Conference noted::hackers_v1

	I have a couple of dictionaries in my account that might be
	useful.  Look in "RAINBO::DVL:[WASSER]DICT*.TXT".  The first
	dictionary (DICTIONARY.TXT I think) has a number of words
	that DECSPELL doesn't know about.  I think it tends toward
	the British spellings.  The second dictionary (DICTIONARY_2.TXT
	I think) has only 15000+ words.  Both are in plain text
	format with spaces or new-lines separating words.

			-John Wasser

	    Is it Huffman encoding?  If so, you can use VMS utility
	routines on it.
			<_Jym_>

    
    Care to expand on Huffman encoding?  I've often thought of writing
    a 'Jumble' solver using the DECspell dictionary ...
    
    John

    Why not use the grep utility on unix?  If /usr/dict/words isn't
    big enough for you, you may be able to get hold of an on-line
    version of Webster's 2nd - just a listing of the words.  This used
    to be available for FTP from some arpa site.  I have a compressed
    version on tape somewhere.
    
    Rich

    The exact format of the Houghton Mifflin dictionary (the dictionary
    used by DECspell) is proprietary to Houghton Mifflin, as is the
    exact word list in the dictionary.
    
    The DECspell dictionary isn't just a list of words.  It's a list
    of the most frequently used words, and apparently a great deal of
    research went into coming up with the list (or so they tell me).
    
    While the word lists do exist in DEC, I'm not sure you can get someone
    to release the list to you ... our CC nearly signed their life away
    for it and the associated decoding routines.

    Given patience, you may be able to come close.  Get a hold of the
    latest WPS kit, and extract the DECspell stuff from it.  Then, start
    poking in the .OBJ's and .EXE (shareable's) with ANAL/OBJ and
    ANAL/IMAGE.  This will give you some entry points and parameter
    lists.  From there, short programs passing the parameters back and
    forth displaying them in between will help.  Note that the routines
    often return error status codes that have little or nothing to do
    with the actuall error.  The debugger is essential, and hacking
    is the only chance of doing this.
    
    It is left as an exorcise to the reader to go from here.

    P.S. It can be done...

    Ermmmm, I've actually done this before, back in my University days.
    I wrote a little program (in C, actually) that produced all the
    permutations and combinations of the letters, and used Unix spell
    to discard the  silly ones. I got about 100 words out of 9 letters
    I think (I didn't have the CPU power available to run the 8 and
    9 letter ones though !). I didn't wi  n  though (and there were
    over 100 prizes). 
    
    
    Trev

    A group I used to work for typed in the entire Scrabble dictionary.
    Last time I checked, I still had access to the files.
    
    If you are interested, send me mail.
    
    	Joe (cookie::) Krantz

There are only 986409 possible words.  This is practically doable
by 'hand'.  What you do is...

Write a program to generate all of them, by choosing and permuting
the letters.  Sort the words.  Flip through the appropriate dictionary
(the contest should say which), matching words.  This is not too bad,
since you skip the thousands of words that start with TG..., or LN...,
et cetera.

It helps if two people do this, one going through the dictionary, and
the other scanning the words (on-line!).  A program (possibly in TPU)
might help, so you can point to a letter and say: "Find the next word
that differs at this position (or left of this position)".

It may be best to 'chop out' the big obvious groups of non-words, and
then go through the smaller list a few separate times (fatigue, y'know).
On the third pass, you might notice something that is really a word
(the ING and ER endings possible with the letters will give you no end
of amusement -- is TREADLING a word? -- check the rules!).

    DECSpell has a callable interface.  See the DECSpell Programmers
    Guide.  However, .-1's solution is more apt to be exhaustive, given
    that the contest official dictionary is known.

Well, I found 2285 words by using a merged copy of the Tops-20 WORDS and
Gorin SPELL dictionaries.  I ran a program that normally finds all the
words that can be built out of a multiset of letters (usually a word, like
"extraterrestrial") without using any letter more than once.  I assumed
that 5 copies each of A, D, E, etc. would suffice to find any such word.

Bear in mind that there were probably some invalid words, like Roman
Numbers and RUNOFF commands found.  The dictionary contains 91159 words
in total.

Using a different dictionary, which has roughly 230000 words, I got
3362 matches.
				/AHM

    Does the contest allow letters to be repeated?

Re .10 - Good idea, this has proved very fruitful.

Re .12 - 3362 words!! - I would be very interested to see the list produced!!

Re .13 - Sorry no it doesn't allow repeats.


Thanks to everybody that has helped, the entry has now been put in, however I 
for one am certainly still interested in this discussion.

Cheers,

Graham Pye (Author of .0)

OK, sorry, I didn't realize from your problem statement that repetition
of the letters was not allowed.  With this change, I was able to find
692 and 858 words from the small and large dictionaries, respectively.
				/AHM

    Given a set of contest letters, and an on-line dictionary, I suggest
    the following method for listing the set of words that can be made
    from the contest letters:
    
    Now, for each word in your dictionary, do this:

    Take a 26-integer array.  In slot "A", put the number
    of As in your contest letters, in slot "B", put the number of
    Bs etc.  (This step can be precompiled as a single MOVC3 instruction).
    
    For every letter of the current dictionary word, decrement the
    corresponding slot in the array.  If the slot becomes negative,
    that dictionary word can not be made.
    
    If you get completely through the dictionary word and no slot
    becomes negative, list that word.
    
    This method is much faster than the permutation business.  You
    only need to go through your dictionary ONCE.
    
    Actually, I've got a program that will do all of this for you.
    It's slightly inappropriately called BOGGLE.EXE on
    RAYNAL::USER$7:[OSMAN.TEST].
    
    The dictionary it reads are AZ.TXT, BZ.TXT, CZ.TXT . . . on
    [OSMAN.WORDS].  Feel free to copy these files (about 80000 words
    total).
    
    Just run it, it will ask you for letters.  When I typed in
    ADEGILNRT, it found about 518 words.
    
    /Eric

Re .16:

Yes, that is how the program GIDNEY::WHITE:<AMARTIN>SUBSEQ.B36 solves
the problem.  I wrote it (no later than) December, 1982, to solve the
Pizza Hut question "How many words can you make out of
``extraterrestrial''".

By the way, four of the words are "extraterrestrial", "extra",
"terrestrial" and "ET".  The last word is either Latin, or an example
of the glitches in my merged dictionary.
				/AHM