UBY
| Company type | Open resource / Software framework |
|---|---|
| Industry | Natural language processing |
| Founder | Ubiquitous Knowledge Processing Lab (UKP), Technische Universität Darmstadt |
| Headquarters | Darmstadt , Germany |
| Products | Lexical-semantic database, UBY-LMF format, DKPro UBY framework, lemonUby variant |
| Website | www |
UBY represents an extensive lexical-semantic database designed for natural language processing (NLP), developed by the Ubiquitous Knowledge Processing Lab (UKP) at the Computer Science faculty of Technische Universität Darmstadt. It merges several expert-curated and collaboratively built lexical resources for English and German, all structured according to the ISO Lexical Markup Framework (LMF) standard.
The resource uses a word sense alignment technique to integrate noun and verb data, improving the accuracy of word sense disambiguation. Presently, UBY consolidates information from 12 distinct resources covering both English and German languages.
Included resources
- English: WordNet, Wiktionary, Wikipedia, FrameNet, VerbNet, OmegaWiki
- German: German Wikipedia, German Wiktionary, OntoWiktionary, GermaNet, IMSLex-Subcat
- Multilingual: OmegaWiki
Format
UBY-LMF is a standardized representation format tailored for lexical resources in NLP that adheres to the ISO Lexical Markup Framework (LMF) established by ISO-TC37. It acts as a concrete serialization of this abstract standard, guaranteeing uniform linguistic feature descriptions through references compliant with ISO 12620 (ISOCat).
Availability and versions
UBY can be accessed through the open resource platform DKPro. DKPro UBY is a Java-based framework designed to create and manipulate sense-linked lexical databases following the UBY-LMF lexicon schema. While UBY’s software is distributed under open licenses such as GPL and CC by SA, some incorporated datasets are governed by different terms, including licenses restricted to academic use.
A Semantic Web-compatible variant named lemonUby is derived from the lemon model developed in the Monnet project. The lemon framework models lexicons and machine-readable dictionaries that integrate with the Semantic Web and Linked Data environments.
UBY vs. BabelNet
BabelNet is an automatically compiled lexical-semantic network connecting Wikipedia with major computational lexicons like WordNet. Although UBY and BabelNet may appear comparable, they are grounded in fundamentally different design approaches.
BabelNet initially concentrated on aligning WordNet with Wikipedia content, focusing primarily on nouns and named entities before broadening its scope to other parts of speech. In contrast, UBY has consistently emphasized verb-related data, particularly syntactic information derived from resources such as VerbNet and FrameNet.
UBY fully models each incorporated resource independently, enabling complete substitution of individual datasets. It offers unified access through resource alignments supported by a common LMF framework that facilitates both combined and standalone resource utilization. Conversely, BabelNet merges selected data into unified Babel Synsets, simplifying retrieval but blending multiple knowledge sources. Additionally, BabelNet enhances multilingual coverage by automatically generating translations for concepts missing in certain languages, which can introduce some inaccuracies.
The choice between these resources depends on specific application requirements; however, integrating both can provide comprehensive lexicographic insights. Their transparent and openly documented architectures represent significant progress toward such interoperability.
Applications
UBY has been utilized across multiple natural language processing applications such as Word Sense Disambiguation, Word Sense Clustering, Verb Sense Labeling, and Text Classification. It has also served as a foundation for initiatives aimed at the automatic generation of lexical semantic databases. Moreover, lemonUby has contributed to advancements in machine translation by enhancing the handling of previously unseen words.