by Steve Lopez

One of the neat features of Fritz5 is the ability to generate training questions as part of its analysis. This generally consists of Fritz adding insult to injury by beating you silly in a game and then asking you questions about your play later. It's like getting a browbeating from your chess coach. I have this mental picture of what it would be like if Fritz could actually speak interactively:

"Do you know why you were stupid here?"

"If I knew that, I wouldn't have made that move!"

"Not good enough -- you lose twenty master points"

"But I'll never get that certificate!"

(It would be cool if Fritz then started sending mass quantities of worthless blank "master certificates" to the printer at this point, but I don't think that Fritz gets out to the movies too often, not even on laptops).

I was thinking about this the other night and wishing that Fritz would ask me a question that I have a reasonable chance of answering. For example, after 23.Nd5 and before 23...Bxd5:

"What was the name of Hannibal's Roman nemesis in the Punic Wars?"

"Scipio Africanus."

"Correct -- you get 10 points."

I then took this a step further. What if Fritz played a new form of chess, combining the Royal Game with something like Jeopardy or Trivial Pursuit? Instead of asking you chess questions after the game, it would ask you trivia questions during the game. You'd go to make a capture or put Fritz in check, making the move on the board, but before the move is registered as completed Fritz would ask you a trivia question. If you answer it correctly, your move is allowed. If you answer incorrectly, you have to take it back and make a different move.

This would be enormously beneficial in a variety of ways. Instead of asking a question about a chess position that you'll never see again in a million years, it would be asking you some actual real-life useful information. By breaking your train of thought in the middle of a game, Fritz would be encouraging and developing non-linear thinking skills (by causing you to completely shift gears in the middle of some deep thinking). Plus it would be a tremendous ego-boost to myself, the uncrowned Trivial Pursuit champion of the Known Universe (it's true; I have never lost a game). My brain is a total garbage dump of completely useless information with no purpose other than to impress the easily-impressed while watching TV shows like Jeopardy. If I can't beat Fritz at chess, I'd at least have the satisfaction of knowing I answered a bunch of obscure questions that would vex much stronger chessplayers than myself.

I can see it now: after a long positional game, capped by a series of spectacular cascading sacrifices in which I had to correctly identify the capital of Venezuela, the author of Pride and Prejudice, and the color of Elizabeth Hurley's dress at last year's Oscars (yeah, like I was looking at the dress), I'm finally ready to deliver mate. I snap my Queen over to e8 for the mate and I suddenly hear my computer's speakers belting out the music to Final Jeopardy:

"He's the author of the book upon which the movie Patton is based."

Rapidly I type out the name: "Ladislas Farago".

To which Fritz replies: "I'm sorry -- you didn't phrase your answer in the form of a question." The program then disallows the mating move, forces me to make a different move, escapes mate, and proceeds to crush me due to its overwhelming material superiority.

You know, on second thought, I think I like Fritz5's training questions just the way they are. But it's amazing what one's mind can come up with after a few Rolling Rocks while a casual-player opponent is taking 25 minutes to figure out how to respond to 1.e4...


by Steve Lopez

Correspondence chess is a whole different world than the over-the-board version. An over-the-board player is called upon to make a relatively snap judgement of a position, discarding 90% of the possible moves to examine only two or three, and those to only a limited extent. A correspondence player, on the other hand, is free to mine the utmost depths of a position, examining a host of candidate moves and the myriad branches of the move tree that grow from those candidates. Correspondence players have the luxury of time and are allowed to write down their analyses before making a move, giving them the opportunity to investigate every subtle nuance of a move before licking a stamp, slapping it on an card, and sending it on its way.

Correspondence chess is a completely different discipline and requires a different set of mental tools than over-the-board chess. The analysis generated by a correspondence player tends to be much more thorough and detailed as well.

Fritz5 contains a new analysis mode, called correspondence mode. Unfortunately a great many people on Usenet have taken this to mean that ChessBase is encouraging people to use Fritz to help them analyze their correspondence games. Nothing could be further from the truth. Using a computer to generate analysis in correspondence play is illegal -- period. You're not allowed to use a playing program to find a move for you, nor are you allowed to use a computer program to check your moves for accuracy. The only exceptions are the rare correspondence events that allow computer assistance. Otherwise the use of a chessplaying program to help one in a postal game is forbidden.

So what does "correspondence mode" mean? It's a reference to the way in which Fritz generates its analysis. Instead of giving you the single best line of play, Fritz will generate a full tree of analysis, including multiple alternatives complete with subvariations.

Here's how it works. You give Fritz a position, fire up analysis mode, and let it think. It first generates a list of initial first moves for the moving side. It takes one of those moves and analyzes it out until it reaches what's known as a quiescent position. This is a position in which there are no checks, captures. obvious tactical shots, or standard positional motives present (Rook to open file, Knight to protected outpost, etc.); in other words, a position in which the moving side has a few unclear, non-forcing alternatives. Fritz then goes back over that line of play and adds a variation at a key point, extending that variation out until it reaches another quiescent position. It then evaluates that position, goes back, and starts to analyze another line. It does this repeatedly, one line at a time, eventually building up a mammoth tree of analysis.

Let's look at an example. I went to my database of Chess Informants and found an incomplete game that ended with a positional evaluation:

Magem Badals,J (2570) - Giorgadze,G (2580) [C68]
Yerevan (ol) 1996

1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Bxc6 dxc6 5.0-0 Qd6 6.Na3 Be6 7.Qe2 f6 8.Nc4 Qd7 9.Rd1 c5 10.c3 Bg4 11.h3 Bxf3 12.Qxf3 Ne7 13.d3 Nc6 14.a3 a5 15.Qg3 g6 16.f4 Bg7 17.fxe5 fxe5 18.a4 0-0 19.Bg5 b6 20.Qe3 Rf7

The Chess Informant evaluated this position as equal, but in the back of my mind I had some doubts. So I fed the game to Fritz5, fired up the correspondence analysis feature, and let Fritz do its stuff. Check out the illustration on page 58 of your Fritz5 manual. You'll see where the program allows you to specify either an alloted time for it to analyze each move or else a fixed ply (half-move) depth for each move. These are mutually exclusive; you can't set both. It's an "either-or" proposition. I set it for 300 seconds (five minutes) per move. Actually, since I have a fairly fast computer, I should probably set it for a fixed depth. At the five-minute mark, Fritz is usually halfway through the eleventh or twelfth ply and it will always complete the analysis of whatever ply it's on before moving onward. The net result is that the program typically takes eight or more minutes per move, even when set to just five minutes. Your mileage may vary; you'll need to fool around with this to see what works best on your system.

Next you set the branching factor. This determines how many alternatives Fritz will examine at each level of the tree. In the example from the manual, Fritz will look at five main alternatives at the starting position. It will generate three alternatives at the first branching point, and two alternatives at any branching points within this subline. (If I've lost you here, I apologize. Check out Think Like a Grandmaster by Alexander Kotov and The Inner Game of Chess by Andy Soltis for more on the tree concept. I'll try to cover this idea myself in a future issue of T-Notes. Also, check out the first three issues of Electronic T-Notes, which may clarify this for you).

When I set Fritz to run on this position, I may have overdone this a bit. I set Fritz to levels of 5, 5, and 3 respectively. Little did I know that this would cause Fritz to run forever and generate a huge tree of analysis.

You can also set the length of each variation. I've found that this is generally correct, but that Fritz may generate longer or shorter lines depending on the position.

The evaluation window is explained in the manual. Generally speaking, the higher you set this number, the more lines of play will be included in the analysis tree. I left this set at 80, but in the future I think I'll set it a bit lower, perhaps 50 or less, which will tend to give me a smaller tree.

I made my settings changes and turned Fritz loose. Nine hours and thirty minutes later I returned and found that Fritz was still going at it! Here's what Fritz came up with, complete with evaluations. Remember, Fritz began analyzing at Black's 20th move:

1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Bxc6 dxc6 5.0-0 Qd6 6.Na3 Be6 7.Qe2 f6 8.Nc4 Qd7 9.Rd1 c5 10.c3 Bg4 11.h3 Bxf3 12.Qxf3 Ne7 13.d3 Nc6 14.a3 a5 15.Qg3 g6 16.f4 Bg7 17.fxe5 fxe5 18.a4 0-0 19.Bg5 b6 20.Qe3 Rf7


      A) 21...Rb8 22.Qe2 (22.Qg3 Rbf8 23.Rf1 Nd8 24.Rxf7 Nxf7 25.Rf1=;
      22.Rf1 Rbf8 23.Rxf7 Rxf7 24.Rd1 Nd8 25.Rd2=) 22...Rbf8 23.Rf1
      Nd8 24.Rxf7 Nxf7 25.Be3+/=; 

      B) 21...Raf8 22.Rf1 (22.Bh4 Bf6 23.Bxf6 Rxf6 24.Rf1 Rf4 25.Qe2=;
      22.Qg3 Nd8 23.Be3 Ne6 24.Ra2 Nf4 25.Rf2=) 22...Rxf1+ 
      23.Rxf1 Rxf1+ 24.Kxf1 Nd8 25.Qd2+/=; 

      C) 21...Rc8; 

      D) 21...Bf6; 

      E) 21...Re8 22.Rf1 (22.Bh4 Ref8 23.Rf1 Bf6 24.Bf2 Qe7 25.Bg3=;
      22.g3 Ref8 23.Nd2 Qxh3 24.Rf1 h6 25.Rxf7-+) 22...Ref8 23.Rxf7 
      Rxf7 24.Rd1 Nd8 25.Rd2=; 





   [21...Nd8 22.Bh6 (22.b3 Rb8 23.Bxd8 Rxf1+ 24.Rxf1 Rxd8 25.Rf3+/=;
   22.Rxf7 Nxf7 23.Bh4 Rf8 24.b3 Nd6 25.Nd2=) 22...Bxh6 23.Qxh6 
   Rxf1+ 24.Rxf1 Nf7 25.Qh4+/-; 

   21...Rxf1+ 22.Rxf1 (22.Kxf1 Nd8 23.Bxd8 Rxd8 24.Ke2 Rf8 25.Qg3=) 
   22...Nd8 23.b3 Nf7 24.Bf6 Bh6 25.Qg3+/=; 

   21...Rff8 22.b3 (22.Qg3 Rae8 23.b3 Nd8 24.Rxf8+ Rxf8 25.Rd1=; 22.Rad1
   Rae8 23.b3 Nd8 24.Kh2 Nf7 25.Bh4=) 22...Rae8 23.Bh4 Nd8 24.Bg3 Nf7 

   21...Rb8 22.Rxf7 (22.Rad1 Rbf8 23.Rxf7 Rxf7 24.b3 Nd8 25.Rd2=; 22.Qg3
   Rbf8 23.Rxf7 Rxf7 24.Rd1 Rf8 25.b3+/=) 22...Qxf7 23.Rf1 Qd7 24.Qg3 Re8 


   [22.Qg3 Nd8 23.Rxf7 Nxf7 24.Bh4 Bh6 25.Nxe5=; 
   22.Rxf7 Qxf7 23.Qg3 Qd7 24.Rf1 Rxf1+ 25.Kxf1=] 
22...Rxf1+ 23.Rxf1 Rxf1+ 24.Kxf1 Nd8 25.Qd2+/=

When I came back to the program after nine and a half hours, it finished up with the 21.b3 Raf8 line as I watched and switched down to start on 21...Rc8. That's where I stopped it.

So from this example we can see that Fritz generated a main line of play from 21.Rf1 to 25.Qd2, evaluating this main line as being slightly better for White (disagreeing with the Informant's assessment of equality in this line). It generated four additional alternatives for White's first move, but after nine and a half hours had not yet finished completely exploring the first of these, namely 21.b3.

Pretty impressive, eh? This sucker spent almost half a day on a single position and yet didn't even finish plumbing its depths. So much for the idea that a computer will one day "solve" chess. But the interesting point here is that Fritz went about analyzing this position in exactly the same way as a correspondence would, in an orderly, structured manner, a single line at a time.

So what's the significance here? What is this feature good for?

Have you ever read a chess book in which the annotator stops his analysis at a certain point and says that the rest is simply "a matter of technique"? Man, I hate that! Personally, I think it means that the annotator hasn't a freaking clue as to what's going on, but wants to make it sound like he knows all about it. We can now test both assertions, his about the win and mine about his knowledge, by using the correspondence analysis feature of Fritz5.

Another good idea is to use this feature in conjuction with ECO, MCO, BCO, or other chess opening books. These volumes give standard lines of play ending with evaluations, such as "White has a large advantage". But why does White have a large advantage? Let Fritz chew on the position and generate a tree of analysis to give you the answer. Maybe Fritz could even punch a hole or two in established theory. Who knows?

Long-time readers of my rambling screeds (going back to the old printed material that ChessBase USA used to send out with catalogs and orders) may recall that I used to keep a couple of databases called "Playbooks" in which I stored my preferred opening lines. Each game in the database was a single line (occasionally containing one or two main variations), a line that I favored in both over-the-board and postal play. Fritz5's correspondence feature would be a good adjunct to my playbooks. I could feed a line to Fritz, let it look at it for a day or so, and show me if my "preferred line" is actually a good idea to play. It could point out any traps or pitfalls in the continuations after my "main line" runs out.

You could even use this feature in combination with ChessBase to see if any of Fritz's generated lines occurred in actual play, "dragging and dropping" the gamescores into Fritz's analysis tree to create the definitive reference giude to particular opening lines.

Grandmasters use chess computers to aid them in their analysis. Garry Kasparov, in the wake of his loss to Deep Blue last spring, admitted to using Hiarcs 6 to help him analyze thorny positions. The champion was in the habit of setting up a position and having Hiarcs chew on it overnight, providing him with the best line of play from that given position. This was a means of checking his analysis of novelties he'd planned to play over the board.

The new correspondence analysis serves this identical function, but now it will generate an entire tree instead of merely a single line. This makes a computer a more useful analysis tool than ever before.

I'm sure that there are ramifications and uses for this tree-style analysis that I haven't thought of yet, but this is where you come in. I'm certain that you can find uses of your own for this wonderful tool. But please don't make analysis of postal games one of them; this is definitely not what the term "correspondence analysis" refers to.

Speaking of correspondence, fire it off to our ChessBase Users Group or you can e-mail me directly.