This paper addresses the task of assigning labels of fine-grained named entity (NE) types to Wikipedia articles. Information of NE types are useful when extracting knowledge of NEs from natural language text. It is common to apply an approach based on supervised machine learning to named entity classification. However, in a setting of classifying into fine-grained types, one big challenge is how to alleviate the data sparseness problem since one may obtain far fewer instances for each fine-grained types. To address this problem, we propose two methods. First, we introduce a multi-task learning framework, in which NE type classifiers are all jointly trained with a neural network. The neural network has a hidden layer, where we expect that effective combinations of input features are learned across different NE types. Second, we propose to extend the input feature set by exploiting the hyperlink structure of Wikipedia. While most of previous studies are focusing on engineering features from the articles' contents, we observe that the information of the contexts the article is mentioned can also be a useful clue for NE type classification. Concretely, we propose to learn article vectors (i.e. entity embeddings) from Wikipedia's hyperlink structure using a Skip-gram model. Then we incorporate the learned article vectors into the input feature set for NE type classification. To conduct large-scale practical experiments, we created a new dataset containing over 22,000 manually labeled articles. With the dataset, we empirically show that both of our ideas gained their own statistically significant improvement separately in classification accuracy. Moreover, we show that our proposed methods are particularly effective in labeling infrequent NE types. We've made the learned article vectors publicly available. The labeled dataset is available if one contacts the authors.

A tremendous amount of knowledge is present in the ever-growing scientific literature. In order to efficiently grasp such knowledge, various computational tasks are proposed that train machines to read and analyze scientific documents. One of these tasks, Scientific Relation Extraction, aims at automatically capturing scientific semantic relationships among entities in scientific documents. Conventionally, only a limited number of commonly used knowledge bases, such as Wikipedia, are used as a source of background knowledge for relation extraction. In this work, we hypothesize that unannotated scientific papers could also be utilized as a source of external background information for relation extraction. Based on our hypothesis, we propose a model that is capable of extracting background information from unannotated scientific papers. Our experiments on the RANIS corpus [1] prove the effectiveness of the proposed model on relation extraction from scientific articles.

In this paper, we propose an answer seeking algorithm for question answering that integrates structural matching and paraphrasing, and report the results of our empirical evaluation conducted with the aim of examining effects of incorporating those two components. According to the results, the contribution of structural matching and paraphrasing was not so large as expected. Based on error analysis, we conclude that structural matching-based approaches to answer seeking require technologies for (a) coreference resolution, (b) processing of parse forests instead of parse trees, and (c) large-scale acquisition of paraphrase patterns.

This paper addresses the issue of modeling the discourse nature of lyrics and presented the first study aiming at capturing the two common discourse-related notions: storylines and themes. We assume that a storyline is a chain of transitions over topics of segments and a song has at least one entire theme. We then hypothesize that transitions over topics of lyric segments can be captured by a probabilistic topic model which incorporates a distribution over transitions of latent topics and that such a distribution of topic transitions is affected by the theme of lyrics. Aiming to test those hypotheses, this study conducts experiments on the word prediction and segment order prediction tasks exploiting a large-scale corpus of popular music lyrics for both English and Japanese (around 100 thousand songs). The findings we gained from these experiments can be summarized into two respects. First, the models with topic transitions significantly outperformed the model without topic transitions in word prediction. This result indicates that typical storylines included in our lyrics datasets were effectively captured as a probabilistic distribution of transitions over latent topics of segments. Second, the model incorporating a latent theme variable on top of topic transitions outperformed the models without such variables in both word prediction and segment order prediction. From this result, we can conclude that considering the notion of theme does contribute to the modeling of storylines of lyrics.