Sharing dictionaries among MT users by common formats and social filtering framework

17303

Authors

nan

Abstract

AbstractMT users have to build "user dictionaries" in order to obtain high-quality translation results. However, building dictionaries needs time and labor. In order to meet the speed of the information flow in the global network society, we need to have common formats for sharing dictionaries among different MT systems, and a new way of dictionary authorization, that is "social filtering".

Keywords

nan

Full Content

MT Summit VII Sept. 1999 Sharing Dictionaries among MT Users By Common Formats and Social Filtering Framework Shin-ichiro Kamei Human Language Technology Group C&C Media Research Laboratories NEC Corporation, Japan kamei@ccm.cl.nec.co.jp Abstract MT users have to build "user dictionaries" in order to obtain high-quality translation results. However, building dictionaries needs time and labor. In order to meet the speed of the information flow in the global network society, we need to have common formats for sharing dictionaries among different MT systems, and a new way of dictionary authorization, that is "social filtering". 1 User dictionaries for MT systems It is essential for MT users to build and use their own "user dictionaries" in order to use MT systems effectively. MT systems have "system dictionaries" which cover basic words and phrases, but these entries are not sufficient for translation in actual situations. Real documents contain topical and technical terms in special fields, and translation quality depends strongly on these terms. However, most of them are not registered in system dictionaries in advance, since they are produced daily in various fields. Hence users need to make new entries for these terms in their own dictionaries and build their own user dictionaries. This is a basic and effective way for obtaining high-quality translation results. Quantity and quality of dictionaries controls the quality of translation results. However, building dictionaries needs much time and labor. Users have to know equivalents of unregistered words and process of making a user dictionary that differs from an MT system to another, and have to spend much time to make new entries of each word and phrase. This means it is difficult for an individual user to make his/her dictionary large and good enough to meet the needs. In order to solve this problem, we need a new framework for building and using MT dictionaries. 1n particular we have to change our way of thinking for authorization of dictionaries. In addition, environments for sharing user dictionaries among users who utilize different MT systems play an important role. 2 Authorizing dictionaries by Social Filtering In general, dictionaries have been approved by authorities of linguistics and published by famous publishers. Users have relied on these authorities when using dictionaries. However, these authorized dictionaries are often useless for MT users in the present day and in future. New words and phrases in documents which users want to translate are not registered in these dictionaries, since speed of generating new terms is getting faster and faster these days, and this way of making dictionaries takes too much time for MT systems. In order to catch up with the speed of the multi- lingual information flow in the global network society, we have to find another way of authorizing dictionaries. One of the choices is voluntary building and social filtering of dictionaries. A trial of exchanging dictionaries was executed by one of MT vendors in Japan, NEC. They have opened a web site [1] for users of their MT products, and users - 180- …

Content truncated to 500 words

Labeled Text

AbstractMT users have to build "user dictionaries" in order to obtain high-quality translation results. However, building dictionaries needs time and labor. In order to meet the speed of the information flow in the global network society, we need to have common formats for sharing dictionaries among different MT systems, and a new way of dictionary authorization, that is "social filtering".

Publication Details

Paper ID:
17303
Year:
1999
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan