Summarized logical forms based on abstract meaning representation and discourse trees
Systems, devices, and methods discussed herein provide improved autonomous agent applications that are configured to generate automated answers to a question using summarized logical forms (SLFs). A myriad of techniques may be utilized to generate one or more summarized logical forms for an answer, where the summarized logical form(s) identifies the main entities/informative portions of the answer. Instead of indexing the whole of the answer as in conventional methods, an answer can be indexed using the summarized logical forms. A subsequent query may be converted to logical form to generate a query representation which can be compared to one or more SLFs of a set of answers. If an SLF of an answer matches the query representation, the answer may be provided in response to the question. By indexing the answer with its informative portions, the speed and accuracy of identifying the answer is improved.
1. A method of generating an automated answer to a question, the method comprising:
generating a plurality of summarized logical forms for text of an answer, the plurality of summarized logical forms comprising:
a first summarized logical form comprising a first syntactically-generalized representation provided in first-order logic and expressing a corresponding main topic of a portion of the text; and
a second summarized logical form representing a second syntactically-generalized representation provided in first-order logic and representing one or more potential questions generated from and answered by the text;
associating a search index corresponding to the text of the answer with the plurality of summarized logical forms;
receiving a natural language query comprising the question;
generating a representation of the natural language query based at least in part on converting the natural language query to logical form, the representation of the natural language query comprising at least one logical predicate and a plurality of variables that express the natural language query in first-order logic;
matching the natural language query to at least one of the plurality of summarized logical forms of the search index associated with the text, the matching being performed based at least in part on comparing the first-order logic of the representation of the natural language query to first-order logic of one or more of the plurality of summarized logical forms of the search index; and
providing the text of the answer in response to the question based at least in part on matching the representation of the natural language query to at least one of the plurality of summarized logical forms.
2. The method of claim 1 , wherein generating the plurality of summarized logical forms further comprises at least one of:
generating the first summarized logical form based at least in part on generating an abstract meaning representation of the text, the abstract meaning representation comprising a directed acyclic graph comprising a plurality of nodes and edges that specify a semantic representation of the text;
generating the second summarized logical form based at least in part on generating a discourse tree from the text, the discourse tree comprising a set of nodes and edges that specify a rhetorical relationship between clauses of the text; or
generating a third summarized logical form based at least in part on one or more generalizations of one or more portions of the text.
3. The method of claim 2 , wherein the plurality of summarized logical forms further identify a corresponding portion of the text that is essential to the answer.
4. The method of claim 1 , wherein generating the plurality of summarized logical forms further comprises:
generating a first abstract meaning representation of a first portion of the text, the first abstract meaning representation comprising a first directed acyclic graph comprising a first plurality of nodes and edges that specify a first semantic representation of the first portion of the text;
generating a second abstract meaning representation of a second portion of the text, the second abstract meaning representation comprising a second directed acyclic graph comprising a second plurality of nodes and edges that specify a second semantic representation of the second portion of the text;
comparing the first abstract meaning representation and the second abstract meaning representation; and
identifying a common portion between the first abstract meaning representation of the first portion of the text and the second abstract meaning representation, wherein the first summarized logical form is generated based at least in part on identifying the common portion between the first abstract meaning representation of the first portion of the text and the second abstract meaning representation.
5. The method of claim 4 , further comprising deriving the one or more potential questions generated from and answered by the text from the common portion identified between the first abstract meaning representation and the second abstract meaning representation, wherein the second summarized logical form is generated based at least in part on the one or more potential questions derived from the common portion identified between the first abstract meaning representation and the second abstract meaning representation.
6. The method of claim 1 , wherein generating the at least one summarized logical form further comprises:
generating a discourse tree of the text, the discourse tree comprising a set of nodes and edges that specify a rhetorical relationship between clauses of the text;
providing the discourse tree as input to a machine-learning classifier, the machine-learning classifier having been previously trained to identify portions of text that are suitable for adding to a corresponding search index associated with the text, the machine-learning classifier being trained to identify the portions of text from input discourse trees based at least in part on a training set comprising text examples for which index suitability for corresponding text portions are known; and
generating the at least one summarized logical form based at least in part on output provided by the machine-learning classifier.
7. The method of claim 1 , wherein generating at least one summarized logical form of the plurality of summarized logical forms further comprises:
performing a pair-wise comparison of pairs of generalizations of the text to identify one or more common terms; and
generating the at least one summarized logical form based at least in part on the one or more common terms.
8. A computing device configured to generate an automated answer to a question, the computing device comprising:
a computer-readable medium storing non-transitory computer-executable program instructions; and
a processing device communicatively coupled to the computer-readable medium for executing the non-transitory computer-executable program instructions, wherein executing the non-transitory computer-executable program instructions with the processing device causes the computing device to perform operations comprising:
generating a plurality of summarized logical forms for text of an answer, the plurality of summarized logical forms comprising:
a first summarized logical form comprising a first syntactically-generalized representation provided in first-order logic and expressing a corresponding main topic of a portion of the text; and
a second summarized logical form representing a second syntactically-generalized representation provided in first-order logic and representing one or more potential questions generated from and answered by the text;
associating a search index corresponding to the text of the answer with the plurality of summarized logical forms;
receiving a natural language query comprising the question;
generating a representation of the natural language query based at least in part on converting the natural language query to logical form, the representation of the natural language query comprising at least one logical predicate and a plurality of variables that express the natural language query in first-order logic;
matching the natural language query to at least one of the plurality of summarized logical forms of the search index associated with the text, the matching being performed based at least in part on comparing the first-order logic of the representation of the natural language query to first-order logic of one or more of the plurality of summarized logical forms of the search index; and
providing the text of the answer in response to the question based at least in part on matching the representation of the natural language query to at least one of the plurality of summarized logical forms.
9. The computing device of claim 8 , wherein generating the plurality of summarized logical forms further comprises at least one of:
generating the first summarized logical form based at least in part on generating an abstract meaning representation of the text, the abstract meaning representation comprising a directed acyclic graph comprising a plurality of nodes and edges that specify a semantic representation of the text;
generating the second summarized logical form based at least in part on generating a discourse tree from the text, the discourse tree comprising a set of nodes and edges that specify a rhetorical relationship between clauses of the text; or
generating a third summarized logical form based at least in part on one or more generalizations of one or more portions of the text.
10. The computing device of claim 8 , wherein generating the plurality of summarized logical forms further comprises:
generating a first abstract meaning representation of a first portion of the text, the first abstract meaning representation comprising a first directed acyclic graph comprising a first plurality of nodes and edges that specify a first semantic representation of the first portion of the text;
generating a second abstract meaning representation of a second portion of the text, the second abstract meaning representation comprising a second directed acyclic graph comprising a second plurality of nodes and edges that specify a second semantic representation of the second portion of the text;
comparing the first abstract meaning representation and the second abstract meaning representation; and
identifying a common portion between the first abstract meaning representation of the first portion of the text and the second abstract meaning representation, wherein the first summarized logical form is generated based at least in part on identifying the common portion between the first abstract meaning representation of the first portion of the text and the second abstract meaning representation.
11. The computing device of claim 10 , wherein the operations further comprise deriving the one or more potential questions generated from and answered by the text from the common portion identified between the first abstract meaning representation and the second abstract meaning representation, wherein the second summarized logical form is generated based at least in part on the one or more potential questions derived from the common portion identified between the first abstract meaning representation and the second abstract meaning representation.
12. The computing device of claim 8 , wherein generating at least one summarized logical form of the plurality of summarized logical forms comprises:
generating a discourse tree of the text, the discourse tree comprising a set of nodes and edges that specify a rhetorical relationship between clauses of the text;
providing the discourse tree as input to a machine-learning classifier, the machine-learning classifier having been previously trained to identify portions of text that are suitable for adding to a corresponding search index associated with the text, the machine-learning classifier being trained to identify the portions of text from input discourse trees based at least in part on a training set comprising text examples for which index suitability for corresponding text portions are known; and
generating the at least one summarized logical form based at least in part on output provided by the machine-learning classifier.
13. The computing device of claim 8 , wherein generating at least one summarized logical form of the plurality of summarized logical forms further comprises:
performing a pair-wise comparison of pairs of generalizations of the text to identify one or more common terms; and
generating the at least one summarized logical form based at least in part on the one or more common terms.
14. A non-transitory computer-readable storage medium storing computer-executable program instructions for generating an automated answer to a question, wherein when executed by a processing device, the program instructions cause the processing device to perform operations comprising:
generating a plurality of summarized logical forms for text of an answer, the plurality of summarized logical forms comprising:
a first summarized logical form comprising a first syntactically-generalized representation provided in first-order logic and expressing a corresponding main topic of a portion of the text; and
a second summarized logical form representing a second syntactically-generalized representation provided in first order logic and representing one or more potential questions generated from and answered by the text;
associating a search index corresponding to the text of the answer with the plurality of summarized logical forms;
receiving a natural language query comprising the question;
generating a representation of the natural language query based at least in part on converting the natural language query to logical form, the representation of the natural language query comprising at least one logical predicate and a plurality of variables that express the natural language query in first-order logic;
matching the natural language query to at least one of the plurality of summarized logical forms of the search index associated with the text, the matching being performed based at least in part on comparing the first-order logic of the representation of the natural language query to first-order logic of one or more of the plurality of summarized logical forms of the search index; and
providing the text of the answer in response to the question based at least in part on matching the representation of the natural language query to at least one of the plurality of summarized logical forms.
15. The non-transitory computer-readable storage medium of claim 14 , wherein generating the plurality of summarized logical forms further comprises at least one of:
generating the first summarized logical form based at least in part on generating an abstract meaning representation of the text, the abstract meaning representation comprising a directed acyclic graph comprising a plurality of nodes and edges that specify a semantic representation of the text;
generating the second summarized logical form based at least in part on generating a discourse tree from the text, the discourse tree comprising a set of nodes and edges that specify a rhetorical relationship between clauses of the text; or
generating a third summarized logical form based at least in part on one or more generalizations of one or more portions of the text.
16. The non-transitory computer-readable storage medium of claim 14 , wherein generating the plurality of summarized logical forms further comprises:
generating a first abstract meaning representation of a first portion of the text, the first abstract meaning representation comprising a first directed acyclic graph comprising a first plurality of nodes and edges that specify a first semantic representation of the first portion of the text;
generating a second abstract meaning representation of a second portion of the text, the second abstract meaning representation comprising a second directed acyclic graph comprising a second plurality of nodes and edges that specify a second semantic representation of the second portion of the text;
comparing the first abstract meaning representation and the second abstract meaning representation; and
identifying a common portion between the first abstract meaning representation of the first portion of the text and the second abstract meaning representation, wherein the first summarized logical form is generated based at least in part on identifying the common portion between the first abstract meaning representation of the first portion of the text and the second abstract meaning representation; and
deriving one or more questions from the common portion identified between the first abstract meaning representation and the second abstract meaning representation, wherein the second summarized logical form is generated based at least in part on the one or more questions derived from the common portion identified between the first abstract meaning representation and the second abstract meaning representation.
17. The non-transitory computer-readable storage medium of claim 14 , wherein generating at least one summarized logical form of the plurality of summarized logical forms further comprises:
performing a pair-wise comparison of pairs of generalizations of the text to identify one or more common terms; and
generating the at least one summarized logical form based at least in part on the one or more common terms.