Two days ago I was contacted by Charles Matthews, former Wikimedian in Residence for the Betty & Gordon Moore Library at Cambridge, saying that Karl Wilcox, developer of DrawShield, wanted volunteers to assist with the next stage. Matthews contacted me because I was, in his mind, “certainly adept with heraldry software”. It was with a heavy heart that I told him the less glamorous truth: I do all of my heraldic art on PowerPoint and Paint. He said it was still interesting to know my methods, so I put together a video of the process by which I make each piece.

For those unfamiliar, blazon is heraldry’s own language, and DrawShield is an attempt at a translator tool. The software allows you to type a blazon into a small box, for which it then generates an illustration of the shield, comprising charges drawn from a bank from Wikimedia Commons. As anyone who has had to use an online translator will know, the subtleties of even an apparently-simple phrase can be difficult to teach to a machine. Drawshield occasionally has difficulty when sentences are factorised for word economy (e.g. noting the tincture at the end of a list of multiple charges, rather than separately for each one) or when it is asked to illustrate a charge not encountered before. Syntax also tends to be a difficulty as blazons are generally written with as little punctuation as possible and attempts to break up the long run-on sentences can drastically change the meaning.

It seems, based on the snippets given to me, that Matthews and Wilcox are attempting to rectify exactly these problems, with the former showing me a list of new charges sought for inclusion. He also boasts a substantial gallery of those he has already made.

More troubling, though, is the artistic element: To depict an aesthetically-pleasing shield requires careful consideration of the relative sizes of elements within the escutcheon. Sometimes the same charge may be depicted in different ways within the same emblazonment, tapering or stretching to fit the curves of the shield. Furthermore an asymmetric charge may have a centre of gravity which is distinct from its centre of width, requiring careful spacing. These decisions are too fine and too subjective for the computer, which instead drops out coats with a lot of empty background and charges too small to properly see. The end result is often an image that is technically correct but looks subjectively cheap and inauthentic. The main advantage, of course, is speed – even my relatively crude pictures take 20-40 minutes depending on complexity while those of Sodacan or RS-Nourse must take a far longer time, but DrawShield can spit out multiple coats of arms in a minute. It would therefore be a very useful tool in filling any gaps or catching up on backlogs in instances where blazons are known but images have not yet been added. I would not, however, recommend it as a permanent solution for the arms of anyone whose page is viewed with a serious degree of regularity. Matthews says “It would be great to work on a reference collection for heraldry that was uniform.” but compared to the works that already exist I fear such uniformity would represent a levelling-down rather than up.