Free Text Entry

Hierarchically, handwriting consists of the following components:

stroke character word sentence paragraph page

Larger components have not been approached in current technology. It should be noted, that apart from the already difficult classification of shape (character, word), also the segmentation preceding it is erroneous. Writers will often write words in a sentence with a horizontal spacing which is of the same magnitude within and between words. The histograms of horizontal distances for within and between-word horizontal spacing will generally have a high overlap.

Difficult issues in current recognition and user interfacing technology are the multitude of styles over writers, the variability of shape within a writer, the processing of delayed pen actions, such as the dotting of i's and j's at a much later stage in the input process, and all kinds of delayed graphical corrections to letters (crossing, adding ink, adding drawings).

The following three categories of Free Text Entry may be defined, depending on the number and nature of existing constraints on the handwriting process.

1. Fully unconstrained (size, orientation, styles)
   This type of input, like on the normal office and household notes or PostIts, cannot be recognized well at this point in time.
2. Lineated form, no prompting, free order of actions
   Here there are weak geometrical constraints, leading to a much more regular input quality. There are no temporal constraints.
3. Prompted
   In order to reduce the error introduced by segmentation into sentences or words, the user is required to indicate the boundaries of an input chunk. In the QWERTY keyboard, a similar segmenting functionality is provided by the <Return> or <Enter> key.

   The following methods are being used:

   • ``OK'' dialog box chunk ending. A ``Cancel'' button allows for rewriting misspelled input. This method works reliable but interrupts the writing process.
   • Time-out. Characters and Words can be segmented by a time out. If the writer lifts the pen and does not produce any ink within T time, a segmentation event is assumed to have occurred. A typical time out value for words is 800 ms, but this varies over users. Long time out values cause merging of ink chunks, short time out values lead to premature processing of unfinished chunks.
   • Gesture (see below). In this case, a simple gesture, like tapping with the pen to the right of the last word will indicate the segmentation event.

Esprit Project 8579/MIAMI (Schomaker et al., '95)
Thu May 18 16:00:17 MET DST 1995