We propose a hybrid approach to coordinate structure analysis that combines a simple grammar to ensure consistent global structure of coordinations in a sentence, and features based on sequence alignment to capture local symmetry of conjuncts. The weight of the alignment-based features, which in turn determines the score of coordinate structures, is optimized by perceptron training on a given corpus. A bottom-up chart parsing algorithm efficiently finds the best scoring structure, taking both nested or non-overlapping flat coordinations into account. We demonstrate that our approach outperforms existing parsers in coordination scope detection on the Genia corpus.
Argumentation in artificial intelligence, often called computational dialectics, is rooted in Aristotle's idea of evaluating argumentation in a dialogue model. In contrast, Chinese traditional philosophy regards dialectics as a style of reasoning that focuses on contradictions and how to resolve them, transcend them or find the truth in both. A compromise is considered one way to resolve conflicts dialectically. In this paper, we formalize reasoning intended to derive a compromise. Both the reasoning and the compromise are defined on abstract lattices procedurally and declaratively, respectively. We prove that the reasoning is sound and complete with respect to the compromise. Then we define the concrete and sound algorithm for the reasoning on the lattice characterized by definite clausal language and generalized subsumption. Under some conditions, the reasoning offers a unified way to reason rationally whether a set of the premises is consistent or not. Such reasoning is outside the scope of logics that have the principle of explosion. Further, the compromise has a unique logical setting compared with other types of reasoning such as deduction, induction, and abduction. We incorporate the reasoning into arguments, and illustrate that the use of arguments with compromise contributes to realizing a compromise-based conflict resolution in argumentation.
Descriptions of attribute and quality are essential elements in ontology developments. Needless to say, science data are description of attributes of target things and it is an important role of ontology to support the validity of and interoperability between the description. Although some upper ontologies such as DOLCE, BFO, etc. are already developed and extensively used, a careful examination reveals some rooms for improvement of them. While each ontology covers quality and quantity, the mutual interchangeability among these ontologies is not considered because each has been designed intended to develop a ``correct'' ontology of quality and quantity. Furthermore, due to variety of ways of data description, no single ontology can cover all the existing scientific data. In this paper, we investigate ``quality'' and ``value'' from an ontological viewpoint and propose a conceptual framework to deal with attribute, property and quality appearing in existing data descriptions in the nanotechnology domain. This framework can be considered as a reference ontology for describing quality with existing upper ontology. Furthermore, on the basis of the results of the consideration, we evaluate and refine a conceptual hierarchy of materials functions which has been built by nanomaterials researchers. Through the evaluation process, we discuss an effect of the definition of a conceptual framework for building/refining ontology. Such conceptual consideration about quality and value is not only the problem in nanomaterials domain but also a first step toward advancement of an intelligent sharing of scientific data in e-Science.
Minimum error rate training (MERT) has been a widely used learning method for statistical machine translation to estimate the feature function weights of a linear model. MERT has an advantage to incorpolate an automatic translation evaluation metrics as BLEU scores to its objective function. Weight vector can directly be optimized with Line search algorithm using error surface on a given set of candidate translations. It efficiently searches the best parameter resulting the highest BLEU scores. In this paper, we presented a new training algorithm for statisitcal machine translation, inspired by MERT and Structural Support Vector Machines. We performed MERT optimization by maximizing the margin between the oracle and incorrect translations under the L2-norm prior. Our experimental results on Japanese-English speech translation task showed that BLEU scores obtained by our proposed method were much better than those obtained by MERT. We achieved the best improvement of BLEU about +3.0 over standard MERT.
In Japan, childhood injury prevention is urgent issue. Safety measures through creating knowledge of injury data are essential for preventing childhood injuries. Especially the injury prevention approach by product modification is very important. The risk assessment is one of the most fundamental methods to design safety products. The conventional risk assessment has been carried out subjectively because product makers have poor data on injuries. This paper deals with evidence-based risk assessment, in which artificial intelligence technologies are strongly needed. This paper describes a new method of foreseeing usage of products, which is the first step of the evidence-based risk assessment, and presents a retrieval system of injury data. The system enables a product designer to foresee how children use a product and which types of injuries occur due to the product in daily environment. The developed system consists of large scale injury data, text mining technology and probabilistic modeling technology. Large scale text data on childhood injuries was collected from medical institutions by an injury surveillance system. Types of behaviors to a product were derived from the injury text data using text mining technology. The relationship among products, types of behaviors, types of injuries and characteristics of children was modeled by Bayesian Network. The fundamental functions of the developed system and examples of new findings obtained by the system are reported in this paper.
The overall goal of this paper is to cross-lingually analyze multilingual blogs collected with a topic keyword. The framework of collecting multilingual blogs with a topic keyword is designed as the blog feed retrieval procedure. In this paper, we take an approach of collecting blog feeds rather than blog posts, mainly because we regard the former as a larger information unit in the blogosphere and prefer it as the information source for cross-lingual blog analysis. In the blog feed retrieval procedure, we also regard Wikipedia as a large scale ontological knowledge base for conceptually indexing the blogosphere. The underlying motivation of employing Wikipedia is in linking a knowledge base of well known facts and relatively neutral opinions with rather raw, user generated media like blogs, which include less well known facts and much more radical opinions. In our framework, first, in order to collect candidates of blog feeds for a given query, we use existing Web search engine APIs, which return a ranked list of blog posts, given a topic keyword. Next, we re-rank the list of blog feeds according to the number of hits of the topic keyword as well as closely related terms extracted from the Wikipedia entry in each blog feed. We compare the proposed blog feed retrieval method to existing Web search engine APIs and achieve significant improvement. We then apply the proposed blog distillation framework to the task of cross-lingually analyzing multilingual blogs collected with a topic keyword. Here, we cross-lingually and cross-culturally compare less well known facts and opinions that are closely related to a given topic. Results of cross-lingual blog analysis support the effectiveness of the proposed framework.
Here is discussed how to learn a large scale of ontology from Japanese Wikipedia. The learned ontology includes the following properties: rdfs:subClassOf (IS-A relationship), rdf:type (class-instance relationship), owl:Object/DatatypeProperty (Infobox triple), rdfs:domain (property domain), and skos:altLabel (synonym). Experimental case studies show us that the learned Japanese Wikipedia Ontology goes better than already existing general linguistic ontologies, such as EDR and Japanese WordNet, from the points of building costs and structure information richness.
A project called `Open life matrix' is not only a research activity but also real problem solving as an action research. This concept is realized by large-scale data collection, probabilistic causal structure model construction and information service providing using the model. One concrete outcome of this project is childhood injury prevention activity in new team consist of hospital, government, and many varieties of researchers. The main result from the project is a general methodology to apply probabilistic causal structure models as servicable knowledge for action research. In this paper, the summary of this project and future direction to emphasize action research driven by artificial intelligence technology are discussed.
Purpose of this study is to explore service design method through the development of support service for prevention and recovery from dementia towards science of lethe. We designed and implemented conversation support service via coimagination method based on multiscale service design method, both were proposed by the author. Multiscale service model consists of tool, event, human, network, style and rule. Service elements at different scales are developed according to the model. Interactive conversation supported by coimagination method activates cognitive functions so as to prevent progress of dementia. This paper proposes theoretical bases for science of lethe. Firstly, relationship among coimagination method and three cognitive functions including division of attention, planning, episodic memory which decline at mild cognitive imparement. Secondly, thought state transition model during conversation which describes cognitive enhancement via interactive communication. Thirdly, Set Theoretical Measure of Interaction is proposed for evaluating effectiveness of conversation to cognitive enhancement. Simulation result suggests that the ideas which cannot be explored by each speaker are explored during interactive conversation. Finally, coimagination method compared with reminiscence therapy and its possibility for collaboration is discussed.