Author: Stan Arts
Date: 04:51:23 01/25/05
Go up one level in this thread
On January 25, 2005 at 01:56:18, Michael Henderson wrote:
>Has anyone tried to make their chess program give good quality annotations? I'm
>not talking about the garbage chessmaster gives. I'm talking about annotations
>similar to a master's annotations, meaning verbal + concrete analysis. Is
>implementing it seen as too difficult, is it not really possible, or do the
>programmers focus on making the program stronger instead? That's a strange
>question for me to ask because I am a chess programmer myself. I think what I
>describe is possible with enough hard work, but I need input from this board.
>
>Michael
Hi Michael
I am experimenting with that since recently.
My goal was to have it kibitz a few lines of sensible comments each move at
CCT7. (as a surprise)
What I came up with is a separate evaluationfunction, which instead of scoring
takes notes of things in "english" and I tried to have a separate scoresystem
to keep track and sort what's important, to be able to talk selective instead of
writing 200 lines of comments.
Instead of calling this evaluationfunction in the searchtree, I call it only
once when the PV has changed and I walked the hashtable. In most positions this
gives it a reasonable idea where the game is heading instead of pure statical.
("your bishop will be out of play" even when that's not the case yet.) and for
material imbalance (material gain or trade beyond the PV, for instance from
Q-search, or mate/draw) I use the actual searchscore.
It works to some extent already, it tells me I have a strong knight on d5, bare
kingsafety or am under attack, etc. Also if it's balanced, or that black is
ahead in material, but white has some compensation.
But I read the CCC thread below which was also about annotating, and my
program fails to the same complaint. It also won't explain the idea behind
a subtle move it plays, or explain a GM's idea in analysis. Unless it's very
lucky..
Also if a piece got badly placed by the end of the PV but is only temporarily
and underway to someplace better it doesn't understand that and will tell it'll
be badly placed, my english-evaluationfunction is far from complete so far, and
trying to sort what's important into a few lines of analysis is really
difficult.
So it doesn't work so good yet and I won't use it at CCT7. :(
But with more time it's comments will become better, and when they are I'll
write about it here. It'll probably be in next Neurosis version anyway.
It's fun anyway, so you should definetly give it a go.
Greetings
Stan
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.