Toshiro Makino

Syntactic Filtering and Content-Based Retrieval of Twitter Sentences for the Generation of System Utterances in Dialogue Systems

Sentences extracted from Twitter have been seen as a valuable resource for response generation in dialogue systems. However, selecting appropriate ones is difficult due to their noise. This paper proposes tackling such noise by syntactic filtering and content-based retrieval. Syntactic filtering ascertains the valid sentence structure as system utterances, and content-based retrieval ascertains that the content has the relevant information related to user utterances. Experimental results show that our proposed method can appropriately select high-quality Twitter sentences, significantly outperforming the baseline.

On the difficulty of improving hand-crafted rules in chat-oriented dialogue systems

In this work, we explore the difficulty of improving hand-crafted rules in chat-oriented dialogue systems. We first created an initial rule set in artificial intelligence markup language and revised it through an iterative cycle. Then, we tested the initial and revised rule sets by using human participants. The dialogue experiment showed that, despite the intensive revision process, the overall performance had little improvement. We investigated the errors made by the systems with each of the two rule sets, leading to the conclusion that it is the tracking of the context and the coverage of questions that are hindering the improvement of hand-crafted rules.

Utterance Selection Using Discourse Relation Filter for Chat-oriented Dialogue Systems

We propose a novel utterance selection method for chat-oriented dialogue systems . Many chat-oriented dialogue systems have huge databases of candidate utterances for utterance generation. However, many of these systems have a critical issue in that they select utterances that are inappropriate to the past conversation due to a limitation in contextual understanding. We solve this problem with our proposed method, which uses a discourse relation to the last utterance when selecting an utterance from candidate utterances. We aim to improve the performance of system utterance selection by preferentially selecting an utterance that has a discourse relation to the last utterance. Experimental results with human subjects showed that our proposed method was more effective than previous utterance selection methods.