Dialogue processing apparatus, learning apparatus, dialogue processing method, learning method and program

1. A dialogue processing apparatus, comprising:

a storage; and

a hardware processor configured to:

take a question Q i that is a word sequence representing a current question in a dialogue, a document P used to generate an answer A i to the question Q i , a question history {Q i-1 , . . . , Q i-k } that is a set of word sequences representing k past questions, and an answer history {A i-1 , . . . , A i-k } that is a set of word sequences representing answers to the k questions as inputs; and

generate the answer A i by machine reading comprehension in an extractive mode or a generative mode using pre-trained model parameters.

2. The dialogue processing apparatus according to claim 1 , wherein features used to generate the answer A i by machine reading comprehension in the extractive mode or the generative mode are characterized in that the number of the features reflecting the question Q i the question history {Q i-1 , . . . , Q i-k }, and the answer history {A i-1 , . . . , A i-k }, is defined as T, and

the hardware processor is configured to calculate features {u i-1 t , . . . , u i-k t } (where t=1, . . . , T) of the document P regarding the question history {Q i-1 , . . . , Q i-k } and features {v i-1 t , . . . , v i-k t } (where t=1, . . . , T) of the document P regarding the answer history {A i-1 , . . . , A i-k }.

3. The dialogue processing apparatus according to claim 2 , wherein the harware processor is configured to:

calculate features u i t (where t=1, . . . , T) of the document P regarding the question Q i ;

combine, for each t, the features u i t (where t=1, . . . , T), the features {u i-1 t , . . . , u i-k t } (where t=1, . . . , T), and the features {v i-1 t , . . . , v i-k t } (where t=1, . . . , T) to calculate features o t (where t=1, . . . , T) as the features reflecting the question Q i the question history {Q i-1 , . . . , Q i-k }, and the answer history {A i-1 , . . . , A i-k }; and

generate the answer A i by the machine reading comprehension in the extractive mode or the generative mode using the features o t (where t=1, . . . , T).

4. A non-transitory computer-readable recording medium having stored threin a program for causing a computer to operate as the dialogue processing apparatus according to claim 1 .

5. A learning apparatus comprising:

a storege; and

a hardware processor is configured to:

take a question Q i that is a word sequence representing a current question in a dialogue, a document P used to generate an answer A i to the question Q i , a question history {Q i-1 , . . . , Q i-k } that is a set of word sequences representing k past questions, and an answer history {A i-1 , . . . , A i-k } that is a set of word sequences representing answers to the k questions as inputs, and to generate the answer A i by machine reading comprehension in an extractive mode or a generative mode using model parameters; and

update the model parameters by supervised learning using the answer A i and a correct answer to the question Q i .

6. A non-transitory computer-readable recording medium having stored therein a program for causing a computer to operate as the learning apparatus according to claim 5 .

7. A dialogue processing method configured to be executed by a computer, the method comprising:

taking a question Q i that is a word sequence representing a current question in a dialogue, a document P used to generate an answer A i to the question Q i , a question history {Q i-1 , . . . , Q i-k } that is a set of word sequences representing k past questions, and an answer history {A i-1 , . . . , A i-k } that is a set of word sequences representing answers to the k questions as inputs, and generating the answer A i by machine reading comprehension in an extractive mode or a generative mode using pre-trained model parameters.