Morphological annotation of a large spontaneous speech corpus in Japanese

要旨

We propose an efficient framework for humanaided morphological annotation of a large spontaneous speech corpus such as the Corpus of Spontaneous Japanese. In this framework, even when word units have several definitions in a given corpus, and not all words are found in a dictionary or in a training corpus, we can morphologically analyze the given corpus with high accuracy and low labor costs by detecting words not found in the dictionary and putting them into it. We can further reduce labor costs by expanding training corpora based on active learning.

主題