A DAML+OIL-Compliant Chinese Lexical Ontology

Yu-Sheng Lai, Ren-Jr Wang, Wei-Tek HsuCOLING 2002

TL;DR

This paper introduces the construction of the Chinese Lexical Ontology (CLO), a DAML+OIL-compliant resource designed to imbue Chinese language data with machine-understandable meaning. By integrating a hierarchical taxonomy derived from HowNet with traditional Chinese lexicons, the CLO aims to bridge the gap in semantic web technologies for Chinese, enabling advanced applications like natural language understanding and information retrieval.

What Problem Does This Paper Address?

The Semantic Web, envisioned as a web of data with well-defined meanings, relies heavily on formal ontologies to structure information for machine comprehension. While significant progress has been made in developing ontologies for English, the development of similar resources for Chinese has lagged. This gap hinders the application of Semantic Web technologies to the vast amount of Chinese language data, limiting capabilities in areas such as natural language understanding, information retrieval, and intelligent agents. The authors aim to address this by creating a comprehensive, machine-readable Chinese ontology that can serve as a foundation for these advanced applications.

Methods

The CLO construction involves three main components:

  1. Hierarchical Taxonomy of Concepts: This is primarily derived from the existing HowNet resource. HowNet's "primary features" are organized into a hierarchical taxonomy, which the authors adopt as the core conceptual structure for the CLO. This taxonomy is then refined by eliminating duplicates, resulting in 1,075 distinct concepts.
  2. Lexical Entries Associated with Concepts and Relations: The CLO integrates lexical entries from two sources:
  3. HowNet: This provides 65,961 simplified Chinese lexical entries, each defined by primary and secondary features. The primary features are used to link these entries to the ontology's concepts.
  4. Sinica Dictionary: To address the lack of traditional Chinese lexicons, the authors incorporate a traditional Chinese dictionary from Academia Sinica. This involves a semi-automatic process to map Sinica's entries to the CLO's taxonomy, handling both completely and incompletely matched entries.
  5. Axioms Describing Constraints: While mentioned as a component, the paper does not delve into the specifics of constructing these axioms.

The core methodology involves leveraging existing resources (HowNet and Sinica Dictionary) and developing a semi-automatic approach for lexicon integration to create a DAML+OIL compliant ontology.

Datasets

  • HowNet: Used as the primary source for the hierarchical taxonomy of concepts and a set of simplified Chinese lexical entries (65,961 entries).
  • Sinica Dictionary: A traditional Chinese dictionary compiled by Academia Sinica, containing 78,322 words (80,491 lexical entries when considering parts-of-speech). This is used to enrich the CLO with traditional Chinese lexicons.

No new benchmarks were introduced; rather, existing resources were adapted and integrated.

Key Insights and Ideas

  • Bridging the Chinese Semantic Web Gap: The CLO project directly addresses the underrepresentation of Chinese in Semantic Web technologies, a critical step for global information integration.
  • Leveraging Existing Resources: The authors demonstrate a practical approach by building upon established resources like HowNet and the Sinica Dictionary, rather than starting from scratch. This is a common and effective strategy in ontology engineering.
  • The Power of Lexical-Conceptual Integration: The CLO's strength lies in its dual nature: a structured conceptual hierarchy combined with rich lexical information. This makes abstract concepts more concrete and facilitates natural language processing.
  • Semi-Automatic Lexicon Mapping: The challenge of mapping a large traditional Chinese lexicon to an ontology is tackled with a semi-automatic approach, acknowledging the need for efficiency while maintaining accuracy. The handling of multi-conceptual entries is a particularly interesting aspect.

Key Results

The CLO, as of the paper's writing, contains:

  • 1,075 concepts.
  • 65,961 lexical entries associated with concepts.
  • 299 relations among concepts (excluding "subClassOf").
  • 27,004 relations between lexical entries and concepts.
  • 79,723 relations associating lexical entries with concepts.

The paper highlights the ongoing work in classifying unmatched traditional Chinese lexical entries and associating relations with lexical entries.

Limitations and Future Work

  • Unmatched Lexical Entries: A significant number of traditional Chinese lexical entries from the Sinica Dictionary remain to be classified into the CLO taxonomy. The authors acknowledge that manual classification of this large volume is challenging and are exploring more efficient approaches, potentially involving mapping tables based on POS.
  • Relation Association: Relations are not yet fully associated with lexical entries, which is identified as a key area for future research.
  • Axiom Construction: The paper mentions axioms as a component but does not detail their construction, suggesting this is a future task.

The future work focuses on completing the lexicon classification and enriching the ontology with relation-lexicon associations, ultimately aiming for a more robust and comprehensive Chinese Semantic Web foundation.

Summary updated

BibTeX
@inproceedings{lai2002daml,
  author = {Yu-Sheng Lai and Ren-Jr Wang and Wei-Tek Hsu},
  title = {{A DAML+OIL-Compliant Chinese Lexical Ontology}},
  booktitle = {International Conference on Computational Linguistics},
  year = {2002},
}
Figure