Search

This paper focuses on the change of named entities over time and its influence on the performance of the named entity tagger. First, we analyze Japanese named entities which appear in Mainichi Newspaper articles published in 1995, 1996, 1997, 1998 and 2005. This analysis reveals that the number of named entity types and the number of named entity tokens are almost steady over time and that 70 ~ 80% of named entity types in a certain year occur in the articles published either in its succeeding year or in its preceding year. These facts lead that 20 ~ 30% of named entity types are replaced with new ones every year. The experiment against these texts shows that our proposing semi-supervised method which combines a small annotated corpus and a large unannotated corpus for training works robustly although the traditional supervised method is fragile against the change of name entity distribution.

This paper proposes a novel method to extract named entities including unfamiliar words which do not occur or occur few times in a training corpus using a large unannotated corpus. The proposed method consists of two steps. The first step is to assign the most similar and familiar word to each unfamiliar word based on their context vectors calculated from a large unannotated corpus. After that, traditional machine learning approaches are employed as the second step. The experiments of extracting Japanese named entities from IREX corpus and NHK corpus show the effectiveness of the proposed method.

This paper proposes an approach of processing Japanese compound functional expressions by identifying them and analyzing their dependency relations through a machine learning technique. First, we formalize the task of identifying Japanese compound functional expressions in a text as a machine learning based chunking problem. Next, against the results of identifying compound functional expressions, we apply the method of dependency analysis based on the cascaded chunking model. The results of experimental evaluation show that, the dependency analysis model achieves improvements when applied after identifying compound functional expressions, compared with the case where it is applied without identifying compound functional expressions.

We propose a novel method to expand a small existing translation dictionary to a large translation dictionary using a pivot language. Our method depends on the assumption that it is possible to find a pivot language for a given language pair on condition that there are both a large translation dictionary from the source language to the pivot language, and a large translation dictionary from the pivot language to the destination language. Experiments that expands the Indonesian-Japanese dictionary using the English language as a pivot language shows that the proposed method can improve performance of a real CLIR system.

Our research aim here is to build a CLIR system that works for a language pair with poor resources where the source language (e.g. Indonesian) has limited language resources. Our Indonesian-Japanese CLIR system employs the existing Japanese IR system, and we focus our research on the Indonesian-Japanese query translation. There are two problems in our limited resource query translation: the OOV problem and the translation ambiguity. The OOV problem is handled using target language's resources (English-Japanese dictionary and Japanese proper name dictionary). The translation ambiguity is handled using a Japanese monolingual corpus in our translation filtering. We select the final translation set using the mutual information score and the TFxIDF score. The result on NTCIR 3 (NII-NACSIS Test Collection for IR Systems) Web Retrieval Task shows that the translation method achieved a higher IR score than the transitive machine translation (using Kataku (Indonesian-English) and Babelfish/ Excite (English-Japanese) engine) result. The best result achieved about 49% of the monolingual retrieval.

This paper proposes an automatic method of detecting grammar elements that decrease readability in a Japanese sentence. The method consists of two components: (1) the check list of the grammar elements that should be detected; and (2) the detector, which is a search program of the grammar elements from a sentence. By defining a readability level for every grammar element, we can find which part of the sentence is difficult to read.