Lexical Databases and Ontologies
Machine-readable inventories of word meanings and their relations — WordNet, FrameNet, and related ontologies — that give computational systems structured access to lexical semantics.
Definition
A lexical database or ontology is a structured resource that records word senses and the semantic relations among them in a form usable by computer programs.
Scope
Covers the design and use of lexical-semantic resources: WordNet's synsets and hierarchical relations, FrameNet's semantic frames and roles, and broader computational ontologies. It addresses how such resources encode synonymy, hypernymy, and predicate-argument structure and how they support disambiguation and inference. Distributional and neural lexical semantics are covered under computational semantics.
Core questions
- How does WordNet organize senses into synsets linked by lexical relations?
- How does FrameNet represent meaning in terms of frames and roles?
- How are these resources used for word-sense disambiguation and similarity?
- What are the limits of hand-built lexical ontologies?
Key concepts
- synset
- hypernymy and hyponymy
- meronymy
- semantic frame
- frame element
- word-sense disambiguation
- lexical ontology
- predicate-argument structure
Key theories
- Synset-based lexical organization
- Modeling the lexicon as sets of synonyms (synsets) connected by relations such as hypernymy and meronymy, as in WordNet, giving a graph structure to word meaning.
- Frame semantics
- Representing word meaning via the situations (frames) it evokes and the participant roles within them, operationalized as a lexical resource in FrameNet.
History
WordNet, begun by George Miller's group in the 1980s and documented in 1995 and 1998, became the most widely used lexical resource in computational linguistics. Fillmore's frame semantics was operationalized as FrameNet in the late 1990s, complementing WordNet with explicit predicate-argument structure.
Debates
- Hand-built resources versus learned representations
- Whether curated lexical databases remain valuable in the era of learned embeddings; many argue they still provide interpretable structure and ground truth that distributional models lack.
Key figures
- George Miller
- Christiane Fellbaum
- Charles Fillmore
- Collin Baker
Related topics
Seminal works
- miller1995
- fellbaum1998
- baker1998
Frequently asked questions
- What is a synset in WordNet?
- A synset is a set of words that are interchangeable in some context — synonyms — grouped as a single concept and linked to other synsets by relations like 'is-a' and 'part-of'.