跨語言知識本體建置實務之探討—以地理空間資訊領域為例

Abstract

知識本體作為語意網的核心元素之一，是知識組織系統工具中結構最為嚴謹、可以呈現物件間複雜關係、且可以供機器所讀取的工具。在巨量資料時代下，知識本體愈漸重要，在地裡空間資訊領域更是如此；由於需要和世界上其他各國不同語言的知識本體進行互通，「跨語言」知識本體的建置也逐漸開始發展，但是目前對於跨語言知識本體的建置方法上，尚未能有一詳細實務架構與操作流程可以依循，基於此，本研究之目的乃探索建構中英文跨語言知識本體的建置方法及實作流程的可行性，期望該方法能夠作為日後國內跨語言知識本體建置之參考架構。 本研究採用三階段的研究設計，首先擬就知識本體之建置方法、及知識本體對應方法進行文獻探討；第二階段將國際間地理空間資訊具代表性的知識本體SWEET知識本體和國內的國家教育院學術名詞網做中英文語言對應的處理，藉此發展一跨語言的知識本體之雛形架構。第三階段經由中英文知識本體間之對應實作，探索此知識本體之可行性。 本研究藉由上述之研究設計，將知識本體建置方法歸納為「從無到有」、「以知識組織系統為基礎的方法」以及「沿用現有知識本體」三種方法；不同語言間知識本體對應的方法則區分為「人工處理」和「自動或半自動化處理」兩種方式。 在第二階段中英文對照的過程中，本研究運用「以知識組織系統為基礎」以及「沿用現有知識本體」的方法，採用半自動化的方式，研究結果模擬出中英文對照的架構及詳細操作步驟。在中英對照的結果中，以Microsoft Access作為輔助工具，成功對照到80.66%完全等同的詞彙。 在第三階段的知識本體實作結果中，本研究亦研擬出實作的「中間轉換檔」架構以及轉製成知識本體.owl檔案於Protégé軟體中實作的詳細操作步驟，並且運用SKOS語言的屬性豐富中文同義詞、近義詞、相關詞之間的關係。本階段最後以簡單查詢中英文類別的方式，確立本研究知識本體中間轉換檔確實可作為後續參考之實務架構。

Ontologies, as the fundamental building blocks for the Semantic Web, are the highest-level classification scheme in the family of knowledge organization systems. With the emergence of big data, ontologies are keys to unraveling the information explosion problems, especially in the geospatial information domain. Under the big data situation, other language cultures, not limited to English, are also in a pressing need to construct ontologies. Many tries to be interoperable with ontologies written in other languages, but what lacks are a successful cross-lingual ontology mapping method, and a detailed mapping model for others to follow. The purpose of this study thus is to investigate such methodology for constructing a cross-lingual ontology, in hoping that the model and constructing steps can be recognized as the de-facto practice for future research. By using a three-phase design methodology, this study begins by reviewing literature on building ontologies and ontology mapping methods. In phase two, we try to map the geospatial information ontology—SWEET ontology—with the termlists from National Academy of Educational Research in Taiwan. In phase three, we model the mapped English/Chinese ontology in Protégé software to explore the prospect of this method. The results in phase one suggests that there are mainly three types of ontology building methods— starting from scratch, KOS-based, and using existing ontologies. As to ontology mapping methods, we divide them by either manual-processing or automatic/semi-automatic processing methods. In phases two and three, we propose a cross-lingual ontology mapping model and provide an actual step-to-step guide to produce a “switch” for connecting ontologies in different formats and languages. We have also used SKOS relationships to Chinese terms in our ontology to express synonyms and related terms. The semi-automatic mapping result from English to Chinese shows 80.66% accuracy on the exact-match terms; and the search process for the Chinese and English classes in Protégé have proven the feasibility of the practice in this study.