Computer
Speech and Language, 28(4), 903-922.</span></li></ul>
<p class="MsoNormal" style="text-align:left;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">Please contact Raveesh Meena (<a href="http://raveeshATcsc.kth.se">raveeshATcsc.kth.se</a>) or <br></span></p><p class="MsoNormal" style="text-align:left;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">Gabriel Skantze
(gskantzeATspeech.csc.kth) if you have any questions.</span></p>
<p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB"><br></span></p><p class="MsoNormal" style="text-align:justify;line-height:normal"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">Further instructions about the corpora are available at </span><a href="http://www.speech.kth.se/maptask/"><span style="font-family:"Times New Roman","serif"" lang="EN-GB">http://www.speech.kth.se/maptask/</span></a><span style="font-family:"Times New Roman","serif"" lang="EN-GB"> </span></p>
<p class=""><span class="">Anderson, A., Bader, M., Bard, E., Boyle, E.,
Doherty, G., Garrod, S., Isard, S., Kowtko, J., McAllister, J., Miller, J.,
Sotillo, C., Thompson, H., & Weinert, R.</span> (1991). The HCRC Map Task
corpus. <span class="">Language and Speech, 34</span>(4),
351-366.</p>
<p class=""><span class="">Cathcart, N., Carletta,
J., & Klein, E.</span> (2003). A shallow model of backchannel continuers in
spoken dialogue. In <span class="">10th Conference of the
European Chapter of the Association for Computational Linguistics</span>.
Budapest.</p>
<p class=""><span class="">Koiso, H., Horiuchi, Y.,
Tutiya, S., Ichikawa, A., & Den, Y.</span> (1998). An analysis of
turn-taking and backchannels based on prosodic and syntactic features in
Japanese Map Task dialogs. <span class="">Language and Speech,
41</span>, 295-321.</p>
<p class=""><span class="">Meena, R., Skantze, G.,
& Gustafson, J.</span> (2013a). A Data-driven Model for Timing Feedback in
a Map Task Dialogue System. In <span class="">14th Annual
Meeting of the Special Interest Group on Discourse and Dialogue - SIGdial</span>
(pp. 375-383). Metz, France.</p>
<p class=""><span class="">Meena, R., Skantze, G.,
& Gustafson, J.</span> (2013b). The Map Task Dialogue System: A Test-bed
for Modelling Human-Like Dialogue. In <span class="">14th
Annual Meeting of the Special Interest Group on Discourse and Dialogue -
SIGdial</span> (pp. 366-368). Metz, France.</p>
<p class=""><span class="">Meena, R., Skantze, G.,
& Gustafson, J.</span> (2014). Data-driven Models for timing feedback
responses in a Map Task dialogue system. <span class="">Computer
Speech and Language, 28</span>(4), 903-922.</p>
<p class=""><span class="">Skantze, G.</span>
(2012). A Testbed for Examining the Timing of Feedback using a Map Task. In <span class="">Proceedings of the Interdisciplinary Workshop on
Feedback Behaviors in Dialog</span>. Portland, OR.</p>
best<br>Raveesh<br clear="all"><div><br>-- <br><div class="gmail_signature">Raveesh Meena<br>PhD / Graduate Student in CS<br><br>Department of Speech, Music and Hearing<br>Royal Institute of Technology (KTH) <br>Lindstedtsvägen 24<br>SE-100 44 Stockholm,<br>Sweden<br><br>Phone: +46-(0)-8-790 7872<br>Fax : +46-(0)-8-790 7854<br><br>Email: raveesh[at]<a href="http://csc.kth.se" target="_blank">csc.kth.se</a><br><a href="http://www.speech.kth.se/%7Eraveesh" target="_blank">http://www.speech.kth.se/~raveesh</a><br></div>
</div></div>