2. Do you mean "dunderhead"?

posted Oct 22, 2010, 5:42 PM by Jeff Ogden   [ updated Apr 5, 2012, 11:32 AM ]
The following note was posted to Risks 19:35

Dave Katz <dkatz@juniper.net>
26 Aug 1997 23:33:17 -0700
The piece about "Semper Fi" being corrected to "Semi-pro fiddles" reminded
me of my favorite spell check faux pas.

One of the then-unique characteristics of MTS (the Michigan Terminal System,
a now-extinct o/s for IBM mainframes in the proud academic tradition) was
that it provided a spelling corrector using a Soundex variant--if you
accidentally typed "sigon", it would respond with, "Do you mean 'Signon'?"

After James Duderstadt was named President of the University of Michigan, it
was discovered that if you typed "Duderstadt", the system would respond, "Do
you mean 'dunderhead'?"  I believe it got a new dictionary entry shortly

Although there was no fallout in this particular case, it does underscore
the political risks of unchaperoned spell checkers.

Dave worked at Michigan State University and later at the Merit Network before going out into the commercial world to make his fortune.

I believe this to be a true story, although two different spelling checkers were involved in the "sigon" vs. "signon" and "Duderstadt" vs. "dunderhead" incidents.

On Apr 3, 2012, at 1:13 PM, Dave Katz wrote among other things:

It would be cool if somebody could type "duderstadt" at the "#" prompt.  I want to know if it still responds, "Do you mean 'dunderhead'?"  (And was that serendipitous or was somebody being clever?)

On Apr 3, 2012, at 2:43 PM, Jeff Ogden wrote:

I don't think this was at the MTS command (#) prompt, but rather something from the MTS $Message system when you entered a name on the To or CC field of a message that wasn't recognized as valid (not in the *Userdirectory database). That program used the Soundex algorithum to give people a list of alternatives. And so what you saw would depend on the names that were and weren't in the database. And the Soundex scheme certainly produced some surprising results, so this story could be true. I'm sure that it was serendipitous rather than deliberate.  We can check with Jim Sterken, the author of $Message, to see if he remembers. And for now you are considered the authority on this topic, see:  http://archive.michigan-terminal-system.org/discussions/anecdotes-comments-observations/doyoumeandunderhead.

On Apr 5, 2012, at 1:26 PM, Jim Sterken wrote:

After reading the comments about $Message and *Userdirectory, I took a trip down memory lane and came up with the two attached documents.
One is the original paper, given to me by Steve Tolkin, describing the Soundex algorithm.  I think I implemented it exactly as described in the paper, so as an exercise for the reader, you could probably verify that “duderstadt” and “dunderhead” both hash to the same value.
The other is a hilarious letter from James Wilkes pointing out some limitations of the Soundex Algorithm.
Jeff, I also found some of our original design notes and sketches for $Message plus some feedback from Fred Swartz and Mike Alexander.  I think I ended up presenting an entire final design with proposed $Message commands and msgxxxx subroutine interface level calls at an MTS Conference, but I don’t have a copy of that paper.

A PDF of the Soundex document is available.

A PDF of the1988 memo from Jim Wilkes is available.
Letter from Jim Wilkes, RE: Algorithms for near-matching of strings

On Apr 5, 2012, at 1:35 PM, Dave Katz wrote:

I seem to remember that the Michigan driver's license number used Soundex as well for the first part (a hash of your last name).  At least I think it did 20 years ago when I last had one.

The scary part is that I can still remember the number after that long (but I can't remember what I had for breakfast...)