Abstract
Modern electronic dictionaries, while standardized for interoperability, often present an overwhelming amount of information to diverse user groups. This paper proposes a systematic approach to generate "personalized views" of these complex resources using the Lexical Markup Framework (LMF). By leveraging user profiles and automated filtering pipelines, the system extracts relevant sub-dictionaries tailored to specific expertise levels and fields of interest. A practical experiment on a large-scale Arabic dictionary demonstrates significant data reduction and successful automation of complex query generation.
Introduction
The normalization of electronic dictionaries according to international standards like ISO 24613 (LMF) has greatly improved data sharing and interoperability. However, this standardization has led to a "Too Much Information" problem. These dictionaries are massive and structurally complex, containing detailed morphological, syntactic, and semantic data that may not be relevant to all users. For instance, a primary school student requires different information than a professional linguist or a software developer. Currently, most systems lack the flexibility to provide customized subsets of data, forcing all users to navigate the same dense information architecture. This research addresses this gap by developing a method to automatically generate personalized "views" based on user-specific criteria.
Methodology
The proposed approach follows a four-step pipeline to transform a comprehensive master dictionary into a tailored sub-dictionary.
1. Projection of the View Model
The process begins by identifying the user's profile. This profile is defined by three primary criteria:
- Kind: The user's role (e.g., lexicographer, developer, student).
- Level: The user's expertise (e.g., beginner, intermediate, expert).
- Field: The specific domain of interest (e.g., general, medicine, biology).
This profile acts as an initial filter, determining which Data Categories (DCs) are mandatory, optional, or hidden for that specific user type.
2. Selection and Coherence Checkout
Users can further refine their view by selecting specific linguistic values (e.g., "Part of Speech = Verb"). To maintain the integrity of the resulting data, the system performs a "coherence checkout." This step uses predefined rules to ensure selections are linguistically logical. For example, in Arabic, if a user selects "Particle," the system automatically disables "Gender" attributes, as particles do not possess gender.
3. Automatic Generation
Once the model is finalized, the system automatically generates a Document Type Definition (DTD) for the new sub-dictionary structure and an XQuery to extract the relevant data from the master XML repository.
4. Refinement
The final step ensures the sub-dictionary is linguistically complete. Because dictionary entries are often interlinked (e.g., verbs linked to their roots), the refinement logic pulls in "neighboring" data necessary to maintain the validity of the lexical resource. The system supports both Virtual Views (calculated dynamically) and Materialized Views (saved as standalone XML files for offline use).
Practical Experiment and Results
The framework was evaluated using a normalized Arabic interactive dictionary containing over 38,000 lexical entries. The dictionary follows the LMF meta-model and includes rich morphological, semantic, and syntactic information.
In a key test case, a user requested a view restricted to Arabic verbs. The system successfully filtered the 38,000-entry database down to a specialized view of 10,000 entries (7,000 verbs and 3,000 associated roots). This demonstrates a significant reduction in data complexity while maintaining linguistic relevance. Furthermore, the automation of XQuery and DTD generation removes the need for users to possess deep technical knowledge of XML or LMF structures.
Conclusion
This research provides a robust solution for managing the complexity of modern electronic dictionaries. By utilizing user profiles and automated refinement logic, the system can deliver personalized, lightweight, and linguistically coherent views. While currently focused on Arabic, the underlying framework is designed for general applicability. Future work will involve testing the system on other languages and deploying it as a full-scale web service with a more intuitive user interface.
Summary updated