20. 11. 2007. Researchers at the Universidad Politécnica de Madrid’s School of Computing have developed an original system for building multilingual dictionaries based on multiple term equivalences from what are known as universal words. System reliability and accuracy is 88%.
The system is based on Princeton University’s WordNet database. WordNet is a lexical database developed by linguists at Princeton’s Cognitive Science Laboratory. The database was designed to inventory, classify and relate the semantic and lexical content of the English language. WordNet is packaged as an electronic database that can be downloaded over the Internet.
WordNet’s underlying foundation is synset (synonym set), a group of interchangeable words that denote a meaning or particular usage. Each synset is one possible meaning of a word, described briefly and concisely. WordNet has a lexicon of over 200,000 perfectly structured and defined English terms. This is one of the pillars of the system conceived by researchers at the UPM’s School of Computing.
The system’s other mainstay are universal words. The concept of universal word came out of the UNL (Universal Networking Language) Project. The aim of this project is to eliminate the barriers of linguistic diversity by creating a medium of information exchange through which users can communicate in their own language
As the UNL Project’s Spanish Language Centre explains, one of the key concepts of UNL is the universal word. A universal word is a word, taken from the English language, to which a number of attributes and constraints are added to disambiguate the term.
The English term plus the attributes and constraints is known as a universal word, because it has an equivalent term in any other language. On account of their accuracy, one of the uses of universal words is to systematically produce multilingual dictionaries.
Researchers at the UPM’s School of Computing have applied an algorithm based on computational models to the WordNet database and expanded the English language lexicons. The constructed universal words can then be used to compile multilingual dictionaries.
UPM engineers created a Universal Words Dictionary. This dictionary can associate the words of each language with the respective disambiguated universal word. These researchers have managed to develop a tool that people can use to enter a word in their original language and select the equivalent of a set of terms written in their own language in another language. This is a breakthrough for multilingualism.
How does it work?
The ultimate aim is to build extremely precise multilingual dictionaries. The system contains universal words in English taken from the WordNet database. The universal words are passed on to lexicographers from different countries. Each lexicographer reads the universal word in English and understands its given meaning. The lexicographers then add the translation of the term in their mother tongue. To do all this, they do not have to be acquainted with the target language.
Even though the universal words are in English, it is not the same as using English language words, as English is just another natural language liable to ambiguity. The addition of attributes and constraints removes all ambiguity, and the equivalence between languages is extremely good.
This is how this multilingual dictionary is being built. The method has already been tried and tested with striking results. The translations from the universal words created using WordNet are 88% accurate and reliable.
Compared to other lexicographical methods, this is an original system, as it can generate bilingual dictionaries without experts having to speak both languages. All they need, apart from their mother tongue, is a good enough level of English to enter the exact translation of the word they are looking for.
Whereas there are plenty of Spanish-English interpreters, for example, it is harder to find Portuguese-Bulgarian translators, a problem that this system developed by researchers at the UPM obviates.
These bilingual dictionaries based on multiple equivalences of terms are not only useful for building dictionaries but also for supporting search systems in different languages.
Spanish cultural heritage multilingual dictionary
The Validation and Business Applications Group, led by Jesús Cardeñosa, a professor at the School of Computing, is using this system to compile a dictionary of multilingual terms on Spain’s cultural heritage, commissioned by the Ministry of Culture under the Patrilex Project. The project is to be completed by the end of 2008.
The goal of this project is to define a methodology and develop tools that support cultural heritage document search based on multilingual lexical resources. To do this, researchers are developing tools to manage lexical resources about Spain’s cultural heritage. The key tool is a multilingual thesaurus (database). A thesaurus is a list of terms, possibly composed of more than one word, related hierarchically to each other (general terms and subordinate terms) and used to index and retrieve documents. The thesaurus will be the core for defining semantic relations to establish the underlying context of a query.
The final result will be a search system based on user keyword entries, capable of putting the query into context and establishing a correspondence to equivalent words in other languages. The system will then be able to return documents in several languages that fit the search terms entered in Spanish with a precision unparalleled by current multilingual systems.
The languages used to build the multilingual thesaurus will be Spanish, English and Russian, and the system’s real test-bed will be the Under-Directorate General of Historical Heritage Conservation’s website, which is now exclusively in Spanish. According to the project brief, the methodology will put the accent on the method’s extensibility to other languages.