This is a copy of the README file for the MONITOR data sample.

The sample MONITOR observation is provided by kind permission 
of the EPSRC(UK)-funded MONITOR project, "Information filtering
in task-orientated dialogue", awarded to Dr. Ellen Bard (University 
of Edinburgh) and Professor Anne H. Anderson (University of Glasgow).

MONITOR is a variant of the HCRC Map Task designed to determine 
whether or not a very minimal level of feedback from a listener 
will affect what a subject says.  In the experiment, the subject
is given the usual kind of Map Task map and told to explain the
route to someone in another room who has a similar map but with
no route on it.  The subject is eye-tracked (the trace shows up
on the video as a white circle), is told the listener is eye-tracked 
as well (and that the red square indicates where they are looking),
and that the listener can hear them but they can't hear the listener.
On some trials (including this sample), the red feedback square 
movements are programmatic and there is no listener. 

The data is orthographically transcribed and segmented into moves (which
carry out one speaker intention) and transactions (which group moves
into larger, plan-oriented chunks).  References to landmarks have
been identified in the transcription and timestamped.  In addition,
there are several different time-aligned gaze codings in terms of
where the feedback square is and whether the subject is looking at it.
There is an affiliated student performing disfluency coding for the
data but (at Dec 2003) is not available in this released sample (and, 
in fact, one of the advertised codings, the feedback coding, isn't 
properly present).

The codings for this data were constructed before NXT was available
and so not even the structural language codings were made using
tools written with NXT.  The time-aligned codings were made using
The Observer and the transcription, move, and transaction codes with
Transcriber (hacking the output to reflect the tags required).  
NXT gives the project an opportunity to put these two sources of data 
together into an analysis that combines timing and structural properties
without having to write special purpose analysis scripts. In fact, the 
display software for the data is their first chance to see all of the 
codings at the same time aligned against the video and judge how well 
their data creation process has worked.

---------------------------------------------
THE DATA
---------------------------------------------

These are the code explanations provided by the project.  We
added durations to all time-aligned codes (a duration attribute).

-----------------------------------------------------
Behavioral Class 1: Eyegaze F (Where the subject is looking with respect
to the moving feedback square)

not-feedback -- not looking at feedback 
feedback -- looking with 9 grid squares of feedback 
offscreen -- looking off screen 
move-to-f -- gaze moving towards feedback 
away-from-f -- gaze moving away from feedback 

-----------------------------------

Behavioral Class 2: Feedback (What the feedback square is doing)

correct --- feedback is where expected
lagging -- legacy code; no longer used 
wrong -- feedback is looking in the wrong place 
lost -- feedback is lost (whatever that means) 
travel -- feedback is moving to the next landmark 
none -- feedback square isn't present 

-----------------------------------

Behavioral Class 3: Eyegaze i (Where the subject is looking with respect
to the route description)

landmark -- at landmark being talked about 
away -- away from landmark being talked about 
no-gaze -- gaze not visible on screen 
future-landmark -- at future (next) landmark on route 
previous-landmark -- about previous landmark on route 
route -- at non-landmark part of the route 
general-landmark -- at any other landmark 

-----------------------------------

Behavioral Class 4: timing

start -- start of session
finish -- end of session
thirty-second-warning --- warning actually given

The language data comes to us undocumented.  The moves and transaction
types relate to those described in Carletta et al. (1997), Computational
Linguistics.

-----------------------------------------
SOME QUERIES THAT WORK WITH THE DATA
------------------------------------------

There are three ways to look at query results:  use the generic
display interface, the generic search interface, or the search
highlighting facility on the tailored MONITOR display.  These
options are available from a menu if you double-click on monitor.bat/
run monitor.sh. 


Test by typing in the following query:

($m move):

It should return fairly quickly with some matches.  You have to hit
the search button to run it; carriage return does not do it.  You can
cut and paste queries into the window.  The results tabbed-pane will
have a list of the results.  This simply tells you the xml file and 
id attribute for things that match by pointing to them, in a format 
that perhaps isn't the best way to look at the data but which is 
convenient for adding information to the data set and can be knit 
(using lib/knit.xsl) to include the data it points to.  

Here are some more simple queries for this data set to get you started.

// ALL FEEDBACK GAZE CODES
($f feedbackgaze):

Try this same form for any  code named in the metadata.

//MOVES WHOSE TYPE ATTRIBUTE (WHICH IS A STRING) BEGINS WITH I

($m move):$m@type ~ /i.*/

//MOVE/TRANSACTION PAIRS
($m move)($t transaction):($t^$m)

//MOVES WITHIN TRANSACTIONS

($m move) (exists $t transaction) :

Note that there are the same number as for the previous query.  That's good.
That means all the moves are within transactions.

//MOVES WITHIN RETRIEVAL TRANSACTIONS
($m move) (exists $t transaction) : ($t ^ $m) && ($t@type ~ /retrieval_transaction/)

//INSTRUCT MOVES WITHIN RETRIEVAL TRANSACTIONS
($m move) (exists $t transaction) : ($t ^ $m) && ($t@type ~ /retrieval_transaction/) && ($m@type ~ /instruct_move/)

//TRANSACTIONS THAT CONTAIN BOTH INSTRUCT AND EXPLAIN MOVES
($t transaction)(exists $m1 move) (exists $m2 move):($t ^ $m1) && ($t ^ $m2) && ($m1@type ~ /explain_move/) && ($m2@type ~ /instruct_move/)

//SAME, BUT THE EXPLAIN MOVE COMES FIRST
($t transaction)(exists $m1 move) (exists $m2 move):($t ^ $m1) && ($t ^ $m2) && ($m1@type ~ /explain_move/) && ($m2@type ~ /instruct_move/) && ($m1 <> $m2)

// FEEDBACK GAZE CODE TEMPORALLY CONTAINED WITHIN A MOVE
($m move)($f feedbackgaze):($m @ $f) && ($f@label= "feedback")

//SAME, BUT FOR INSTRUCT MOVES ONLY
($m move)($f feedbackgaze ):($m@type ~ /instruct_move/) && ($m @ $f) && ($f@label= "feedback")

//FEEDBACK GAZE CODE OVERLAPS WITH MOVE
($m move)($f feedbackgaze): ($m # $f) && ($f@label= "feedback")

// MOVES DURING WHICH 30 SECOND WARNINGS ARE GIVEN
($m move)(exists $w timing):($m # $w) && ($w@label= "thirty-second-warning")

// MOVES THAT START AFTER 30 SECOND WARNING
($m move)(exists $w timing):($w << $m) && ($w@label= "thirty-second-warning")