Specializing disambiguation of a natural language expression
Disambiguation of the meaning of a natural language expression proceeds by constructing a natural language expression, and then incrementally specializing the meaning representation to more specific meanings as more information and constraints are obtained, in accordance with one or more specialization hierarchies between semantic descriptors. The method is generalized to disjunctive sets of interpretations that can be specialized hierarchically.
1. A method of semantically parsing a natural language expression, comprising:
constructing, by a processor, a first ambiguous meaning representation for a first natural language expression;
fully or partially disambiguating, by a processor, the first meaning representation by specializing it by replacing a first semantic descriptor in it by a second, more specific semantic descriptor;
associating with at least one semantic descriptor in the meaning representation a weight indicating an evaluation of how good an alternative it is; and
adjusting at least one such weight in response to a later parsing or disambiguation action.
2. The method of claim 1 , wherein the second semantic descriptor better matches constraints imposed by semantic and syntactic aspects of the natural language expression parsed so far in the discourse context where the parsing takes place.
3. The method of claim 1 , further comprising further disambiguating a meaning representation that has already been partially disambiguated by further specializing it by replacing the second semantic descriptor in it by a third, even more specific semantic descriptor.
4. The method of claim 1 , wherein the second semantic descriptor is more specific than any dictionary word sense stored in the lexicon of the computer.
5. The method of claim 1 , further comprising selecting the second semantic descriptor based in part on syntactic-lexical constraints.
6. The method of claim 5 , wherein at least one syntactic-lexical constraint requires that the clause contains a certain kind of lexical particle.
7. The method of claim 5 , wherein at least one syntactic-lexical constraint requires that the expression be preceded by a certain lexical word.
8. The method of claim 1 , further comprising:
dynamically generating at least one specialization relation in at least one specialization hierarchy by one or more hierarchy generating rules.
9. A method of semantically parsing a natural language expression, comprising:
constructing, by a processor, a first ambiguous meaning representation for a first natural language expression; and
fully or partially disambiguating, by a processor, the first meaning representation by specializing it by replacing a first semantic descriptor in it by a second, more specific semantic descriptor;
wherein:
the constructed meaning representation comprises a first pointer to a first semantic descriptor that is logically organized in one or more specialization hierarchies;
the disambiguating comprises changing the pointer to point to the second semantic descriptor; and
the second semantic descriptor is a specialization of the first semantic descriptor according to a first specialization hierarchy.
10. The method of claim 9 , further comprising including in the meaning representation a second pointer to a second semantic descriptor that is not a generalization or specialization of the first semantic descriptor in the first specialization hierarchy, and treating the first and second pointer disjunctively.
11. The method of claim 10 , further comprising:
adding a new pointer to the meaning representation, the new pointer pointing to a specialization of the first semantic descriptor but not a specialization or generalization of the second semantic descriptor.
12. A method of semantically parsing a natural language expression, comprising:
constructing, by a processor, a first ambiguous meaning representation for a first natural language expression;
fully or partially disambiguating, by a processor, the first meaning representation by specializing it by replacing a first semantic descriptor in it by a second, more specific semantic descriptor; and
combining the first meaning representation with a second meaning representation associated with a second natural language expression to form a third meaning representation comprising a relation between the first and second meaning representations, the relation type identified by a fourth meaning representation.
13. The method of claim 12 , wherein at least one semantic descriptor for the fourth meaning representation is specified in the grammar.
14. The method of claim 12 , further comprising:
fully or partially disambiguating the fourth meaning representation by replacing a semantic descriptor in it by a more specific semantic descriptor.
15. The method of claim 12 , wherein the second natural language expression and the second meaning representation are inserted as part of processing an elliptic or zero-realized construction.
16. A method of semantically parsing a natural language expression, comprising:
constructing, by a processor, a first ambiguous meaning representation for a first natural language expression;
fully or partially disambiguating, by a processor, the first meaning representation by specializing it by replacing a first semantic descriptor in it by a second, more specific semantic descriptor; and
dynamically generating at least one specialization relation in at least one specialization hierarchy by one or more hierarchy generating rules;
wherein at least one of the
hierarchy generating rules generates a metaphoric interpretation for a class of natural language expressions.
17. An apparatus comprising:
a means for constructing a first ambiguous meaning representation for a first natural language expression; and
a means for fully or partially disambiguating the first meaning representation by specializing it by replacing a first semantic descriptor in it by a second, more specific semantic descriptor;
wherein:
the constructed meaning representation comprises a first pointer to a first semantic desriptor that is logically organized in one or more specialization hierarchies;
the disambiguating comprises changing the pointer to point to the second semantic descriptor; and
the second semantic descriptor is a specialization of the first semantic descriptor according to a first specialization hierarchy.
18. A non-transitory computer-readable medium operable to cause a computer to
perform semantic parsing, comprising:
computer readable program code means for constructing a first ambiguous meaning representation for a natural language expression; and
computer readable program code means for fully or partially disambiguating the first meaning representation by specializing it by replacing a first semantic descriptor in it by a second, more specific semantic descriptor;
wherein:
the constructed meaning representation comprises a first pointer to a first semantic descriptor that is logically organized in one or more specialization hierarchies;
the disambiguating comprises changing the pointer to point to the second semantic descriptor; and
the second semantic descriptor is a specialization of the first semantic descriptor according to a first specialization hierarchy.