MODEL LEARNING DEVICE, FILTER DEVICE, METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To learn a model for determining whether a sentence is suitable as a speech sentence with high precision.SOLUTION: For each input sentence to which information on a positive example representing that a sentence is qualified as a speech sentence or information on a negative example representing that the sentence is not qualified as a speech sentence is added, a modification analysis section 40 generates, as a tree structure which includes word nodes representing parts of speech of the words included in the input sentence corresponding to the respective words and has the word nodes connected by edges corresponding to modification relations between the words, a tree structure to which at least one of a node representing a notation of a word, a node representing a standard notation of the word, and a node representing a termination form of the word is added as a child node of each word node for the word node. On the basis of a plurality of subtrees obtained from the generated tree structure and information on a positive example or information on a negative example added to each of the input sentence, a model learning section 50 learns a model for determining whether the sentence corresponding to the tree structure is qualified as a speech sentence.