1. A textual entailment recognition apparatus (2) for determining whether a
first text entails a second text, comprising:
a vector generation unit (21) configured to acquire a predicate-argument structure that includes a predicate included in the first text or the second text, a word serving as an argument of the predicate, and a word indicating the type of argument, from each of the first text and the second text, and
generate, for each of the first text and the second text, by using the predicate included in the predicate argument structure and the word serving as an argument of the predicate, a vector whose dimension is the value obtained by subtracting the number of common character strings from a total value of character strings included in each predicate argument structure,
set a character string corresponding to each dimension in generating each of the vectors,
set a component in which the character string exists to 1 and a component in which the character string does not exist to 0;
a combination identification unit (22) configured to calculate the similarity between the vector generated for each predicate-argument structure for the first text and the vector generated for each predicate-argument structure for the second text, and identifies combinations of the predicate-argument structures of the first text and the predicate-argument structure of the second text based on the similarity; and
an entailment determination unit (23) configured to obtain for each of the combinations, one of a degree of word coverage and a degree of word matching for only the word serving as the argument between the predicate-argument structure of the first text and the predicate-argument structure of the second text, as a feature amount, based on a word other than a word indicating the type of argument of a predicate in the predicate-argument structure, and
determine whether the first text entails the second text based on the obtained feature amounts.
2. The textual entailment recognition apparatus as claimed in claim 1,

wherein the combination identification unit (22) performs normalization processing during calculation of the similarity, in accordance with an amount of information of the vector.
3. The textual entailment recognition apparatus as claimed in claim 1,
wherein the entailment determination unit (23) uses, in addition to the
feature amount, a structural feature of the predicate-argument structure to determine whether the first text entails the second text.
4. The textual entailment recognition apparatus as claimed in claim 3,
wherein the entailment determination unit (23) performs determination with
preference to one of the feature amount and the structural feature of the predicate-argument structure, in accordance with a structural similarity between the predicate-argument structure of the first text and the predicate-argument structure of the second text.
5. The textual entailment recognition apparatus as claimed in claim 1,
wherein the entailment determination unit (23) adds a weight to the feature
amount based on data obtained through machine learning, when obtaining the feature amount.
6. A textual entailment recognition method for determining whether a first text
entails a second text by an electronic device, comprising the steps of:
(a) acquiring a predicate-argument structure that includes a predicate included in the first text or the second text, a word serving as an argument of the predicate, and a word indicating the type of argument, from each of the first text and the second text,
generating, for each of the first text and the second text, by using the predicate included in the predicate argument structure and the word serving as an argument of the predicate, a vector whose dimension is the value obtained by subtracting the number of common character strings from a total value of character strings included in each predicate argument structure,

setting a character string corresponding to each dimension in generating each of the vectors,
setting a component in which the character string exists to 1 and a component in which the character string does not exist to 0;
(b) calculating the similarity between the vector generated for each predicate-argument structure for the first text and the vector generated for each predicate-argument structure for the second text, and identifying combinations of the predicate-argument structures of the first text and the predicate-argument structure of the second text based on the similarity; and
(c) obtaining for each of the combinations, one of a degree of word coverage and a degree of word matching for only the word serving as the argument between the predicate-argument structure of the first text and the predicate-argument structure of the second text, as a feature amount, based on a word other than a word indicating the type of argument of a predicate in the predicate-argument structure, and
determining whether the first text entails the second text based on the obtained feature amounts.
7. The textual entailment recognition method as claimed in claim 6,
wherein, in the step (b), normalization processing is performed during
calculation of the similarity in accordance with an amount of information of the vector.
8. The textual entailment recognition method as claimed in claim 6,
wherein, in the step (c), in addition to the feature amount, a structural feature
of the predicate-argument structure is used to determine whether the first text entails the second text.
9. The textual entailment recognition method as claimed in claim 8,
wherein, in the step (c), determination is performed with preference to one of the feature amount and the structural feature of the predicate-argument structure, in accordance with a structural similarity between the predicate-argument structure of

the first text and the predicate-argument structure of the second text.
10. The textual entailment recognition method as claimed in claim 7,
wherein, in the step (c), a weight is added to the feature amount based on data obtained through machine learning, when obtaining the feature amount.