(I sent this quite some time ago, but .Mac mail has been having problems and I haven't seen it come back from the list, so, trying again.)
I think I recall that there's an unused bit in the dictionary data structure. Where there's currently a bit that says "a word ends here", this could mean "a legal word (for adjudication) ends here", and where this bit is set, the unused bit could then be set or not to mean "a word the program currently 'knows' for playing with ends here". You could even go as far as the player being able to specify the program's playing vocabulary on a per-word basis (or supply lists of words to add or remove from the current playing vocab). Or it could be done purely according to which (legal) words the player has used against Quackle so far - maybe the player would play against the full dictionary until they had, e.g., played all the 2s at least once, and would then get the option of playing the full vocab or the limited vocab (with any further new legal words they play in either mode being added to Quackle's limited vocab). I agree that the most learning/analysis way to provide an 'easier' opponent in Quackle would be to only change the vocab. Learning new words when and only when an opponent happens to play them against us is a fairly inefficient way for a human to learn words, but would be a way for the computer to build a fairly human vocab. When using Quackle to analyse games that I've played against human opponents, I often want to know "what's the best move I could have played given my current word knowledge" - i.e. I want to learn to make the best use of what I've got (being already aware of where the gaps in my word knowledge are, and what I should be studying next). Steven. On Thursday, September 06, 2007, at 03:00AM, "Martin DeMello" <[EMAIL PROTECTED]> wrote: >On 9/5/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: >> No perfect way of doing this. There are several sources of words in the >> average person's vocabulary. >> >> But most of us have a scrabble vocabulary that far outstrips the normal >> person's. Do you >> want to give up RETINAE or NEROLIS or QI or ZA, just because they are not in >> most >> folks vocabs or spellcheckers? > >*You* wouldn't need to give it up - what is being proposed is to >separate the move-generation lexicon from the adjudication lexicon. >Thus, NEROLIS, QI and ZA could remain in the word judge, but only be >added to the move generation lexicon after, say, the human player had >played it thrice. > >Ideally you'd want the move generation lexicon to be per-player. > >martin
