######################################
#                                    #
#  The SamPDA cursive word database  #
#                                    #
#           - release 1.0 -          #
#   containing sets SamPDA_[0,1,2]   #
#                                    #
#          - DOCUMENTATION -         #
#                                    #
######################################

## - Contact information - ##

Collector: Jakob Sternby
			  Centre for Mathematical Sciences
			  Lund Institute of Technology
           jakob@maths.lth.se

## - SamPDA Concept - ##

For evaluating a cursive recognition engine one would 
optimally want to try all plausible letter combinations. 
Aiming at this bigram-, trigram-, and tetragram letter 
statistics were generated from the dictionary. This was then
used to produce a compact list of words containing as many 
as possible of the statistical letter combinations. 

The SamPDA data sets collected here are samples from this 
list. The SamPDA collection strategy also involves playing 
audio files of the words to be written instead of displaying
a font. This is to avoid influencing the way the user writes
by showing the word in a computer font. Each of the sets 
SamPDA_[0,1,2] contains 20 different words from the list 
described above. 

All of the data was collected on a COMPAQ iPAQ PDA, with a
SamPDA software that played an audio file of the word to be
written as described above. Only a baseline in a box was
displayed for reference. This baseline corresponds to the
y-coordinate 64.

## - Database contents - ##

This release of the SamPDA contains three different datasets
based on lists of 20 different words. Due to some 
misinterpretation of the audio file a few other words have
also been written, such as 'lamp' or 'lamb' instead of 'land'
, 'wart' instead of 'want', 'rage' instead of 'range'. Some 
misspellings are also present such as 'business' 
written as 'buisness' and 'address' as 'adress'.

 -------------------------------------------------------
| Dataset  |  Numer of writers |  Unique words  | Total |
|-------------------------------------------------------
| SamPDA_0 |         29        |       23       |  580  |
| SamPDA_1 |         14        |       23       |  280  |
| SamPDA_2 |          7        |       23       |  140  |
 -------------------------------------------------------

Since this database was not originally created in the UNIPEN
format it is basically only the coordinate information that
has been included into each datafile. However some extra
information can be extracted from the filenames which have
the following structure.

## - Filename structure - ##

The number of the extension _[0, 1, 2] in the 
SamPDA_[0, 1, 2] set corresponds to the [wordlist] marker
in each filename.

sampda_[lang]_[type][wordlist]_[writername].unipen

# lang #

ae = American English

# type #

mw = users were instructed to write neatly in 'as much 
     cursive writing as possible'. Plenty of writers without
	  cursive tradition are present. 

# wordlist #

The number [0, 1, 2] corresponds to the three different
wordlists used. Each wordlist contains 20 words.

# writername #

The initials of the test subject.