TL;DR
Building ontologies is hard, especially for domain experts without deep knowledge engineering skills. This paper introduces ROO, a Protégé plugin that uses a controlled natural language (Rabbit) and intelligent user interface techniques to make ontology construction more accessible to these experts, aiming to improve both the process and the quality of the resulting ontologies.
What Problem Does This Paper Address?
Creating comprehensive ontologies, which are structured representations of knowledge, is a crucial but often time-consuming and complex task. A major hurdle is the significant effort and specialized knowledge required, which often deters large organizations. Current ontology construction tools tend to cater to knowledge engineers, leaving domain experts—who possess the crucial subject matter knowledge—with limited options to contribute effectively. This gap can lead to ontologies that are either incomplete or don't accurately reflect the domain's nuances. The paper addresses the need for intuitive tools that empower domain experts to actively participate in ontology construction, leveraging their expertise without requiring extensive knowledge engineering training.
Methods
The core of ROO is its use of a controlled natural language called Rabbit. This language is designed to be intuitive for domain experts while still being formal enough for machine processing. ROO acts as a Protégé plugin, integrating with an existing, widely-used ontology editor.
Key techniques employed by ROO include:
- Intelligent User Interface: ROO provides task suggestions based on the current state of the ontology and user activity. For instance, it might prompt users to define the scope of the ontology or add descriptions to concepts.
- Natural Language Parsing: It leverages robust language technologies to parse English sentences entered by the user and translate them into OWL (Web Ontology Language), the standard for semantic web ontologies. This parsing is based on an extension of the GATE (General Architecture for Text Engineering) framework.
- Controlled Language (Rabbit): The tool guides users to input knowledge using Rabbit, which has a defined set of patterns (currently nine basic patterns are supported) that are then converted to OWL 1.1.
- User Guidance: A rule system based on the JBoss Drools engine implements the user guidance and task suggestion features.
- Immediate Feedback: The system provides instant feedback on user input, including error messages specific to the task at hand.
Datasets
Not discussed in the paper. The evaluation study involved users defining and extending an ontology of musical instruments, but no specific datasets were mentioned.
Key Insights and Ideas
- Bridging the Domain Expert-Knowledge Engineer Divide: The most significant insight is the potential for controlled natural languages and intelligent interfaces to democratize ontology creation. By abstracting away much of the complexity of formal ontology languages, ROO allows those with deep domain knowledge to contribute directly.
- Ontology Authoring Methodology: The tool's design is informed by practical experience from Ordnance Survey's methodology for building large ontologies with domain expert involvement. This suggests that practical, real-world workflows are crucial for developing effective ontology tools.
- Iterative Improvement: The evaluation process highlights the importance of iterative development. Initial usability issues were identified and addressed through cognitive walkthroughs and feedback from domain experts, leading to an improved version of the tool.
- Task Suggestions as a Guiding Force: The proactive task suggestions are a key feature, helping users navigate the ontology construction process and ensuring that important steps are not overlooked.
Key Results
The paper reports on an evaluation study with eight users defining and extending a musical instrument ontology. While initial feedback indicated that the music expert struggled with defining Rabbit sentences due to limited help, an improved version of ROO received positive feedback from both a music expert and a knowledge engineer. They found the task suggestions and the help with Rabbit particularly useful, indicating improved usability. Further studies are planned to evaluate efficiency and quality with domain experts at Ordnance Survey.
Limitations and Future Work
- Limited Rabbit Syntax Support: Currently, the Rabbit language processor only parses and converts nine basic Rabbit patterns into OWL. Future work includes extending support to the full Rabbit syntax.
- Ontology Validation and Inspection: While the tool assists in construction, comprehensive ontology validation and inspection are currently handled by Protégé. Future plans include implementing additional support for these aspects within ROO.
- Further Evaluation: The authors plan to conduct more extensive evaluations with domain experts at Ordnance Survey to assess the efficiency and quality of the ontology authoring process.
Summary updated