To obtain a zip file containing the handwriting samples, please send the following information via email to math-corpus@scg.cs.uwaterloo.ca.

• Name and affiliation
• Email address to which the corpus will be sent
• Planned uses for the corpus

If you make use of the corpus in published work, please cite the paper in which it was published:

S. MacLean, G. Labahn, E. Lank, M. Marzouk, and D. Tausky. Grammar-based techniques for creating ground-truthed sketch corpora. International Journal of Document Analysis and Recognition, 2010.

The contents of the corpus are described in the next section.

The corpus consists of samples collected from 20 study participants. Each participant's data is stored in a directory named usern, where n ranges from 1 to 20. Within each of these directories is an auto-labels subdirectory containing the automatically-generated ground-truth labels from our experiment using 50 symbol candidates per ink group without recognizer pre-training.

In the directories, each sample consists of 6 files with a common prefix. The files are as follows:

*.spec

Contains a parenthesized parse-tree representation of the template expression.

*.links

Contains the derivation string of the template expression (refer to the paper for details).

*.tex

Contains a LaTeX representation of the template expression.

*.gif

Contains a GIF image of the template expression.

*.msink

Contains the expression as transcribed by the participant, in Microsoft's Tablet PC ink format.

*.ink

Contains the expression as transcribed by the participant, in SCG ink format (see below).

The SCG ink format is a simple text-based representation of the (x,y)-coordinates drawn by the user. If you require pressure, tilt, etc. information, you should use the Microsoft format. However, the SCG format includes ground-truth labels, whereas Microsoft's format only includes ink data. Each SCG ink file is organized as follows:

1. The string "SCG_INK" followed by a newline.
2. A number indicating the number of strokes appearing in the ink, followed by a newline.
3. For each stroke:
   1. A number indicating the number of points appearing in the stroke, followed by a newline.
   2. For each point, two numbers indicating the x- and y-coordinates, separated by a space, and followed by a newline.
4. The string "ANNOTATIONS" followed by a newline.
5. A number of annotations, each of one of the following three forms:

   Symbol label

   A symbol label associates a symbol name with a group of ink strokes in the sample. These labels consist of the string "SYMBOL" followed by a stroke list, a symbol name, and a newline.

   e.g.

   SYMBOL <1, 2> x

   Symbol mapping

   A symbol mapping associates a group of ink strokes in the sample with a terminal symbol in the associated .links file (i.e. the derivation string). These labels consist of the string "SYMBOLMAP" followed by a stroke list, a symbol index, and a newline.

   e.g.

   SYMBOLMAP <1, 2> 3

   Relationship label

   A relationship label indicates how two groups of ink are structurally connected in the mathematical expression. These labels consist of the string "LINK" followed by a stroke list, a relation name, a stroke list, and a newline. The relations indicated are as follows:

   • R: "Right" - horizontal adjacency, first ink group leftmost
   • AR: "Above-Right" - superscript position
   • BR: "Below-Right" - subscript position
   • B: "Below" - vertical adjacency, first ink group topmost
   • C: "Contains" - containment, first ink group the container

   e.g.

   LINK <1, 2> R <3>

   A stroke list is a comma-separated list of numbers enclosed in angle brackets (eg. <1, 2>) that specifies a group of ink strokes by their zero-based offsets in the ink section of the file.

   The end of the annotation section is indicated by end-of-file.