Ryohei Sasano

A Fully-Lexicalized Probabilistic Model for Japanese Zero Anaphora Resolution

This paper presents a probabilistic model for Japanese zero anaphora resolution. First, this model recognizes discourse entities and links all mentions to them. Zero pronouns are then detected by case structure analysis based on automatically constructed case frames. Their appropriate antecedents are selected from the entities with high salience scores, based on the case frames and several preferences on the relation between a zero pronoun and an antecedent. Case structure and zero anaphora relation are simultaneously determined based on probabilistic evaluation metrics.

Controlling Output Length in Neural Encoder-Decoders

Neural encoder-decoder models have shown great success in many sequence generation tasks. However, previous work has not investigated situations in which we would like to control the length of encoder-decoder outputs. This capability is crucial for applications such as text summarization, in which we have to generate concise summaries with a desired length. In this paper, we propose methods for controlling the output sequence length for neural encoder-decoder models: two decoding-based methods and two learning-based methods. Results show that our learning-based methods have the capability to control length without degrading summary quality in a summarization task.

Generating Live Sports Updates from Twitter by Finding Good Reporters

Twitter has emerged as a platform for crowds to express their opinions. Many Twitter users post their opinions, impressions, and statuses of televised events such as sports events. However, since the volume of such posts is extremely huge, it requires a lot of time and effort to understand what happens within events. We propose a method of generating live sports updates from Twitter posts on an event. Our method selects descriptive and prompt tweets that are posted within a short time after important sub events by exploiting users called good reporters, who promptly explain what is happening at each moment throughout the event. The experimental results indicated that our new technique generated more comprehensive updates than other methods presented in previous work.

The Effect of Corpus Size on Case Frame Acquisition for Discourse Analysis

This paper reports the effect of corpus size on case frame acquisition for discourse analysis in Japanese. For this study, we collected a Japanese corpus consisting of up to 100 billion words, and constructed case frames from corpora of six different sizes. Then, we applied these case frames to syntactic and case structure analysis, and zero anaphora resolution. We obtained better results by using case frames constructed from larger corpora; the performance was not saturated even with a corpus size of 100 billion words.

A Probabilistic Model for Associative Anaphora Resolution

This paper proposes a probabilistic model for associative anaphora resolution in Japanese. Associative anaphora is a type of bridging anaphora, in which the anaphor and its antecedent are not coreferent. Our model regards associative anaphora as a kind of zero anaphora and resolves it in the same manner as zero anaphora resolution using automatically acquired lexical knowledge. Experimental results show that our model resolves associative anaphora with good performance and the performance is improved by resolving it simultaneously with zero anaphora.

Automatic construction of nominal case frames and its application to indirect anaphora resolution

This paper proposes a method to automatically construct Japanese nominal case frames. The point of our method is the integrated use of a dictionary and example phrases from large corpora. To examine the practical usefulness of the constructed nominal case frames, we also built a system of indirect anaphora resolution based on the case frames. The constructed case frames were evaluated by hand, and were confirmed to be good quality. Experimental results of indirect anaphora resolution also indicated the effectiveness of our approach.

Context Representation with Word Embeddings for WSD

Word embeddings obtained through neural language models developed recently can capture semantic and grammatical behaviors of words and very capably find relationships between words. Such word embeddings are shown to be effective for various NLP tasks. In this paper, we develop a supervised method for word sense disambiguation (WSD) that employs word embeddings as local context features. Our experiments show the usefulness of word embeddings in the WSD task. We also compare the methods with different vector representations and reveal their effects on the WSD task.

Improving coreference resolution using bridging reference resolution and automatically acquired synonyms

We present a knowledge-rich approach to Japanese coreference resolution. In Japanese, proper noun coreference and common noun coreference occupy a central position in coreference relations. To improve coreference resolution for such language, wide-coverage knowledge of synonyms is required. We first acquire knowledge of synonyms from large raw corpus and dictionary definition sentences, and resolve coreference relations based on the knowledge. Furthermore, to boost the performance of coreference resolution, we integrate bridging reference resolution system into coreference resolver.

Context-Dependent Automatic Response Generation Using Statistical Machine Translation Techniques

Developing a system that can automatically respond to a user’s utterance has recently become a topic of research in natural language processing. However, most works on the topic take into account only a single preceding utterance to generate a response. Recent works demonstrate that the application of statistical machine translation (SMT) techniques towards monolingual dialogue setting, in which a response is treated as a translation of a stimulus, has a great potential, and we exploit the approach to tackle the context-dependent response generation task. We attempt to extract relevant and significant information from the wider contextual scope of the conversation, and incorporate it into the SMT techniques. We also discuss the advantages and limitations of this approach through our experimental results.

Generating Personalized Snippets for Web Page Recommender Systems

Web page recommender systems provide users with web pages they might be interested in. Users then select some of the recommended web pages that catch their interest by making a relevance judgment. However, if web page recommender systems do not offer enough useful information for the relevance judgment, users would end up reading irrelevant web pages or overlook relevant web pages. To provide information for the relevance judgment, we propose a novel method for generating personalized snippets for web page recommender systems. Our method directly uses the reasons the web pages are recommended to the user. This use of reasons enables snippets to be selected that better reflect the interest of the user. Moreover, our method can work with various web page recommender systems. It also leverages the maximum coverage summarization model to generate personalized snippets. The results of an experiment on a manually created dataset show that our method is more effective than a personalized summarization model.