Go talking software

Although of no interest to muscle-car revheads fixated on copy-and-paste MonteCarlo steamrollers, anyone who would like to get a box of software tricks to be able to talk meaningfully about Go might like to take a second look at the observations of Redmond, Kim & Hajin, and An Younggil, which have been collated into a single sgf showing the common and different variations they explored after move 77 in game 4 of the Alphago - Lee Sedol match, supplemented with Lee's own reflections.

game4commented.sgf

The commentators all focussed on achieving and/or refuting the same overall goals of black and white, and looked at various variations within that common theme.

How the goals and theme could be conceived, expressed, thought about and talked about by a computer program will be the subject of a future posting, coming soon.

sgf does not recognise transpositions of subsequences nor different variations leading to the same position, producing a much bushier variation sgf tree than the network of relatively few key lines collectively examined by the commentators. 

a game variation network record format (backwards-compatible with sgf) plus an offline editor that could identify and display transpositions, convergence of variations, authors of variations (by colour-coding of node links), occurrences of common subsequences across variations and key moves transported from within one variation to the start of another (by superimposed graphics), that would show, for example, how Hajin Lee's musings prompted Myungwan Kim to discover the correct defence to Lee Sedol's magic wedge, would be not all that difficult to produce, so would be an interesting and marketable computer science student programming project.

PS the editor also ought to be able to cycle through the endpoints of variations, so its user can immediately see where they each lead so as to compare them with a view to making a choice, and enable hopping from anywhere within one branch to the start of the next, so as to facilitate such things as the learning of joseki, fuseki and yose, and, as the sgf record example provided doesn't really show in an easy way to compare with a tree crawler like cgoban3 or gogui, the ultimate outcomes of different attack/defence tactics.