ChaseKolozsy / TreeLex

(In Development) Automates the process of finding definitions and phrases where only one unknown word exists in the phrase and all words are known in the definition by having AI generate the definitions or modify the phrases to 1 unknown word. Keeps track of all phrases encountered and sorts them based on how many unknown words are in them.
1 stars 0 forks source link

add logic to update enumerations #35

Closed ChaseKolozsy closed 1 week ago

ChaseKolozsy commented 2 weeks ago

in definition genreator, before a definition is generated, it first searches for a base_lemma for a given word. Then it matches the lemmas that have the same part of speech as the word, and checks to see if the words match based on existing definition and pos. If they don't, then a new definition is generated. This means that a matcher will have to adjudicate IFF there is a matching part of speech. If the pos doesn't match then it is safe to assume that they definitions don't match. If the adjudicator says they are not the same intended meaning, then a new definition is generated based on the phrase and the part of speech and the assertion that none of the other definitinos with the same pos are the correct definition so it should produce a definition that is not the others.

ChaseKolozsy commented 2 weeks ago

It will be requested of the adjudicator to rate the match on a scale of 0 to 100, where 0 means there is no conceptual overlap whatsoever, and 100 means the definition is a perfect match. If the definition scores an 85 or higher, then it is matched. If the score is less than 85, then a new definition is genreated for it. IF there are multiple definitions, the one with the highest score is chosen. It will be instructed to not given any of the definitions the same score to avoid tie breakers, but if there is a tie, then the two will be passed through the same filter, and the adjudicator will be asked to break the tie.