I just came home from a sabre refereeing seminar, held by none other than Medhat El-Bakry, referee of the Men’s sabre final in the 2016 Olympic games. As far as teaching sabre refereeing, it does not get better than that.
He had prepared a very thorough seminar document, replete with lots of video snippets where we could work on assigning priority in lots of different types of cases, and at the end there were some snippets covering penalty situations. All in all, three very rewarding hours for everyone who referees sabre, or is interested in how it is refereed.
Given the teacher, it is not surprising that the seminar soon reached the very technical aspects, so those of you who are not already familiar with with sabre refereeing would find that quite complicated.
A lot of time was spent going over how priority should be assigned for in-the-box sabre fencing, that is sabre where the entire fencing phrase consists of both fencers advancing quickly from the guard line until they meet close to the middle line, and neither of the fencers retreating at any time during the phrase. Given how much of competitive sabre consists of those phrases, it makes perfect sense to spend so much effort on it. An unexpected takeaway: at higher levels of refereeing, there is a lot of effort going in to assigning priority – one way or another – in those phrases, so that one gets as few as possible simultaneous actions called. I had not thought about that in such detail beforehand, but if I would have been pressed I would have guessed that refereeing would leave more phrases to be called as simultaneous, thereby making refereeing somewhat easier and forcing the fencers to come up with tactics that would result in phrases where the priority is easier to assign.
Another thing that struck my mind: there are several aspects of sabre refereeing that, given Moore’s law, might be done well by AI in the future. One would have to define the details of what kind of human kinematics result in various defined actions (point in line/attack/etc) which would then give the priorities, but that is a doable endeavour. With a line of cameras along the side of the piste one would capture all of the activity, and the camera feed would then go right into the computer. In-the-box fencing would be the first to be AI-analyzed, since it is so common, comparatively kinematically simple, and would require so few cameras.
Another approach would be to feed the AI system with thousands of video snippets, all of which have been classified by high-level human referees as to what the phrase should be called as, and to which side priority should be given. Then the AI would be left to find similarities and differences ad-hoc, and come up with its own understanding of sabre phrase analysis.
Blade contacts – how many they are, and where on the blade they occur – would be something that AI would become better than humans at doing. Camera frame rates can be increased, but human eyes have the physiologies that they have, and there is not much that can be done to make them drastically quicker. That would lead to better discerning between cases of attacque-de-fer versus parries, and it would also ensure that phrases which include counterparries are not missed.
In the long run, the work of a sabre referee would be more similar to that of an epee referee, with the referee calling priority only in cases where the AI obviously gets it wrong, or if there is a video call.
I am completely aware that this prediction is contentious to quite a few people. Let us hear in the comments section!