Browse Prior Art Database

Learning Method for Correction of Affix-Related Errors

IP.com Disclosure Number: IPCOM000036066D
Original Publication Date: 1989-Sep-01
Included in the Prior Art Database: 2005-Jan-28
Document File: 2 page(s) / 75K

Publishing Venue

IBM

Related People

Nozaki, N: AUTHOR [+3]

Abstract

This article describes a new learning method applied to a process for correcting affix-related errors in Kana-to-Kanji Conversion (KKC). With the new method, the KKC system learns both an affix and also a compound word consisting of a content word and an affix.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 57% of the total text.

Page 1 of 2

Learning Method for Correction of Affix-Related Errors

This article describes a new learning method applied to a process for correcting affix-related errors in Kana-to-Kanji Conversion (KKC). With the new method, the KKC system learns both an affix and also a compound word consisting of a content word and an affix.

Since affixes are short and have many homonyms, they often cause conversion errors. For an affix which is used with a limited number of contentwords, an adequate approach is to keep the combinations as compound words in a dictionary. However, most affixes are used with so many content words that it is impossible to include all their possible combinations in a dictionary. A typical approach to this problem is to apply a learning method when an affix-related error is corrected, so that the corrected words can appear correctly later on. Conventional learning methods, however, have disadvantages, and often cause conversion errors that are contrary to the user's expectations.

(Image Omitted)

Conventional methods:

There are two typical learning methods for correction of affix- related errors; one is to learn an affix (Method 1), and the other to learn a compound word (Method 2). Both the methods have disadvantages; the learning effect of Method 1 is cancelled out by the learning of another homonym affix, and Method 2 may require frequent homonym selection even for the same affix. 4D The left half of Fig. 1 shows a case where Method 1 has a side- effect. Supp...