WordNets? SignNets!

Ineke Schuurman, Vincent Vandeghinste, Caro Brosens, Margot Janssens and Thierry Declerck

WordNets exist for many languages, but they are rather scarce for Sign Languages. There have been attempts to build a WordNet for Italian Sign language (LIS) and for American Sign Language (ASL). Building a WordNet for Sign Languages, i.e. a SignNet, entails more than adding videos of signs to an existing WordNet as Sign Languages are independent natural languages.

We are developing such SignNets, starting with one for VGT (Flemish Sign Language), within the framework of the SignON project (https://signon-project.eu/). The approach should be Sign Language independent.

Sign Languages are multimodal natural languages, having a gestural and a spatial mode. One issue when constructing a SignNet is that there is no generally accepted writing system. For linguistic purposes, glosses are being used. Such a gloss is similar to a lemma. These glosses only refer to the manual component of a sign (not taking into account non-manual features). They present a shortcut to create a new SignNet, because a WordNet may be available in the spoken language used in the glosses. We thus don’t have to start completely ‘from scratch’, for example to find obvious synsets/indices to link with.

Often there is a single gloss for a sign, even when it refers to both a noun and a verb (’afbraak’ (demolition) vs ’afbreken’ (to demolish)), but there can be several synonym glosses when the concept expressed by the sign varies per region or age. Such glosses can be made part of a SignNet synset. Many glosses also have a broader meaning than their counterpart in spoken language, for example the gloss for ’aanhaken’ (to hook up) en ’aanhangen’ (to hitch on) is the same in VGT: AANHAKEN. Also one synset in SignNet (with its own index), but two in Open Dutch WordNet. The indices used in Open Dutch Wordnet (ODWN) and Open English Wordnet (OEWN)/Princeton Wordnet (PWN) will be added in our SignNet, as will the collaborative interlingual indices (CILI). We link videos to the glosses, plus the possibility to include notations in Sign Language notation systems, such as HamNoSys, Sign_A, SignWriting or SigML (SiGn Mark-up Language), taking not just manual aspects into account. Apart from the indices to the synsets, the contents of these synsets will also be accessible. And one last crucial component: examples will be presented not just in the spoken language used in the glosses as well as in English, but especially also in the relevant SLs (video format): the one represented in the SignNet concerned and in International Sign.

We are working on encoding those SL specific synsets etc within the scope of the OntoLex-Lemon framework, as this model is able to differentiate the senses and ontological references to be attached to the glosses, when those are used for transcribing Sign Languages lexical data. Lexical semantics relations can then be used for describing the relations between the lexical elements of the glosses used as transcription of a SL data and the same lexical elements when encoded as part of the spoken or written form.