TL;DR
This paper introduces a novel "Sign Expansion Approach" to build more compact and flexible lexicons. Instead of storing individual word entries, it uses abstract "minimal signs" (conceptual/phonological frames) and a set of expansion rules to dynamically generate word entries, leading to reduced redundancy and improved maintenance.
What Problem Does This Paper Address?
Traditional lexicon systems often suffer from two major drawbacks: they become excessively large and are restricted to specific domains. This bloat arises from the need to repeat lexical information to ensure fast access, leading to maintenance headaches and obscuring important lexical relationships. While inheritance hierarchies offer a path to compactness, existing syntactic hierarchies are often unclear, and semantic hierarchies introduce their own complexities due to the intricate interplay between semantic and syntactic properties. Furthermore, many lexicons are heavily tied to specific syntactic theories, limiting their flexibility.
Methods
The core of this paper is the "Sign Expansion Approach," built upon the "Sign Model" (SM) by Hellan and Dimitrova-Vulchanova. This approach views lexicons as dynamic rule systems.
- Minimal Signs: Instead of word entries, the lexicon stores abstract "minimal signs." These signs consist of a semantic (conceptual) part and a realizational part (graphemic representation). The semantic part captures grammar-relevant meaning, analyzing situations with participants (arguments) and their roles (e.g., SOURCE, CONTROLLER), aspectual values, and selectional properties.
- Lexical Expansion Rules: A suite of expansion rules is used to generate dynamic entries for actual words from these minimal signs. These rules are categorized into:
- Morpho-syntactic augmentations: Add word category and inflectional paradigms.
- Inflections: Generate specific inflected word forms.
- Conceptual expansions: Extend the semantic part of signs, explaining how a sign can support various subcategorization frames.
- Syntactic mappings: Derive syntactic properties from conceptual structures, mapping them to grammatical functions (e.g., using LFG).
- Compositions: Combine two sign structures to create a new compound structure.
This system aims to be theory-independent, though specific syntactic theories (like LFG) are used for mapping rules.
Datasets
Not discussed in the paper. The paper focuses on the theoretical framework and implementation of the TROLL lexicon project, which is building a lexicon of Norwegian verbs.
Key Insights and Ideas
- Shifting from Word Entries to Conceptual Frames: The fundamental shift from storing individual words to abstract conceptual frames is key to achieving compactness and flexibility. This allows a single frame to support multiple word forms and categories derived from the same semantic basis.
- Dynamic Generation of Lexical Entries: The lexicon is not a static collection but a dynamic system. Word entries are generated on demand by applying expansion rules to minimal signs, making the lexicon adaptable and reducing redundancy.
- Semantic-Centric Organization: The approach prioritizes semantic relationships as the primary organizing principle, which is argued to be more robust and generalizable than purely syntactic hierarchies.
- Theory Independence: While specific syntactic theories are used for implementation details (like LFG), the core Sign Model and expansion approach are designed to be adaptable to various syntactic frameworks.
Key Results
The paper presents a theoretical framework and outlines its implementation in the TROLL lexicon project. Specific quantitative results are not detailed, but the approach promises:
- More Compact Lexicons: By avoiding redundancy and storing abstract frames, lexicons are significantly smaller.
- Increased Flexibility: The dynamic generation of entries allows for multi-purpose use and easier adaptation to different linguistic phenomena.
- Improved Maintenance: A more structured and less redundant lexicon is easier to maintain and update.
- Demonstrated Generation: Examples are provided showing how minimal signs like "PAINT" and "WALK" can be expanded into various word forms (e.g., "painty," "paint," "painting," "walked") with their associated semantic, syntactic, and morphological information.
Limitations and Future Work
- Psychological and Perceptual Matters: The authors acknowledge that the theory is still in development concerning psychological and perceptual aspects of language.
- Interface with Syntactic Theories: Future work includes establishing proper interfaces with various syntactic theories to integrate the system with existing parsers and generators.
- Evaluation: The ongoing TROLL project aims to evaluate the lexicon structures and the Sign Model through the construction of a complete Norwegian lexicon.
Summary updated