AI's inability to grasp implicit meanings limits its role as a translator - Expert

This shortcoming keeps AI limited to serving as an intelligent assistant to translators and prevents it from taking over the role of experts in the field, said Prof Dr. Surinderpal Kaur Chanan Singh, Dean of the Faculty of Languages and Linguistics at Universiti Malaya.

She stated that although the quality of AI improves with advancements in algorithms, its inability to grasp the underlying meanings of words--especially within cultural and literary contexts--prevents it from producing high-quality translations.

"Even though AI is artificial intelligence, it is not truly intelligent yet. It struggles to capture implicit meanings, such as proverbs and idiomatic expressions, tone, or style. For instance, if I am being sarcastic or joking with someone, AI will just translate it literally.

"This can lead to miscommunication across cultures. If I make a joke, AI may fail to convey it, and the person receiving the message might become upset with me," she said during an interview about AI and the future of translation during Apa Khabar Malaysia, produced by Bernama TV, today.

This issue is evident in the AI-generated translations for subtitles in the film industry, which often do not resonate with the audience's understanding, as many movie dialogues touch on nuances and idiomatic expressions.

She emphasised that translations should also be tailored to the context of the local community to ensure they are easily understood.

"For instance, we translated a Peruvian poet's work from Spanish to Malay, which talks about hope, and we had to work hard to ensure the message was accurately conveyed to Malaysia's diverse audience.

"We cannot just do a direct translation; we need to adapt it to our cultural understanding," she added.

Sharing insights on challenges in the translation field in the digital age, she observed that many multinational companies are adopting AI for translation tasks due to its speed.

However, she advised these companies to consistently review the accuracy of the language generated by these systems.

-- BERNAMA