The Day Type-in Mania Died: Joseph Roehrig's Scrabble (1981)

I recently worked on typing in a program from BYTE Magazine. It was for a Scrabble game "simulation"/AI opponent.  From the article accompanying the program you could tell that the author Joseph Roehrig was very proud of his creation.  Working completely in BASIC he had managed to create an AI opponent for a very complex game.  He had also created a version specifically for the TRS-80 that loaded as one file (i.e. could work from tape), only required a 32K machine (the oncoming standard replacing the 16K and 4K base versions first produced in 1977), and included a dictionary of over 700 words.  Compromises had to be made for the sake of speed and storage.  He limited the AI opponent to using only 2 and 3 letter words. There were other compromises made in terms of search rigor, and also you could choose a FAST game that lowered these standards even further.  But it could play a basic game that certainly might challenge a kid or maybe neophyte players (it's hard for to tell as I'm a neophyte).

The year of 1981 was still firmly a year of the "type-in mania" early informal 8-bit computing period.  Lots of home coders were making programs and sharing them with the wider world.  The period of selling cassette tapes from home sold in a plastic baggies with some simple reproduced documentation folded in was still a reality, exemplified by Scott Adam's Adventure International company.  At that point his "company" was essentially him and a loose group of collaborators.  However, by 1983, the company had become a proper software company and had a location and 40 employees (by 1986 it would be bankrupt).  It was somewhere in that period that computing pivoted from a do-it-yourself hobby, to an industry overseen by organized corporate actors serving clients.  In researching about Roehrig's program in subsequent issues of BYTE to check for possible bug reports, I cam across the following letter:

We were pleased to see an article discussing the feasibility of a computer opponent for Selchow & Righter's popular Scrabble word game (see "Computer Scrabble," December 1981 BYTE, page 320). Others who are intrigued by this concept will appreciate knowing that the state of the art in microcomputer Scrabble has made a great leap forward. It is far beyond the boundaries that Mr. Roehrig tells us will not be broken by anything less than a new, superior generation of microcomputers.

"Monty plays the Scrabble Brand Crossword Game" (a computer-opponent program available on disk for the Apple II and TRS-80 Models I and III from Ritam Corporation for $39.95) demonstrates both speed and ability, within the constraints of today's microcomputers. Monty spends an average of only 4 1/2 minutes per move at the highest skill level, and yet it uses an extensive word list (over 50,000), based in part on the Official Scrabble Players Dictionary. 

As for memory, the program requires no more than 48K bytes for Apple and 32K bytes for TRS-80 versions, much of which is ,devoted to machine-language graphics, music, and other user-interface requirements. The dictionary is accessed from disk and is stored in an average of only two bytes per word (with an average length of 6 or 7 letters) by use of advanced compression techniques. In addition, Monty is capable of challenging other players' words, based on linguistic analysis, without accessing the disk.

To give Mr. Roehrig's efforts due credit, the "game's complexities" do offer a challenge I It took us several major design breakthroughs, over four man-years of programming (for three different computers), and a lot of determination to develop "Monty plays Scrabble" without conceding to "certain constraints" on word length, search, and placement.

Although his conclusion that "improved computerized Scrabble will require a faster host computer with more memory capacity" has been disproved by example, we thank Mr. Roehrig for his article. It makes our endeavor seem quite worthwhile when we learn that we've achieved the impossible!

By the way, Mr. Roehrig neglected to properly acknowledge that Scrabble is a trademark of the Selchow & Righter Company, and to disclaim, as does Ritam, any sponsorship or endorsement by Selchow & Righter.

Robert Wall
Ritam Corporation
POB 921
Fairfield, IA 52556

* Scrabble is a registered trademark of Selchow & Righter Company. We apologize for not acknowledging this in a prior article ... MH 

(BYTE: Small Systems Journal, April 1982 Vol 7 No. 4., p. 20.)

This letter perfectly exemplifies the transition. The condescension is palpable.  You can sense the hand of Ritam Corp. patting poor Mr. Roehrig on the head for a his "good effort."  But the message is clear-- real programs require many "man-years" of programing using a real language like Assembly, and not a kiddie language like BASIC.   

The fact that I can't find anything else on the Net about Mr. Roehrig's program indicates his efforts were not rewarded with sales of his own via plastic baggies and reproduced sheets of paper documentation.  Around this time people started buying disk drives in North America and purchasing their programs from stores. There would still be a brief sunset for folks like me, getting our first machines in 1984, to "do it yourself" (my MC-10 cost $79.99 just twice the price of Ritam's $39.99 Scrabble program).  But if you had more than lawnmowing money, you wouldn't have to tarry with the type-in age very much longer. The software industry was on the march.

To make the program work on MC-10 I had to switch some of the larger two-dimensional numeric array's to string arrays.  Since the program only stored 0s or 1s in these two-dimensional matrixes, using strings and string operations like MID$ to look horizontally meant each "item" only will take a single byte to store.  This is much better than the 5 bytes needed for each floating-point numeric item in a numeric array. The TRS-80 Model I/III has a DEFINT command that allows such variables to be designated as INTs, saving memory, but the MC-10 only has floating point. But by making this change  to string arrays and searching and replacing multi-character variable names (some had more than 6 character long names!) to single character variable names allowed me to get the program to work in 20K.

Hopefully all this also speeds up the searches for the computer's move. Roehrig mentions that some of them can take over 15 minutes.  It's hard to tell how speed will translate from the TRS-Senior to the MC-10.  The Senior has a higher clock speed, but the MC-10 has a later more efficient version of Microsoft BASIC.  It also uses a Motorola MC6803 rather than the Senior's Z80.  My sense is that the MC-10 is normally faster.  A sign of this is that the standard delay for displaying a brief message on screen on the TRS-Senior is FORDL=1TO1000:NEXT, whereas for the MC-10 you need FORDL=1TO2500.  But who knows.  Perhaps the string array searches and string manipulations will trigger lots of garbage collection, which can slow things down.  I did most of my testing using the 8X normal speed mode of the VMC10 emulator.

If you have the patience and interest in seeing an old early BASIC AI at work, SCRABB2 can be played online using an MC-10 emulator (Just search for "SCRABB2"):

https://archive.org/details/@james_gerrie

The TRS source ("SCRABB2TRS80.TXT") and the source for my MC-10 version can be found here:

https://github.com/jggames/trs80mc10/tree/master/quicktype/Board%20Games/Scrabb2

TRS-80 Model I/III Version

The original game used a system of numbers (1-225) listed on the TRS-80 Model I/III screen, that you typed in to enter words. I switched it to using arrow keys to move a cursor and added the color markers for word and letter bonuses to allow it to fit the smaller screen of the MC-10 and to ease use for those playing it online today.  I also added some checking to prevent errors if you try to enter characters other than letters.