Scientific and Societal
Problem Statement
Language technology is transforming people’s lives.
The language sciences have taken an enormous flight over the last 20 years. New AI methods, larger datasets, and increasing computing power have enabled a wide range of applications, ranging from machine translation, speech recognition and synthesis to recommender systems, chatbots and personal assistants. The language technology industry is one of the most rapidly growing technology sectors and is radically transforming people’s everyday lives.
Major methodological limitations
No transparency
These developments bring much good to society. However, the methods that underlie these new technologies have two fundamental drawbacks. First, they are not transparent. Current systems are able to learn, for instance, mappings from one language to another, from one format (e.g. speech) to another (e.g. text), or from large amounts of linguistic data to (e.g., legal or medical) decisions or recommendations. However, they do not yield explicit, auditable representations of these mappings. Many end-users are therefore still reluctant to trust currently available language technologies, especially if they are used to inform high-stake decisions or recommendations (e.g. in medical, legal or human resource procedures).
Skewed towards mainstream language varieties
A second fundamental limitation of current language technologies is that they are skewed towards mainstream language varieties. The major developments that we have witnessed over the last two decades have mainly focused on just a few widely spoken languages, and within those languages, just on the main varieties. For instance, speech recognition systems struggle with minority languages and regional accents, as well as differences in language use associated with age, gender, and language modality (spoken vs signed). As a result, the benefits of current language technologies are not equally distributed, and their impact could be even much larger if the underlying methods were enhanced to deal with variation within and across languages and language modalities.
Language technology for social good
Besides these two methodological limitations of current language technologies, there are also important gaps in the range of applications that have been explored. While most applications to date have been driven by economic gain, language technologies also have the potential to yield tremendous societal gain.
Inclusive Society
Societies are becoming ever more multilingual due to globalisation and migration. Moreover, people with limited hearing- or reading-ability and people with language-related cognitive disorders have limited access to public services, including education and healthcare. At the individual level, these language barriers lead to reduced social participation, unequal opportunities, and higher susceptibility to misinformation. At the collective level, they are a major impeding factor for societal cohesion.
Safe Society
Besides fostering a more inclusive society by reducing language barriers, the language sciences can also contribute to a safer and more resilient society. In a world where online communication has become ubiquitous, hazardous information spreads faster than a wildfire and can form a major threat for safety and societal resilience. Automatic detection of suspicious, offensive, and biased language is essential to warrant a safe, respectful, and undamaging social environment and to mitigate polarization.