We propose Integrated Meta Translation - to integrate the output of multiple different translation engines - and present a sample implementation. Our goal is to improve the quality of machine translation by combining multiple machine translation outputs, and therewith hopefully to reduce translation costs. In our sample implementation, we combined six online machine translation services and made the result available as a Web service. Integration of the different translations is achieved by deriving the optimal path from a graph based on an N-gram rule. We explored the various ways to further improve quality: combining word nodes, extracting and translating partial sentences, and estimating translation quality but back-translation.

This paper describes a method of concept extraction using editing differences between pairs of SVG graphics and between the corresponding textual metadata. Differences for graphics are extracted based on the analysis of the tree structure of SVG, and are generalized to abstract from the specifics of targets and document structure. We verify the effectiveness of our method with an experiment generating new SVG graphics from textual descriptions, existing graphics and corresponding textual metadata.

Pictograms should be composed of several components, with an appropriate layout. However, it is difficult for non-experts to create pictograms that conform to established design principles and can be understood world-wide. We postulate that this problem is best solved by a framework that automatically eliminates inappropriate layouts and supports the creation of suitable layouts. In this paper, we propose an algorithm that extracts rules for pictogram layout based on input from a large number of people. We verify the effectiveness of our algorithm based on layout data obtained from several test subjects.

There are a lot of reports about using pictograms as an universal language. However, because pictograms can be laid out in various ways, without the knowledge of design principles it is difficult to create pictograms that can be universally understood. To solve this problem, pictogram creators should be aided by a system which automatically rejects inappropriate layouts and assists in creating suitable layouts. In this paper, we propose an algorithm that extracts rules for pictogram layout assuming input from a large number of people. This algorithm extracts layout rules for relative location, angle, size, and mirroring of pictogram components. In addition, by using trust values for input and output, our algorithm is able to work with small datasets and can be adapted to various needs. The effectiveness of the algorithm is verified based on the layout data created by several test subjects.

http://www.svgopen.org/2007/paperAbstracts/SVGOpen2007abstract.html