We propose an efficient method to disambiguate English nouns using scene information as context. Recent research directions into context dependent analysis point out the difficulty of defining context and acquiring knowledge. Another problem is the efficiency of resolving semantic constraints. To resolve these problems, we use knowledge from a pictorial dictionary available on the market. It provides spatialscene information as extra-lingual knowledge required for discourse analysis. One of its targets is to disambiguate words. We compared our method with random search approach on a narrative story. The experimental analysis supports that the worddisambiguating speed with our method is over 1.5 times faster than that with the random search. Also it indicates that the disambiguating rate with our method is 83%, higher than that with the random search (51%). We discuss the importance of the representational type for the scene information, evaluate our method's limits, and argue future technologies required to analyze the spatial-scene.

View full abstract

Analyzing compound nouns is one of the crucial issues for natural language processing systems. Registering all compound nouns in a dictionary is an impractical approach, since we can create a new compound noun by combining nouns. Therefore, a mechanism to analyze the structure of a compound noun from the individual nouns is necessary. However, the analysis are difficult only when using syntactic knowledge. Therefore, we have to use semantic knowledge. It is hard to construct and maintain a large semantic knowledge base, so we need a method to acquire semantic knowledge and use such the knowledge for the analysis. In this paper, we propose a method to analyze structures of Japanese compound nouns by using word collocational information and a thesaurus. The collocational information is acquired from a corpus of four kanzi character words. For each possible structure of a compound noun, the preference is calculated based on this collocational information. An experiment is conducted with 160, 000 word collocations to analyze compound nouns of with an average length of 5.5 characters. The accuracy of this method is about 78%.

View full abstract

This paper concentrates on research into the translation of compounds in a transfer based MT system. We determine how contemporary linguistic theory can contribute to the characterisation, representation and translation of compounds. We determine productive, compositional and translatable compounds. We discuss evolving strategies for effecting translation of these compound types. The results obtained are readily adaptable to models which have a stratificational linguistic framework are able to emulate feature value percolation. The research undertaken was reductionist in nature, leading to a set of compositionally translatable productive compound types being isolated.

View full abstract

It is important to clarify referents of nouns in machine translation and conversational processing. We present a method for estimating referents of nouns in Japanese sentences using referential properties, modifiers, and owners of nouns. Since there is no article in Japanese language, it is difficult to decide whether two nouns have the same referent in Japanese. To solve this difficult problem, we had made a research to estimate referential properties of nouns that correspond to articles using words in the sentences, and we estimate referents of nouns using these referential properties. For example if the referential property of a noun is definite, the noun can refer to nouns that appear ahead, and if the referential property of a noun is indefinite, the noun can not refer to nouns that appear ahead. Furthermore we estimate referents of nouns using modifiers and owners of nouns more precisely. As a result, we obtained a precision rate of 82% and a recall rate of 85% in the estimation of referent of nouns that have antecedents on training sentences, and obtained a precision rate of 79% and a recall rate of 77% on held-out test sentences. We verified that it is effective to use referential properties, modifiers, and owners of nouns through control experiments.

View full abstract