Systems and methods for a reading and comprehension assistance tool
Embodiments described herein provide a method and system for generating a reading interface for a user. The method includes receiving a first text passage from a first data source and a second text passage from a second data source. The method also includes generating a candidate question relating to contents of the first and the second text passages. The method further includes generating a first answer to the candidate question and a second answer to the candidate question. The method further includes determining that the candidate question qualifies as a discord question when the first answer and the second answer are both available and exhibit semantic diversity. The method further includes constructing a reading interface comprising a first component displaying the discord question and one or more user-engageable components each displaying the first or the second answer and linking to the first or the second data source, respectively.
1 . A system for generating a reading interface for a user, the system comprising:
a data interface that receives a first text passage from a first data source and a second text passage from a second data source;
a memory storing a question generation model, a question answering model and an answer consolidation model that are implemented on one or more neural networks, and a plurality of processor-executable instructions for operating the models; and
a processor executing instructions to perform operations comprising:
generating, by the question generation model, a candidate question relating to contents of the first and the second text passages;
generating, by the question answering model, a first answer to the candidate question conditioned on the first text passage and a second answer to the candidate question conditioned on the second text passage;
determining, by the answer consolidation model, that the candidate question qualifies as a discord question when the first answer and the second answer are both available and exhibit semantic diversity; and
constructing a reading interface comprising a first component displaying the discord question and one or more user-engageable components each displaying the first or the second answer and linking to the first or the second data source, respectively; and a hardware user interface that presents the reading interface to a user.
2 . The system of claim 1 , wherein the operations further comprise:
generating, by the question answering model, a plurality of answers to the candidate question based on relevant text passages, respectively; and
determining, by the answer consolidation model, a first semantic group of semantically similar answers and a second semantic group of semantically similar answers,
wherein a similarity score between two of the semantically similar answers in a same semantic group is higher than a threshold score.
3 . The system of claim 2 , wherein an operation of determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
determining that each answer from the plurality of answers comprises content for more than a first pre-defined percentage of a corresponding text passage based on which each answer is generated.
4 . The system of claim 2 , wherein an operation of determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
determining that a larger one of the first sematic group and the second semantic group contains less than or equal to a second pre-defined percentage of the plurality of answers.
5 . The system of claim 2 , wherein an operation of determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
obtaining one or more distractor text passages that are different from the first text passage or the second text passage;
generating, by the question answering model, one or more answers to the candidate question based on the one or more distractor text passages, respectively;
computing a question specificity score based on a number of available answers generated from distractor text passages and a number of the plurality of answers generated from relevant text passages;
determining that the question specificity score is greater than a pre-defined threshold.
6 . The system of claim 2 , further comprising an operation of obtaining a summary text relating to the first text passage and the second text passage, wherein
generating, by the question generation model, a candidate question relating to contents of the first and the second text passages comprises generating the candidate question based on the summary text.
7 . The system of claim 1 , wherein the question generation model, the question answering model, and the answer consolidation model are each trained separately.
8 . The system of claim 1 , wherein the first text passage and the second text passage are associated with a ground-truth question for training, and wherein the question generation model, the question answering model, and the answer consolidation model are jointly trained, by:
computing a training objective based on the discord question and the ground-truth question; and
jointly updating the question generation model, the question answering model, and the answer consolidation model in an end-to-end manner based on the training objective via backpropagation.
9 . A method for generating a reading interface for a user, the method comprising:
receiving, via a communication interface, a first text passage from a first data source and a second text passage from a second data source;
generating, by the question generation model, a candidate question relating to contents of the first and the second text passages;
generating, by the question answering model, a first answer to the candidate question conditioned on the first text passage and a second answer to the candidate question conditioned on the second text passage;
determining, by the answer consolidation model, that the candidate question qualifies as a discord question when the first answer and the second answer are both available and exhibit semantic diversity; and
constructing a reading interface comprising a first component displaying the discord question and one or more user-engageable components each displaying the first or the second answer and linking to the first or the second data source, respectively.
10 . The method of claim 9 , further comprise:
generating, by the question answering model, a plurality of answers to the candidate question based on relevant text passages, respectively; and
determining, by the answer consolidation model, a first semantic group of semantically similar answers and a second semantic group of semantically similar answers,
wherein a similarity score between two of the semantically similar answers in a same semantic group is higher than a threshold score.
11 . The method of claim 10 , wherein determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
determining that each answer from the plurality of answers comprises content for more than a first pre-defined percentage of a corresponding text passage based on which each answer is generated.
12 . The method of claim 10 , wherein determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
determining that a larger one of the first sematic group and the second semantic group contains less than or equal to a second pre-defined percentage of the plurality of answers.
13 . The method of claim 10 , wherein determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
obtaining one or more distractor text passages that are different from the first text passage or the second text passage;
generating, by the question answering model, one or more answers to the candidate question based on the one or more distractor text passages, respectively;
computing a question specificity score based on a number of available answers generated from distractor text passages and a number of the plurality of answers generated from relevant text passages;
determining that the question specificity score is greater than a pre-defined threshold.
14 . The method of claim 10 , further comprising obtaining a summary text relating to the first text passage and the second text passage, wherein
generating, by the question generation model, a candidate question relating to contents of the first and the second text passages comprises generating the candidate question based on the summary text.
15 . The method of claim 9 , wherein the question generation model, the question answering model, and the answer consolidation model are each trained separately.
16 . The method of claim 9 , wherein the first text passage and the second text passage are associated with a ground-truth question for training, and wherein the question generation model, the question answering model, and the answer consolidation model are jointly trained, by:
computing a training objective based on the discord question and the ground-truth question; and
jointly updating the question generation model, the question answering model, and the answer consolidation model in an end-to-end manner based on the training objective via backpropagation.
17 . A non-transitory processor-readable storage medium storing a plurality of processor-executable instructions for code program synthesis for a target problem, the instructions being executed by one or more hardware processors to perform operations comprising:
receiving, via a communication interface, a first text passage from a first data source and a second text passage from a second data source;
generating, by the question generation model, a candidate question relating to contents of the first and the second text passages;
generating, by the question answering model, a first answer to the candidate question conditioned on the first text passage and a second answer to the candidate question conditioned on the second text passage;
determining, by the answer consolidation model, that the candidate question qualifies as a discord question when the first answer and the second answer are both available and exhibit semantic diversity; and
constructing a reading interface comprising a first component displaying the discord question and one or more user-engageable components each displaying the first or the second answer and linking to the first or the second data source, respectively.
18 . The non-transitory processor-readable storage medium of claim 17 , wherein the operations further comprise:
generating, by the question answering model, a plurality of answers to the candidate question based on relevant text passages, respectively; and
determining, by the answer consolidation model, a first semantic group of semantically similar answers and a second semantic group of semantically similar answers,
wherein a similarity score between two of the semantically similar answers in a same semantic group is higher than a threshold score.
19 . The non-transitory processor-readable storage medium of claim 18 , wherein determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
determining that each answer from the plurality of answers comprises content for more than a first pre-defined percentage of a corresponding text passage based on which each answer is generated.
20 . The non-transitory processor-readable storage medium of claim 18 , wherein determining, by the answer consolidation model, that the candidate question qualifies as a discord question comprises:
determining that a larger one of the first sematic group and the second semantic group contains less than or equal to a second pre-defined percentage of the plurality of answers.