Nowadays, new domain-specific knowledge is created as new technology is developed. Domain-specific knowledge is represented with relationships between domain-specific concepts that are represented with technical terms in texts. Therefore, domain-specific knowI-edge acquisition can be defined as recognizing technical terms and relationships between them. We need consistent knowledge about technical terms for effective technical term recognition and effective domain-specific knowledge acquisition. Terminology management is a technology for consistently collecting, analyzing, and publishing technical terms. For terminology management, it is important to recognize technical terms that is to be man-aged and to localize them. When technical terms are localized, translation knowledge about existing technical terms and characteristics of localized technical terms should be considered. This paper describes an English term recognition model for acquisition of coined terms. Then a transliteration model and a term constituent alignment model for acquisition of English-Korean translation knowledge for localizing recognized technical terms are described.
"Term recognition" is a task to find technical terms for given domain and domain-specific text. It is composed of `term extraction' that extracts candidates to be a technical terms from texts and `term selection' that finds technical terms in the list of terms derived from `term extraction' step. `Term selection' is a task to rank a term list depending on termhood (ranking problem) and to find the boundary between term and non-term (boundary finding problem). To effectively handle the two problems related to term selection, I propose a term selection model making use of binary classification and various features such as orthographical, morphological, lexical, syntactic, and dictionary feature.
Two transliteration models are used for transliteration knowledge acquisition. Correspondence based transliteration model, which utilizes correspondence between source language grapheme and phoneme, and ensemble transliteration model, which uses ensemble of grapheme-based transliteration model, phoneme-based transliteration model, and correspondence transliteration model make it possible to generate transliteration more correctly for a given English word. The ensemble transliteration model first generates transliterations using three different transliteration models and then ranks the transliterations depending on the occurrence in web data.
A term constituent alignment model recognizes translation of term constituents in English Korean technical term pairs. Characteristics of translation between term constituents de-rived from data analysis are used for modelling term constituent alignment. Our model can handle cross alignment, 1:n alignment, and n:1 alignment.
기술의 발전함에 따라 전문분야의 방대한 지식이 새롭게 생성되고 있다. 전문분야 지식은 주로 전문분야 개념간의 관계로 표현되기 때문에 전문분야 지식의 획득은 전문분야 개념을 파악하고 이들간의 관계를 파악하는 작업으로 정의된다. 효과적인 전문분야 지식 획득을 위해서는 전문분야 개념을 표현하는 전문용어를 효과적으로 파악하는 작업이 선행되어야 하며, 이를 위해서는 일관성 있는 전문용어 지식이 필요하다. 전문용어 관리(Terminology management)란 이러한 전문용어 지식을 일관되고 효율적으로 관리하는 기술로서 전문용어의 수집, 분석, 배포에 관련된 일련의 작업으로 정의된다. 전문용어관리를 위해서는 새로운 전문용어를 파악하는 작업이 선행되어야 하며, 파악된 전문용어를 자국어로 번역하는 것이 중요하다. 이 때, 새로운 전문용어의 번역은 기존 전문용어의 번역 지식과 자국어의 특성을 고려하는 것이 중요하다. 본 논문에서는 이러한 일련의 과정에서 새로운 전문용어를 문서에서 자동으로 파악하는 "전문용어 인식 모델", 전문용어의 번역지식을 자동으로 획득하기 위한 "자동 음차표기 모델"과 "조어단위 정렬 모델"을 제안한다.
"전문용어 인식"이란 일정 분야의 문서들에서 그 분야 특성을 반영하는 전문용어를 자동으로 파악하는 작업으로 정의된다.
