WordNet-based Relations between Words, Signs and Concepts
Ineke Schuurman
Centrum voor Computerlinguïstiek, KU Leuven
Caro Brosens
Vlaams GebarentaalCentrum
Kris Heylen
Instituut voor de Nederlandse Taal
Vincent Vandeghinste
Instituut voor de Nederlandse Taal
Bram Vanroy
Centrum voor Computerlinguïstiek, KU Leuven
WordNets are more and more being used when dealing with Sign Languages (SLs), to link words used in spoken languages (SpLs) to signs, for example for translation purposes. The corpora available for SLs are so small that modern techniques like ChatGPT are not an option for the foreseeable future.
Just like in dictionaries of spoken languages, sign language dictionaries (and their underlying databases, e.g. Signbank) document the lexemes of said language. In other words, signs in said dictionaries do not represent words in spoken languages but concepts, just like in other reference works.
Wordnets also deal with concepts, represented by the series of synonyms, the synset, coming with a given word.
Both SLs and SpLs are independent natural languages, and there are not per se close relations between SLs that have a common surrounding SpL. NGT (Sign Language of the Netherlands) and VGT (Flemish Sign Language), for example, are quite different languages.
SLs being full natural languages means that they too deserve their own ‘word’net, or rather signnet.
The concept represented by a sign is often broader than a ‘corresponding’ one in a surrounding SpL, i.e. Dutch when VGT and NGT are concerned.
In most SL dictionaries and Signbanks, the surrounding SpL is used to indicate (possible) translations and meaning.
Besides the aim of being a bilingual resource, these words in SpL are also there to assist people who are not familiar with (the notation of) SLs.
After all, SLs have no widely adopted script the way SpLs do.
In VGT, there are often several signs expressing the same concept, these often originate from different regions and the institutes for the deaf they hosted. Such signs constitute a synset at the SL level (SUPPORTER, consisting of SUPPORTER-A and SUPPORTER-E).
In the dictionary, both versions of SUPPORTER come with possible translations supporteren, aanhanger, fan, aanmoedigen, supporter. SUPPORTER (capitals) is just a gloss, a label.
One and the same SL concept often covers several different synsets in SpL. How to handle this without adapting SL synsets to SpL synsets, i.e. SLs have their own ontology. Signs, or rather SLs, should not be seen as additions to SpL wordnets (as is the case with for example pictographs).
Sometimes when translating between different languages, there is no one on one relation, i.e. there is no one lexeme to express the meaning of a lexeme in the other language straightforwardly. The same goes for translating or interpreting between SLs and SpL, like the sign expressing applause in SL (not clapping but waving as a way to make the applause visual, a ‘silent applause’).
How to deal with this? How to systematically handle other divisions in synsets, like splitting some? How to link Synsets and (corrected) WordNet entries? Adding separate identifiers in SignNet, linking both using hypernym and hyponym (and sometimes just synonym) relations? And how to link with SpL variants (like BE vs NL)? What about the Interlingual Index? Come and see our poster!