Yosuke Takashima

Multimodal definite clause grammar

This paper describes the first reported grammatical framework for a multimodal interface. The authors have developed Multimodal Definite Clause Grammar (MM-DCG), an extension of Definite Clause Grammar. The major features of MM-DCG include the capability to handle an arbitrary number of modes and temporal information in grammar rules. Further, the techniques for translating MM-DCG rules into Prolog predicates are described in detail.

Design tool combining keyword analyzer and case-based parser for developing natural language database interfaces

We have designed and experimentally implemented a tool for developing a natural language systems that can accept extra-grammatical expressions, keyword sequences, and linguistic fragments, as well as ordinary natural language queries. The key to this tools efficiency is its effective use of a simple keyword analyzer in combination with a conventional case-based parser. The keyword analyzer performs a majority of those queries which are simple data retrievals. Since it uses only keywords in any query, this analyzer is robust with regard to extra-grammatical expressions. Since little labor is required of the application designer in using the keyword analyzer portion of the tool, and since the case-based parser processes only those queries which the keyword analyzer fails to interpret, total labor required of the designer is less than that for a tool which employs a conventional case-based parser alone.

Lessons Learned from Deployed CBR Systems and Design Decisions Made in Building a Commercial CBR Tool

This paper reports our experiences in several CBR system building projects. Based on the lessons learned in these projects, we have developed a help desk support tool, Help Desk Builder. It consists of a knowledge management tool and a customer information management tool. The knowledge management tool includes case-based retrieval functions. This paper focuses on the design decisions and architecture of the case-based functions in Help Desk Builder.

Texture based 3D shape reconstruction from multiple stereo images

Describes a 3D shape reconstruction method including an improved stereo matching technique and a new shape connection technique. To reconstruct the 3D shape for an object, a stereo camera is moved around the object and several stereo images of the object are captured from different view points. Each stereo image is then analyzed to obtain the range data using the improved stereo matching technique. Each obtained range data represents shape for a part of the object. These object part shapes are connected to reconstruct the 3-D shape for the object, using the proposed shape connection technique. The shape connection technique reconstructs the shape for an object using object part shapes themselves and textures on the object part shapes. It is not necessary for this technique to measure camera positions and directions accurately when stereo images are captured.<<ETX>>

Multi-Modal-Method: a design method for building multi-modal systems

This paper describes Multi-Modal-Method, a design method for building grammar-based multi modal systems. Multi-Modal-Method defines the procedure which interface designers may follow in developing multi-modal systems, and provides MM-DCG, a grammatical framework for multi-modal input interpretation. Multi-Modal-Method has been inductively defined through several experimental multi-modal interface system developments. A case study of a multi-modal drawing tool development along with Multi-Modal-Method is reported.

Understanding of stories for animation

This paper presents the story understanding mechanism for creating computer animation scenarios. The story understanding mechanism reads a natural language story and creates its scenario for realistic graphic animations. This paper presents three types of hidden actions and relations of actions that must be discovered for realistic animations of stories but which are not explicitly described in the stories. They are: 1) causality check among actions; 2) interpolation of a continuous action beyond a sentence; 3) interpolation of hidden actions between neighboring sentences. This paper also describes the inference mechanism which recognizes the need for interpolation of these hidden actions. Multiple TMS is introduced in the mechanism. The knowledge base is action-oriented, hence it is independent of individual stories domains.

Personal computer-music system-song transcription and its application

A description is given of an easy-to-use computer music system using a personal computer, which can transcribe a song into a music score and generate accompaniment to match the melody. The system consists of a 16 bit personal computer, an A/D (analog to digital) converter, a digital signal processor, a microphone, and a music synthesizer. The sounds of a song being sung or hummed are analyzed for pitch and strength. Then, the melody of the song is extracted using the computers preprogrammed musical knowledge. After the transcription is completed, a chord progression is determined automatically and three-part accompaniment is generated on the rhythm pattern selected interactively. The melody and generated accompaniment are played together by musical synthesizer. >

A construction tool for context-sensitive guidance system

We have designed and experimentally implemented a tool for developing intelligent on-line guidance systems for electronic appliances and software programs. The key to this tool’s efficiency is its capability to generate effectively context-sensitive answers to users’ queries. Since the guidance system holds the statetransition representation of its target systems and receives all user operational inputs, it can simulate the internal states of the target systems. Any user’s query is interpreted as a user goal ;vithin a specific context, and an internal planner generates the best plans to meet the goal. The planner’s knowledge is defined declaratively for easy extension.

An Integrated Computer Animation Environment

For those who are not familiar with computers and animation, there are three problems in making animation easily, i.e. making actors’ shapes, defining actors’ actions and directing animation. This paper introduces generating actors model, which consists of the shape module and the action module, and camera work model. We realized an integrated animation environment based on these models. The environment consists of three subsystems, making actors’ shape, defining actors’ actions, and directing animation (Fig. 1). Because three subsystems are tightly connected, user can browse among the subsystems.