| CPC G06F 40/35 (2020.01) [G06N 20/00 (2019.01); G09B 7/00 (2013.01)] | 13 Claims |
|
1. A knowledge-grounded dialogue system for language learning, comprising:
a dialogue corpus generator configured to generate a language learning dialogue corpus from reading comprehension data, comprising a passage and a passage-related exercise, the passage-related exercise comprising at least one of a question, a candidate answer, and a correct answer, wherein the language learning dialogue corpus comprises: a passage; at least one of a question, a correct answer, and a candidate answer comprised in the passage-related exercise; a teacher utterance generated from the question comprised in the reading comprehension data; and a student utterance generated from at least one of the correct answer and the candidate answer comprised in the reading comprehension data;
a knowledge-grounded dialogue model trainer configured, to train a dialogue model using the generated language learning dialogue corpus;
a dialogue evaluation model trainer configured, to train a dialogue evaluation model using the generated language learning dialogue corpus;
a dialogue evaluator configured, to receive the passage, the passage-related exercise, and a dialogue history that includes a dialogue conducted between a user and a system about a passage and evaluate a latest utterance received by the system as an input, using the trained dialogue evaluation model; and
a dialogue generator configured, to receive the passage, the exercise, the dialogue history, and an evaluation result of the dialogue evaluator and generate a system utterance,
wherein the dialogue corpus generator comprises a supporting sentence searcher configured, to support the sentence to the correct answer in the passage by referring to the question and the correct answer of the reading comprehension data; and
a dialogue converter configured, to convert the at least one of the question, the correct answer, and the candidate answer comprised in the reading comprehension data into a dialogue for the language learning dialogue corpus, and
wherein the knowledge-grounded dialogue model trainer is configured, to train the dialogue model using the dialogue history with an evaluation result that is the dialogue history to which a result of the dialogue evaluator has been added.
|