| 研究実績の概要 |
The research aims to develop methods of visualising and making accessible semantic information, e.g., predicate argument information, but also higher levels of analysis, such as propositional connectives as well as modal and negation operators. Such information has enabled, for example, the mapping out of language binding dependencies, which has proved particularly relevant as a method to reconstruct unpronounced argument information (zero pronouns) for Japanese.
To carry out this work it has been necessary to continue developing a method for reaching semantic representations automatically from syntactic parsed representations and to create a large base of already analysed and human checked syntactic structures that can be transformed to semantic representations. The establishment of such a base forms training data for creating yet more like data, with the potential to scale to large volumes of data.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
1: 当初の計画以上に進展している
理由
The work on visualisation has focused on ways to add descriptive power. Innovations have included (i) adding coded indexing carrying ontological information and (ii) folding represented material around "inverse roles that compact hierarchical structure. The presence of indexing has to a degree cluttered the visualisation but it has made possible flipping between different views of dimensions of content, e.g., with a dimension to best capture sentence content, and a dimension to best capture an overview of total discourse content. In contrast to indexing, the addition of inverse roles has greatly simplified visualisation and accessibility. As a side effect, this has created the foundation for supporting a method of natural language generation, that is, a way to get back to natural language from a semantic representation.
|
| 今後の研究の推進方策 |
Now that research is making semantic representations available of a high quality it has become possible to think about supporting natural language generation from meaning representations. This has enabled taking a natural language sentence, going to its meaning representation, and then back to a natural language sentence. Keeping to the same language tests the combined success of building meaning representations and of generating output. Switching languages when manipulating meaning representations would achieve machine translation. I am currently investigating transfer shortfall seen with meaning representations built from parsed parallel corpora data of Japanese-English.
I also aim to start supporting the creation of searching for semantic information by transforming semantic information into search patterns of tree extraction languages that are able to recover related semantic information.
|
