GWA Base Concepts
Introduction
The notion of Base Concepts was introduced in the EuroWordNet project to reach maximum overlap and compatibility across wordnets in different languages, while at the same time, allow for the distributive development of wordnets in the world, each wordnet being a language specific structure and lexicalization pattern. The Base Concepts are supposed to be the concepts that play the most important role in the various wordnets of different languages. This role was measured in terms of two main criteria:
- A high position in the semantic hierarchy;
- Having many relations to other concepts;
The Base Concepts are thus the fundamental building blocks for establishing the relations in a wordnet and give information about the dominant lexicalization patterns in languages. Base Concepts should not be confused with Basic Level Concepts as defined by Rosch (1977). Basic Level Concepts are the result of a compromise between two conflicting principles of categorization:
- Apply to as many concepts as possible;
- Apply as many features as possible;
As a result of this, Basic Level Concepts typically occur in the middle of hierarchies and less than the maximum number of relations. Base Concepts mostly involve the first principle only. They are generalizations of features or semantic components and thus apply to a maximum number of concepts.
Universality
The following types of Base Concepts have been distinguished:
- Common Base Concepts (CBC): concepts that act as Base Concepts in at least two languages;
- Local Base Concepts (LBC): concepts that act as Base Concepts in only a single language;
- Global Base Concepts (GBC): concepts that act as Base Concepts in all languages of the world;
The selection of the Base Concepts is an approximation based on:
- Structural properties of the wordnets in different languages;
- An equivalence mapping of the synset in a wordnet in Language X to a synset in the Princeton WordNet, initially to version 1.5 currently to version 2.0.
The structural properties of wordnets are partially arbitrary and thus only weakly indicative. The idea has been so far that independent selections from a large number of languages will still give a good approximation. As properties have been used:
- Position in the hierarchy and number or relations, if available;
- Frequency in the definitions or glosses, if available;
- Morphological complexity and dependency;
Sense frequencies are not available and word frequencies were shown to be unreliable. Furthermore, it should be noted that many wordnets are developed by expanding from Princeton WordNet and therefore do not contribute to the definition of the Global Base Concepts. Roughly two approaches have been followed for building wordnets:
- Expand approach: translate the synsets in the Princeton WordNet to your own language, take over the relations from Princeton and revise;
- Merge approach: define synsets and relations in your own language and then align your wordnet with the Princeton WordNet using equivalence relations;
Obviously, the merge aproach would give more independent suggestions for BCs. However, the expand approach can still contribute if the resulting local wordnet structure is revised and validated in a later phase and afterwards makes a selection according to the same criteria.
In EuroWordNet, an initial set of 1024 Common Base Concepts (CBCs) were selected and defined as Princeton WordNet1.5 synsets. These CBCs play a BC role in at least two independent wordnets. The languages in EuroWordNet are: English, Dutch, German, French, Spanish, Italian, Czech and Estonian, but for the initial selection only English, Dutch, Spanish and Italian were used. For the 1012 CBCs, EuroWordNet defined a top-ontology that has been the common semantic framework for defining the relations in each individual wordnet separately. On the next page you can find a definition of the EuroWordNet CBCs and the top-ontology classification: EuroWordNet Base Concepts and Top-Ontology
In the BalkaNet project, a similar approach was applied to another set of languages: Greek, Romanian, Serbian, Turkish, Bulgarian. BalkaNet extended the set to 4689 synsets and upgraded the mapping of the CBCs to Princeton WordNet 2.0. The 5000 CBCs as WordNet2.0 synsets can be downloaded here:
4689 Common Base Concepts from EuroWordNet and BalkaNet as Wordnet2.0 synsets
The role of Base Concepts for Building wordnets
The Base Concepts have been defined so far in two European projects, EuroWordNet and BalkaNet. They played a crucial role in building the wordnets. More information can be found in this powerpoint presentation: Building Wordnets, by Piek Vossen. Below is a short description of the approach.
Each wordnet was developed in two phases according to a top-down approach:
- Develop a core wordnet around the CBCs that is highly compatible in coverage and semantic interpretation;
- Extend the core wordnet semi-automatically top-down, given the semantic basis of the core wordnets;
Using this approach we guarantee that the cores of the wordnets are highly compatible and comparable, but at the same time language-specific structures and lexicalizations can be expressed. For developing the core wordnets, we followed the next approach:
- Represent the 1024 CBCs by one or more synsets in the local language that are either equivalent or most closely approximate the CBCs in Princeton WordNet. These are the CBC-representatives in the local language;
- Add Base Concepts that are important to the local wordnet but not in the CBC set. These are Local Base Concepts (LBCs)
- Add hyperonyms of the BCs (=CBC-representatives+LBCs) to form a closed and consistent hyperonym hierarchy that is compatible with the EuroWordNet top-ontology
- Add further (horizontal) relations that are necessary to specify the semantics of the BCs.
- Extend the BCs with hyponyms one level down;
For developing the extended wordnets, various criteria were used, among which:
- Adding most frequent words from a corpus that are not included;
- Add words and synsets that are easily included due to the specific properties of the language;
- Add words and synsets that are easily linkable to the Princeton Wordnet with equivalence relations (automatically) derived from bilingual dictionaries;
- Add synsets that increase the intersection with other wordnets;
- Synsets that can be added because of specifically available resources, e.g. monolingual lexicons, ontologies, thesauri, etc.;
Core wordnets are typicaly between 5,000 and 10,000 synsets. Extended wordnets go beyond 20,000 synsets.