natural logic
Tags: nlp
monotonicity
- key phoneoma that natural logic attempts to capture
- substituting a phrase has predictable entailment relationships
- example:
- a reptile moved -> upward monotone
- if we replace a superset with a subset, it’s entailed
- a reptile moved -> a turtle moved
- no reptile moved -> downward monotone
- no turtle moved -> replacing a superset with a subset, negation has flipped the monotone
- non-montone: do no preserve entailment in either direction
- a reptile moved -> upward monotone
- example:
langauge
- upward monotone is sort of the default for lexical items
- i.e. most determiners (a, some, at least), second argument of every (danced in every), etc
- downward monotone is usually negation, first argument of every, determiners like most, few, conditional antecedents (if-clauses), etc
limitations
- de morgan’s laws cannot be parsed by natural logic