Carole Tiberius appointed professor of Computational Linguistics: 'If you know how systems work, you can better assess their limitations'
photography: Anna Loh
ChatGPT, translation machines and bots: for Carole Tiberius, they are a piece of cake. On 1 January, she was appointed professor of Computational Linguistics. 'There ae two elements to the field: computer science and linguistics.'
Tiberius had a love of computers and language from an early age. After studying French-Russian at the Higher Institute for Translators and Interpreters in Antwerp, she moved to Nijmegen to study computer linguistics. 'Not translating yourself, but letting the computer do the work. I found that very interesting,' she says.
Man and the machine
As interesting as the subject was, Tiberius' timing was not optimal. 'In the late 1990s, machine translation had reached its limit for a while,' she says. She did a PhD on a related topic ('multilingual lexical knowledge representation') and worked as a linguist at the University of Surrey for a few years, until she joined the Institute of Dutch Lexicology (now the Institute of Dutch Language) in 2006. Slowly, the computer crept back into her work. It resulted last January in an appointment as professor for one day a week. 'The work within the INT is mainly applied science,' she says. 'I’m looking forward to also being able to do more fundamental research again, especially into multi-word expressions. These are combinations of words whose meaning or behaviour does not quite follow the rules of language. For example, an expression such as 'having no clue about anything' should be interpreted figuratively rather than literally, and is always used in the completed tense. Because of this anomalous character, multi-word expressions are still often a stumbling block for language-technology applications.'
Indeed, the texts that roll out of online translation machines often look good, but in practice that isn’t always the case, Tiberius argues. 'The output looks error-free, until you compare the result with the input. Then you see that errors have crept in after all. Much the same applies to ChatGPT. There you have no original to compare the output with, but that is precisely why you have to be extremely critical yourself. What does it actually say?'
Staying needed
She would also like to instill this critical view in her future students. 'Leiden University Centre for Linguistics wants to focus more on computational linguistics in education,' she explains. 'At master's level, the Computational Linguistics track has already started and, if all goes well, more attention will soon be paid to computational linguistics in the bachelor's programme.' Tiberius wants to do her bit in both programmes. 'I want to teach students how to build a corpus on which they can perform linguistic analysis.'
The choices students make in preparing their corpus are crucial for the further course of the research. Tiberius: 'We may now be in a phase with a lot of focus on what computers can do, but I believe that humans and linguistics will also be and remain necessary. That means I’m very happy with my appointment at LUCL.'