Improving Text-to-Pictograph Translation Through Word Sense Disambiguation

Proceedings of the Fifth Joint Conference on Lexical and Computational Semantics (*SEM 2016), pages 131–135, Berlin, Germany, August 11-12, 2016. Improving Text-to-Pictograph Translation Through Word Sense Disambiguation Leen Sevens*, Gilles Jacobs**, Vincent Vandeghinste*, Ineke Schuurman*, Frank Van Eynde* *Centre for Computational Linguistics (KU Leuven) firstname@ccl.kuleuven.be **Language and Translation Technology Team (Universiteit Gent) gillesm.jacobs@ugent.be Abstract We describe the implementation of a Word Sense Disambiguation (WSD) tool in a Dutch Text-to-Pictograph translation sys- tem, which converts textual messages into sequences of pictographic images. The system is used in an online platform for Augmentative and Alternative Communi- cation (AAC). In the original translation process, the appropriate sense of a word was not disambiguated before converting it into a pictograph. This often resulted in incorrect translations. The implemen- tation of a WSD tool provides a better semantic understanding of the input mes- sages. 1 Introduction In today’s digital age, people with Intellectual Dis- abilities (ID) often have trouble partaking in on- line activities such as email, chat, and social net- work websites. Not being able to access or use information technology is a major form of social exclusion. There is a dire need for digital commu- nication interfaces that enable people with ID to contact one another. Vandeghinste et al. (2015) are developing a Text-to-Pictograph and Pictograph-to-Text trans- lation system for the WAI-NOT1 communication platform. WAI-NOT is a Flemish non-profit or- ganization that gives people with severe commu- nication disabilities the opportunity to familiar- ize themselves with the Internet. Their safe web- site environment offers an email client that makes use of the Dutch pictograph translation solutions. The Text-to-Pictograph translation system (Van- deghinste et al., 2015; Sevens et al., 2015a) au- 1http://www.wai-not.be/ tomatically augments written text with Beta2 or Sclera3 pictographs and is primarily conceived to improve the comprehension of textual content. The Pictograph-to-Text translation system (Sevens et al., 2015b) allows the user to insert a series of Beta or Sclera pictographs, automatically translat- ing this image sequence into natural language text where possible. This facilitates the construction of textual content. The Text-to-Pictograph translation process did not yet perform Word Sense Disambiguation (WSD) to select the appropriate sense of a word before converting it into a pictograph. Instead, the most frequent sense of the word was chosen. This sometimes resulted in incorrect pictograph trans- lations (see Figure 1). Figure 1: Example of Dutch-to-Sclera translation. The word bloem means both flower and flour. The most common sense is flower, which would be the wrong choice within the context of baking. Note that the pictograph language is a simplified lan- guage. Function words and number information are not represented. We describe the implementation of a WSD tool 2The Beta set consists of more than 3,000 coloured pic- tographs: https://www.betasymbols.com/ 3Sclera pictographs are mainly black-and-white pic- tographs. Over 13,000 pictographs are available and more are added upon user request: http://www.sclera.be/ 131 in the Dutch Text-to-Pictograph translation sys- tem. After a discussion of related work (section 2), we present both the Text-to-Pictograph transla- tion tool and the WSD …

Content truncated to 500 words